Author response:

The following is the authors’ response to the original reviews.

Public Reviews:

Reviewer #1 (Public Review):

Strengths:

Overall there are some very interesting results that make an important contribution to the field. Notably, the results seem to point to differential recruitment of the PL-DMS pathway in goal-tracking vs sign-tracking behaviors.

Thank you.

Weaknesses:

There is a lot of missing information and data that should be reported/presented to allow a complete understanding of the findings and what was done. The writing of the manuscript was mostly quite clear, however, there are some specific leaps in logic that require more elaboration, and the focus at the start and end on cholinergic neurons and Parkinson's disease are, at the moment, confusing and require more justification.

In the revised paper, we provide additional graphs and information in support of results, and we further clarify procedures and findings. Furthermore, we expanded the description of the proposed interpretational framework that suggests that the contrasts between the cortical-striatal processing of movement cues in sign- versus goal trackers are related to previously established contrasts between the capacity for the  cortical cholinergic detection of attention-demanding cues.

Reviewer #2 (Public review):

Strengths:

The power of the sign- and goal-tracking model to account for neurobiological and behavioral variability is critically important to the field's understanding of the heterogeneity of the brain in health and disease. The approach and methodology are sound in their contribution to this important effort.

The authors establish behavioral differences, measure a neurobiological correlate of relevance, and then manipulate that correlate in a broader circuitry and show a causal role in behavior that is consistent with neurobiological measurements and phenotypic differences.

Sophisticated analyses provide a compelling description of the authors' observations.

Thank you.

Weaknesses:

It is challenging to assess what is considered the "n" in each analysis (trial, session, rat, trace (averaged across a session or single trial)). Representative glutamate traces (n = 5 traces (out of hundreds of recorded traces)) are used to illustrate a central finding, while more conventional trial-averaged population activity traces are not presented or analyzed. The latter would provide much-needed support for the reported findings and conclusions. Digging deeper into the methods, results, and figure legends, provides some answers to the reader, but much can be done to clarify what each data point represents and, in particular, how each rat contributes to a reported finding (ie. single trial-averaged trace per session for multiple sessions, or dozens of single traces across multiple sessions).

Representative traces should in theory be consistent with population averages within phenotype, and if not, discussion of such inconsistencies would enrich the conclusions drawn from the study. In particular, population traces of the phasic cue response in GT may resemble the representative peak examples, while smaller irregular peaks of ST may be missed in a population average (averaged prolonged elevation) and could serve as a rationale for more sophisticated analyses of peak probability presented subsequently.

We have added two new Tables to clarify the number of rats per phenotype and sex used for each experiment described in the paper (Table 1), and the number of glutamate traces (range, median and total number) extracted for each analysis of performance-associated glutamate levels and the impact of CNO-mediated inhibition of fronto-striatal glutamate (Table 3).

As the timing of glutamate peaks varies between individual traces and subjects, relative to turn and stop cue onset or reward delivery, subject-and trial-averaged glutamate traces would “wash-out” the essential findings of phenotype- and task event-dependent patterns of glutamate peaks. In the detailed responses to the reviewers, we illustrate the results of an analysis of averaged traces to substantiate this view. Furthermore, as detailed in the section on statistical methods, and as mentioned by the reviewer under Strengths, we used advanced statistical methods to assure that data from individual animals contribute equally to the overall result, and to minimize the possibility that an inordinate number of trials obtained from just one or a couple of rats biased the overall analysis.

Reviewer #3 (Public review):

Strengths:

Overall these studies are interesting and are of general relevance to a number of research questions in neurology and psychiatry. The assessment of the intersection of individual differences in cue-related learning strategies with movement-related questions - in this case, cued turning behavior - is an interesting and understudied question. The link between this work and growing notions of corticostriatal control of action selection makes it timely.

Thank you.

Weaknesses:

The clarity of the manuscript could be improved in several places, including in the graphical visualization of data. It is sometimes difficult to interpret the glutamate results, as presented, in the context of specific behavior, for example.

We appreciate the reviewer’s concerns about the complexity of some of the graphics, particularly the results from the arguably innovative analysis illustrated in Figure 6. Figure 6 illustrates that the likelihood of a cued turn can be predicted based on single and combined glutamate peak characteristics. The revised legend for this figure provides additional information and examples to ease the readers’ access to this figure. In addition, as already mentioned above, we have added several graphs to further illustrate our findings.

(Recommendations for the authors)

Reviewer #1 (Recommendations for the authors):

(1) The differences in behavioral phenotype according to vendor (Figure 1c) are slightly concerning, could the authors please elaborate on why they believe this difference is? Are there any other differences in these stocks- i.e. weight, appearance, other types of behaviors?

Differences in PCA behavior across vendors or specific breeding colonies were documented previously and may reflect the impact of environmental, developmental and genetic factors (references added in the revised manuscript). We included animals from both vendors to increase phenotypic variability and due to animal procurement constraints during COVID-related restrictions.

(2) Possibly related to the above, the rats in Figure 1a and Figure 2 are different strains. Please clarify.

In the revised legend of Figure 2 we clarify that the rat shown in the photographs is a Long-Evans rat that was not part of the experiments described in this paper. This rat was used to generate these photos as the black-spotted fur provided better contrast against the white treadmill belt.

(3) Figure 3c, the pairwise comparison showing a significant increase from Day 1 to Day 3 is hard to understand unless this is a lasting change. Is this increase preserved at Day 4? Examination of either a linear trend across days or a simple comparison of either Day 1 & 2 against Day 3 & 4 or, minimally Day 1 against Day 4 would communicate this message. Otherwise, there doesn't seem to be much of a case for improvement across test sessions, which would also be fine in my view.

As the analysis of post-criterion performance also revealed an effect of DAY, we felt compelled to report and illustrate the results of pairwise comparisons in Fig. 3c. In agreement with the reviewer’s point, we did not further comment on this finding in the manuscript.

(4) Figure 4e. I find it extremely unlikely that every included electrode was located exactly at anterior 0.5mm. Please indicate the range - most anterior and most posterior of the included electrodes in the study.

The schematic section shown in Fig. 4e depicted that AP level of that section and collapsed all placements onto that level. As detailed in Methods, electrode placements needed to be within the following stereotaxic space: AP: -0.3 to 0.6 mm, ML: 2 to 2.5 mm, and DV: -4.2 to -5 mm (see Methods). To clarify this issue, the text in Results and the legend was modified and the 0.5 mm label was removed from Fig. 4e.

(5) The paper generally is quite data light and there are a lot of extra results reported that aren't shown in the figures. There are 17 instances of the phrase "not shown", some are certainly justified, but a lot of results are missing…

We followed the reviewer’s suggestion and added several graphs. The revised Figure 5 includes the new graph 5d that shows the number of glutamate traces with just 1, 2 or 3 peaks occurring during cue presentation period. Likewise, the revised Figure 7 includes the new graph 7h that shows the number of glutamate traces with just 1, 2 or 3 peaks following the administration of CNO or its vehicle. In both cases, we also revised the analysis of peak number data, by counting the number of cases (or traces) with just 1, 2 or 3 peaks and using Chi-squared tests to determine the impact of phenotype and, in the latter case, of CNO. In addition, the revised Figure 7 now includes a graph showing the main effects of phenotype and CNO in reward delivery-locked glutamate maximum peak concentrations (Fig. 7k). In revising these sections, we also removed the prior statement about glutamate current rise times as this isolated observation had no impact on subsequent analyses or the discussion.

Concerning the reviewer’s point 5d (DMS eGFP transfection correlations Figure 8), the manuscript clarifies that the absence of such a correlation was expected given that eGFP expression in the DMS does not accurately reproduce the prelimbic-DMS projection space that was inhibited by CNO. In contrast, the correlations between the efficacy of CNO and DREADD expression measures in prelimbic cortex were significant and are graphed (Figs. 8g and 8j).

(6) Please clarify the exact number of animals in each experiment. The caption of Figure 3 seems to suggest there are 29 GTs and 22 STs in the initial experiment, but the caption of Figure 5b seems to suggest there are N=30 total rats being analyzed (leaving 21 un-accounted for), or is this just the number of GTs (meaning there is one extra)?

We have added Table 1 to clarify the number of animals used across different experiments and stages. Additionally, we have included a new Table 3 that identifies, for each graph showing results from the analyses of glutamate concentrations, the number of rats from which recordings were obtained and the number of traces per rat (range, median, and total).

(7) Relatedly, in Figures 5c-f and Figures 7g-i, the data seem to be analyzed by trial rather than subject-averaged, please clarify and what is the justification for this?

As detailed Experimental design and statistical analyses, we employed linear mixed-effects modeling to analyze the amperometric data that generated figures 5 and 7 to minimize the risk of bias due to an excessive number of trials obtained from specific rats. LMMs were chosen to analyze these repeated (non-independent) data to address issues that may be present with subject-averaged data. For clarity, throughout the results for these figures, the numerator in the F-ratio reflects the degrees of freedom from the fixed effects (phenotype/sex) and the denominator reflects the error term influenced by the number of subjects and the within-subject variance.

Concerning the illustration and analysis of trial- or subject-averaged glutamate traces please see reviewer 2, point 1 and the graph in that section. Within a response bin, such as the 2-s period following turn cues, glutamate peaks – as defined in Methods - occur at variable times relative to cue onset. Averaging traces over a population of rats or trials would “wash-out” the phenotype- and task event-dependent patterns of glutamate concentration peaks, yielding, for example, a single, nearly 2-s long plateau for cue-locked glutamate recordings from STs (see Figure 5b versus the graph shown in response to reviewer 2, point 1).

(8) Likewise on page 22, the number of animals from which these trials were taken should be stated "The characteristics of glutamate traces (maximum peak concentration, number of peaks, and time to peak) were extracted from 548 recordings of turn cue trials, 364 of which yielded a turn (GTs: 206, STs: 158) and 184 a miss (GTs: 112, STs: 72).".

The number of animals is now included in the text and listed in Table 3.

(9) The control group for Figure 7 given the mCherry fluorophore - given the known off-target effects of CNO, this is a very important control. Minimally, this data should be shown, but it is troubling that the ST group has n=2, I don't really understand how any sort of sensible stats can be conducted with a group this size, and obviously it's too small to find any significant differences if they were there.

As discussed on p. 14-15 in the manuscript under the section Clozapine N-Oxide, the conversion rate of CNO to clozapine suggests that approximately 50-100 times the dose of clozapine (compared to our 5.0 mg/kg CNO dosage) would be required to produce effects on rodent behavior (references on p. 14-15).

Regarding evidence from control rats expressing the empty construct, the revised manuscript clarifies that no effects of CNO on cued turns were found in 5 GTs expressing the empty control vector. Although CNO had no effects in STs expressing the DREADD, we also tested the effects of CNO in 2 STs expressing the empty control vector (individual turn rates following vehicle and CNO are reported for these 2 STs). Moreover, we extracted turn cue-locked glutamate traces (vehicle: 18 traces; 16 CNO traces) from an empty vector-expressing GT and found that administration of CNO neither reduced maximum glutamate peak concentrations nor the proportion of traces with just one peak. The absence of effects of CNO on cued turning performance and on turn-cue locked glutamate dynamics are consistent with prior studies showing no effects of 5.0 mg/kg CNO in rats not expressing the DREADD vector (references in manuscript).

(10) Figure 8b - the green circle indicated by 1 is definitely not the DMS, this is the DLS, and animals with virus placement in this region should be excluded.

The reviewer of course is correct and that exactly was the point of that illustration, as such a transfection space would have received the lowest possible rating (as indicated by the “1” in the green space). Fig. 8b was intended to illustrate expression efficacy ratings and does not indicate actual viral transfection spaces. Because the results described in the manuscript did not include data from a brain with a striatal transfection space as was illustrated in green in the original Fig. 8b, we removed that illustration of an off-target transfection space.  

(11) Figure 8j, the correlation specifically counts double-labeled PL hM4Di + eGFP neurons. Separating dual-labeled cells from all mCherry-labeled cells seems very strange given the nature of the viral approach. There seems to be an assumption that there are some neurons that express the mCherry-hM4Di that don't also have the AAV-Cre (eGFP). Obviously, if that were true this poses a huge problem for your viral approach and would mean that you're inhibiting a non-selective population of neurons. More likely, the AAV-Cre (eGFP) is present in all of your mCherry-hM4Di cells, just not at levels visible without GFP antibody amplification. Ideally, staining should be done to show that all cells with mCherry also have eGFP, but minimally this correlation should include all cells expressing mCherry with the assumption that they must also have the AAV-Cre.

As noted on page 15 in the Visualization and Quantification of eGFP/mCherry-Expressing Neurons section, eGFP expression in our viral approach was notably bright and did not necessitate signal enhancement. Furthermore, given the topographic organization of prelimbic-DMS projections on the on hand, and the variable transfection spaces in cortex and striatum on the other hand, the speculation that AAV-Cre may have been present in all mCherry cells is without basis. Second, there certainly are mCherry-positive cells that do not also express the retrogradely transported AAV-Cre, and that therefore were not affected by CNO. Third, the entire point of this dual vector strategy was to selectively inhibit prelimbic-striatal projections, and the strong correlation between double-labeled neuron numbers and cued turn scores substantiates the usefulness of this approach.

(12) Discussion, a bit more interpretation of the results would be good. Specifically - does the PL-DMS inhibition convert GTs to STs? There were several instances where the behavior and glutamate signals seemed to be pushed to look like STs but also a lot of missing data so it is hard to say. One would assume this kind of thing if, as I think is being said (please clarify), the ST phenotype is being driven by glutamatergic drive either locally or from sources other than PL cell bodies, presumably silencing the PL cell body inputs in GTs also leaves other glutamatergic inputs as the primary sources?

We agree with the reviewer that one could say, perhaps somewhat colloquially, that PL-DMS inhibition turns GTs to STs, in terms of turning performance and associated glutamate peak dynamics. The newly added data graphs are consistent with this notion. However, there are of course numerous other neurobiological characteristics which differ between GTs and STs and are revealed in the context of other behavioral or physiological functions.  In the Discussion, and as noted by the reviewer, we discuss alternative sources of glutamatergic control in STs and the functional implications of bottom-up mechanisms. In the revised manuscript, we have updated references and made minor revisions to improve this perspective.

(13) I found the abstract really detailed and very dense, it is pretty hard to understand in its current form for someone who hasn't yet read the paper. At this level, I would recommend more emphasis on what the results mean rather than listing the specific findings, given that the task is still quite opaque to the reader.

We revised the abstract, in part by deleting two rather dense but non-essential statements of results and by adding a more accessible conclusion statement.

(14) There are a lot of abbreviations: CTTT, PD, PCA, GT, ST, MEA, GO, LMM, EMMs, PL, DMS. Some of these are only mentioned a few times: MEA, LMM, and EMMs are all mentioned less than 5 times. To reduce mental load for the reader, you could spell these ones out, or include a table somewhere with all of the abbreviations.

We added a list of Abbreviations and Acronyms and eliminated abbreviations that were used infrequently.

(15) Generally, the logic that cortico-striatal connections contribute to GT vs ST seems easy to justify, however, the provided justification is missing a line of connection: "As such biases of GTs and STs were previously shown to be mediated in part via contrasting cholinergic capacities for the detection of cues (Paolone et al., 2013; Koshy Cherian et al., 2017; Pitchers et al., 2017a; Pitchers et al., 2017b), we hypothesized that contrasts in the cortico-striatal processing of movement cues contribute to the expression of these opponent biases." Please elaborate on why specifically cholinergic involvement suggests corticostriatal involvement. I think there are probably more direct reasons for the current hypothesis.

Done – see p. 4-5.

(16) Along the same line, paragraph 3 of the intro about Parkinson's disease and cholinergics seems slightly out of place. This is because the specific or hypothesized link between these things and corticostriatal glutamate has not been made clear. Consider streamlining the message specifically to corticostriatal projections in the context of the function you are investigating.

Done – see p. 4-5.

(17) Page 8, paragraph 2. There is a heading or preceding sentence missing from the start of this paragraph: "Contrary to the acclimation training phase, during which experimenters manually controlled the treadmill, this phase was controlled entirely by custom scripts using Med-PC software and interface (MedAssociates).".

Revised and clarified.

(18) Page 13 "We utilized a pathway-specific dual-vector chemogenetic strategy (e.g., Sherafat et al., 2020) to selectively inhibit the activity of fronto-cortical projections to the DMS". The Hart et al (2018) reference seems more appropriate being both the same pathway and viral combination approach.

Yes, thank you, we’ve updated the citation.

(19) Pages 20-21: "Maximum glutamate peak concentrations recorded during the cue period were significantly higher in GTs than in STs (phenotype: F(1,28.85)= 8.85, P=0.006, ηp 2=0.23; Fig. 5c). In contrast, maximum peak amplitudes locked to other task events all were significantly higher in STs." The wording here is misleading, both Figures 5c and 5d report glutamate peaks during the turn cue, the difference is what the animal does. So, it should be something like "Maximum glutamate peak concentrations recorded during the cue period were significantly higher in GTs than in STs when the animal correctly made a turn (stats) but this pattern reversed on missed trials when the animal failed to turn (stats)..." or something similar.

Yes, thank you. We have revised this section accordingly.  

(20) Same paragraph: "Contingency tables were used to compare phenotype and outcome-specific proportions and to compute the probability for turns in GTs relative to STs." What is an outcome-specific proportion?

This has been clarified.

.

(21) Page 22 typo: "GTs were only 0.74 times as likely as GTs to turn".

Fixed.

(22) The hypothesis for the DREADDs experiment isn't made clear enough. Page 23 "In contrast, in STs, more slowly rising, multiple glutamate release events, as well as the presence of relatively greater reward delivery-locked glutamate release, may have reflected the impact of intra-striatal circuitry and ascending, including dopaminergic, inputs on the excitability of glutamatergic terminals of corticostriatal projections" As far as I can understand, the claim seems to be that glutamate release might be locally modulated in the case of ST, on account of the profile of glutamate release- more slowly rising, multiple events, and reward-locked. Please clarify why these properties would preferentially suggest local modulation.

We have revised and expanded this section to clarify the basis for this hypothesis.

(23) The subheadings for the section related to Figure 7 "CNO disrupts..." "CNO attenuates..." presumably you mean fronto-striatal inhibition disrupts/attenuates. As it stands, it reads like the CNO per se is having these effects, off-target.

Fixed.

(24) The comparison of the results in the discussion against a "hypothetical" results section had the animals not been phenotyped behaviorally is unnecessary and overly speculative, given that 30-40% of rats don't fall into either of these two categories. I think the point here is to emphasize the importance of taking phenotype into account. This point can surely be made directly in its own sentence, probably somewhere towards the end of the discussion).

We have partly followed the reviewer’s advice and separated the discussion of the hypothetical results from the summary of main findings. However, we did not move this discussion toward the end of the Discussion section as we believe that it justifies the guiding focus of the discussion on the impact of phenotype.

(25) The discussion, like the introduction, talks a lot about cholinergic activity. As noted, this link is unclear - particularly how it links with the present results, please clarify or remove. Likewise high-frequency oscillations.

We have revised relevant sections in the Introduction (see above) and Discussion sections. However, given the considerable literature indicating contrasts between the cortical cholinergic-attentional capacities of GTs and STs, the interpretation of the current findings in that larger context is justified.

(26) Typo DSM in the discussion x 2.

Thanks, fixed.

Reviewer #2 (Recommendations for the authors):

(1) As mentioned in the Public Review, it is challenging to assess what is considered the "n" in each analysis, particularly for the glutamate signal analysis (trial, session, rat, trace (averaged across session or single trial)). Representative glutamate traces are used to illustrate a central finding, while more conventional trial-averaged population activity traces are not presented or analyzed. For example, n = 5 traces, out of hundreds of recorded traces, with each rat contributing 1-27 traces across multiple sessions suggests ~1-2% of the data are shown as time-resolved traces. Representative traces should in theory be consistent with population averages within phenotype, and if not, discussion of such inconsistencies would enrich the conclusions drawn from the study. In particular, population traces of the phasic cue response in GT may resemble the representative peak examples, while smaller irregular peaks of ST may be missed in a population average (averaged prolonged elevation in signal) and could serve as rationale for more sophisticated analyses of peak probability presented subsequently (and relevant to opening paragraph of discussion where hypothetical data rationale is presented).

We have added the new Table 1 to provide a complete account of the number of rats, per phenotype and sex, for each component of the experiments. In addition, the new Table 3 provides the range, median and total number of glutamate traces that were analyzed and formed the foundation of the individual data graphs depicting the results of glutamate concentration analyses.

We chose not to present trial- or subject-averaged traces, as glutamate peaks occur at variable times relative to the onset of turn and stop cues and reward delivery, and therefore averaging across a population of rats or trials would obscure phenotype- and task event-dependent patterns of glutamate peaks. The attached graph serves to illustrate this issue. The graph shows turn cue-locked glutamate concentrations (M, SD) from trials that yielded turns, averaged over all traces used for the analysis of the data shown in Fig. 5d (see also Table 3, top row). Because of the variability of peak times, trial- and subject-averaging of traces from STs yielded a nearly 2-s long elevated plateau of glutamate concentrations (red triangles), contrasting with the presence single and multiple peaks in STs as illustrated in Figs. 5b and 5e. Furthermore, averaging of traces from GTs obscured the presence of primarily single turn cue-locked peaks. Because of the relatively large variances of averaged data points, again reflecting the variability of peak times, analysis of glutamate levels during the cue period did not indicate an effect of phenotype (F(1,190)=1.65, P\=0.16). Together, subject- or trial-averaged traces would not convey the glutamate dynamics that form the essence of the amperometric findings obtained from our study. We recognize, as inferred by the reviewer, that smaller irregular peaks in STs may have been missed given the definition of a glutamate peak (see Methods). It is in part for that reason that we conducted a prospective analysis of the probability for turns given a combination of peak characteristics (maximum peak concentration and peak numbers; Fig. 6).

(2)To this latter point, the relationship between the likelihood to turn and the size of glutamate peak is focused on the GT phenotype, which limits understanding of how smaller multiple peaks relate to variables of interest in ST (missed turns, stops, reward). If it were possible to determine the likelihood for each phenotype, without a direct contrast of one phenotype relative to the other, this would be a more straightforward description of how signal frequency and amplitude relate to relevant behaviors in each group. Depending on the results, this could be done in addition to or instead of the current analysis in Figure 6.

We considered the reviewer’s suggestion but could not see how attempts to analyze the role of maximum glutamate concentrations and number of peaks within a single phenotype would provide any significant insights beyond the current description of results. Moreover, as stressed in the 2nd paragraph of the Discussion (see Reviewer 1, point 24), the removal of the phenotype comparison would nearly completely abolish the relationships between glutamate dynamics and behavior from the current data set.

Author response image 1.

(3) If Figure 6 is kept, a point made in the text is that GT is 1.002x more likely than ST to turn at a given magnitude of Glu signal. 1.002 x more likely is easily (perhaps mistakenly) interpreted as nearly identical likelihood. Looking closely at the data, perhaps what is meant is @ >4uM the difference between top-line labeled {b} and bottom-line labeled {d,e} is 1.002? If not, there may be a better way to describe the difference as 1x could be interpreted as the same/similar.

Concerning the potential for misinterpretation, the original manuscript stated (key phrase marked here in red font): Comparing the relative turn probabilities at maximum peak concentrations >4 µM, GTs were 1.002 times more likely (or nearly exactly twice as likely) as STs to turn if the number of cue-evoked glutamate peaks was limited to one (rhombi in Fig. 6a)  when compared to the presence of 2 or 3 peaks (triangles in Fig. 6a). However, we appreciate the reviewer’s concern about the complexity of this statement and, as it merely re-emphasized a result already described, it was deleted.

(4) For Figure 7e, the phenotype x day interaction is reported, but posthocs are looking within phenotype (GT) at treatment effects. Is there a phenotype x day x treatment, or simply phenotype x treatment (day collapsed) to justify within-group treatment posthocs?

We have revised the analysis and illustration of the data shown in Figs 7e and 7f, by averaging the test scores from the two tests, per animal, of the effects of vehicle and CNO, to be able to conduct a simpler 2-way analysis of the effects of phenotype and treatment.

(5) Ideally, viral control is included as a factor in this analysis as well. The separate analysis for viral controls was likely done due to low n, however negative findings from an ANOVA in which an n=2 (ST) should be interpreted with extreme caution. The authors already have treatment control (veh, CNO) and may consider dropping the viral controls completely due to the lack of power to perform appropriate analyses.

This issue has been clarified – see reviewer 1, point 9.

Minor:

(1) In the task description, it could be clearer how reward delivery relates to turns and stops. For example, does the turn cue indicate the rat will be rewarded at the port behind it? Does the stop cue indicate that the rat will be rewarded at the port in front of it? This makes logical sense, but the current text does not describe the task in this way, instead focusing on what is the correct action (seemingly but unlikely independent of reinforcement).

We have updated the task description in Methods and the legend of Figure 2 to indicate the location of reward delivery following turns and stops.

(2) For the peak analysis, what is the bin size for determining peaks? It is indicated that the value before and after the peak is >1 SD below the peak value, so it is helpful to know the temporal bin resolution for this definition.

As detailed on p 11-12 under Amperometry Data Processing and Analysis of Glutamate Peaks, we analyzed glutamate concentrations recorded at a frequency of 5 Hz (200 ms bins) throughout the 2-second-long presentation of turn and stop cues and for a 2-second period following reward delivery.

(3) Long Evans rats are pictured in Figure 2 (presumably contrast with a white background is better here), while SD rats are pictured in Figure 1. Perhaps stating why LE rats are pictured would help clear up any ambiguity about the strains used, as a quick look gives the impression two strains are used in two different tasks.

Yes, see reviewer 1, point 2.

(4) In Figure 7e, the ST and GT difference in turns/turn cue does not seem to replicate prior findings for tracking differences for this measure (Figure 3b). ST from the chemogenetic cohort seems to perform better than rats whose behavior was examined prior to glutamate sensor insertion. What accounts for this difference? Training and testing conditions/parameters?

The reviewer is correct. The absence of a significant difference between vehicle-treated GTs and vehicle-treated STs in Fig. 7e reflects a relatively lower turn rate in GTs than was seen in the analysis of baseline behavior (Fig. 3b; note the different ordinates of the two figures, needed to show the impact of CNO in Fig. 7e). Notably, the data in Fig. 7e are based on fewer rats (12 versus 29 GTs and 10 versus 22 STs; Table 1) and on rats which at this point had undergone additional surgeries to infuse the DREADD construct and implant electrode arrays. We can only speculate that these surgeries had greater detrimental effects in GTs, perhaps consistent with evidence suggesting that immune challenges trigger a relatively greater activation of their innate immune system (Carmen et al., 2023). We acknowledged this issue in the revised Results.

(5) The authors are encouraged to revise for grammar (are vs. is, sentence ending with a preposition, "not only" clause standing alone) and word choice (i.e. in introduction: insert, import, auditorily). Consider revising the opening sentence on page 5 for clarity.

We have revised the entire text to improve grammar and word choice.

(6) Do PD fallers refer to rats or humans? if the latter, this may be a somewhat stigmatizing word choice.

We have replaced such phrases using more neutral descriptions, such as referring to people with PD who frequently experience falls.

(7) Page 27 What does "non-instrumental" behavior mean?

We have re-phrased this statement without using this term.

(8) The opening paragraph of the discussion is focused on comparing reported results (with phenotype as a factor) to a hypothetical description of results (without phenotype as a factor) that were not presented in the results section. There is one reference to a correlation analysis on collapsed data, but otherwise, no reporting of data overall rats without phenotype as a factor. If this is a main focus, including these analyses in the results would be warranted. If this is only a minor point leading to discussion, authors could consider omitting the hypothetical comparison.

We have revised this section - see reviewer 1 point 24.

Reviewer #3 (Recommendations for the authors):

(1) These are really interesting studies. I think there are issues in data presentation/analysis that make it difficult to parse what exactly is happening in the glutamate signals, and when. Overall the paper is just a bit of a difficult read. A generally standard approach for showing neural recording data of many kinds, including, for example, subject-averaged traces, peri-event histograms, heatmaps, etc summarizing and quantifying the results - would be helpful. Beyond the examples in Figure 5, I would suggest including averaged traces of the glutamate signals and quantification of those traces.

We have addressed these issues in multiple ways, see the response to several points of reviewers 1 and 2, particularly reviewer 2, point 1.

(2) Figure 6 (and the description in the response letter) is also very non-intuitive. It's unclear how the examples shown relate to the reported significance indicators/labels/colors etc in the figure. I would suggest rethinking this figure overall, and if there is a more direct quantitative way to connect signal features with behavior. Again, drawing from standard visualization approaches for neural data could be one approach.

See also reviewer 2 points 1 and 3. Furthermore, we have revised the text in Results and the legend to improve the accessibility of Fig. 6.

(3) As far as I can tell, all of the glutamate sensor conclusions reflect analysis collapsed across 100s of trials. Do any of the patterns hold for a subjects-wise analysis? How variable are individual subjects?

We employed linear mixed-effect model analyses and added a random subject intercept to account for subject variability outside fixed effects (phenotype and treatment). The variance of the intercept ranged 0.01-1.71 SEM across outcome (cued turns/cued stops/misses). See also reviewer 1, point 7 and reviewer 2, point 1.

Author Response

The following is the authors’ response to the original reviews.

eLife assessment

This important study reports a novel measurement for the chemotactic response to potassium by Escherichia coli. The authors convincingly demonstrate that these bacteria exhibit an attractant response to potassium and connect this to changes in intracellular pH level. However, some experimental results are incomplete, with additional controls/alternate measurements required to support the conclusions. The work will be of interest to those studying bacterial signalling and response to environmental cues.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

This paper shows that E. coli exhibits a chemotactic response to potassium by measuring both the motor response (using a bead assay) and the intracellular signaling response (CheY phosporylation level via FRET) to step changes in potassium concentration. They find increase in potassium concentration induces a considerable attractant response, with an amplitude larger than aspartate, and cells can quickly adapt (but possibly imperfectly). The authors propose that the mechanism for potassium response is through modifying intracellular pH; they find both that potassium modifies pH and other pH modifiers induce similar attractant responses. It is also shown, using Tar- and Tsr-only mutants, that these two chemoreceptors respond to potassium differently. Tsr has a standard attractant response, while Tar has a biphasic response (repellent-like then attractant-like). Finally, the authors use computer simulations to study the swimming response of cells to a periodic potassium signal secreted from a biofilm and find a phase delay that depends on the period of oscillation.

Strengths:

The finding that E. coli can sense and adapt to potassium signals and the connection to intracellular pH is quite interesting and this work should stimulate future experimental and theoretical studies regarding the microscopic mechanisms governing this response. The evidence (from both the bead assay and FRET) that potassium induces an attractant response is convincing, as is the proposed mechanism involving modification of intracellular pH.

Weaknesses:

The authors show that changes in pH impact fluorescent protein brightness and modify the FRET signal; this measurement explains the apparent imprecise adaptation they measured. However, this effect reduces confidence in the quantitative accuracy of the FRET measurements. For example, part of the potassium response curve (Fig. 4B) can be attributed to chemotactic response and part comes from the pH modifying the FRET signal. Measuring the full potassium response curve of the no-receptor mutants as a control would help quantify the true magnitude of the chemotactic response and the adaptation precision to potassium.

Response: We thank the reviewer for the suggestion. We have now measured the full potassium response curve for the no-receptor mutant (HCB1414-pVS88), as shown in Fig. S4. We characterized the pH effects on CFP and YFP channels at different concentrations of KCl, and the relationship between the ratio of the signal post- to pre-KCl addition and the KCl concentration was established for both channels, as shown in Fig. S4C. The pH-corrected signal after KCl addition for strains with receptors was obtained by dividing the original signal after KCl addition by this ratio at the specific KCl concentration. This was done for both CFP and YFP channels. The pH-corrected responses for the Tar-only and Tsr-only strains are represented by red dots in Fig. 5BC. The recalculated response curve and adaptation curve for the wild-type strain are shown in Fig. S5. The same correction was applied to Fig. 3 as well. We also re-performed the simulations using the corrected dose-response curve and replotted Fig. 6, though the simulation results did not change much.

We have now added a subsection “Revised FRET responses by correcting the pH effects on the brightness of eCFP and eYFP” at line 296 in “Results” to describe this.

The measured response may also be impacted by adaptation. For other strong attractant stimuli, the response typically shows a low plateau before it recovers (adapts). However, in the case of Potassium, the FRET signal does not have an obvious plateau following the stimuli. Do the authors have an explanation for that? One possibility is that the cells may have already partially adapted when the response reaches its minimum, which could indicate a different response and/or adaptation dynamics from that of a regular chemo-attractant? In any case, directly measuring the response to potassium in mutants without adaptation enzymes (CheR, CheB) and with the receptors in different methylation levels would shed more light on the problem.

Response: We appreciate the reviewer’s insightful questions. To observe the low plateau before adaptation, a saturating amount of attractant should be added in a stepwise manner. According to the dose-response curve we measured for potassium, a saturating amount of potassium would be close to 100 mM. In fact, there is a small segment of the low plateau in the step response to 30 mM KCl (Fig. 4C or Fig. S5A). To observe more of this low plateau, we could have used a higher concentration of KCl. However, a stimulation higher than 30 mM KCl will induce substantial physiological changes in the cell, resulting in a significant decrease in fluorescence for both channels (Fig. S7). Therefore, the range of KCl concentration that can be reliably applied in FRET measurements is limited.

The half-time of adaptation at 30 mM KCl was measured to be approximately 80 s, demonstrating a faster adaptation than 0.1 mM MeAsp, which induced a similar magnitude of response. Nevertheless, this is still significantly slower than the time required for medium exchange in the flow chamber, which takes less than 10 s to replace 99% of the medium. Thus, the effect on the measured response magnitude due to adaptation should be small (less than 10%).

We thank the reviewer for the suggestion of measuring the response to potassium in mutants without adaptation enzymes (CheR, CheB) and with the receptors in different methylation levels. However, these mutants are typically less sensitive than the wild-type, exhibiting higher values of K0.5 (Sourjik & Berg, PNAS 99:123, 2002), and thus require an even higher KCl concentration to see the low plateau. Consistent with this, we attempted to measure the response to potassium in a cheRcheB mutant (HCB1382-pVS88). As shown in Fig. R1 below, there is no response to up to 30 mM KCl, suggesting that the sensitive region of the mutant is beyond 30 mM KCl.

The relevant text was added at line 413-424.

Author response image 1.

The response of the cheRcheB mutant (HCB1382-pVS88) to different concentrations of KCl. The blue solid line denotes the original signal, while the red dots represent the pH-corrected signal. The vertical purple (green) dashed lines indicate the moment of adding (removing) 0.01 mM, 0.1 mM, 0.3 mM, 1 mM, 3 mM, 10 mM and 30 mM KCl, in chronological order.

There seems to be an inconsistency between the FRET and bead assay measurements, the CW bias shows over-adaptation, while the FRET measurement does not.

Response: We thank the reviewer for pointing this out. We have now demonstrated that the imprecise adaptation shown in the FRET assay primarily resulted from the pH-induced intensity change of the fluorescent proteins. As shown in Fig. S5A&C, the FRET signal also shows over-adaptation, similar to the bead assay, when we recalculated the response by correcting the CFP and YFP channels.

Now we clarified it at line 315.

The small hill coefficient of the potassium response curve and the biphasic response of the Tar-only strain, while both very interesting, require further explanation since these are quite different than responses to more conventional chemoattractants.

Response: We thank the reviewer for pointing this out. We have now recalculated the pH-corrected results for the dose-response curve (Fig. S5) and the biphasic response of the Tar-only strain (Fig. 5C). The new Hill coefficient is 0.880.14 (meanSD), which is close to the response to MeAsp (1.2) (ref. 46). We suspected that this Hill coefficient of slightly less than 1 resulted from the different responses of Tar and Tsr receptors to potassium.

The Tar-only strain exhibits a repellent response to stepwise addition of low concentrations of potassium less than 10 mM, and a biphasic response above (Fig. 5C). This biphasic response might result from additional pH-effects on the activity of intracellular enzymes such as CheRB and CheA, which may have a different timescale and response from the Tar receptor. We have now added the penultimate paragraph in “Discussion” to talk about the response of the Tar-only strain.

Reviewer #2 (Public Review):

Summary:

Zhang et al investigated the biophysical mechanism of potassium-mediated chemotactic behavior in E coli. Previously, it was reported by Humphries et al that the potassium waves from oscillating B subtilis biofilm attract P aeruginosa through chemotactic behavior of motile P aeruginosa cells. It was proposed that K+ waves alter PMF of P aeruginosa. However, the mechanism was this behaviour was not elusive. In this study, Zhang et al demonstrated that motile E coli cells accumulate in regions of high potassium levels. They found that this behavior is likely resulting from the chemotaxis signalling pathway, mediated by an elevation of intracellular pH. Overall, a solid body of evidence is provided to support the claims. However, the impacts of pH on the fluorescence proteins need to be better evaluated. In its current form, the evidence is insufficient to say that the fluoresce intensity ratio results from FRET. It may well be an artefact of pH change. Nevertheless, this is an important piece of work. The text is well written, with a good balance of background information to help the reader follow the questions investigated in this research work.

In my view, the effect of pH on the FRET between CheY-eYFP and CheZ-eCFP is not fully examined. The authors demonstrated in Fig. S3 that CFP intensity itself changes by KCl, likely due to pH. They showed that CFP itself is affected by pH. This result raises a question of whether the FRET data in Fig3-5 could result from the intensity changes of FPs, but not FRET. The measured dynamics may have nothing to do with the interaction between CheY and CheZ. It should be noted that CFP and YFP have different sensitivities to pH. So, the measurement is likely confounded by the change in intracellular pH. Without further experiments to evaluate the effect of pH on CFP and YFP, the data using this FRET pair is inconclusive.

Response: We thank the reviewer for pointing this out. We have now measured the full potassium response curve for the no-receptor mutant (HCB1414-pVS88), as shown in Fig. S4. We characterized the pH effects on CFP and YFP channels at different concentrations of KCl, and the relationship between the ratio of the signal post- to pre-KCl addition and the KCl concentration was established for both channels, as shown in Fig. S4C. The pH-corrected signal after KCl addition for strains with receptors was obtained by dividing the original signal after KCl addition by this ratio at the specific KCl concentration. This was done for both CFP and YFP channels. The pH-corrected responses for the Tar-only and Tsr-only strains are represented by red dots in Fig. 5BC. The recalculated response curve and adaptation curve for the wild-type strain are shown in Fig. S5. The same correction was applied to Fig. 3 as well. We also re-performed the simulations using the corrected dose-response curve and replotted Fig. 6, though the simulation results did not change much.

We have now added a subsection “Revised FRET responses by correcting the pH effects on the brightness of eCFP and eYFP” at line 296 in “Results” to describe this.

The data in Figure 1 is convincing. It would be helpful to include example videos. There is also ambiguity in the method section for this experiment. It states 100mM KCl was flown to the source channel. However, it is not clear if 100 mM KCl was prepared in water or in the potassium-depleted motility buffer. If KCl was prepared with water, there would be a gradient of other chemicals in the buffer, which confound the data.

Response: We apologize for the ambiguity. The KCl solution used in this work was prepared in the potassium-depleted motility buffer. We have now clarified this at both lines 116 and 497. We now provided an example video, Movie S1, with the relevant text added at line 123.

The authors show that the FRET data with both KCl and K2SO4, and concluded that the chemotactic response mainly resulted from potassium ions. However, this was only measured by FRET. It would be more convincing if the motility assay in Fig1 is also performed with K2SO4.

Response: We thank the reviewer for the suggestion. The aim of comparing the responses to KCl and K2SO4 was to determine the role of chloride ions in the response and to prove that the chemotactic response of E. coli to KCl comes primarily from its response to potassium ions. It is more sensitive to compare the responses to KCl and K2SO4 by using the FRET assay. In contrast, the microfluidic motility assay is less sensitive in revealing the difference in the chemotactic responses, making it difficult to determine the potential role of chloride ions.

Methods:

• Please clarify the promotes used for the constitutive expression of FliCsticky and LacI.

Response: The promoters used for the constitutive expression of LacIq and FliCsticky were the Iq promoter and the native promoter of fliC, respectively (ref. 57).

Now these have been clarified at line 471.

• Fluorescence filters and imaging conditions (exposure time, light intensity) are missing.

Response: Thank you for the suggestion. We have now added more descriptions at lines 535-546: The FRET setup was based on a Nikon Ti-E microscope equipped with a 40× 0.60 NA objective. The illumination light was provided by a 130-W mercury lamp, attenuated by a factor of 1024 with neutral density filters, and passed through an excitation bandpass filter (FF02-438/24-25, Semrock) and a dichroic mirror (FF458-Di02-25x36, Semrock). The epifluorescent emission was split into cyan and yellow channels by a second dichroic mirror (FF509-FDi01-25x36, Semrock). The signals in the two channels were then filtered by two emission bandpass filters (FF01-483/32-25 and FF01-542/32-25, Semrock) and collected by two photon-counting photomultipliers (H7421-40, Hamamatsu, Hamamatsu City, Japan), respectively. Signals from the two photomultipliers were recorded at a sampling rate of 1 Hz using a data-acquisition card installed in a computer (USB-1901(G)-1020, ADlink, New Taipei, Taiwan).

• Please clarify if the temperature was controlled in motility assays.

Response: All measurements in our work were performed at 23 ℃. It was clarified at line 496.

• L513. It is not clear how theta was selected. Was theta set to be between 0 and pi? If not, P(theta) can be negative?

Response: The θ was set to be between 0 and π. This has now been added at line 581.

• Typo in L442 (and) and L519 (Koff)

Response: Thank you. Corrected.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

(1) From the motor measurements the authors find that the CW bias over-adapts to a level larger than prestimulus, but this is not seen in the FRET measurements. What causes this inconsistency? Fig. 2D seems to rule out any change in CheY binding to the motor.

Response: We thank the reviewer for pointing this out. We have now demonstrated that the imprecise adaptation shown in the FRET assay primarily resulted from the pH-induced intensity change of the fluorescent proteins. As shown in Fig. S5A&C, the FRET signal also shows over-adaptation, similar to the bead assay, when we recalculated the response by correcting the CFP and YFP channels.

We now clarified it at line 315.

(2) It would be useful to compare the response amplitude for potassium (Fig. 3C) to a large concentration of both MeAsp and serine. This is a fairer comparison since your work shows potassium acts on both Tar and Tsr. Alternatively, testing a much larger concentration (~10^6 micromolar) at which MeAsp also binds to Tsr would also be useful.

Response: We thank the reviewer for pointing this out. We have now recalculated the response to potassium by correcting the pH-induced effects on fluorescence intensity of CFP and YFP. The response to 30 mM KCl was 1.060.10 times as large as that to 100 μM MeAsp. The aim of the comparison between the responses to potassium and MeAsp was to provide an idea of the magnitude of the chemotactic response to potassium. The stimulus of 100 μM MeAsp is already a saturating amount of attractant and induces zero-kinase activity, thus using a higher stimulus (adding serine or a larger concentration of MeAsp) is probably not needed. Moreover, a larger concentration (~10^6 micromolar) of MeAsp would also induce an osmotactic response.

(3) The fitted Hill coefficient (~0.5) to the FRET response curve is quite small and the authors suggest this indicates negative cooperativity. Do they have a proposed mechanism for negative cooperativity? Have similar coefficients been measured for other responses?

Response: We thank the reviewer for pointing this out. We have now recalculated the pH-corrected results for the dose-response curve (Fig. S5). The new Hill coefficient is 0.880.14 (meanSD), which is close to the response to MeAsp (1.2) (ref. 46). We suspect that this Hill coefficient of slightly less than 1 results from the differing responses of Tar and Tsr receptors to potassium.

(3a) The authors state a few times that the response to potassium is "very sensitive", but the low Hill coefficient indicates that the response is not very sensitive (at least compared to aspartate and serine responses).

Response: We apologize for the confusion. We described the response to potassium as “very sensitive” due to the small value of K0.5. This has now been clarified at line 236.

(3b) Since the measurements are performed in wild-type cells the response amplitude following the addition of potassium may be biased if the cell has already partially adapted. This seems to be the case since the FRET time series does not plateau after the addition of the stimulus. The accuracy of the response curve and hill coefficient would be more convincing if the experiment was repeated with a cheR cheB deficient mutant.

Response: We thank the reviewer for raising these questions. To observe the low plateau before adaptation, a saturating amount of attractant should be added in a stepwise manner. According to the dose-response curve we measured for potassium, a saturating amount of potassium would be close to 100 mM. In fact, there is a small segment of the low plateau in the step response to 30 mM KCl (Fig. 4C or Fig. S5A). To observe more of this low plateau, we could have used a higher concentration of KCl. However, a stimulation higher than 30 mM KCl will induce substantial physiological changes in the cell, resulting in a significant decrease in fluorescence for both channels (Fig. S7). Therefore, the range of KCl concentration that can be reliably applied in FRET measurements is limited.

The half-time of adaptation at 30 mM KCl was measured to be approximately 80 s, demonstrating a faster adaptation than 0.1 mM MeAsp, which induced a similar magnitude of response. Nevertheless, this is still significantly slower than the time required for medium exchange in the flow chamber, which takes less than 10 s to replace 99% of the medium. Thus, the effect on the measured response magnitude due to adaptation should be small (less than 10%).

We thank the reviewer for the suggestion of measuring the response to potassium in mutants without adaptation enzymes (CheR, CheB) and with the receptors in different methylation levels. However, these mutants are typically less sensitive than the wild-type, exhibiting higher values of K0.5 (ref. 46), and thus require an even higher KCl concentration to see the low plateau. Consistent with this, we attempted to measure the response to potassium in a cheRcheB mutant (HCB1382-pVS88). As shown in Fig. R1, there is no response to up to 30 mM KCl, suggesting that the sensitive region of the mutant is beyond 30 mM KCl.

The relevant text was added at line 413-424.

(4) The authors show that the measured imprecise adaptation can be (at least partially) attributed to pH impacting the FRET signal by changing eCFP and eYFP brightness.

(4a) Comparing Fig. 5C and D, the chemosensing and pH response time scales look similar. Therefore, does the pH effect bias the measured response amplitude (just as it biases the adapted FRET level)?

Response: We agree with the reviewer that the pH effect on CFP and YFP biases the measured response amplitude. We have now performed the measurement of dose-response curve to potassium for the no-receptor mutant (HCB1414-pVS88), as shown in Fig. S4. The pH effects on CFP and YFP were corrected. The dose-response curve and adaptation curve were recalculated and plotted in Fig. S5.

(4b) It would help to measure a full response curve (at many concentrations) for the no-receptor strain as a control. This would help distinguish, as a function of concentration, how much response can be attributed to pH impacting the FRET signal versus the true chemotactic response.

Response: We thank the reviewer for the suggestion. We have now performed the measurements for the no-receptor strain. The impact of pH on CFP and YFP has been corrected. The pH-corrected results, previously in Fig.3-5, are now presented in Fig. 3, Fig. S5 and Fig. 5, respectively.

(5) The biphasic response of Tar is strange and warrants further discussion. Do the authors have any proposed mechanisms that lead to this behavior? For the 10mM and 30mM KCl measurements there is a repellent response followed by an attractant response for both adding and removing the stimuli, why is this?

Response: We thank the reviewer for pointing this out. The Tar-only strain exhibits a repellent response to stepwise addition of low concentrations of potassium less than 10 mM, and a biphasic response above (Fig. 5C). This biphasic response might result from additional pH-effects on the activity of intracellular enzymes such as CheRB and CheA, which may have a different timescale and response from the Tar receptor. We have now added the penultimate paragraph in “Discussion” to talk about the response of the Tar-only strain.

(5a) The fact that Tar and Tsr are both attractant (after the initial repellant response in Tar) appears to be inconsistent with previous work on pH response (Ref 52, Yang and Sourjik Molecular Microbiology (2012) 86(6), 1482-1489). This study also didn't see any biphasic response.

Response: We thank the reviewer for pointing this out. The Tar-only strain shows a repellent response to stepwise addition of low concentrations of potassium, specifically less than 10 mM. This is consistent with previous observations of the response of Tar to changes in intracellular pH (refs. 44,45) and also with the work of Yang and Sourjik (new ref. 53), although the work in ref. 53 dealt with the response to external pH change, and bacteria were known to maintain a relatively stable intracellular pH when external pH changes (Chen & Berg, Biophysical Journal (2000) 78:2280-2284). Interestingly, the Tar-only strain exhibits a biphasic response to high potassium concentrations of 10 mM and above. This biphasic response might result from additional pH-effects on the activity of intracellular enzymes such as CheRB and CheA (ref. 56), which may have a different timescale and response from the Tar receptor. We have now added the penultimate paragraph in “Discussion” to talk about the response of the Tar-only strain.

(5b) The response of Tar to the removal of sodium benzoate (Fig. S2) seems to be triphasic, is there any explanation for this?

Response: We thank the reviewer for pointing this out. We have now acknowledged in the legend of Fig. S2 that this response is interesting and warrants further exploration: “The response to the removal of sodium benzoate seems to be a superposition of an attractant and a repellent response, the reason for which deserves to be further explored.”

(6) Fitting the MWC model leads to N=0.35<1. It is fine to use this as a phenomenological parameter, but can the authors comment on what might be causing such a small effective cluster size for potassium response?

Response: We thank the reviewer for pointing this out. We have now recalculated the pH-corrected results for the dose-response curve (Fig. S5). The new Hill coefficient is 0.880.14 (meanSD), which is close to the response to MeAsp (1.2) (ref. 46). We now refit the MWC model to the pH-corrected dose-response curve, obtaining N of 0.85. We think the small N is due partly to the fact that we are fitting the curve with four parameters: N, Kon, Koff, and fm, while only three features of the sigmoid does-response curve are relevant (the vertical scale, the midpoint concentration, and the slope of the sigmoid). Future experiments may determine these parameters more accurately, but they should not significantly affect the simulation results as long as the wild-type dose-response curve is accurate.

(7) The results of the modeling are closely related to Zhu et. al. Phys. Rev. Lett. 108, 128101. Is the lag time for large T related to the adaptation time?

Response: We thank the reviewer for pointing this out. We used a similar framework of modeling as Zhu et. al. The potassium response was also analogous to the chemotactic response to MeAsp. Thus, the results are closely related to Zhu et al. We have now cited Zhu et al. (Ref. 52) and noted this at line 366.

The lag time for large T is related to the adaptation time. We have now simulated the chemotaxis to potassium for large T with different adaptation time by varying the methylation rate kR. The results are shown in Fig. S8. The simulated lag time decreases with the methylation rate kR, but levels off at high values of kR. Now this has been added at line 603.

Minor issues:

• Fig. 1C: should the axis label be y?

Response: Yes, thank you. Now corrected.

• Line 519: Koff given twice, the second should be Kon.

Response: Thank you. Corrected.

• When fitting the MWC model (Eq. 3 and Fig. 6B) did you fix a particular value for m?

Response: m was treated as a fitting parameter, grouped in the parameter fm.

Reviewer #2 (Recommendations For The Authors):

Minor points: - I suggest explaining the acronyms when they first appear in the text (eg CMC, CW, CCW).

Response: Thank you. Now they have been added.

• L144. L242. "decrease" is ambiguous since membrane potential is negative. I understand the authors meant less negative (which is an increase). I suggest to avoid this expression.

Response: Thank you for the suggestion. Now they have been replaced by “The absolute value of the transmembrane electrical potential will decrease”.

• For Fig 1b - it says the shaded area is SEM in the text, but SD in the legend. Please clarify.

Response: Thank you. The annotation in the legend has now been revised as SEM.

• Fig 1C label of x axis should be "y" instead of "x" to be consistent with Fig 1A.

Response: Thank you. It has now been revised.

• In Figure 2, the number of independent experiments as well as the number of samples should be included.

Response: Thank you. The response in Fig. 2C is the average of 83 motors from 5 samples for wild-type strain (JY26-pKAF131). The response in Fig. 2D is the average of 22 motors from 4 samples for the chemotaxis-defective strain (HCB901-pBES38). They have now been added to the legend.

• Regarding the attractant or repelling action of potassium and sucrose, it would be important to have a move showing the cells' behaviours.

Response: We thank the reviewer for the suggestion. We have now provided Movie S1 to show the cells’ behavior to potassium. As shown in Fig. 3B, the chemotactic response to 60 mM sucrose is very small compared to the response to 30 mM KCl. This implies that a noticeable response to sucrose necessitates higher concentrations of stimulation. However, Jerko et al. [Rosko, J., Martinez, V. A., Poon, W. C. K. & Pilizota, T. Proc. Natl Acad. Sci. USA 114, E7969-E7976 (2017).] have shown that high concentrations of sucrose lead to a significant reduction in the speed of the flagella motor. Thus, in a motility assay for sucrose, the osmolarity-induced motility effect may overwhelm the minor repellent-like response.

Author response:

The following is the authors’ response to the original reviews.

eLife Assessment

This well-written report uses functional neuroimaging in human observers to provide convincing evidence that activity in the early visual cortex is suppressed at locations that are frequently occupied by a task-irrelevant but salient item. This suppression appears to be general to any kind of stimulus, and also occurs in advance of any item actually appearing. The work in its present form will be valuable to those examining attention, perception, learning and prediction, but with a few additional analyses could more informatively rule out potential alternative hypotheses. Further discussion of the mechanistic implications could clarify further the broad extent of its significance. 

We thank the editor and the reviewers for the positive evaluation of our manuscript and the thoughtful comments. Below we provide a detailed point-by-point reply to the reviewers’ comments.

In addition to addressing the reviewers' comments, we have improved the figure legends by explicitly describing the type of error bars depicted in the figures, information which was previously only listed in the Materials and Methods section. Specifically, the statement: “Error bars denote within-subject SEM” was added to several figures, as applicable. We believe that briefly reiterating this information in the figure legends enhances clarity and enables readers to interpret the results more accurately and efficiently. We also updated our code and data sharing statement, as well as opened the repository for the public: “Analysis and experiment code, as well as data required to replicate the results reported in this manuscript are available here: https://doi.org/10.17605/OSF.IO/G4RXV. Raw MRI data is available upon request.”

Public Reviews

Reviewer #1 (Public review): 

Summary: 

The authors investigated if/how distractor suppression derived from statistical learning may be implemented in early visual cortex. While in a scanner, participants conducted a standard additional singleton task in which one location more frequently contained a salient distractor. The results showed that activity in EVC was suppressed for the location of the salient distractor as well as for neighbouring neutral locations. This suppression was not stimulus specific - meaning it occurred equally for distractors, targets and neutral items - and it was even present in trials in which the search display was omitted. Generally, the paper was clear, the experiment was well-designed, and the data are interesting. Nevertheless, I do have several concerns mostly regarding the interpretation of the results. 

(1) My biggest concern with the study is regarding the interpretation of some of the results. Specifically, regarding the dynamics of the suppression. I appreciate that there are some limitations with what you might be able to say here given the method but I do feel as if you have committed to a single interpretation where others might still be at play. Below I've listed a few alternatives to consider. 

We agree with the reviewer that there are important alternatives to consider. Adequately addressing these alternatives will substantially increase the inferences we can draw from our data. Therefore, we address each alternative interpretation in detail below.

(a) Sustained Suppression. I was wondering if there is anything in your results that would speak for or against the suppression being task specific. That is, is it possible that people are just suppressing the HPDL throughout the entire experiment (i.e., also through ITI, breaks, etc., rather than just before and during the search). Since the suppression does not seem volitional, I wonder if participants might apply a blanket suppression to HPDL un l they learn otherwise. Since your localiser comes a er the task you might be able to see hints of sustained suppression in the HPDL during these trials.  

It is indeed possible that participants suppressed the HPDL throughout the entire experiment, instead of proactively instantiating suppression on each trial. While possible, we believe that this account is less likely to explain the present results, given the utilized analysis approach, a voxel-wise GLM fit to the BOLD data per run (see Materials and Methods for details). Specifically, we derived parameter estimates from this GLM per location to estimate the relative suppression. Sustained suppression would modulate BOLD responses throughout the run, i.e. presumably also during the implicit baseline period used to estimate the contrast parameter estimates per location. Hence, sustained suppression should not result in a differential modulation between locations, as the BOLD response at the HPDL during the baseline period would be equally suppressed as during the trial. Inspired by the reviewer’s comment, we now clarify this critical point in the manuscript’s Discussion section:

“Third, participants might have suppressed the HPDL consistently throughout the experiment. This sustained suppression account differs from the proactive suppression proposed here. While this alternative is plausible, we believe that it is less likely to account for the present results, given the analysis conducted. Specifically, we computed voxel-wise parameter estimates and contrasted the obtained betas between locations. Under a sustained suppression account, the HPDL would show suppression even during the implicit baseline period, which would obscure the observed BOLD suppression at and near the HPDL.” 

(b) Enhancement followed by suppression. Another alternative that wasn't discussed would be an initial transient enhancement of the HPDL which might be brought on by the placeholders followed by more sustained suppression through the search task. Of course, on the whole this would look like suppression, but this still seems like it would hold different implications compared to simply "proactive suppression". This would be something like search and destroy however could be on the location level before the actual onset of the search display.  

R1 correctly points out that BOLD data, given the poor temporal resolution, do not allow for the detection of potential transient enhancements at the HPDL followed by a later and more pronounced suppression (akin to “search and destroy”). We fully agree with this assessment. However, we also argue that a transient enhancement followed by sustained suppression before search display onset constitutes proactive suppression in line with our interpretation, because suppression would still arise proactively (i.e., before search, and hence distractor, onset). Whether transient enhancement precedes suppression cannot be elucidated by our data, but we believe that it constitutes an interesting avenue for future studies using me-resolved and spatially specific recording methods. We now clarify this important implementational variation in the updated manuscript.

“Finally, due to the limited temporal resolution of BOLD data, the present data do not elucidate whether the present suppression is preceded by a brief attentional enhancement of the HPDL, as implied by some prior work (Huang et al., 2024). On this account the HPDL would see transient enhancement, followed by sustained suppression, akin to a ‘search and destroy’ mechanism. Critically, we believe that this variation would nonetheless constitute proactive distractor suppression as the suppression would still arise before search onset. Using temporally and spatially resolved methods to explore potential transient enhancements preceding suppression is a promising avenue for future research charting the neural mechanisms underlying distractor suppression.”

(2) I was also considering whether your effects might be at least partially attributable to priming type effects. This would be on the spatial (not feature) level as it is clear that the distractors are switching colours. Basically, is it possible that on trial n participants see the HPDL with the distractor in it and then on trial n+1 they suppress that location. This would be something distinct from the statistical learning framework and from the repetition suppression discussion you have already included. To test for this, you could look at the trials that follow omission or trials. If there is no suppression or less suppression on these trials it would seem fair to conclude that the suppression is at least in part due to the previous trial. 

We agree with the reviewer that it is plausible that participants particularly suppress locations which on previous trials contained a distractor. To address this possibility, we conducted a new analysis and adjusted the manuscript accordingly:

“Second, participants may have suppressed locations that contained the distractor on the previous trial, reflecting a spatial priming effect. This account constitutes a complementary but different perspective than statistical learning, which integrates implicit prior knowledge across many trials. We ruled out that spatial priming explains the present results by contrasting BOLD suppression magnitudes on trials with the distractor at the HPDL and trials where the distractor was not at the HPDL on the previous trial. Results, depicted in Supplementary Figure 4 showed that distractor suppression was statistically significant across both trial types, including trials without a distractor at the HPDL on the preceding trial. This indicates that the observed BOLD suppression is unlikely to be driven by priming and is instead more consistent with statistical learning. Moreover, results did not yield a statistically significant difference between trial types based on the distractor location in the preceding trial. However, these results should not be taken to suggest that spatial priming cannot contribute to distractor suppression; for details see: Supplementary Figure 4.” (p. 13).

We note that this analysis approach slightly differs from the reviewer’s suggestion, which considered omission trials. However, we decided to exclude trials immediately following an omission to ensure that both conditions were matched as closely as possible. In particular, omission trials represent extended rest periods, which could alter participants’ state and especially modulate the visually evoked BOLD responses (e.g., potentially increasing the dynamic range) compared to trials that did not follow omissions. Our analysis approach avoids this difference while still addressing the hypothesis put forward by the reviewer. We now provide the full explanation and results figure of this priming analysis in the figure text of Supplementary Figure 4: 

Reviewer #2 (Public review): 

The authors of this work set out to test ideas about how observers learn to ignore irrelevant visual information. Specifically, they used fMRI to scan participants who performed a visual search task. The task was designed in such a way that highly salient but irrelevant search items were more likely to appear at a given spatial location. With a region-of-interest approach, the authors found that activity in visual cortex that selectively responds to that location was generally suppressed, in response to all stimuli (search targets, salient distractors, or neutral items), as well as in the absence of an anticipated stimulus. 

Strengths of the study include: A well-written and well-argued manuscript; clever application of a region of interest approach to fMRI design, which allows articulating clear tests of different hypotheses; careful application of follow-up analyses to rule out alternative, strategy-based accounts of the findings; tests of the robustness of the findings to detailed analysis parameters such as ROI size; and exclusion of the role of regional baseline differences in BOLD responses. 

We thank the reviewer for the positive evaluation of our manuscript.

The report might be enhanced by analyses (perhaps in a surface space) that distinguish amongst the multiple "early" retinotopic visual areas that are analysed in the aggregate here. 

We agree with the reviewer that an exploratory analysis separating early visual cortex (EVC) into its retinotopic areas could be an interesting addition. Our reasoning to combine early visual areas into one mask in the original analyses was two-fold: First, we did not have an a priori reason to expected distinct neural suppression between these early ROIs. Therefore, we did not acquire retinotopy data to reliably separate early visual areas (e.g. V1, V2 and V3), instead opting to increase the number of search task trials. The lack of retinotopy data inherently limits the reliability of the resulting cortical segmentation. However, we now performed an analysis separating early visual cortex into V1 and V2 and report the details as Supplementary Text 1:

“In an exploratory analysis we investigated whether subdivisions of EVC exhibit different representations of priority signals. In brief, we used FreeSurfer to reconstruct brain surfaces (recon-all) from each subject’s anatomical scan. From these reconstructions we derived V1_exvivo and V2_exvivo labels, which were transformed into volume space using ‘mri_label2vol’ and merged into a bilateral mask for each ROI. We then selected the voxels within each ROI that were most responsive to the four stimulus locations, based on independent localizer data. This voxel selection followed the procedure outlined in the Materials and Methods: Region of Interest (ROI) Definition. To accommodate the subdivision into two ROIs (V1 and V2) compared to the single EVC ROI in the main analysis, we halved the number of voxels selected per location. Finally, we applied the same ROI analysis to investigate distractor suppression during search and omission trials, following the procedure described in Materials and Methods: Statistical Analysis. 

Results of this more fine-grained ROI analyses are depicted in Supplementary Figure 1. First, the results from V2 qualitatively mirrored our primary ROI analysis. BOLD responses in V2 differed significantly between stimulus types (main effect of stimulus type: F_(2,54) = 31.11, p < 0.001, 𝜂 = 0.54). Targets elicited larger BOLD responses compared to distractors (t₍₂₇₎ = 3.05, p_holm = 0.004, d = 0.06) and neutral stimuli (t₍₂₇₎ = 7.82, p_holm < 0.001, d = 0.14). Distractors also evoked larger responses than neutral stimuli (t₍₂₇₎ = 4.78, p_holm < 0.001, d = 0.09). These results likely reflect top-down modulation due to target relevance and bo om-up effects of distractor salience. Consistent with the primary ROI analysis, the manipula on of distractor predictability showed a distinct pattern of location specific BOLD suppression in V2 (main effect of location: F_(1.1,52.8) = 5.01, p = 0.030, 𝜂 = 0.16). Neural populations with receptive fields at the HPDL showed significantly reduced BOLD responses compared to the diagonally opposite neutral location (NL-far; post hoc test HPDL vs NL-far: t₍₂₇₎ = 2.69, p_holm = 0.022, d = 0.62). Again, this suppression was not confined to the HPDL but also extended to close by neutral locations (NL-near vs NL-far: t₍₂₇₎ = 2.79, p_holm = 0.022, d = 0.65). BOLD responses did not differ between HPDL and NL-near locations (HPDL vs NL-near: t₍₂₇₎ = 0.11, p_holm = 0.915, d = 0.03; BF₁₀ = 0.13). As in the EVC ROI analysis, this suppression pattern was consistent across distractor, target, and neutral stimuli presented at the HPDL and NL-near locations compared to NL-far. In sum, neural responses in V2 were significantly modulated by the distractor contingencies, evident as reduced BOLD responses in neural populations with receptive fields at the HPDL and neutral locations near the location of the frequent distractor (NL-near), relative to the neutral location diagonally across the HPDL (NL-far). 

In V1, BOLD responses also differed significantly between stimulus types (main effect of stimulus type: F_(1.3,35.6) = 6.69, p = 0.009, 𝜂 = 0.20). Targets elicited larger BOLD responses compared neutral stimuli (t₍₂₇₎ = 3.52, p_holm = 0.003, d = 0.12) and distractors evoked larger responses than neutral stimuli (t₍₂₇₎ = 2.62, p_holm = 0.023, d = 0.09). However, no difference between targets and distractors was observed (t₍₂₇₎ = 0.90, p_holm = 0.375, d = 0.03; BF₁₀ = 0.17), suggesting reduced sensitivity to task-related effects in V1. Indeed, analyzing the effect of distractor predictability for BOLD responses in V1 showed a different result than in V2 and the combined EVC ROI. There was no significant main effect of location (F_(2,54) = 2.20, p = 0.120, 𝜂 = 0.08; BF₁₀ = 0.77). BOLD responses at NL-near and NL-far were similar (BF₁₀ = 0.171), with the only reliable difference found between target stimuli at the HPDL and NL-far locations (W = 94, p_holm = 0.012, r = 0.54).”  

We include the new result figure as Supplementary Figure 5

We now include reference to these results in the manuscript’s Discussion section:

“Are representations of priority signals uniform across EVC? A priori we did not have any hypotheses regarding distinct neural suppression profiles across different early visual areas, hence our primary analyses focused stimulus responses neural populations in EVC, irrespective of subdivision. However, an exploratory analysis suggests that distractor suppression may show different patterns in V1 compared to V2 (Supplementary Figure 5 and Supplementary Text 1). In brief, results in V2 mirrored those reported for the combined EVC ROI (Figure 4). In contrast, results in V1 appeared to be only partially modulated by distractor contingencies, and if so, the modulation was less robust and not as spatially broad as in V2. This suggests the possibility of different effects of distractor predictability across subdivisions of early visual areas. However, these results should be interpreted with caution. First, our design did not optimize the delineation of early visual areas (e.g., no functional retinotopy), limiting the accuracy of V1 and V2 segmentation. Additionally, analyses were conducted in volumetric space, which further reduces spatial precision. Future studies could improve this by including retinotopy runs to accurately delineate V1, V2, and V3, and by performing analyses in surface space. Higher-resolution functional and anatomical MRI sequences would also help elucidate how distractor suppression is implemented across EVC with greater precision.”

Furthermore, the study could benefit from an analysis that tests the correlation over observers between the magnitude of their behavioural effects and their neural responses. 

R2 highlights that behavioral facilitation and neural suppression could be correlated across participants. The rationale is that if neural suppression in EVC is related to the facilitation of behavioral responses, we should expect a positive relationship between neural suppression at the HPDL and RTs across participants. In this analysis we focused on the contrast between HPDL and NL-far, as this contrast was statistically significant in both the RT (Figure 2) and the neural suppression analysis (Figure 4). First, we computed for each participant the behavioural benefit of distractor suppression as: RT_facilitation = RT_NL-far – RT_HPDL. Thereby RT facilitation reflects the response speeding due to a distractor appearing at the high probability distractor location compared to the far neutral location. Next, we computed neural suppression as: BOLD_suppression = BOLD_NL-far – BOLD_HPDL Thus, positive values reflect the suppression of BOLD responses at the HPDL comparted to the NL-far location. The BOLD suppression index was computed for each stimulus type separately, as in the main ROI analysis (i.e. for Targets, Neutrals and Distractors). Finally, we correlated RT_facilitation with BOLD_suppression across participants using Pearson correlation. Results showed a small, but not statistically significant correlation between RT facilitation and BOLD suppression for distractor (r₍₂₆₎ = 0.22, p = 0.257), target (r₍₂₆₎ = 0.10, p = 0.598) and neutral (r₍₂₆₎ = 0.13, p = 0.519) stimuli. Thus, while the direc on of the correlation was in line with the specula on by the reviewer in the “ Recommendations for the authors”, results were not statistically reliable and therefore inconclusive. As also noted in our preliminary reply to the reviewer comments, it was a priori unlikely that this analysis would yield a statistically significant correlation. An a priori power analysis suggested that, to reach a power of 0.8 at a standard alpha of 0.05, given the present sample size of n=28, the effect size would need to exceed r > 0.75, which seemed unlikely for the correlation of behavioural and neural difference scores. Given the inconclusive nature of the results, we prefer to not include this additional analysis in the manuscript, as we believe that it does not add to the main message of the paper but have it accessible to the interested reader in the public “peer review process”.

The study provides an advance over previous studies, which iden fied enhancement or suppression in visual cortex as a function of search target/distractor predictability, but in less spatially-specific way. It also speaks to open questions about whether such suppression/enhancement is observed only in response to the arrival of visual information, or instead is preparatory, favouring the la er view. The theoretical advance is moderate, in that it is largely congruent with previous frameworks, rather than strongly excluding an opposing view or providing a major step change in our understanding of how distractor suppression unfolds. 

We agree with the reviewer that our results are an advancement of prior work, particularly with respect to narrowing down the role of sensory areas and the proactive nature of distractor suppression. However, we argue that this represents a significant step forward for several reasons. First, to our knowledge, the literature on distractor suppression, and visual search in general, is by no means unanimous with respect to the conclusion that distractor suppression is instantiated proactively (Huang et al., 2021, 2022). Indeed, there are several studies suggesting the opposite account; reactive suppression (Chang et al., 2023) or contributions by both proactive and reactive mechanisms (Sauter et al., 2021; Wang et al., 2019). Moreover, studies in support of proactive distractor suppression did not investigate the involvement of (early) sensory areas during suppression. Conversely, to our knowledge most studies investigating the involvement of sensory cortex during distractor suppression did not address the question whether suppression arises proactive or reactively.

Recommendations for the authors: 

Reviewer #1 ( Recommendations for the authors): 

Minor Points: 

(1) There are several disconnects between the behaviour and the MR results - i.e. not stimulus specific yet there are no deficits for targets appearing the HPDL, also no behavioural suppression for the NLNear but neural suppression found. Nevertheless, the behaviour is used as a way to rule out potential attentional strategies when considering whether there is enhancement in the NL-Far condition. I realise you have a few other points here, but I think it's worth addressing what could be seen as a double standard.

The reviewer points out an important concern, which we feel could have better been addressed in the manuscript. From our point of view a partial dissociation between neural modulations in EVC and eventual behavioural facilitation is not surprising, given the extensive neural processing beyond EVC required for behaviour. However, this assessment may differ, if one stresses an explicit volitional attentional strategy over an implicit statistical learning account. That said, we clearly do not want to create the impression of using a double standard. The lack of behavioural facilitation for targets at NLfar is not a critical part of our argument against explicit attentional strategies. Therefore, we rephrased the relevant paragraph in the Discussion section to now emphasize the importance of the control analysis excluding participants who reported the correct HPDL in the questionnaire (Figure 5), but nonetheless yielded qualitatively identical results to the main ROI analysis (Figure 4). In our opinion, this control analysis provides more compelling evidence against a volitional attentional strategy account without the risk of crea ng the impression of applying a double standard in the interpretation of behavioural data. Additionally, we now acknowledge the limitation of relying on behavioral data in ruling out volitional attentional strategies in the updated manuscript:

“It is well established that attention enhances BOLD responses in visual cortex (Maunsell, 2015; Reynolds & Chelazzi, 2004; Williford & Maunsell, 2006). If participants learned the underlying distractor contingencies, they could deploy an explicit strategy by directing their attention away from the HPDL, for example by focusing attention on the diagonally opposite neutral location. This account provides an alternative explanation for the observed EVC modulations. However, while credible, the current findings are not consistent with such an interpretation. First, there was no behavioral facilitation for target stimuli presented at the far neutral location, contrary to what one might expect if participants employed an explicit strategy. However, given the partial dissociation between neural suppression in EVC and behavioral facilitation, additional neural data analyses are required to rule out volitional attention strategies. Thus, we performed a control analysis that excluded all participants that indicated the correct HPDL location in the questionnaire, thereby possibly expressing explicit awareness of the contingencies. This control analysis yielded qualitatively identical results to the full sample, showing significant distractor suppression in EVC. Therefore, it is unlikely that explicit attentional strategies, and the enhancement of locations far from the HPDL, drive the results observed here. Instead the current finding are consistent with an account emphasizing the automa c deployment of spatial priors (He et al., 2022) based on implicitly learned statistical regularities.”

(2) Does the level of suppression change in any way through the experiment? I.e., does it get stronger in the second vs. first half of the experiment? 

The reviewer askes an interesting question, whether BOLD suppression may change across the experiment. To address this question, we performed an additional analysis testing BOLD suppression in EVC during the first compared to second half of the MRI experiment. Here we defined BOLD suppression as: BOLD_suppression = ((BOLD_NL-far – BOLD_HPDL) + (BOLD_NL-far – BOLD_NL-near)) / 2. Thus, in this formula on of BOLD suppression we summarize the two primary BOLD suppression effects observed in our main results (Figure 4). Additionally, as we previously did not observe any significant differences in BOLD suppression magnitudes between different stimulus types (i.e. suppression was similar for target, distractor and neutral stimuli), we collapsed across stimulus types in this analysis.

Results, depicted below, showed that during both the initial (Run 1+2) and later part (Run 4+5) of the MRI experiment BOLD suppression was statistically significant (BOLD suppression Run 1+2: W = 331, p = 0.003, r = 0.63; BOLD suppression Run 4+5: W = 320, p = 0.007, r= 0.58) , confirming our main results of reliable distractor suppression even in this subset of trials. However, we did not observe any statistically significant differences between early and late runs of the experiment (t₍₂₇₎ = -0.21, p = 0.835, d = -0.04). In fact, a Bayesian paired t-test provided evidence for the absence of a difference in BOLD suppression between early compared to later runs (BF₁₀ = 0.205), suggesting that distractor suppression in EVC was stable throughout the experiment. A qualitatively similar, pattern was evident during omission trials, with significant distractor suppression during early runs (t₍₂₇₎ = 2.70, p = 0.012, d = 0.51), but not quite a statistically significant modulation for later runs (t₍₂₇₎ = 1.97, p = 0.059, d = 0.37). Again, there was no evidence for a difference in suppression magnitudes across the experiment (W = 198, p = 0.920, d = -0.025) and support for the absence of a difference in BOLD suppression between early and late runs (BF₁₀ = 0.278).

Author response image 1.

Analysis of BOLD suppression magnitudes in EVC across the MRI experiment phases. BOLD suppression was comparable between early (Run 1+2) and late (Run 4+5) phases of the MRI experiment, suggesting consistent suppression in EVC following statistical learning. Error-bars denote within-subject SEM. * p < 0.05, ** p < 0.01, = BF₁₀ < 1/3.

In sum, results suggest that distractor suppression in EVC was stable across runs and did not change significantly throughout the experiment. This result was a priori likely, given that participants already underwent behavioral training before entering the MRI. This enabled them to establish modified spatial priority maps, containing the high probability distractor location contingencies, already before the first MRI run. While specula ve, it is possible that participants may still have consolidated the spatial priority maps during the initial runs, but that this additional consolation is not evident in the data, as later runs may see less engagement by participants due to increasing fa gue towards the end of the MRI experiment. Indeed, rapid learning and stable suppression throughout the remainder of the experiment is also reported by prior work (Lin et al., 2021). We believe that it is highly interesting for future studies to investigate the development of distractor suppression across learning, with initial exposure to the contingencies inside the MRI. However, as the present results are inconclusive, we prefer to not include this analysis in the main manuscript, as it may not provide significant additional insight into the neural mechanisms underlying distractor suppression. 

(3) In the methods vs. results you have reported the probabili es slightly differently. In the methods you say the HPDL was 6x more likely to contain a distractor whereas in the results you say 4x. Based on the reported trial numbers I think it should be 4, but probably you want to double check that this is consistent and correct throughout. 

We thank the reviewer for bringing this inconsistency to our attention. We have corrected this oversight in the adjusted manuscript: 

“One of the four locations of interest was designated the high probability distractor location (HPDL), which contained distractor stimuli (unique color) four mes more o en than any of the remaining three locations of interest. In other words, if a distractor was present on a given trial (42 trials per run), the distractor appeared 57% (24 trials per run) at the HPDL and at one of the other three locations with equal probability (i.e., 14% or 6 trials per run per location).” 

Reviewer #2 ( Recommendations for the authors): 

The authors have performed their analyses in the volume rather than the surface, and have grouped together V1, V2, and V3 as "early visual cortex". As the authors' claims lean heavily on the idea that they are measuring "early" visual responses, the study would be improved by delinea ng the ROIS within these different retinotopic regions. Such an approach might be facilitated by analysing data on the reconstructed surface. 

Please refer to our reply to this analysis suggested in the Public review.

The authors rightly tread carefully on the causal link between their neural findings and the behavioural outcomes. The picture might be clarified somewhat further by testing for a positive relationship between behavioural effect sizes and neural effect sizes across participants. e.g. to what extent is the search advantage when distractors are presented at the "HPDL" linked to greater suppression of BOLD at the HDPL region of early visual cortex? 

Please refer to our reply to this analysis suggested in the Public review.

Some of the claims based on null hypotheses would be better supported by Bayesian tests e.g. page 6 "This pattern of results was the same regardless whether the distractor, target, or a neutral stimulus presented at the HPDL and NL-near locations compared to NL-far ..." and "BOLD responses between HPDL and NL-near locations did not reliably differ ..." This is similar to the approach that the authors adopted later in the section "Ruling out attentional modulation".

We agree with the reviewer that our ROI analyses would benefit from providing evidence for the absence of a modulation. Accordingly, we updated our results by adding equivalent Bayesian tests. Bayes Factors were computed using JASP 0.18.2 (JASP Team, 2024; RRID:SCR_015823) with default settings; i.e. for Bayesian paired t-tests with a Cauchy prior width of 0.707. Qualitative interpretations of BFs were based on Lee and Wagenmakers (2014). We now report the obtained BF in the Results section. 

“BOLD responses between HPDL and NL-near locations did not reliably differ (HPDL vs NL-near: t₍₂₇₎ = 0.47, p_holm = 0.643, d = 0.08; BF₁₀ = 0.19).”

And:

“Neural responses at HPDL and NL-near did not reliably differ (t₍₂₇₎ = 0.21, p_holm = 0.835 d = 0.04; BF₁₀ = 0.21).”

Moreover, we now denote any equivalent results (defined as BF₁₀<1/3) in Fig. 4 and Fig. 5, and included the descrip on of the associated symbol in the figure text (“ = BF₁₀ < 1/3”).

Additionally, we now also report the BF for all paired t-tests reported in Supplementary Table 1.

Finally, we addressed the statement: “This pattern of results was the same regardless whether the distractor, target, or a neutral stimulus presented at the HPDL and NL-near locations compared to NLfar”. Our inten on was to emphasize that the pattern of results reported in the sentence preceding it was evident for distractor, target, or neutral stimulus, and not to suggest that the magnitude of the effect is the same. Hence, to more accurate reflect the results, we changed this sentence to:  “This pattern of results was present regardless whether the distractor, target, or a neutral stimulus presented at the HPDL and NL-near locations compared to NL-far”

Author response:

The following is the authors’ response to the original reviews.

Public Reviews:

Reviewer #1 (Public Review):

Based on previous publications suggesting a potential role for miR-26b in the pathogenesis of metabolic dysfunction-associated steatohepatitis (MASH), the researchers aim to clarify its function in hepatic health and explore the therapeutical potential of lipid nanoparticles (LNPs) to treat this condition. First, they employed both whole-body and myeloid cell-specific miR-26b KO mice and observed elevated hepatic steatosis features in these mice compared to WT controls when subjected to WTD. Moreover, livers from whole-body miR-26b KO mice also displayed increased levels of inflammation and fibrosis markers. Kinase activity profiling analyses revealed distinct alterations, particularly in kinases associated with inflammatory pathways, in these samples. Treatment with LNPs containing miR-26b mimics restored lipid metabolism and kinase activity in these animals. Finally, similar anti-inflammatory effects were observed in the livers of individuals with cirrhosis, whereas elevated miR-26b levels were found in the plasma of these patients in comparison with healthy control. Overall, the authors conclude that miR-26b plays a protective role in MASH and that its delivery via LNPs efficiently mitigates MASH development.

The study has some strengths, most notably, its employ of a combination of animal models, analyses of potential underlying mechanisms, as well as innovative treatment delivery methods with significant promise. However, it also presents numerous weaknesses that leave the research work somewhat incomplete. The precise role of miR-26b in a human context remains elusive, hindering direct translation to clinical practice. Additionally, the evaluation of the kinase activity, although innovative, does not provide a clear molecular mechanisms-based explanation behind the protective role of this miRNA.

Therefore, to fortify the solidity of their conclusions, these concerns require careful attention and resolution. Once these issues are comprehensively addressed, the study stands to make a significant impact on the field.

We would like the reviewer for his/her careful evaluation of our manuscript and appreciate his/her appraisal for the strengths of our study. Regarding the weaknesses, we have addressed these as good as possible during the revision of our manuscript.

We can already state that miR-26b has clear anti-inflammatory effects on human liver slices, which is in line with our results demonstrating that miR-26b plays a protective role in MASH development in mice. The notion that patients with liver cirrhosis have increasing plasma levels of miR-26b, seems contradictory at first glance. However, we believe that this increased miR-26b expression is a compensatory mechanism to counteract the MASH/cirrhotic effects. However, the exact source of this miR-26b remains to be elucidated in future studies.

The performed kinase activity analysis revealed that miR-26b affects kinases that particularly play an important role in inflammation and angiogenesis. Strikingly and supporting these data, these effects could be inverted again by LNP treatment. Combined, these results already provide strong mechanistic insights on molecular and intracellular signalling level. Although the exact target of miR-26b remains elusive and its identification is probably beyond the scope of the current manuscript due to its complexity, we believe that the kinase activity results already provide a solid mechanistic basis.

Reviewer #1 (Recommendations For The Authors):

A list of recommendations for the authors is presented below:

(1) The title should emphasize that the majority of experiments were conducted in mice to accurately reflect the scope of the study.

As suggested we have updated our title to include the statement that we primarily used a murine model:

“MicroRNA-26b protects against MASH development in mice and can be efficiently targeted with lipid nanoparticles.”

(2) It would be useful to know more about miR-26b function, including its target genes, tissue-specific expression, and tissue vs. circulating levels. Is it expected that the two strains of the miRNA (i.e., -3p and -5p) act this similarly? Also, miR-26b expression in the liver of individuals with cirrhosis should be determined.

The function of miR-26b is still rather elusive, making functional studies using this miR very interesting. In a previous study, describing our used mouse model (Van der Vorst et al. BMC Genom Data, 2021) we have eluded several functions of miR-26b that are already investigated. This was particularly already described in carcinogenesis and the neurological field.

Target gene wise, there are already several targets described in miRbase. However, for our experiments we feel that determination of the specific target genes is beyond the scope of the current manuscript and rather a focus of follow-up projects.

Regarding the expression of miR-26b, the liver and blood have rather high and similar expressions of both miR-26b-3p and miR-26b-5p as shown in Author response image 1.

Author response image 1.

Expression of miR-26b-3p and -5p. Expression of miR-26b-3p (left) and miR-26b-5p (right), generated by using the miRNATissueAtlas 2025 (Rishik et al. Nucleic Acids Research, 2024). Unfortunately, due to restrictions in tissue availability and the lack of stored RNA samples, we are unable to measure miR-26b expression in the human livers. However, based on the potency of the miR-26b mimic loaded LNPs in the mice (Revised Supplemental Figure 2A-B), we are confident that these LNPs also resulted in a overexpression of miR-26b in the human livers.

(3) Please explain the rationale behind primarily using whole-body miR-26b KO mice rather than the myeloid cell-specific KO model for the studies.

The main goal of our study is the elucidation of the general role of miR-26b in MASH formation. Therefore, we decided to primarily focus on the whole-body KO model. While we used the myeloid cell-specific KO model to highlight that myeloid cells play an important role in the observed phenotypes, we believe the whole-body KO model is more appropriate as main focus, particularly also in light of the used LNP targeting that also provides a whole-body approach. Furthermore, this focus on the whole-body model also reflects a more therapeutically relevant approach.

(4) The authors claim that treatment with LNPs containing miR-26b "replenish the miR-26b level in the whole-body deficient mouse" but the results of this observation are not presented.

This is indeed a valid point that we have now addressed. We have measured the mir26b-3p and mir26b-5p expression levels in livers from mice after 4-week WTD with simultaneous injection with either empty LNPs as vehicle control (eLNP) or LNPs containing miR-26b mimics (mLNP) every 3 days. As shown in Revised Supplemental Figure 2A-B, mLNP treatment clearly results in an overexpression of the mir26b in the livers of these mice. We have rephrased the text accordingly by stating that mLNP results in an “overexpression” rather than “replenishment”.

(5) The number of 3 human donors for the precision-cut liver slices is clearly insufficient and clinical parameters need to be shown. Additionally, inconsistencies in individual values in Figures 8B-E need clarification.

Unfortunately, due to restrictions in tissue availability, we are unable to increase our n-number for these experiments. Clinical parameters are not available, but the liver slices were from healthy tissue.

We have performed these experiments in duplicates for each individual donor. We have now specified this also in the figure legend to explain the individual values in the graphs:

“…(3 individual donors, cultured in duplicates).”

(6) Figure 2D: Please include representative images.

As suggested we have included representative images in our revised manuscript.

(7) Address discrepancies in the findings across different experimental settings. For example, the expression levels of the lipid metabolism-related genes are not significantly modulated in whole-body miR-26b KO mice (except for Sra), but they are in the myeloid cell-specific model (but not Sra), and none of them are restored after LNPs injections.

Although Cd36 is not significantly increased in the whole-body miR-26b KO mice, there is a clear tendency towards increased expression, which is now also validated on protein level (Revised Figure 1K-L). In the myeloid cell-specific model we see a similar tendency, although the gene expression difference of Sra is not significantly changed. This could be due to the difference in the model, since only myeloid cells are affected, suggesting that the effects on Sra are to a large extend driven by non-myeloid cells. This would also fit to the tendency to decreased Sra expression in the mimic-LNP treated mice. Due to the larger variation, this difference did not reach significance, which is rather a statistical issue due to relatively small n-numbers. At this moment, we cannot exclude that these receptors are differentially regulated by different cell-types. For this, future studies are needed focussing on cell-specific targeting of miR-26b in somatic cells, like hepatocytes.

(8) Figure 4A the images are not representative of the quantification.

We have selected another representative image that is exactly reflecting the average Sirius red positive area, to reflect the quantification appropriately.

(9) Figures 5 and 7: Are there not significantly decreased/increased kinases? A deeper analysis of these kinase alterations is necessary to understand how miR-26b exerts its role. A comparison analysis of these two datasets might clarify this regard.

We indeed very often see in these kinome analysis that the general tendency of kinase activity is unidirectional. We believe that this is caused by the highly interconnected nature of kinases. Activation of one signalling cascade will also results in the activation of many other cascades. However, it is interesting to see which pathways are affected in our study and we find it particularly interesting to see that the tendencies is exactly opposite between both comparisons as KO vs. WT shows increase kinase activities, while KO-LNP vs. KO shows a decrease again. Further showing that the method is reflecting a true biological effect that is mediated by miR26b.

(10) Determinations of the effect of LNPs containing miR-26b in the KO mice are limited to only a few observations (that are not only significant). More extensive findings are needed to conclusively demonstrate the effectiveness of this treatment method. Similar to the experiments with human liver samples (Figures 8A-E).

We have now elaborated our observations in the mouse model using LNPs by also analysing the effects on inflammation and fibrosis. However, it seems that the treatment time was not long enough to see pronounced changes on these later stages of disease development. Interestingly, the expression of Tgfb was significantly reduced, suggesting at least that the LNPs on genetic levels have an effect already on fibrotic processes. Thereby, it can be suggested that longer mLNP treatment may result in more effects on protein level as well, which remains to be determined in future studies.

Unfortunately, due to restrictions in tissue availability, we are unable to increase our n-number or read-outs for these experiments at this moment.

(11) In Figures 8F-H, the observed increase in circulating miR-26b levels in the plasma of cirrhotic individuals seems contradictory to its proposed protective role. This discrepancy requires clarification.

In the revised discussion (second to last paragraph), we have now elaborated more on the findings in the plasma of cirrhotic individuals in comparison to our murine in-vivo results, to highlight and discuss this discrepancy.

(12) Figures 8F-H legend mentions using 8-11 patients per group, but the methods section lacks corresponding information about these individuals.

These patients, together with inclusion/exclusion criteria and definition of cirrhosis are described in the method section 2.14.

(13) Figure 8G has 7 data points in the cirrhosis group, instead of 8. Any data exclusion should be justified in the methods section.

As defined in method section 2.15, we have identified outliers using the ROUT = 1 method, which is the reason why Figure 8G only has 7 data points instead of 8.

Reviewer #2 (Public Review):

Summary:

This manuscript by Peters, Rakateli, et al. aims to characterize the contribution of miR-26b in a mouse model of metabolic dysfunction-associated steatohepatitis (MASH) generated by a Western-type diet on the background of Apoe knock-out. In addition, the authors provide a rescue of the miR-26b using lipid nanoparticles (LNPs), with potential therapeutic implications. In addition, the authors provide useful insights into the role of macrophages and some validation of the effect of miR-26b LNPs on human liver samples.

Strengths:

The authors provide a well-designed mouse model, that aims to characterize the role of miR-26b in a mouse model of metabolic dysfunction-associated steatohepatitis (MASH) generated by a Western-type diet on the background of Apoe knock-out. The rescue of the phenotypes associated with the model used using miR-26b using lipid nanoparticles (LNPs) provides an interesting avenue to novel potential therapeutic avenues.

Weaknesses:

Although the authors provide a new and interesting avenue to understand the role of miR-26b in MASH, the study needs some additional validations and mechanistic insights in order to strengthen the author's conclusions.

(1) Analysis of the expression of miRNAs based on miRNA-seq of human samples (see https://ccb-compute.cs.uni-saarland.de/isomirdb/mirnas) suggests that miR-26b-5p is highly abundant both on liver and blood. It seems hard to reconcile that despite miRNA abundance being similar in both tissues, the physiological effects claimed by the authors in Figure 2 come exclusively from the myeloid (macrophages).

We agree with the reviewer that the effects observed in the whole-body KO model are most likely a combination of cellular effects, particularly since miR-26b is also highly expressed in the liver. However, with the LysM-model we merely want to demonstrate that the myeloid cells at least play an important, though not exclusive, role in the phenotype. In the discussion, we also further elaborate on the fact that the observed changes in the liver can me mediated by hepatic changes.

To stress this, we have adjusted the conclusion of Figure 2:

“Interestingly, mice that have a myeloid-specific lack of miR-26b also show increased hepatic cholesterol levels and lipid accumulation demonstrated by Oil-red-O staining, coinciding with an increased hepatic Cd36 expression (Figure 2), demonstrating that myeloid miR-26b plays a major, but not exclusive, role in the observed steatosis.”

(2) Similarly, the miRNA-seq expression from isomirdb suggests also that expression of miR-26a-5p is indeed 4-fold higher than miR-26b-5p both in the liver and blood. Since both miRNAs share the same seed sequence, and most of the supplemental regions (only 2 nt difference), their endogenous targets must be highly overlapped. It would be interesting to know whether deletion of miR-26b is somehow compensated by increased expression of miR-26a-5p loci. That would suggest that the model is rather a depletion of miR-26.

UUCAAGUAAUUCAGGAUAGGU mmu-miR-26b-5p mature miRNA

UUCAAGUAAUCCAGGAUAGGCU mmu-miR-26a-5p mature miRNA

This is a very valid point raised by the reviewer, which we actually already explored in a previous study, describing our used mouse model (Van der Vorst et al. BMC Genom Data, 2021). In this manuscript, we could show that miR-26a is not affected by the deficiency of miR-26b (Figure 1G in: Van der Vorst et al. BMC Genom Data, 2021).

(3) Similarly, the miRNA-seq expression from isomirdb suggests also that expression of miR-26b-5p is indeed 50-fold higher than miR-26b-3p in the liver and blood. This difference in abundance of the two strands is usually regarded as one of them being the guide strand (in this case the 5p) and the other being the passenger (in this case the 3p). In some cases, passenger strands can be a byproduct of miRNA biogenesis, thus the rescue experiments using LNPs with both strands in equimolar amounts would not reflect the physiological abundance miR-26b-3p. The non-physiological overabundance of miR-26b-3p would constitute a source of undesired off-targets.

We agree with the reviewer on this aspect and this is something we had to consider while generating the mimic LNPs. However, we believe that we do not observe and undesired off-target effects, as the effects of the mimic LNPs at least on functional outcomes are relatively mild and only restricted to the expected effects on lipids. Furthermore, the effects on the kinase profile due to the mimic LNP treatment are in line with our expectations. Combined these results suggest at least that potential off-target effects are minor.

(4) It would also be valuable to check the miRNA levels on the liver upon LNP treatment, or at least the signatures of miR-26b-3p and miR-26b-5p activity using RNA-seq on the RNA samples already collected.

This is indeed a valid point that we have now addressed. We have measured the mir26b-3p and mir26b-5p expression levels in livers from mice after 4-week WTD with simultaneous injection with either empty LNPs as vehicle control (eLNP) or LNPs containing miR-26b mimics (mLNP) every 3 days. As shown in Supplemental Figure 2A-B, mLNP treatment clearly results in an overexpression of the mir26b in the livers of these mice. We have rephrased the text accordingly by stating that mLNP results in an “overexpression” rather than “replenishment”.

(5) Some of the phenotypes described, such as the increase in cholesterol, overlap with the previous publication by van der Vorst et al. BMC Genom Data (2021), despite in this case the authors are doing their model in Apoe knock-out and Western-type diet. I would encourage the authors to investigate more or discuss why the initial phenotypes don't become more obvious despite the stressors added in the current manuscript.

In our previous publication (BMC Genom Data; 2021), we actually did not see any changes in circulating lipid levels. However, in that study we did not evaluate the livers of the mice, so we do not have any information about the hepatic lipid levels.

As mentioned by the reviewer, we believe that we see much more pronounced phenotypes in the current model because we use the combined stressor of Apoe-/- and Western-type diet, which cannot be compared to the wildtype and chow-fed mice used in the BMC Genom Data manuscript.

(6) The authors have focused part of their analysis on a few gene makers that show relatively modest changes. Deeper characterization using RNA-seq might reveal other genes that are more profoundly impacted by miR-26 depletion. It would strengthen the conclusions proposed if the authors validated that changes in mRNA abundance (Sra, Cd36) do impact the protein abundance. These relatively small changes or trends in mRNA expression, might not translate into changes in protein abundance.

As suggested by the reviewer we have now also confirmed that the protein expression of CD36 and SRA is significantly increased upon miR-26b depletion, visualized as Figure 1K-L in the revised manuscript. Unfortunately, we do not have enough material left to perform similar analysis for the LysM-model or the LNP-model, although based on the whole-body effects we are confident that at least for CD36/SRA in this case the gene expression matches effects observed on protein level.

(7) In Figures 5 and 7, the authors run a phosphorylation array (STK) to analyze the changes in the activity of the kinome. It seems that a relatively large number of signaling pathways are being altered, I think that should be strengthened by further validations by Western blot on the collected tissue samples. For quite a few of the kinases, there might be antibodies that recognise phosphorylation. The two figures lack a mechanistic connection to the rest of the manuscript.<br /> On this point we respectfully have to disagree with the reviewer. We have used a kinase activity profiling approach (PamGene) to analyse the real-time activity of kinases in our lysates. This approach is different than the classical Western blot approach in which only the presence or absence of a specific phosphorylation is detected. Thereby, Western blot analysis does not analyse phosphorylation in real-time, but rather determines whether there has been phosphorylation in the past. Our approach actually determines the real-time, current activity of the kinases, which we believe is a different and perhaps even more reliable read-out measurement. Therefore, validation by Western blot would not strengthen these observations.

We have particularly tried to connect these observations to the rest of the manuscript by highlighting the observed signalling cascades that are affected, highlighting a role in inflammation and angiogenesis, thereby providing some mechanistic insights.

Reviewer #2 (Recommendations For The Authors):

I would encourage the authors to follow-up on some of the more miRNA focused comments made above, which would strengthen the mechanistic part of the work presented.

I suggest the authors tone down some of some of the claims made (eg. "clearly increased expression", "exacerbated hepatic fibrosis"), given that some of it might need further validation.

Wherever needed we have tuned down the tone of some claims, although we believe that most claims are already written carefully enough and in line with the observed results.

Some of the panels that are supposed to have the same amount of animals have variable N, despite they come from the same exact number of RNA samples or tissue lysates (eg. 1G and 1H, vs 1I and 1J).

This is indeed correct and caused by the fact that some analysis resulted in statistical outliers as identified using the ROUT = 1 method, as also specified in section 2.15 of the method section.

It would be nice to have representative images of oil-red-o in all the figures where it is quantified (or at least in the supplementary figures).

As suggested by the reviewer, we have now included representative images for the LysM-model (Revised Figure 2D) and the LNP-model (Revised Figure 6D) as well.

Author response:

The following is the authors’ response to the original reviews

Public Reviews:

Reviewer #1 (Public Review):

In this study, the authors aim to understand why decision formation during behavioural tasks is distributed across multiple brain areas. They hypothesize that multiple areas are used in order to implement an information bottleneck (IB). Using neural activity recorded from monkey DLPFC and PMd performing a 2-AFC task, they show that DLPFC represents various task variables (decision, color, target configuration), while downstream PMd primarily represents decision information. Since decision information is the only information needed to make a decision, the authors point out that PMd has a minimal sufficient representation (as expected from an IB). They then train 3-area RNNs on the same task and show that activity in the first and third areas resemble the neural representations of DLPFC and PMd, respectively. In order to propose a mechanism, they analyse the RNN and find that area 3 ends up with primarily decision information because feedforward connections between areas primarily propagate decision information.

The paper addresses a deep, normative question, namely why task information is distributed across several areas.

Overall, it reads well and the analysis is well done and mostly correct (see below for some comments). My major problem with the paper is that I do not see that it actually provides an answer to the question posed (why is information distributed across areas?). I find that the core problem is that the information bottleneck method, which is evoked throughout the paper, is simply a generic compression method.

Being a generic compressor, the IB does not make any statements about how a particular compression should be distributed across brain areas - see major points (1) and (2).

If I ignore the reference to the information bottleneck and the question of why pieces of information are distributed, I still see a more mechanistic study that proposes a neural mechanism of how decisions are formed, in the tradition of RNN-modelling of neural activity as in Mante et al 2013. Seen through this more limited sense, the present study succeeds at pointing out a good model-data match, and I could support a publication along those lines. I point out some suggestions for improvement below.

We thank the reviewer for their comments, feedback and suggestions. We are glad to hear you support the good model-data match for this manuscript.  With your helpful comments, we have clarified the connections to the information bottleneck principle and also contrasted it against the information maximization principle (the InfoMax principle), an alternative hypothesis. We elaborate on these issues in response to your points below, particularly major points (1) and (2). We also address all your other comments below.

Major points

(1) It seems to me that the author's use of the IB is based on the reasoning that deep neural networks form decisions by passing task information through a series of transformations/layers/areas and that these deep nets have been shown to implement an IB. Furthermore, these transformations are also loosely motivated by the data processing inequality.

On Major Point 1 and these following subpoints, we first want to make a high-level statement before delving into a detailed response to your points as it relates to the information bottleneck (IB). We hope this high-level statement will provide helpful context for the rest of our point-by-point responses. 

We want to be clear that we draw on the information bottleneck (IB) principle as a general principle to explain why cortical representations differ by brain area. The IB principle, as applied to cortex, is only stating that a minimal sufficient representation to perform the task is formed in cortex, not how it is formed. The alternative hypothesis to the IB is that brain areas do not form minimal sufficient representations. For example, the InfoMax principle states that each brain area stores information about all inputs (even if they’re not necessary to perform the task). InfoMax isn’t unreasonable: it’s possible that storing as much information about the inputs, even in downstream areas, can support flexible computation and InfoMax also supports redundancy in cortical areas. Indeed, many studies claim that action choice related signals are in many cortical areas, which may reflect evidence of an InfoMax principle in action for areas upstream of PMd.

While we observe an IB in deep neural networks and cortex in our perceptual decision-making task, we stress that its emergence across multiple areas is an empirical result. At the same time, multiple areas producing an IB makes intuitive sense: due to the data processing inequality, successive transformations typically decrease the information in a representation (especially when, e.g., in neural networks, every activation passes through the Relu function, which is not bijective). Multiple areas are therefore a sufficient and even ‘natural’ way to implement an IB, but multiple areas are not necessary for an IB. That we observe an IB in deep neural networks and cortex emerge through multi-area computation is empirical, and, contrasting InfoMax, we believe it is an important result of this paper. 

Nevertheless, your incisive comments have helped us to update the manuscript that when we talk about the IB, we should be clear that the alternative hypothesis is non-minimal representations, a prominent example of which is the InfoMax principle. We have now significantly revised our introduction to avoid this confusion. We hope this provides helpful context for our point-by-point replies, below.

However, assuming as a given that deep neural networks implement an IB does not mean that an IB can only be implemented through a deep neural network. In fact, IBs could be performed with a single transformation just as well. More formally, a task associates stimuli (X) with required responses (Y), and the IB principle states that X should be mapped to a representation Z, such that I(X;Z) is minimal and I(Y,Z) is maximal. Importantly, the form of the map Z=f(X) is not constrained by the IB. In other words, the IB does not impose that there needs to be a series of transformations. I therefore do not see how the IB by itself makes any statement about the distribution of information across various brain areas.

We agree with you that an IB can be implemented in a single transformation. We wish to be clear that we do not intend to argue necessity: that multiple areas are the only way to form minimal sufficient representations. Rather, multiple areas are sufficient to induce minimal sufficient representations, and moreover, they are a natural and reasonably simple way to do so. By ‘natural,’ we mean that minimal sufficient representations empirically arise in systems with multiple areas (more than 2), including deep neural networks and the cortex at least for our task and simulations. For example, we did not see minimal sufficient representations in 1- or 2-area RNNs, but we did see them emerge in RNNs with 3 areas or more. One potential reason for this result is that sequential transformations through multiple areas can never increase information about the input; it can only maintain or reduce information due to the data processing inequality.

Our finding that multiple areas facilitate IBs in the brain is therefore an empirical result: like in deep neural networks, we observe the brain has minimal sufficient representations that emerge in output areas (PMd), even as an area upstream (DLPFC) is not minimal. While the IB makes a statement that this minimal sufficient representation emerges, to your point, the fact that it emerges over multiple areas is not a part of the IB – as you have pointed out, the IB doesn’t state where or how the information is discarded, only that it is discarded. Our RNN modeling later proposes one potential mechanism for how it is discarded. We updated the manuscript introduction to make these points:

“An empirical observation from Machine Learning is that deep neural networks tend to form minimal sufficient representations in the last layers. Although multi-layer computation is not necessary for an IB, they provide a sufficient and even “natural” way to form an IB. A representation z = f(x) cannot contain more information than the input x itself due to the data processing inequality[19]. Thus, adding additional layers typically results in representations that contain less information about the input.”

And later in the introduction:

“Consistent with these predictions of the IB principle, we found that DLPFC has information about the color, target configuration, and direction. In contrast, PMd had a minimal sufficient representation of the direction choice. Our recordings therefore identified a cortical IB. However, we emphasize the IB does not tell us where or how the minimal sufficient representation is formed. Instead, only our empirical results implicate DLPFC-PMd in an IB computation. Further, to propose a mechanism for how this IB is formed, we trained a multi-area RNN to perform this task. We found that the RNN faithfully reproduced DLPFC and PMd activity, enabling us to propose a mechanism for how cortex uses multiple areas to compute a minimal sufficient representation.”

In the context of our work, we want to be clear the IB makes these predictions:

Prediction 1: There exists a downstream area of cortex that has a minimal and sufficient representation to perform a task (i.e.,. I(X;Z) is minimal while preserving task information so that I(Z;Y) is approximately equal to  I(X;Y)). We identify PMd as an area with a minimal sufficient representation in our perceptual-decision-making task. 

Prediction 2 (corollary if Prediction 1 is true): There exists an upstream brain area that contains more input information than the minimal sufficient area. We identify DLPFC as an upstream area relative to PMd, which indeed has more input information than downstream PMd in our perceptual decision-making task. 

Note: as you raise in other points, it could have been possible that the IB is implemented early on, e.g., in either the parietal cortex (dorsal stream) or inferotemporal cortex (ventral stream), so that DLPFC and PMd both contained minimal sufficient representations. The fact that it doesn’t is entirely an empirical result from our data. If DLPFC had minimal sufficient representations for the perceptual decision making task, we would have needed to record in other regions to identify brain areas that are consistent with Prediction 2. But, empirically, we found that DLPFC has more input information relative to PMd, and therefore the DLPFC-PMd connection is implicated in the IB process.

What is the alternative hypothesis to the IB? We want to emphasize: it isn’t single-area computation. It’s that the cortex does not form minimal sufficient representations. For example, an alternative hypothesis (“InfoMax”) would be for all engaged brain areas to form representations that retain all input information. One reason this could be beneficial is because each brain area could support a variety of downstream tasks. In this scenario, PMd would not be minimal, invalidating Prediction 1. However, this is not supported by our empirical observations of the representations in PMd, which has a minimal sufficient representation of the task. We updated our introduction to make this clear:

“But cortex may not necessarily implement an IB. The alternative hypothesis to IB is that the cortex does not form minimal sufficient representations. One manifestation of this alternative hypothesis is the “InfoMax” principle, where downstream representations are not minimal but rather contain maximal input information22. This means information about task inputs not required to perform the task are present in downstream output areas. Two potential benefits of an InfoMax principle are (1) to increase redundancy in cortical areas and thereby provide fault tolerance, and (2) for each area to support a wide variety of tasks and thereby improve the ability of brain areas to guide many different behaviors. In contrast to InfoMax, the IB principle makes two testable predictions about cortical representations. Prediction 1: there exists a downstream area of cortex that has a minimal and sufficient representation to perform a task (i.e., I(X; Z) is minimal while preserving task information so that I(Z; Y) ≈ I(X; Y)). Prediction 2 (corollary if Prediction 1 is true): there exists an upstream area of cortex that has more task information than the minimal sufficient area.”

Your review helped us realize we should have been clearer in explaining that these are the key predictions of the IB principle tested in our paper. We also realized we should be much clearer that these predictions aren’t trivial or expected, and there is an alternative hypothesis. We have re-written the introduction of our paper to highlight that the key prediction of the IB is minimal sufficient representations for the task, in contrast to the alternative hypothesis of InfoMax.

A related problem is that the authors really only evoke the IB to explain the representation in PMd: Fig 2 shows that PMd is almost only showing decision information, and thus one can call this a minimal sufficient representation of the decision (although ignoring substantial condition independent activity).

However, there is no IB prediction about what the representation of DLPFC should look like.

Consequently, there is no IB prediction about how information should be distributed across DLPFC and PMd.

We agree: the IB doesn’t tell us how information is distributed, only that there is a transformation that eventually makes PMd minimal. The fact that we find input information in DLPFC reflects that this computation occurs across areas, and is an empirical characterization of this IB in that DLPFC has direction, color and context information while PMd has primarily direction information. To be clear: only our empirical recordings verified that the DLPFC-PMd circuit is involved in the IB. As described above, if not, we would have recorded even further upstream to identify an inter-areal connection implicated in the IB.

We updated the text to clearly state that the IB predicts that an upstream area’s activity should contain more information about the task inputs. We now explicitly describe this in the introduction, copy and pasted again here for convenience.

“In contrast to InfoMax, the IB principle makes two testable predictions about cortical representations. Prediction 1: there exists a downstream area of cortex that has a minimal and sufficient representation to perform a task (i.e., I(X; Z) is minimal while preserving task information so that I(Z; Y) ≈ I(X; Y)). Prediction 2 (corollary if Prediction 1 is true): there exists an upstream area of cortex that has more task information than the minimal sufficient area.

Consistent with the predictions of the IB principle, we found that DLPFC has information about the color, target configuration, and direction. In contrast, PMd had a minimal sufficient representation of the direction choice. Our recordings therefore identified a cortical IB. However, we emphasize the IB does not tell us where or how the minimal sufficient representation is formed. Instead, only our empirical results implicate DLPFC-PMd in an IB computation Further, to propose a mechanism for how this IB is formed, we trained a multi-area RNN to perform this task.”  

The only way we knew DLPFC was not minimal was through our experiments. Please also note that the IB principle does not describe how information could be lost between areas or layers, whereas our RNN simulations show that this may occur through preferential propagation of task-relevant information with respect to the inter-area connections.  

(2) Now the authors could change their argument and state that what is really needed is an IB with the additional assumption that transformations go through a feedforward network. However, even in this case, I am not sure I understand the need for distributing information in this task. In fact, in both the data and the network model, there is a nice linear readout of the decision information in dPFC (data) or area 1 (network model). Accordingly, the decision readout could occur at this stage already, and there is absolutely no need to tag on another area (PMd, area 2+3).

Similarly, I noticed that the authors consider 2,3, and 4-area models, but they do not consider a 1-area model. It is not clear why the 1-area model is not considered. Given that e.g. Mante et al, 2013, manage to fit a 1-area model to a task of similar complexity, I would a priori assume that a 1-area RNN would do just as well in solving this task.

While decision information could indeed be read out in Area 1 in our multi-area model, we were interested in understanding how the network converged to a PMd-like representation (minimal sufficient) for solving this task. Empirically, we only observed a match between our model representations and animal cortical representations during this task when considering multiple areas. Given that we empirically observed that our downstream area had a minimal sufficient representation, our multi-area model allowed how this minimal sufficient representation emerged (through preferential propagation of task-relevant information).

We also analyzed single-area networks in our initial manuscript, though we could have highlighted these analyses more clearly to be sure they were not overlooked. We are clearer in this revision that we did consider a 1-area network (results in our Fig 5). While a single-area RNN can indeed solve this task, the single area model had all task information present in the representation, and did not match the representations in DLPFC or PMd. It would therefore not allow us to understand how the network converged to a PMd-like representation (minimal sufficient) for solving this task. We updated the schematic in Fig 5 to add in the single-area network (which may have caused the confusion).

We have added an additional paragraph commenting on this in the discussion. We also added an additional supplementary figure with the PCs of the single area RNN (Fig S15). We highlight that single area RNNs do not resemble PMd activity because they contain strong color and context information. 

In the discussion:

“We also found it was possible to solve this task with single area RNNs, although they did not resemble PMd (Figure S15) since it did not form a minimal sufficient representation. Rather, for our RNN simulations, we found that the following components were sufficient to induce minimal sufficient representations: (1) RNNs with at least 3 areas, following Dale’s law (independent of the ratio of feedforward to feedback connections).”

I think there are two more general problems with the author's approach. First, transformations or hierarchical representations are usually evoked to get information into the right format in a pure feedforward network. An RNN can be seen as an infinitely deep feedforward network, so even a single RNN has, at least in theory, and in contrast to feedforward layers, the power to do arbitrarily complex transformations. Second, the information coming into the network here (color + target) is a classical xor-task. While this task cannot be solved by a perceptron (=single neuron), it also is not that complex either, at least compared to, e.g., the task of distinguishing cats from dogs based on an incoming image in pixel format.

An RNN can be viewed as an infinitely deep feedforward network in time. However, we wish to clarify two things. First, our task runs for a fixed amount of time, and therefore this RNN in practice is not infinitely deep in time. Second, if it were to perform an IB operation in time, we would expect to see color discriminability decrease as a function of time. Indeed, we considered this as a mechanism (recurrent attenuation, Figure 4a), but as we show in Supplementary Figure S9, we do not observe it to be the case that discriminability decreases through time. This is equivalent to a dynamical mechanism that removes color through successive transformations in time, which our analyses reject (Fig 4). We therefore rule out that an IB is implemented through time via an RNN’s recurrent computation (viewed as feedforward in time). Rather, as we show, the IB comes primarily through inter-areal connections between RNN areas. We clarified that our dynamical hypothesis is equivalent to rejecting the feedforward-in-time filtering hypothesis in the Results: 

“We first tested the hypothesis that the RNN IB is implemented primarily by recurrent dynamics (left side of Fig. 4a). These recurrent dynamics can be equivalently interpreted as the RNN implementing a feedforward neural network in time.”  

The reviewer is correct that the task is a classical XOR task and not as complex as e.g., computer vision classification. That said, our related work has looked at IBs for computer vision tasks and found them in deep feedforward networks (Kleinman et al., ICLR 2021). Even though the task is relatively straightforward, we believe it is appropriate for our conclusions because it does not have a trivial minimal sufficient representation: a minimal sufficient representation for XOR must contain only target, but not color or target configuration information. This can only be solved via a nonlinear computation. In this manner, we favor this task because it is relatively simple, and the minimal sufficient representations are interpretable, while at the same time not being so trivially simple (the minimal sufficient representations require nonlinearity to compute).  

Finally, we want to note that this decision-making task is a logical and straightforward way to add complexity to classical animal decision-making tasks, where stimulus evidence and the behavioral report are frequently correlated. In tasks such as these, it may be challenging to untangle stimulus and behavioral variables, making it impossible to determine if an area like premotor cortex represents only behavior rather than stimulus. However, our task decorrelates both the stimulus and the behaviors. 

(3) I am convinced of the author's argument that the RNN reproduces key features of the neural data. However, there are some points where the analysis should be improved.

(a) It seems that dPCA was applied without regularization. Since dPCA can overfit the data, proper regularization is important, so that one can judge, e.g., whether the components of Fig.2g,h are significant, or whether the differences between DLPFC and PMd are significant.

We note that the dPCA codebase optimizes the regularization hyperparameter through cross-validation and requires single-trial firing rates for all neurons, i.e., data matrices of the form (n_Neurons x Color x Choice x Time x n_Trials), which are unavailable for our data. We recognized that you are fundamentally asking whether differences are significant or not. We therefore believe it is possible to address this through a statistical test, described further below. 

In order to test whether the differences of variance explained by task variables between DLPFC and PMd are significant, we performed a shuffle test. For this test, we randomly sampled 500 units from the DLPFC dataset and 500 units from the PMd dataset. We then used dPCA to measure the variance explained by target configuration, color choice, and reach direction (e.g., Var^True_DLPFC,Color, Var^True_PMd,Color).

To test if this variance was significant, we performed the following shuffle test. We combined the PMd and DLPFC dataset into a pool of 1000 units and then randomly selected 500 units from this pool to create a surrogate PMd dataset and used the remaining 500 units as a surrogate DLPFC dataset. We then again performed dPCA on these surrogate datasets and estimated the variance for the various task variables (e.g., Var_{ShuffledDLPFC,Color}  ,Var_{ShuffledPMd,Color}).

We repeated this process for 100 times and estimated a sampling distribution for the true difference in variance between DLPFC and PMd for various task variables (e.g., Var^True_DLPFC,Color - Var^True_PMd,Color). At the same time, we estimated the distribution of the variance difference between surrogate PMd and DLPFC dataset for various task variables (e.g., Var_{ShuffleDLPFC,Color} - Var_{ShufflePMd,Color}). 

We defined a p-value as the number of shuffles in which the difference in variance was higher than the median of the true difference and divided it by 100. Note, for resampling and shuffle tests with n shuffles/bootstraps, the lowest theoretical p-value is given as 2/n, even in the case that no shuffle was higher than the median of the true distribution. Thus, the differences were statistically significant (p < 0.02) for color and target configuration but not for direction (p=0.72). These results are reported in Figure S6 and show both the true sampling distribution and the shuffled sampling distributions.

(b) I would have assumed that the analyses performed on the neural data were identical to the ones performed on the RNN data. However, it looked to me like that was not the case. For instance, dPCA of the neural data is done by restretching randomly timed trials to a median trial. It seemed that this restretching was not performed on the RNN. Maybe that is just an oversight, but it should be clarified. Moreover, the decoding analyses used SVC for the neural data, but a neural-net-based approach for the RNN data. Why the differences?

Thanks for bringing up these points. We want to clarify that we did include SVM decoding for the multi-area network in the appendix (Fig. S4), and the conclusions are the same. Moreover, in previous work, we also found that training with a linear decoder led to analogous conclusions (Fig. 11 of Kleinman et al, NeurIPS 2021).  As we had a larger amount of trials for the RNN than the monkey, we wanted to allow a more expressive decoder for the RNN, though this choice does not affect our conclusions. We clarified the text to reflect that we did use an SVM decoder.

“We also found analogous conclusions when using an SVM decoder (Fig. S4).”

dPCA analysis requires trials of equal length. For the RNN, this is straightforward to generate because we can set the delay lengths to be equal during inference (although the RNN was trained on various length trials and can perform various length trials). Animals must have varying delay periods, or else they will learn the timing of the task and anticipate epoch changes. Because animal trial lengths were therefore different, their trials had to be restretched. We clarified this in the Methods.

“For analyses of the RNN, we fixed the timing of trials, obviating the need to to restretch trial lengths. Note that while at inference, we generated RNN trials with equal length, the RNN was trained with varying delay periods.” 

(4) The RNN seems to fit the data quite nicely, so that is interesting. At the same time, the fit seems somewhat serendipitous, or at least, I did not get a good sense of what was needed to make the RNN fit the data. The authors did go to great lengths to fit various network models and turn several knobs on the fit. However, at least to me, there are a few (obvious) knobs that were not tested.

First, as already mentioned above, why not try to fit a single-area model? I would expect that a single area model could also learn the task - after all, that is what Mante et al did in their 2013 paper and the author's task does not seem any more complex than the task by Mante and colleagues.

Thank you for bringing up this point. As mentioned in response to your prior point, we did analyze a single-area RNN (Fig. 5d). We updated the schematic to clarify that we analyzed a single area network. Moreover, we also added a supplementary figure to qualitatively visualize the PCs of the single area network (Fig. S15). While a single area network can solve the task, it does not allow us to study how representations change across areas, nor did it empirically resemble our neural recordings. Single-area networks contain significant color, context, and direction information. They therefore do not form minimal representations and do not resemble PMd activity.

Second, I noticed that the networks fitted are always feedforward-dominated. What happens when feedforward and feedback connections are on an equal footing? Do we still find that only the decision information propagates to the next area? Quite generally, when it comes to attenuating information that is fed into the network (e.g. color), then that is much easier done through feedforward connections (where it can be done in a single pass, through proper alignment or misalignment of the feedforward synapses) than through recurrent connections (where you need to actively cancel the incoming information). So it seems to me that the reason the attenuation occurs in the inter-area connections could simply be because the odds are a priori stacked against recurrent connections. In the real brain, of course, there is no clear evidence that feedforward connections dominate over feedback connections anatomically.

We want to clarify that we did pick feedforward and feedback connections based on the following macaque atlas, reference 27 in our manuscript: 

Markov, N. T., Ercsey-Ravasz, M. M., Ribeiro Gomes, A. R., Lamy, C., Magrou, L., Vezoli, J., Misery, P., Falchier, A., Quilodran, R., Gariel, M. A., Sallet, J., Gamanut, R., Huissoud, C., Clavagnier, S., Giroud, P., Sappey-Marinier, D., Barone, P., Dehay, C., Toroczkai, Z., … Kennedy, H. (2014). A weighted and directed interareal connectivity matrix for macaque cerebral cortex. Cerebral Cortex , 24(1), 17–36.

We therefore believe there is evidence for more feedforward than feedback connections. Nevertheless, as stated in response to your next point below, we ran a simulation where feedback and feedforward connectivity were matched.

More generally, it would be useful to clarify what exactly is sufficient:

(a) the information distribution occurs in any RNN, i.e., also in one-area RNNs

(b) the information distribution occurs when there are several, sparsely connected areas

(c) the information distribution occurs when there are feedforward-dominated connections between areas

We better clarify what exactly is sufficient. 

- We trained single-area RNNs and found that these RNNs contained color information; additionally two area RNNs also contained color information in the last area (Fig 5d). 

- We indeed found that the minimal sufficient representations emerged when we had several areas, with Dale’s law constraint on the connectivity. When we had even sparser connections, without Dale’s law, there was significantly more color information, even at 1% feedforward connections; Fig 5a.

- When we matched the percentage of feedforward and feedback connections with Dale’s law constraint on the connectivity (10% feedforward and 10% feedback), we also observed minimal sufficient representations (Fig S9). 

Together, we found that minimal sufficient representations emerged when we had several areas (3 or greater), with Dale’s law constraint on the connectivity, independent of the ratio of feedforward/feedback connections. We thank the reviewer for raising this point about the space of constraints leading to minimal sufficient representations in the late area. We clarified this in the Discussion.

“We also found it was possible to solve this task with single area RNNs, although they did not resemble PMd (Figure S15) since it did not form a minimal sufficient representation. Rather, for our RNN simulations, we found that the following components were sufficient to induce minimal sufficient representations: RNNs with at least 3 areas, following Dale’s law (independent of the ratio of feedforward to feedback connections).”

Thank you for your helpful and constructive comments!

Reviewer #2 (Public Review):

Kleinman and colleagues conducted an analysis of two datasets, one recorded from DLPFC in one monkey and the other from PMD in two monkeys. They also performed similar analyses on trained RNNs with various architectures.

The study revealed four main findings. (1) All task variables (color coherence, target configuration, and choice direction) were found to be encoded in DLPFC. (2) PMD, an area downstream of PFC, only encoded choice direction. (3) These empirical findings align with the celebrated 'information bottleneck principle,' which suggests that FF networks progressively filter out task-irrelevant information. (4) Moreover, similar results were observed in RNNs with three modules.

We thank the reviewer for their comments, feedback and suggestions, which we address below.

While the analyses supporting results 1 and 2 were convincing and robust, I have some concerns and recommendations regarding findings 3 and 4, which I will elaborate on below. It is important to note that findings 2 and 4 had already been reported in a previous publication by the same authors (ref. 43).

Note the NeurIPS paper only had PMd data and did not contain any DLPFC data. That manuscript made predictions about representations and dynamics upstream of PMd, and subsequent experiments reported in this manuscript validated these predictions. Importantly, this manuscript observes an information bottleneck between DLPFC and PMd.

Major recommendation/comments:

The interpretation of the empirical findings regarding the communication subspace in relation to the information bottleneck theory is very interesting and novel. However, it may be a stretch to apply this interpretation directly to PFC-PMd, as was done with early vs. late areas of a FF neural network.

In the RNN simulations, the main finding indicates that a network with three or more modules lacks information about the stimulus in the third or subsequent modules. The authors draw a direct analogy between monkey PFC and PMd and Modules 1 and 3 of the RNNs, respectively. However, considering the model's architecture, it seems more appropriate to map Area 1 to regions upstream of PFC, such as the visual cortex, since Area 1 receives visual stimuli. Moreover, both PFC and PMd are deep within the brain hierarchy, suggesting a more natural mapping to later areas. This contradicts the CCA analysis in Figure 3e. It is recommended to either remap the areas or provide further support for the current mapping choice.

We updated the Introduction to better clarify the predictions of the information bottleneck (IB) principle. In particular, the IB principle predicts that later areas should have minimal sufficient representations of task information, whereas upstream areas should have more information. In PMd, we observed a minimal sufficient representation of task information during the decision-making task. In DLPFC, we observed more task information, particularly more information about the target colors and the target configuration.

In terms of the exact map between areas, we do not believe or intend to claim the DLPFC is the first area implicated in the sensorimotor transformation during our perceptual decision-making task. Rather, DLPFC best matches Area 1 of our model. It is important to note that we abstracted our task so that the first area of our model received checkerboard coherence and target configuration as input (and hence did not need to transform task visual inputs). Indeed, in Figure 1d we hypothesize that the early visual areas should contain additional information, which we do not model directly in this work. Future work could model RNNs to take in an image or video input of the task stimulus. In this case, it would be interesting to assess if earlier areas resemble visual cortical areas. We updated the results, where we first present the RNN, to state the inputs explicitly and be clear the inputs are not images or videos of the checkerboard task.

“The RNN input was 4D representing the target configuration and checkerboard signed coherence, while the RNN output was 2D, representing decision variables for a left and right reach (see Methods).”

Another reason that we mapped Area 1 to DLPFC is because anatomical, physiological and lesion studies suggest that DLPFC receives inputs from both the dorsal and ventral stream (Romanski, et, al, 2007; Hoshi, et al, 2006; Wilson, at al, 1993). The dorsal stream originates from the occipital lobe, passes through the posterior parietal cortex, to DLPFC, which carries visuospatial information of the object. The ventral stream originates from the occipital lobe, passes through the inferior temporal cortex, ventrolateral prefrontal cortex to DLPFC, which encodes the identity of the object, including color and texture. In our RNN simulation, Area 1 receives processed inputs of the task: target configuration and the evidence for each color in the checkerboard. Target configuration contains information of the spatial location of the targets, which represents the inputs from the dorsal stream, while evidence for each color by analogy is the input from the ventral stream. Purely visual areas would not fit this dual input from both the dorsal and ventral stream. A potential alternative candidate would be the parietal cortex which is largely part of the dorsal stream and is thought to have modest color inputs (although there is some shape and color selectivity in areas such as LIP, e.g., work from Sereno et al.). On balance given the strong inputs from both the dorsal and ventral stream, we believe Area 1 maps better on to DLPFC than earlier visual areas.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

(1) Line 35/36: Please specify the type of nuisance that the representation is robust to. I guess this refers to small changes in the inputs, not to changes in the representation itself.

Indeed it refers to input variability unrelated to the task. We clarified the text.

(2) For reference, it would be nice to have a tick for the event "Targets on" in Fig.2c.

In this plot, the PSTHs are aligned to the checkerboard onset. Because there is a variable time between target and checkerboard onset, there is a trial-by-trial difference of when the target was turned on, so there is no single place on the x-axis where we could place a “Targets on” tick. In response to this point, we generated a plot with both targets on and check on alignment, with a break in the middle, shown in Supplementary Figure S5. 

(3) It would strengthen the comparison between neural data and RNN if the DPCA components of the RNN areas were shown, as they are shown in Fig.2g,h for the neural data.

We include the PSTHs plotted onto the dPCA components here for Area 1 of the exemplar network. Dashed lines indicate a left reach, while solid lines indicate a right reach, and the color corresponds to the color of the selected target. As expected, we find that the dPCA components capture the separation between components. We emphasize that the trajectory paths along the decoder axes are not particularly meaningful to interpret, except to demonstrate whether variables can be decoded or not (as in Fig 2g,h, comparing DLPFC and PMd). The decoder axes of dPCA are not constrained in any way, in contrast to the readout (encoder) axis (see Methods). This is why our manuscript focuses on analyzing the readout axes. However, if the reviewer strongly prefers these plots to be put in the manuscript, we will add them.   

Author response image 1.

(4) The session-by-session decode analysis presented in Fig.2i suggests that DLPFC has mostly direction information while in Area 1 target information is on top, as suggested by Fig.3g. An additional decoding analysis on trial averaged neural data, i.e. a figure for neural data analogous to Fig.3g,h, would allow for a more straightforward and direct comparison between RNN and neural data. 

We first clarify that we did not decode trial-averaged neural data for either recorded neural data or RNNs. In Fig 3g, h (for the RNN) all decoding was performed on single trial data and then averaged. We have revised the main manuscript to make this clear. Because of this, the mean accuracies we reported for DLPFC and PMd in the text are therefore computed in the same way as the mean accuracies presented in Fig 3g, h. We believe this likely addresses your concern: i.e., the mean decode accuracies presented for both neural data and the RNN were computed the same way. 

If the above paragraph did not address your concern, we also wish to be clear that we presented the neural data as histograms, rather than a mean with standard error, because we found that accuracies were highly variable depending on electrode insertion location. For example, some insertions in DLPFC achieved chance-levels of decoding performance for color and target configuration. For this reason, we prefer to keep the histogram as it shows more information than reporting the mean, which we report in the main text. However, if the reviewer strongly prefers us to make a bar plot of these means, we will add them.

(5) Line 129 mentions an analysis of single trials. But in Fig.2i,j sessions are analyzed. Please clarify.

For each session, we decode from single trials and then average these decoding accuracies, leading to a per-session average decoding accuracy. Note that for each session, we record from different neurons. In the text, we also report the average over the sessions. We clarified this in the text and Methods.

(6) Fig.4c,f show how color and direction axes align with the potent subspaces. We assume that the target axis was omitted here because it highly aligns with the color axis, yet we note that this was not pointed out explicitly.

You are correct, and we revised the text to point this out explicitly.

“We quantified how the color and direction axis were aligned with these potent and null spaces of the intra-areal recurrent dynamics matrix of Area 1 ($\W^1_{rec}$). We did not include the target configuration axis for simplicity, since it highly aligns with the color axis for this network.”

(7) The caption of Fig.4c reads: "Projections onto the potent space of the intra-areal dynamics for each area." Yet, they only show area 1 in Fig.4c, and the rest in a supplement figure. Please refer properly.

Thank you for pointing this out. We updated the text to reference the supplementary figure.

(8) Line 300: "We found the direction axis was more aligned with the potent space and the color axis was more aligned with the null space." They rather show that the color axis is as aligned to the potent space as a random vector, but nothing about the alignments with the null space. Contrarily, on line 379 they write "...with the important difference that color information isn't preferentially projected to a nullspace...". Please clarify.

Thank you for pointing this out. We clarified the text to read: “We found the direction axis was more aligned with the potent space”. The text then describes that the color axis is aligned like a random vector: “In contrast, the color axis was aligned to a random vector.”

(9) Line 313: 'unconstrained' networks are mentioned. What constraints are implied there, Dale's law? Please define and clarify.

Indeed, the constraint refers to Dale’s law constraints. We clarified the text: “Further, we found that W₂₁ in unconstrained 3 area networks (i.e., without Dale's law constraints) had significantly reduced…”

(10) Line 355 mentions a 'feedforward bottleneck'. What does this exactly mean? No E-I feedforward connections, or...? Please define and clarify.

This refers to sparser connections between areas than within an area, as well as a smaller fraction of E-I connections. We clarified the text to read:

“Together, these results suggest  that a connection bottleneck in the form of neurophysiological architecture constraints (i.e., sparser connections between areas than within an area, as well as a smaller fraction of E-I connections) was the key design choice leading to RNNs with minimal color representations and consistent with the information bottleneck principle.”

(11) Fig.5c is supposedly without feedforward connections, but it looks like the plot depicts these connections (i.e. identical to Fig.5b).

In Figure 5, we are varying the E to I connectivity in panel B, and the E-E connectivity in panel C. We vary the feedback connections in Supp Fig. S12. We updated the caption accordingly. 

(12) For reference, it would be nice to have the parameters of the exemplar network indicated in the panels of Fig.5.

We updated the caption to reference the parameter configuration in Table 1 of the Appendix.

(13) Line 659: incomplete sentence

Thank you for pointing this out. We removed this incomplete sentence.

(14) In the methods section "Decoding and Mutual information for RNNs" a linear neural net decoder as well as a nonlinear neural net decoder are described, yet it was unclear which one was used in the end.

We used the nonlinear network, and clarified the text accordingly. We obtained consistent conclusions using a linear network, but did not include these results in the text. (These are reported in Fig. 11 of Kleinman et al, 2021). Moreover, we also obtain consistent results by using an SVM decoder in Fig. S4 for our exemplar parameter configuration.

(15) In the discussion, the paragraph starting from line 410 introduces a new set of results along with the benefits of minimal representations. This should go to the results section.

We prefer to leave this as a discussion, since the task was potentially too simplistic to generate a clear conclusion on this matter. We believe this remains a discussion point for further investigation.

(16) Fig S5: hard to parse. Show some arrows for trajectories (a) (d) is pretty mysterious: where do I see the slow dynamics?

Slow points are denoted by crosses, which forms an approximate line attractor. We clarified this in the caption.

Reviewer #2 (Recommendations For The Authors):

Minor recommendations (not ordered by importance)

(1) Be more explicit that the recordings come from different monkeys and are not simultaneously recorded. For instance, say 'recordings from PFC or PMD'. Say early on that PMD recordings come from two monkeys and that PFC recordings come from 1 of those monkeys. Furthermore, I would highlight which datasets are novel and which are not. For instance, I believe the PFC dataset is a previously unpublished dataset and should be highlighted as such.

We added: “The PMd data was previously described in a study by Chandrasekaran and colleagues” to the main text which clarifies that the PMd data was previously recorded and has been analyzed in other studies.

(2) I personally feel that talking about 'optimal', as is done in the abstract, is a bit of a stretch for this simple task.

In using the terminology “optimal,” we are following the convention of IB literature that optimal representations are sufficient and minimal. The term “optimal” therefore is task-specific; every task will have its own optimal representation. We clarify in the text that this definition comes from Machine Learning and Information Theory, stating:

“The IB principle defines an optimal representation as a representation that is minimal and sufficient for a task or set of tasks.”

In this way, we take an information-theoretic view for describing multi-area representations. This view was satisfactory for explaining and reconciling the multi-area recordings and simulations for this task, and we think it is helpful to provide a normative perspective for explaining the differences in cortical representations by brain area. Even though the task is simple, it still allows us to study how sensory/perceptual information is represented, and well as how choice-related information is being represented.

(3) It is mentioned (and even highlighted) in the abstract that we don't know why the brain distributes computations. I agree with that statement, but I don't think this manuscript answers that question. Relatedly, the introduction mentions robustness as one reason why the brain would distribute computations, but then raises the question of whether there is 'also a computational benefit for distributing computations across multiple areas'. Isn't the latter (robustness) a clear 'computational benefit'?

We decided to keep the word “why” in the abstract, because this is a generally true statement (it is unclear why the brain distributes computation) that we wish to convey succinctly, pointing to the importance of studying this relatively grand question (which could only be fully answered by many studies over decades). We consider this the setting of our work. However, to avoid confusion that we are trying to give a full answer to this question, we are now more precise in the first paragraph of our introduction as to the particular questions we ask that will take a step towards this question. In particular, the first paragraph now asks these questions, which we answer in our study.

“For example, is all stimuli and decision-related information present in all brain areas, or do the cortical representations differ depending on their processing stage? If the representations differ, are there general principles that can explain why the cortical representations differ by brain area?”

We also removed the language on robustness, as we agree it was confusing. Thank you for these suggestions. 

(4) Figure 2e and Fig. 3d, left, do not look very similar. I suggest zooming in or rotating Figure 2 to highlight the similarities. Consider generating a baseline CCA correlation using some sort of data shuffle to highlight the differences.

The main point of the trajectories is to demonstrate that both Area 1 and DLPFC represent both color and direction. We now clarify this in the manuscript. However, we do not intend for these two plots to be a rigorous comparison of similarity. Rather, we quantify similarity using CCA and our decoding analysis. We also better emphasize the relative values of the CCA, rather than the absolute values.

(5) Line 152: 'For this analysis, we restricted it to sessions with significant decode accuracy with a session considered to have a significant decodability for a variable if the true accuracy was above the 99th percentile of the shuffled accuracy for a session.' Why? Sounds fishy, especially if one is building a case on 'non-decodability'. I would either not do it or better justify it.

The reason to choose only sessions with significant decoding accuracy is that we consider those sessions to be the sessions containing information of task variables. In response to this comment, we also now generate a plot with all recording sessions in Supplementary Figure S7. We modified the manuscript accordingly.

“For this analysis, we restricted it to sessions with significant decode accuracy with a session considered to have a significant decodability for a variable if the true accuracy was above the 99th percentile of the shuffled accuracy for a session. This is because these sessions contain information about task variables. However, we also present the same analyses using all sessions in Fig. S7.”

(6) Line 232: 'The RNN therefore models many aspects of our physiological data and is therefore'. Many seems a stretch?

We changed “many” to “key.”

(7) The illustration in Fig. 4B is very hard to understand, I recommend removing it.

We are unsure what this refers to, as Figure 4B represents data of axis overlaps and is not an illustration. 

(8) At some point the authors use IB instead of information bottleneck (eg line 288), I would not do it.

We now clearly write that IB is an abbreviation of Information Bottleneck the first time it is introduced.  

(9) Fig. 5 caption is insufficient to understand it. Text in the main document does not help. I would move most part of this figure, or at least F, to supplementary. Instead, I would move the results in S11 and S10 to the main document.

We clarified the caption to summarize the key points. It now reads: 

“Overall, neurophysiological architecture constraints in the form of multiple areas, sparser connections between areas than within an area, as well as a smaller fraction of E-I connections lead to a minimal color representation in the last area.”

(10) Line 355: 'Together, these results suggest that a connection bottleneck in the form of neurophysiological architecture constraints was the key design choice leading to RNNs with minimal color representations and consistent with the information bottleneck principle.' The authors show convincingly that increased sparsity leads to the removal of irrelevant information. There is an alternative model of the communication subspace hypothesis that uses low-rank matrices, instead of sparse, to implement said bottlenecks (https://www.biorxiv.org/content/10.1101/2022.07.21.500962v2)

We thank the reviewer for pointing us to this very nice paper. Indeed, a low-rank connectivity matrix is another mechanism to limit the amount of information that is passed to subsequent areas. In fact, the low-rank matrix forms a hard-version of our observations as we found that task-relevant information was preferentially propagated along the top singular mode of the inter-areal connectivity matrix. In our paper we observed this tendency naturally emerges through training with neurophysiological architecture constraints. In the paper, for the multi-area RNN, they hand-engineered the multi-area network, whereas our network is trained. We added this reference to our discussion. 

Thank you for your helpful and constructive comments.

Author response:

The following is the authors’ response to the original reviews.

Response to Public Reviewer Comments:

Reviewer 1:

In this work, Veseli et al. present a computational framework to infer the functional diversity of microbiomes in relation to microbial diversity directly from metagenomic data. The framework reconstructs metabolic modules from metagenomes and calculates the per-population copy number of each module, resulting in the proportion of microbes in the sample carrying certain genes. They applied this framework to a dataset of gut microbiomes from 109 inflammatory bowel disease (IBD) patients, 78 patients with other gastrointestinal conditions, and 229 healthy controls. They found that the microbiomes of IBD patients were enriched in a high fraction of metabolic pathways, including biosynthesis pathways such as those for amino acids, vitamins, nucleotides, and lipids. Hence, they had higher metabolic independence compared with healthy controls. To an extent, the authors also found a pathway enrichment suggesting higher metabolic independence in patients with gastrointestinal conditions other than IBD indicating this could be a signal for a general loss in host health. Finally, a machine learning classifier using high metabolic independence in microbiomes could predict IBD with good accuracy. Overall, this is an interesting and well-written article and presents a novel workflow that enables a comprehensive characterization of microbiome cohorts.

We thank the reviewer for their interest in our study, their summary of its findings, and their kind words about the manuscript quality.

Reviewer 2:

This study builds upon the team's recent discovery that antibiotic treatment and other disturbances favour the persistence of bacteria with genomes that encode complete modules for the synthesis of essential metabolites (Watson et al. 2023). Veseli and collaborators now provide an in-depth analysis of metabolic pathway completeness within microbiomes, finding strong evidence for an enrichment of bacteria with high metabolic independence in the microbiomes associated with IBD and other gastrointestinal disorders. Importantly, this study provides new open-source software to facilitate the reconstruction of metabolic pathways, estimate their completeness and normalize their results according to species diversity. Finally, this study also shows that the metabolic independence of microbial communities can be used as a marker of dysbiosis. The function-based health index proposed here is more robust to individuals' lifestyles and geographic origin than previously proposed methods based on bacterial taxonomy.

The implications of this study have the potential to spur a paradigm shift in the field. It shows that certain bacterial taxa that have been consistently associated with disease might not be harmful to their host as previously thought. These bacteria seem to be the only species that are able to survive in a stressed gut environment. They might even be important to rebuild a healthy microbiome (although the authors are careful not to make this speculation).

This paper provides an in-depth discussion of the results, and limitations are clearly addressed throughout the manuscript. Some of the potential limitations relate to the use of large publicly available datasets, where sample processing and the definition of healthy status varies between studies. The authors have recognised these issues and their results were robust to analyses performed on a per-cohort basis. These potential limitations, therefore, are unlikely to have affected the conclusions of this study.

Overall, this manuscript is a magnificent contribution to the field, likely to inspire many other studies to come.

We thank the reviewer for their endorsement of our study and their precision regarding the evaluation of its strengths. We also appreciate their high expectations for its impact in the field.

Reviewer 3:

The major strength of this manuscript is the "anvi-estimate-metabolism' tool, which is already accessible online, extensively documented, and potentially broadly useful to microbial ecologists.

We thank the reviewer for their recognition of the computational advances in this study. We also thank the reviewer for their suggestions that we have addressed below, which allowed us to strengthen our manuscript.

However, the context for this tool and its validation is lacking in the current version of the manuscript. It is unclear whether similar tools exist; if so, it would help to benchmark this new tool against prior methods.

The reviewer brings up a very good point about the lack of context for the `anvi-estimate-metabolism` program. While our efforts that led to the emergence of this software included detailed benchmarking efforts, a formal assessment of its performance and accuracy was indeed lacking. We are thankful for our reviewer to point this out, which motivated us to perform additional analyses to address such concerns. Our revision contains a new, 34-page long supplementary information file (Supplementary File 2) that includes a section titled “Comparison of anvi-estimate-metabolism to existing tools for metabolism reconstruction”. The text therein describes the landscape of currently available software for metabolism reconstruction and describes the features that make `anvi-estimate-metabolism` unique – namely, (1) its implementation of metrics that make it suitable for metagenome-level analyses (i.e., pathway copy number and stepwise interpretation of pathway definitions) and (2) its ability to process user-defined metabolic pathways rather than exclusively relying on KEGG. As described in that section, there is currently no other tool that can compute copy numbers of metabolic pathways from metagenomic data. Hence, it is not quite possible to benchmark the copy number methodology used in our study against prior methods; however, our benchmarking of this functionality with synthetic genomes and metagenomes (described later in this document) does provide necessary quantitative insights into its accuracy and efficiency.

While comparison of the copy number calculations to other tools was not possible due to the unique nature of this functionality, it was possible to benchmark our gene function annotation methodology against existing tools that also annotate genes with KEGG KOfams, which is a step commonly used by various tools that aim to estimate metabolic potential in genomes and metagenomes. In the anvi’o software ecosystem the annotation of genes for metabolic reconstruction is implemented in `anvi-run-kegg-kofams`, and represents a step that is required by `anvi-estimate-metabolism`. As our comparisons were quite extensive and involved additional researchers, we described them in another study which we titled “Adaptive adjustment of significance thresholds produces large gains in microbial gene annotations and metabolic insights” (doi:10.1101/2024.07.03.601779) that is now cited from within our revision in the appropriate context. Briefly, our comparison of anvi’o, Kofamscan, and MicrobeAnnotator using 396 publicly-available bacterial genomes from 11 families demonstrated that `anvi-run-kegg-kofams` is able to identify an average of 12.8% more KO annotations per genome than the other tools, especially in families commonly found in the gut environment (Figure 1). Furthermore, anvi’o recovered the highest proportion of annotations that were independently validated using eggNOG-mapper. Our comparisons also showed that annotations from anvi’o yield at least 11.6% more complete metabolic modules than Kofamscan or MicrobeAnnotator, including the identification of butyrate biosynthesis in Lachnospiraceae genomes at rates similar to manual identification of this pathway in this clade (Figure 2a). Overall, our findings that are now described extensively in DOI:10.1101/2024.07.03.601779 show that our method captures high-quality annotations for accurate downstream metabolism estimates.

We hope these new data help increase the reviewer’s confidence in our results.

Simulated datasets could be used to validate the approach and test its robustness to different levels of bacterial richness, genome sizes, and annotation level.

We thank the reviewer for this suggestion. It was an extremely useful exercise that not only helped us elucidate the nuances of our approach, but also enabled us to further highlight its strengths in our manuscript. We created simulated datasets including a total of 409 synthetic metagenomes that we used to test the robustness of our approach to different genome sizes, community sizes, and levels of diversity. Overall, our tests with these synthetic metagenomes demonstrated that our approach of computing PPCN values to summarize the metabolic capacity within a metagenomic community is accurate and robust to differences in all three critical variables. Most of these variables were weakly correlated between PPCN or PPCN accuracy, and the few correlations that were stronger in fact further supported our original hypothesis that we generated from our comparisons of healthy and IBD gut metagenomes. The methods and results of our validation efforts are explained in detail in our new Supplementary File 2 (see the section titled “Validation of per-population copy number (PPCN) approach on simulated metagenomic data”), but we copy here the subsection that summarizes our findings for the reviewer’s convenience:

Overall impact on the comparison between healthy and IBD gut metagenomes

“In summary, our validation strategy revealed good accuracy at estimating metagenome-level metabolic capacity relative to our genome-level knowledge in the simulated data. While it often underestimated average genomic completeness by ignoring partial copies of metabolic pathways and often overestimated average genomic copy number due to the effect of pathway complementarity between different community members, the magnitude of error was overall limited in range and the error distributions were centered at or near 0. Furthermore, we observed these broad error trends in all cases we tested, and therefore we expect that they would also apply to both sample groups in our comparative analysis. Thus, we next considered how the PPCN approach might have influenced our analyses that considered metagenomes from healthy individuals and from those who have IBD – two groups that differed from one another with respect to some of the variables considered in our tests.

Most of the correlations between PPCN or PPCN accuracy and sample parameters were weak, yet significant (Table 1). They showed that community size and diversity level have limited influence on the PPCN calculation, while genome size does not influence its accuracy. The only exception was the moderate correlation between PPCN and genome size, particularly for the subset of IBD-enriched pathways. It was a negative correlation with the proportion of small genomes in a metagenome, indicating that PPCN values for these pathways are larger when there are more large genomes in the community and suggesting that these pathways tend to occur frequently in larger genomes. This is in line with our observation that IBD communities contain more large genomes and therefore confirms our interpretation that the populations surviving in the IBD gut microbiome are those with the genomic space to encode more metabolic capacities.

If we consider even the weak correlations, two of those relationships indicate that our approach would be more accurate for IBD metagenomes than for healthy metagenomes. For instance, PPCN accuracy was slightly higher for smaller communities (as in IBD samples), with a weakly positive correlation between PPCN error and community size. It was also slightly more accurate for less diverse communities (as in IBD samples), with a weakly positive correlation between PPCN error and number of phyla. The only opposing trend was the weakly positive correlation between PPCN error and proportion of smaller genomes, which favors higher accuracy in communities with smaller genomes (as in healthy samples). Given that our analysis focuses on the pathways enriched in IBD samples, an overall higher accuracy in IBD samples would increase the confidence in our enrichment results.

We also examined the accuracy of our method to predict the number of populations within a metagenome based on the distribution and frequency of single-copy core genes (i.e., the denominator in the calculation of PPCN). Our benchmarks show that the estimates are overall accurate, where most errors reflect a negligible amount of underestimations of the actual number of populations. Errors occurred more frequently for the realistic synthetic assemblies generated from simulated short read data than for the ideal synthetic assemblies generated from the combination of genomic contigs. The correlations between estimation accuracy and sample parameters indicated that the population estimates are more accurate for smaller communities and communities with more large genomes, as in IBD samples (Table 2). Thus, this method is more likely to underestimate the community size in healthy samples, and these errors could lead to overestimation of PPCN in healthy samples relative to IBD samples. Thus, the enrichment of a given pathway in the IBD samples would have to overcome its relative overestimation in the healthy sample group, making it more likely that we identified pathways that were truly enriched in the IBD communities.

Overall, the consideration of our simulations in the context of healthy vs IBD metagenomes suggest that slight biases in our estimates as a function of unequal diversity with sample groups should have driven PPCN calculations towards a conclusion that is opposite of our observations under neutral conditions. Thus, clear differences between healthy vs IBD metagenomes that overcome these biases suggest that    biology, and not potential bioinformatics artifacts, is the primary driver of our observations.”

Accordingly, we have added the following sentence summarizing the validation results to our paper:

“Our validation of this method on simulated metagenomic data demonstrated that it is accurate in capturing metagenome-level metabolic capacity relative to genome-level metabolic capacity estimated from the same data (Supplementary File 2, Supplementary Table 6).”

Early in this process of validation, we identified and fixed two minor bugs in our codebase. The bugs did not affect the results of our paper and therefore did not warrant a re-analysis of our data. The first bug, which is detailed in the Github issue https://github.com/merenlab/anvio/issues/2231 and fixed in the pull request https://github.com/merenlab/anvio/pull/2235, led to the overestimation of the number of microbial populations in a metagenome when the metagenome contains both Bacteria and Archaea. None of the gut metagenomes analyzed in our paper contained archaeal populations, so this bug did not affect our community size estimates.

The second bug, which is detailed in the Github issue https://github.com/merenlab/anvio/issues/2217 and fixed in the pull request https://github.com/merenlab/anvio/pull/2218, caused inflation of stepwise copy numbers for a specific type of metabolic pathway in which the definition contained an inner parenthetical clause. This bug affected only 3 pathways in the KEGG MODULE database we used for our analysis, M00083, M00144, and M00149. It is worth noting that one of those pathways, M00083, was identified as an IBD-enriched module in our analysis. However, the copy number inflation resulting from this bug would have occurred equivalently in both the healthy and IBD sample groups and thus should not have impacted our comparative analysis.

Regardless, we are grateful for the suggestion to validate our approach since it enabled us to identify and eliminate these minor issues.

The concept of metabolic independence was intriguing, although it also raises some concerns about the overinterpretation of metagenomic data. As mentioned by the authors, IBD is associated with taxonomic shifts that could confound the copy number estimates that are the primary focus of this analysis. It is unclear if the current results can be explained by IBD-associated shifts in taxonomic composition and/or average genome size. The level of prior knowledge varies a lot between taxa; especially for the IBD-associated gamma-Proteobacteria.

The reviewer brings up an important point, and we are thankful for the opportunity to clarify the impact of taxonomy on our analysis. Though IBD has been associated with taxonomic shifts in the gut microbiome, a major problem with such associations is that the taxonomic signal is extremely variable, leading to inconsistency in the observed shifts across different studies (doi:https://doi.org/10.3390/pathogens8030126). Indeed, one of the most comprehensive prior studies into this topic demonstrated that inter-individual variation is the largest contributor to all multi-omic measurements aiming to differentiate between the gut microbiome of individuals with IBD from that of healthy individuals, including taxonomy (doi:10.1038/s41586-019-1237-9). We therefore took a different approach to study this question that is independent of taxonomy, by focusing on metabolic potential estimated directly from metagenomes to elucidate an ecological explanation behind the reduced diversity of the IBD gut microbiome, which studies of taxonomic composition alone are not able to provide. Furthermore, the variability inherent to taxonomic profiles of the gut microbiome makes it unlikely that taxonomic shifts could confound our analysis, especially given our large sample set encompassing a variety of individuals with different origins, ages, and genders.

We agree with the reviewer that our level of prior knowledge varies substantially across taxa. Regardless, the only prior knowledge with any bearing on our ability to estimate metabolic capacity in a taxonomy-independent manner is the extent of sequence diversity captured by our annotation models for the enzymes used in metabolic pathways. During our analysis, we had observed that metagenomes in the healthy group had fewer gene annotations than those in the IBD group and we therefore shared the reviewer’s concern about potential annotation bias, whereby less-studied genomes are not always incorporated into the Hidden Markov Models for annotating KEGG Orthologs, perhaps making it more likely for us to miss annotations in these genomes (and leading to lower completeness scores for metabolic pathways in the healthy samples). Our annotation method partially addresses this limitation by taking a second look at any unannotated genes and mindfully relaxing the bit score similarity thresholds to capture annotations for any genes that are slightly too different from reference sequences for annotation with default thresholds. As mentioned previously, our recent preprint demonstrates the efficacy of this strategy (doi:10.1101/2024.07.03.601779). To further address this concern, we also investigated the extent of distant homology in these metagenomes using AGNOSTOS (doi:https://doi.org/10.7554/eLife.67667), which showed a higher proportion of unknown genes in the healthy metagenomes and suggested that a substantial portion of the unannotated genes are not distant homologs of known enzymes that we failed to annotate due to lack of prior knowledge about them, but rather are completely novel functions. To describe these results, we added the following paragraph and two accompanying figures (Supplementary Figure 4g-h) to the section “Differential annotation efficiency between IBD and Healthy samples” in Supplementary File 1:

“To understand the potential origins of the reduced annotation rate in healthy metagenomes, we ran AGNOSTOS (Vanni et al. 2022) to classify known and unknown genes within the healthy and IBD sample groups. AGNOSTOS clusters genes to contextualize them within an extensive reference dataset and then categorizes each gene as ‘known’ (has homology to genes annotated with Pfam domains of known function), ‘genomic unknown’ (has homology to genes in genomic reference databases that do not have known functional domains), or ‘environmental unknown’ (has homology to genes from metagenomes or MAGs that do not have known functional domains). The resulting classifications confirm that healthy metagenomes contain fewer ‘known’ genes than metagenomes in the IBD sample group – the proportion of ‘known’ genes classified by AGNOSTOS is about 3.0% less in the healthy metagenomes than in the IBD sample group, which is similar to the ~3.5% decrease in the proportion of ‘unannotated’ genes observed by simply counting the number of genes with at least one functional annotation (Supplementary Figure 4g-h, Supplementary Table 1e). Furthermore, the majority of the unannotated genes in either sample group were categorized by AGNOSTOS as ‘genomic unknown’ (Supplementary Figure 4g), suggesting that the unannotated sequences are genes without biochemically-characterized functions currently associated with them and are thus legitimately lacking a functional annotation in our analysis, rather than representing distant homologs of known protein families that we failed to annotate. Based upon the classifications, a systematic technical bias is unlikely driving the annotation discrepancy between the sample groups.”

Furthermore, we have already discussed this limitation and its implications in our manuscript (see section “Key biosynthetic pathways are enriched in microbial populations from IBD samples”). To further clarify that our approach is independent of taxonomy, we have now also amended the following statement in our introduction:

“Here we implemented a high-throughput, taxonomy-independent strategy to estimate metabolic capabilities of microbial communities directly from metagenomes and investigate whether the enrichment of populations with high metabolic independence predicts IBD in the human gut.”

Finally, the reviewer is also correct that genome size is a part of the equation, as genome size and level of metabolic capacity are inextricable. In fact, we observed this in our analysis, as already stated in our paper:

“HMI genomes were on average substantially larger (3.8 Mbp) than non-HMI genomes (2.9 Mbp) and encoded more genes (3,634 vs. 2,683 genes, respectively)”

Since larger genomes have the space to encode more functional capacity, it follows that having higher metabolic independence would require a microbe to have a larger genome. The validation of our method on simulated metagenomic data supported this idea by demonstrating that the IBD-enriched metabolic pathways are commonly identified in large genomes. The validation also proved that genome size does not influence the accuracy of our approach (Supplementary File 2).

It can be difficult to distinguish genes for biosynthesis and catabolism just from the KEGG module names and the new normalization tool proposed herein markedly affects the results relative to more traditional analyses.

We agree with the reviewer that KEGG module names do not clearly indicate the presence of biosynthetic genes of interest. That said, KEGG is a commonly-used and extensively-curated resource, and many biologists (including ourselves) trust their categorization of genes into pathways. We hope that readers who are interested in specific genes within our results would make use of our publicly-available datasets (which include gene annotations) to conduct a targeted analysis based on their expertise and research question.

However, we would like to respectfully note that the ability to distinguish the genes within each KEGG module may not be very useful to most readers, and is unlikely to have a meaningful impact in our findings. As the reviewer most likely appreciates, the presence of individual genes in isolation can be insufficient to indicate biosynthetic capacity, considering that 1) most biosynthetic pathways involve several biochemical conversions requiring a series of enzymes, 2) enzymes are often multi-functional rather than exclusive to one pathway, and 3) different organisms in a community may utilize enzymes encoded by different genes to perform the same or similar biochemical reaction in a pathway. We therefore made the choice to analyze metabolic capacity at the pathway level, because this would better reflect the biosynthetic abilities encoded by the multiple microbial populations within each metagenome.

The reviewer also suggests that our novel normalization method affects our results, yet we believe that this normalization strategy is one of the strengths of our study in comparison to ‘more traditional analyses’ as it enables an appropriate comparison between metagenomes describing microbial communities of dramatically different degrees of richness. Indeed, we suspect that the lack of normalization in more traditional analyses may be one reason why prior analyses have so far failed to uncover any mechanistic explanation for the loss of diversity in the IBD gut microbiome. We hope that our validation efforts were sufficiently convincing in demonstrating the suitability of our approach, and copy here a particularly illuminating section of the validation results that we have added to Supplementary Information File 2:

“As expected, we observed a significant positive correlation between metagenomic copy number (the numerator of PPCN) and community size in each group, likely driven by the increase in the copy number of core metabolic pathways in larger communities (Supplementary Figure 18). Interestingly, this correlation was much stronger for the subset of IBD-enriched pathways (0.49 <= R <= 0.67) than for all modules (0.12 <= R <=0.13).

“However, the correlation was much weaker and often nonsignificant for the normalized PPCN data in both groups of modules (all modules: 0.01 < R < 0.04, enriched modules: 0.04 < R < 0.09, Supplementary Table 6b, Supplementary Figure 19), which demonstrates the suitability of our normalization method to remove the effect of community size in comparisons of metagenome-level metabolic capacity.”

As such, it seems safer to view the current analysis as hypothesis-generating, requiring additional data to assess the degree to which metabolic dependencies are linked to IBD.

We certainly agree with the reviewer that our study, similar to the vast majority of studies published every year, is a hypothesis-generating work. Any idea proposed in any scientific study in life sciences will certainly benefit from additional data analyses, and therefore we respectfully do not accept this as a valid criticism of our work. The inception of this study is linked to an earlier work that hypothesized high metabolic independence as a determinant of microbial fitness in stressed gut communities (doi:10.1186/s13059-023-02924-x), which lacked validation on larger sets of data. Our study tests this original hypothesis using a large number of metagenomes, and lends further support for it with approaches that are now better validated. Furthermore, there are other studies that agree with our interpretation of the data (doi:10.1101/2023.02.17.528570, doi:10.1038/s41540-021-00178-6), and we look forward to more computational and/or experimental work in the future to generate more evidence to evaluate these insights further.

Response to Recommendations for the Authors

Reviewer 1:

My main comments include:

- From the results reported in lines 178-185, it seems that metabolic pathways in general were enriched in IBD microbiomes, not specifically biosynthetic pathways. Can we really say then that the signal is specific for biosynthesis capabilities?

We apologize for the confusion here. When we read the text again, we ourselves were confused with our phrasing.

The reviewer is correct that a similar proportion of both biosynthetic and non-biosynthetic pathways had elevated per-population copy number (PPCN) values in the IBD samples. However, the low microbial diversity associated with IBD and the on average larger genome size of individual populations contributes to this relative enrichment of the majority of metabolic modules. To remove this bias and identify specific modules whose enrichment was highly conserved across microbial populations associated with IBD, we implemented two criteria: 1) we selected modules that passed a high statistical significance threshold in our enrichment test (Wilcoxon Rank Sum Test, FDR-adjusted p-value < 2e-10), and 2) we accounted for effect size by ranking these modules according to the difference between their median PPCN in IBD samples and their median PPCN in healthy samples, and keeping only those in the top 50% (which translated to an effect size threshold of > 0.12).

This analysis revealed a set of metabolic modules that were consistently and highly significantly enriched in microbial communities associated with IBD. The majority of these metabolic modules encode biosynthesis pathways. Our use of the terms “elevated”, “enriched”, and “significantly enriched” in the previous version of the text was confusing to the reader. We thank the reviewer for pointing this out, and we hope that our revision of the text clarifies the analysis strategy and observations:

“To gain insight into potential metabolic determinants of microbial survival in the IBD gut environment, we assessed the distribution of metabolic modules within samples from each group (IBD and healthy) with and without using PPCN normalization. Without normalizing, module copy numbers were overall higher in healthy samples (Figure 2a) and modules exhibited weak differential occurrence between cohorts (Figure 2b, 2c, Supplementary Figure 3). The application of PPCN reversed this trend, and most metabolic modules were elevated in IBD (Supplementary Figure 5). This observation is influenced by two independent aspects of the healthy and IBD microbiota. The first one is the increased representation of microbial organisms with smaller genomes in healthy individuals (Watson et al. 2023), which increases the likelihood that the overall copy number of a given metabolic module is below the actual number of populations. In contrast, one of the hallmarks of the IBD microbiota is the generally increased representation of organisms with larger genomes (Watson et al. 2023). The second aspect is that the generally higher diversity of microbes in healthy individuals increases the denominator of the PPCN. This results in a greater reduction in the PPCN of metabolic modules that are not shared across all members of the diverse gut microbial populations in health.

To go beyond this general trend and identify modules that were highly conserved in the IBD group, we first selected those that passed a relatively high statistical significance threshold in our enrichment test (Wilcoxon Rank Sum Test, FDR-adjusted p-value < 2e-10). We then accounted for effect size by ranking these modules according to the difference between their median PPCN in IBD samples and their median PPCN in healthy samples, and keeping only those in the top 50% (which translated to an effect size threshold of > 0.12). This stringent filtering revealed a set of 33 metabolic modules that were significantly enriched in metagenomes obtained from individuals diagnosed with IBD (Figure 2d, 2e), 17 of which matched the modules that were associated with high metabolic independence previously (Watson et al. 2023) (Figure 2f). This result suggests that the PPCN normalization is an important step in comparative analyses of metabolisms between samples with different levels of microbial diversity.”

Lines 178-185 from our original submission have been removed to avoid further confusion. These results can be found in Supplementary File 1 (section “Module enrichment without consideration of effect size leads to nonspecific results”).

It is not entirely clear to me what is meant by PPCN normalization. Normalize the number of copy numbers to the overall number of genes?

The idea behind using per-population copy number (PPCN) is to normalize the prevalence of each metabolic module found in an environment with the number of microbial populations within the same sample. PPCN achieves this by dividing the pathway copy numbers by the number of microbial populations in a given metagenome, which we estimate from the frequency of bacterial single-copy core genes. We have updated the description of the per-population copy number (PPCN) calculation to clarify its use:

“Briefly, the PPCN estimates the proportion of microbes in a community with a particular metabolic capacity (Figure 1, Supplementary Figure 2) by normalizing observed metabolic module copy numbers with the ‘number of microbial populations in a given metagenome’, which we estimate using the single-copy core genes (SCGs) without relying on the reconstruction of individual genomes.”

We also note that the equation for PPCN is shown in Figure 1.

It is also not clear to me how the classifier predicts stress on microbiomes rather than dysbiosis.

The reviewer asks an interesting question since it is true that we could also use the term “dysbiosis” rather than “stress”. Yet we refrained from the use of dysbiosis as it is considered a poorly-defined term to describe an altered microbiome often associated with a specific disease (doi:https://doi.org/10.3390/microorganisms10030578), such as IBD, relative to another poorly-defined state, “healthy microbiome” (doi:https://doi.org/10.1002/phar.2731). We do consider that stress is not necessarily a term that is less vague than dysbiosis, yet it has the advantage of being more common in studies of ecology compared to dysbiosis. Our relatively neutral stance towards which term to use has shifted dramatically due to one critical observation in our study: the identical patterns of enrichment of HMI microbes in individuals diagnosed with IBD as well as in healthy individuals treated with antibiotics. We appreciate that the observed changes in the antibiotics case can also fulfill the definition of “dysbiosis”, but the term “stress response” more accurately describes what the classifier identifies in our opinion.

What is the advantage of using the estimate-metabolism pipeline presented in this article over workflows such as those using genome-scale models, which are repeatedly cited and discussed?

Genome-scale models are often appropriate for a big-picture view of metabolism, and especially when the capability to perform quantitative simulations like flux-balance analysis is needed. For our investigation, we wanted a more specific and descriptive summary of metabolic capacity, so we focused on individual KEGG modules, which qualitatively describe subsets of the vast metabolic network with pathway names that all readers can understand, rather than working with an abstract model of the entire network. Furthermore, genome-scale models would have prevented us from assessing the redundancy (copy number) of metabolic pathways, as these networks usually focus on the presence-absence of gene annotations for enzymes in the network rather than the copy number of these annotations. The copy number metric has been critical for our analyses, considering that we are focusing on metabolic capacity at the community level and require the ability to normalize this metabolic capacity by the size of the community described by each metagenome. Finally, assessing a discrete set of metabolic pathways yielded a corresponding set of features that we used to create the machine learning classifier, whereas data from genome-scale models would not be as easily transferable into classifier features.

Minor comments:

Figure 2d and e are mentioned in the text before Figure 2a.

We thank the reviewer for catching this. We have rewritten the section as follows to put the figure references in numerical order:

!To gain insight into potential metabolic determinants of microbial survival in the IBD gut environment, we assessed the distribution of metabolic modules within samples from each group (IBD and healthy) with and without using PPCN normalization. Without normalizing, module copy numbers were overall higher in healthy samples (Figure 2a) and modules exhibited weak differential occurrence between cohorts (Figure 2b, 2c, Supplementary Figure 3). After the application of PPCN, most metabolic modules were elevated in IBD (Supplementary Figure 5). This observation is a product of two independent aspects of the healthy and IBD microbiota. The first one is the increased representation of microbial organisms with smaller genomes in healthy individuals (Watson et al. 2023), which increases the likelihood that the overall copy number of a given metabolic module is below the actual number of populations. In contrast, one of the hallmarks of the IBD microbiota is the generally increased representation of organisms with larger genomes (Watson et al. 2023). The second aspect is that the generally higher diversity of microbes in healthy individuals increases the denominator of the PPCN due to the higher number of populations detected in these samples. This results in a greater reduction in the PPCN of metabolic modules that are not shared across all members of the diverse gut microbial populations in health. To go beyond this general trend and identify modules that were highly conserved in the IBD group, we first selected those that passed a relatively high statistical significance threshold in our enrichment test (Wilcoxon Rank Sum Test, FDR-adjusted p-value <2e-10). We then accounted for effect size by ranking these modules according to the difference between their median PPCN in IBD samples and their median PPCN in healthy samples, and keeping only those in the top 50% (which translated to an effect size threshold of > 0.12). This stringent filtering revealed a set of 33 metabolic modules that were significantly enriched in metagenomes obtained from individuals diagnosed with IBD (Figure 2d, 2e), 17 of which matched the modules that were associated with high metabolic independence previously (Watson et al. 2023) (Figure 2f). This result suggests that the PPCN normalization is an important step in comparative analyses of metabolisms between samples with different levels of microbial diversity.!

How much preparation is needed for users that want to apply the estimate-metabolism pipeline to their own datasets? From the documentation at anvi'o, it still seems like a significant effort.

We thank the reviewer for this important question. The use of anvi-estimate-metabolism is simple, but the concept it makes available and the means it offers its users to interact with their data are not basic, thus its use requires some effort. Anvi’o provides users with the ability to directly interact with their data at each step of the analysis to have full control over the analysis and to make informed decisions on the way. In comparison to pre-defined analysis pipelines that often require no additional input from the user, this approach requires some level of involvement of the user throughout the process – namely, they must run a few programs in series rather than running just one pipeline command that quietly handles everything on their behalf. The most basic workflow for using `anvi-estimate-metabolism` is quite straightforward and requires four simple steps following the installation of anvi’o: 1. Run the program `anvi-setup-kegg-data` to download the KEGG data. 2. Convert the assembly FASTA file into an anvi’o-compatible database format with gene calls by running `anvi-gen-contigs-database`. 3. Annotate genes with KOs with the program `anvi-run-kegg-kofams`. 4. Get module completeness scores and copy numbers by running `anvi-estimate-metabolism`. In addition, we provide simple tutorials (such as the one at https://anvio.org/tutorials/fmt-mag-metabolism/) and reproducible bioinformatics workflows online (including for this study at https://merenlab.org/data/ibd-gut-metabolism/) which helps early career researchers to apply similar strategies to their own datasets. We are happy to report that we have been using this tool in our undergraduate education, and observed that students with no background in computation were able to apply it to their questions without any trouble.

Reviewer 2:

Congratulations on this great work, the manuscript is a pleasure to read. Minor questions that the authors might want to clarify:

L 275: Why use reference genomes from the GTDB (for only 3 phyla) instead of using MAGs reconstructed from the data? I understand that assemblies based on individual samples would probably not yield enough complete MAGs, but I would expect that co-binning the assemblies for the entire dataset would.

We thank the reviewer for their kind words. We certainly agree that metagenome assembled genomes (MAGs) reconstructed directly from the assemblies would by nature represent the populations in these communities better than reference genomes. However, one of our aims in this study was to avoid the often error-prone and time-consuming step of reconstructing MAGs. Most automatic binning algorithms inevitably make mistakes, and especially for metabolism estimation, low quality MAGs can introduce a bias in the analysis. At the same time the manual curation of each bin to remove any contamination would require a substantial effort and make the workflow less accessible for others to use. As an example, in our previous work (doi:10.1186/s13059-023-02924-x), careful refinement of MAGs from just two co-assemblies took two months. Here, we developed the PPCN workflow as a more scalable, assembly-level analysis to avoid the need for binning in the first place.

To supplement and confirm the metagenome-level results, we decided to run a genome-level analysis. We used the GTDB since it represents the most comprehensive, dereplicated collection of reference genomes across the tree of life. We chose those 3 phyla in particular because of their ecological relevance in the human gut environment. Bacteroidetes and

Firmicutes together represent the majority (up to ~90%) of the populations in healthy individuals (doi:10.1038/nature07540), and Proteobacteria represent the next most abundant phylum on average (2% ± 10%) (doi:10.1371/journal.pone.0206484).

L 403: Should the Franzosa and Papa papers be referenced as numbers?

Thanks for pointing this out. The rogue numerical citation was actually an artifact of the submission and was corrected to a long-format citation in the online version of the manuscript on the eLife website.

Reviewer 3:

The lack of any experimental validation contributes to the tentative nature of the conclusions that can be drawn at this time. Numerous studies have looked at the metabolism of gut bacterial species during in vitro growth, which could be mined to test if the in silico predictions of metabolism can be supported. Alternatively, the authors could isolate key strains of interest and study them in culture or in mouse models of IBD.

We appreciate these suggestions and agree with the reviewer that experimental validation is important. However, we do not agree that either the use of mouse models or the isolation of individual microbial strains would be an appropriate experimental test in this case. The use of humanized gnotobiotic mice has critical limitations (see doi:10.1016/j.cell.2019.12.025 and references within the section on “human microbiota-associated murine models”). As it is not possible to establish a mouse model whose gut microbiota fully reflect the human gut microbiome, such an approach would neither be appropriate to validate our findings, nor would it have been possible to produce the insights we have gained based on environmental data. We are not sure how exactly a mouse model, even when ignoring the well established limitations, could improve or validate a comprehensive analysis of a large “environmental” datasets that resulted in highly significant signals.

We are also not sure that we understand how the reviewer believes that the isolation of individual strains would aid in validating our findings. While we appreciate that not all relevant genes are captured by the available annotation routines and that some genes may be misannotated, the large dataset used here renders these concerns negligible. Isolating a small subset of bacterial populations would hardly lead to a representative sample and testing their metabolic capacities in vitro would not improve the reliability of our analysis.

Boilerplate suggestions as vague as “isolate key strains of interest” or “experiment in mouse models of IBD” do not add or retract anything from our findings. Our findings and hypotheses are well supported by our data and extensive analyses.

Line 9 - not sure this approach is hypothesis testing in the traditional sense, you might reword.

Hypothesis testing occurs when one makes an observation, develops an hypothesis that explains the observation, and then gathers and analyzes data to investigate whether additional data support or disprove the hypothesis. We are not convinced a reword is necessary.

Line 40 - the lack of consistent differences in IBD and healthy individuals does not mean that the microbiome doesn't impact disease. It's important to consider all the mechanistic studies in animal models and other systems.

Our study does not claim that microbiome has no impact on the course of disease.

Line 50 - this seemed out of place and undercuts the current findings. Upon checking Ref. 31, the analysis seems distinct enough to not mention in the introduction.

We disagree. Ref 31 uses genome-scale metabolic models to identify the loss of cross-feeding interactions in the gut microbiome of individuals with IBD, which is another way of saying that the microbes in IBD no longer rely on their community for metabolic exchange – in other words, they are metabolically independent. This is an independent observation that is parallel to our results and confirms our analysis; hence, it is important to keep in our introduction.

Line 55 - Ref. 32 looked at FMT, which should be explicitly stated here.

The reviewer’s suggestion is not helpful. Ref 32 has a significant focus on IBD as it compares a total of 300 MAGs generated from individuals with IBD to 264 MAGs from healthy individuals and shows differences in metabolic enrichment between healthy and IBD samples independent of taxonomy, thus setting the stage for our current work. What model has been used to generate the initial insights that led to the IBD-related conclusion in Ref 32 has no significance in this context.

Lines 92-107 - this text is out of place in the Results section and reads more like a review article. Please trim it down and move it to the introduction.

We would like to draw the reviewer’s attention to the fact that this is a “Result and Discussion” section. In this specific case it is important for readers to appreciate the context for our new tool, as the reviewer commented in the public review. We kindly disagree with the reviewer’s suggestion to remove this text as that would diminish the context.

Line 107 - is "selection" the word you meant to use?

If the frequency of a given metabolic module remains the same or increases despite the decreasing diversity of the microbial community, it is conceivable to assume that its enrichment indicates the presence of a selective process to which the module responds. It is indeed the word we meant to use.

Line 110 - this is the first mention of this new method, need to add it to the abstract and introduction.

The reviewer must have overlooked the text passages in which we mention the strategy we developed within the abstract:

“Here, we tested this hypothesis on a large scale, by developing a software framework to quantify the enrichment of microbial metabolisms in complex metagenomes as a function of microbial diversity.”

And in the last paragraph of the introduction:

“Here we implemented a high-throughput, taxonomy-independent strategy to estimate metabolic capabilities of microbial communities directly from metagenomes…”

Figure 1 - a nice summary, but no data is shown to support the validity of this model. Consider shrinking the cartoon and adding validation with simulated datasets.

We hope we have addressed this recommendation with the extensive validation efforts summarized above.

Line 134 - need to state the FDR and effect size cutoffs used.

We have reworded this sentence as follows to clarify which thresholds were used:

“We identified significantly enriched modules using an FDR-adjusted p-value threshold of p < 2e-10 and an effect size threshold of > 0.12 from a Wilcoxon Rank Sum Test comparing IBD and healthy samples.”

I'm also concerned about the simple comparison of IBD to healthy without adjusting for confounders like study, geographical location, age, sex, drug use, diet, etc. More text is needed to explain the nature of these data, how much metadata is available, and which other variables distinguish IBD from healthy.

The reviewer is correct that there is a large amount of interindividual variation between samples due to host and environmental factors. However, the lack of adjusting for confounders was intentional, and in fact one of the critical strengths of our study. We observe a clear signal between healthy individuals and individuals diagnosed with IBD, despite the amount of interindividual variation in our diverse set of samples from 13 different studies (details of which are summarized in Supplementary Table 1). The clear increase in predicted metabolic capacity that we consistently observe in IBD patients using both metagenomes and genomes across diverse cohorts points to metabolic independence as a high-level trend that is predictive of microbial prevalence in stressed gut environments irrespective of host factors.

Line 145 - calling PPCN normalization an "essential step" is a huge claim and requires a lot more data to back it up. Might be best to qualify this statement.

We hope we have addressed this recommendation with our validation efforts. Supplementary Figures 18 and 19 in particular show evidence for the necessity of the normalization step. It is indeed an essential step if the purpose is to compare metabolic enrichment between cohorts of highly different microbial diversity.

Figure 2a - the use of a 1:1 trend line seems potentially misleading. I would replace it with a best-fit line.

Our purpose here was not to show the best fit. Instead, the 1:1 trend line separates the modules based on their relative abundance distribution between healthy individuals and individuals diagnosed with IBD. If the module is to the left of the line, it has a higher median copy number in healthy individuals and if the module is to the right, it has a higher median copy number in individuals with IBD. The line also helps to demonstrate the shift that occurs between the unnormalized data in Figure 2a. Without the normalization, more modules occur to the left of the

1/1 line as a result of the higher raw copy numbers in healthy metagenomes which simply contain more microbial populations. With the normalization (Figure 2d), more modules fall on the right side of the 1/1 line due to higher PPCN values. A best-fit line would not serve well for these purposes.

The text should be revised to state that this analysis actually did find many significant differences and to discuss whether they were the same modules identified in Figure 2d.

We apologize for the confusion and thank the reviewer for bringing this issue to our attention. As mentioned above, the disparate levels of microbial diversity between healthy individuals and individuals with IBD resulted in much larger copy numbers of metabolic modules in healthy samples reflecting the often much larger communities. Hence, we ran statistical tests only on normalized (PPCN) data. The p-values associated with each module in Figure 2a, as well as the colors of each point, are based on the PPCN data in Figure 2d. We aimed to improve the clarity of the visual comparison between normalized and unnormalized results by identifying the same set of IBD-enriched modules in plots a-c and plots d-f.

That being said, the reviewer’s comment made us realize the potential for confusion when using the normalized data’s statistical results in Figure 2a that otherwise shows results from unnormalized data. We have now run the same statistical test on the unnormalized (raw copy number) data and re-generated Figure 2a with the new FDR-adjusted p-values and points colored based on the statistical tests using unnormalized data. We’ve also removed the arrow connecting to Figure 2b (since we no longer show the same set of IBD-enriched modules in Figures 2a and 2b), and added a dashed line to indicate the effect size threshold (similar to the one in Figure 2d). We have updated the legend for Figure 2a-d to reflect these changes:

When we used the same p-value threshold (p < 2e-10) as before and also filtered for an effect size larger than the mean (the same strategy used to set our effect size threshold for the normalized data), there are 10 modules that are significantly enriched based on the unnormalized data. Of course, it is difficult to gauge the relevance of these 10 modules to microbial fitness in the IBD gut environment since their raw copy numbers do not tell us anything about the relative proportion of community members that harbor these modules. Therefore, we are reluctant to add these modules to the results text. For the record, only 3 of those modules were also significantly enriched based on the normalized PPCN values: M00010 (Citrate cycle, first carbon oxidation), M00053 (Pyrimidine deoxyribonucleotide biosynthesis), and M00121 (Heme biosynthesis).

Figure 2c,f - these panels raise a lot of concerns given that the choice of method inverts the trend. Without additional data/validation, it's hard to know which method is right.

We hope we have addressed this recommendation with the extensive validation efforts summarized above. Inversion of the trend is an expected outcome, because the raw copy numbers of most metabolic modules are much lower in the IBD sample group due to lower community sizes.

Line 167 - Need to take the KEGG names with a grain of salt, just because it says "biosynthesis" doesn't mean that the pathway goes in that direction in your bacterium of interest.

We believe the reviewer is under a misapprehension regarding the general reversibility of KEGG metabolic modules, or indeed of metabolic pathways. Most metabolic pathways have one or several (practically) irreversible reactions. To demonstrate this for the 33 IBD-enriched modules, we evaluated their reversibility based upon their corresponding KEGG Pathway Maps, which indicate reaction reversibility via double-sided arrows. Aside from the signature modules M00705 and M00627, in 26 out of 31 pathway modules one or more irreversible reactions render these pathways one-directional. Indeed, on average the majority (54%) of the reactions in a given module are irreversible. When focusing on the 23 “biosynthesis” modules, 22 out of 23 (96%) modules have at least one irreversible reaction, and on average 64% of a given module’s reactions are irreversible. These data (which can be accessed at doi:10.6084/m9.figshare.27203226 for the reviewer’s convenience) challenge the reviewer’s notion that pathway directionality is free to change arbitrarily, since the presence of even one irreversible reaction effectively blocks the flux in the opposing direction. Thus, “biosynthesis” is indeed a meaningful term in KEGG module names.

That said, KEGG Pathway Maps, though highly curated, are likely not the final word on whether a given reaction in a metabolic pathway can be considered reversible or irreversible in each microbial population and under all conditions. And our analysis, like many others that rely on metagenomic data, does not consider the environmental conditions in the gut such as temperature or metabolite concentrations that might influence the Gibbs free energy and thus the directionality of these reactions in vivo. However, even assuming general reversibility of metabolic pathways, this would not invalidate the fact that these microbes have the metabolic capacity to synthesize the respective molecules. In other words, the potential reversibility of pathways is irrelevant to our analysis since we are describing metabolic potential. The lac operon in E. coli might only be expressed in the absence of glucose, but E. coli always has the capability to degrade lactose regardless of whether that pathway is active. Thus, our overall conclusion that gut microbes associated with IBD are metabolically self-sufficient (encoding the enzymatic capability to synthesize certain key metabolites) remains valid irrespective of fixed or flexible pathway directionality.

It's also important to be careful not to conflate KEGG modules (small subsets of a pathway) with the actual metabolic pathway. It's possible to have a module change in abundance while not altering the full pathway. Inspection of the individual genes could help in this respect - are they rate-limiting steps for biosynthesis or catabolism?

The reviewer is absolutely correct that KEGG modules do not necessarily represent full pathways. We have updated the language in our manuscript to explicitly refer to “modules” rather than “pathways” whenever appropriate, to restrict the scope of the analysis to metabolic modules rather than full pathways.

That said, we do not see how “inspection of individual genes” would improve our analysis. The strength of looking at complete modules rather than individual genes is that we can gain conclusive insights into a certain metabolic capacity. Of course, no pathway or module stands alone. However, the enrichment of metabolic modules does conclusively indicate that these modules are beneficial under the given conditions, such as stress caused by inflammation or antibiotic use. Whether a certain step in a module or pathway is rate limiting is completely irrelevant for this analysis.

Line 177 - I'm not a big fan of the HMI acronym. Is there a LMI group? It seems simplistic to lump all of metabolism into dependent or independent, which in reality will differ depending on the specific substrate, the growth condition, and the strain.

While we are sorry that our study failed to provide the reviewer with a term they could be a fan of, their input did not change our view that HMI, an acronym we have adapted from a previously peer-reviewed study (doi:10.1186/s13059-023-02924-x), is a powerfully simplistic means to describe a phenomenon we observe and demonstrate in multiple different ways with our extensive analyses. The argument that HMI or LMI status will differ given the growth condition, substrate availability, or strain differences is not helping this case either: our analyses cut across a large number of humans and naturally occurring microbial systems in their guts that are exposed to largely variable ‘growth conditions’ and ‘substrates’ and composed of many strain variants of similar populations. Yet, we observe a clear role for HMI despite all these differences. Perhaps it is because HMI simply describes a higher metabolic capacity based on a defined subset of largely biosynthetic pathways that we observe to be consistently enriched in a large dataset covering a large variety of host, environmental and diet factors and indicates that a population has a higher metabolic capacity to not rely on ecosystem services. We show in our analysis that in the inflamed gut these capacities are indeed required, which is why HMI populations are enriched in IBD samples. HMI has no relation to any of the constraints mentioned by the reviewer, which is one of the major strengths of this metric.

Line 198 - It seems like a big assumption to state that efflux and drug resistance are unrelated to biosynthesis, as they could be genetically or even phenotypically linked.

We agree with the reviewer and are thankful for their input. We have weakened the assertion in this statement.

“These capacities may provide an advantage since antibiotics are a common treatment for IBDs (Nitzan et al. 2016), but are not necessarily related to the systematic enrichment of biosynthesis modules that likely provide resilience to general environmental stress rather than to a specific stressor such as antibiotics.”

Lines 202-218 - I'd suggest removing this paragraph. The "non-IBD" data introduces even more complications to the meta-analysis and seems irrelevant to the current study.

We thank the reviewer for this suggestion. Non-IBD data is important, but its relevance to the primary aims of the study is indeed negligible. We now have moved this paragraph to Supplementary File 1 (under the section “‘Non-IBD’ samples are intermediate to IBD and healthy samples”).

The health gradient is particularly problematic, putting cancer closer to healthy than IBD.

We took the reviewer’s advice and have swapped the order of the studies in Supplementary Figure 6 to place the cancer samples from Feng et al. closer to the IBD samples, on the other side of the non-IBD samples from the IBD studies.

Lines 235-257 - should trim this down and move to the discussion.

As mentioned above, we have opted for a “Results and Discussion format” for our manuscript, so we believe this discussion is in the correct place. We find it important to clearly highlight the limitations and potential biases of our work and trimming this text would take away from that goal.

Figure 3 - panels are out of order. Need to put the current panel D below current panel C. Also, relabel panel letters to go top to bottom (the bottom panel should be D). Could change current panel 3D to a violin plot to match current 3C.

We have updated Figure 3 by converting panel A into a new supplementary figure (Supplementary Figure 8), moving panels C and D below panel B, and relabeling the panels accordingly.

Figure 3B - this panel was incredibly useful and quite surprising to me in many respects. I would have assumed that the Bacteroides would be in the "HMI" bin. Is this a function of the specific strains included here? Was B. theta or B. fragilis included?

The reviewer makes an excellent observation that has been keeping us awake at night, yet somehow was not appropriately discussed in the text until their input. We are very thankful for their attention to detail here.

It is indeed true that Bacteroides genomes are often detected with increased abundance in individuals with IBD and likely have a survival advantage in the IBD gut environment, Bacteroides fragilis and Bacteroides thetaiotaomicron being some of the most dominant residents of the IBD gut. Their non-HMI status is not a function of which strains were included, since all taxa here are represented by the representative genomes available in the publicly available Genome Taxonomy Database. Their non-HMI status comes from the fact that they have HMI scores of around 24 to 26, which fall slightly below the threshold score of 26.4 that we used to classify genomes as HMI. This threshold is back-calculated from the metabolic completion requirement of at least 80% average completion of all 33 metabolic modules that are significantly enriched in IBD. So these genomes are right there at the edge, but not quite over it.

Thanks to this comment by our reviewer, we started wondering whether we should follow a more ‘literature-driven’ approach to set the threshold for HMI, rather than the 80% cutoff, and in fact attempted to lower the HMI score threshold to see if we could include more of the IBD-associated Bacteroides in the HMI bin. Author response table 1 below shows the relevant subset of our new Supplementary Table 3h, which describes the data from our tests on different thresholds.

Author response table 1.

Number and proportion of Bacteroides genomes classified as HMI at each HMI score threshold. There were 20 total Bacteroides genomes in the set of 338 gut microbes identified from the GTDB. The HMI score is computed by adding the percent completeness of all 33 IBD-enriched KEGG modules. The full table can be viewed in Supplementary Table 3h.

Lowering the threshold to 24.75, which corresponds to an average of 75% completeness in the 33 IBD-enriched modules, enabled the classification of 6 Bacteroides genomes as HMI, including B. fragilis, B. intestinalis, B. theta, and B. faecis. However, it also identified several microbes that are not IBD-associated as HMI, including 75 genomes from the Lachnospiraceae family and 18 genomes from the Ruminococcaceae family. In the latter family, several Faecalibacterium genomes, including 10 representatives of Faecalibacterium prausnitzii, were considered HMI using this threshold. These microbes are empirically known to decrease in abundance during inflammatory gastrointestinal conditions (doi:10.3390/microorganisms8040573, doi:10.1093/femsre/fuad039), and therefore these genomes should not be considered HMI – at least not under the working definition of HMI used in our study. To avoid including such a large number of obvious false positives in the HMI bin, we decided to maintain a higher threshold despite the exclusion of Bacteroides genomes.

This outcome demonstrates that our reductionist approach does not successfully capture every microbial population that is associated with IBD. Nevertheless, and in our opinion very surprisingly, the metric does capture a very large proportion of genomes with increased detection and abundance in IBD samples, as demonstrated by the peaks of detection/abundance that match to HMI status Author response image 1.

Author response image 1.

Screenshots of Figure 3 that demonstrate the overlapping signal between HMI status and genome detection/abundance in IBD.

Furthermore, the violin plots in Figure 3B (formerly Figure 3C) clearly reflect the increased representation of HMI populations in IBD metagenomes. Although our classification method is imperfect, it still demonstrates the predictive power of metabolic competencies in identifying which microbes will survive in stressful gut environments. To ensure that readers recognize the crude nature of this classification strategy and the possibility that high metabolic independence can be achieved in different ways, we have added the following sentences to the relevant section of our manuscript:

“Given the number of ways a genome can pass or fail this threshold, this arbitrary cut-off has significant shortcomings, which was demonstrated by the fact that several species in the Bacteroides group were not classified as HMI despite their frequent dominance of the gut microbiome of individuals with IBD (Saitoh et al. 2002; Wexler 2007; Vineis et al. 2016) (Supplementary File 1). That said, the genomes that were classified as HMI by this approach were consistently higher in their detection and abundance in IBD samples (Figure 3a). It is likely that there are multiple ways to have high metabolic independence which are not fully captured by the 33 IBD-enriched metabolic modules identified in this study.”

We have also included a discussion of these findings in Supplementary Information File 1 (see section “Examining the impact of different HMI score thresholds on genome-level results”).

This panel also makes it clear that many of these modules are widespread in all genomes and thus unlikely to meaningfully differ in the microbiome. It would be interesting to use this type of analysis to identify a subset of KEGG modules with high variability between strains.

The figure makes it ‘look like’ many of these modules are widespread in all genomes and thus unlikely to meaningfully differ in the microbiome, but our quantitative analyses clearly demonstrate that these modules indeed differ meaningfully between microbiomes of healthy individuals and those diagnosed with IBD. For instance, the classifier that we built relying exclusively upon these modules’ PPCN values was able to reliably distinguish between the healthy and IBD sample groups in our dataset. The fact that the differentiating signal does not rely on rare metabolic or signature modules is what makes the classifier powerful enough to differentiate between “healthy” and “stressed” microbiomes in 86% of cases. Modules that are by nature less common could not serve this purpose. That said, we do agree with the reviewer that it might be interesting to study variability of KEGG modules as a function of variability between strains. This does not fall into the scope of this work, but we hope to assist others with the technical aspects of such work.

Considering the entirety of the exchange in this section, perhaps there is a broader discussion to be had around this topic. In retrospect, not being able to perfectly split microbes into two groups that completely recapitulate their enrichment in healthy or IBD samples by a crude metric and an arbitrary threshold is not surprising at all. What is surprising is that such a crude metric in fact works for the vast majority of microbes and predicts their increased presence in the IBD gut by only considering their genetic make up. In some respects, we believe that the inability of this cutoff to propose a perfect classifier is similar to the limited power of metabolic independence concept and the classes of HMI or LMI to capture and fully explain microbial fitness in health and disease. What is again surprising here is that these almost offensively simple classes do capture more than what one would expect. We can envision a few ways to implement a more sophisticated HMI/LMI classifier, and it is certainly an important task that is achievable. However, we are hopeful that this technical work can also be done better by others in our field, and that step forward, along with further scrutinizing the relevance of HMI/LMI classes to understand metabolic factors that contribute to the biodiversity of stressful environments, will have to remain as future work.

We thank the reviewer again for their comment here and pushing us to think more carefully and address the oddity regarding the poor representation of Bacteroides as HMI by our cutoff.

Given that a lot of the gaps are in the Firmicutes, this panel also makes me more concerned about annotation bias. How many of these gaps are real?

Analyses relying on gene annotations all suffer equally from the potential for missannotation or missing annotations, which primarily result from limitations in our reference databases for functional data. For instance, the Hidden Markov models for microbial genes in the KEGG Ortholog database are generated from a curated set of gene sequences primarily originating from cultivable microorganisms and particularly from commonly-used model organisms; hence, they do not capture the full extent of sequence diversity observed in populations that are less well-represented in reference databases – a category which includes several Firmicutes, as the reviewer points out. For KEGG KOfams in particular, the precomputed bit score thresholds for distinguishing between ‘good’ and ‘bad’ matches to a given model are often too stringent to enable annotation of genes that are just slightly too divergent from the set of known sequences, thus resulting in missing annotations. Based on our experience with these sorts of issues, we implemented a heuristic that reduces the number of missing annotations for KOs and captures significantly more homologs than other state-of-the-art approaches, as described in doi:10.1101/2024.07.03.601779. We refer the reviewer to our response to the related public comment about annotation bias above, which includes additional details about our investigations of annotation bias in our data. In comparison to the current standard, the heuristic we implemented improves functional annotation results. However, neither our nor any other bioinformatic study that relies on functional gene annotation can exclude the potential for annotation bias.

Figure 3B plotting issues - need to use the full names of the modules; for example, M00844 is "arginine biosynthesis, ornithine => arginine", which changes the interpretation. Need a key for the heatmap on the figure. The tree is difficult to see, needs a darker font.

We have darkened the lines of the tree and dendrogram, and added a legend for the heatmap gradient (see new version of Figure 3 above). Unfortunately, we could not fit the full names of the modules into the figure due to space constraints. However, the full module name and other relevant information can be found in Supplementary Table 2a, and the matrix of pathway completeness scores in these genomes (e.g., the values plotted in the heatmap) can be found in Supplementary Table 3b. We are not sure what the reviewer refers to when stating that “for example, M00844 is "arginine biosynthesis, ornithine => arginine", which changes the interpretation”. There is no ambiguity regarding the identity of KEGG module M00844, which is arginine biosynthesis from ornithine.

Line 321 - more justification for the 80% cutoff is needed along with a sensitivity analysis to see if this choice matters for the key results.

Inspired by this comment, and the one above regarding the classification of Bacteroides genomes, we tested several HMI score thresholds ranging from 75% to 85% average completeness of the 33 IBD-enriched modules. For each threshold, we computed all the key statistics reported in this section of our paper, including the statistical tests. We found that the choice of HMI score threshold does not influence the overall conclusions drawn in this section of our manuscript. Author response table 2 below shows the relevant subset of our new Supplementary Table 3h, which describes the results for each threshold:

Author response table 2.

Key genome-level results at each HMI score threshold. The HMI score is computed by adding the percent completeness of all 33 IBD-enriched KEGG modules. WRS – Wilcoxon Rank Sum test; KW – Kruskal-Wallis test. The full table can be viewed in Supplementary Table 3h

We’ve summarized these findings in a new section of Supplementary File 1 entitled “Examining the impact of different HMI score thresholds on genome-level results”. We copy below the relevant text for the reviewer’s convenience:

“Determining the HMI status of a given genome required us to set a threshold for the HMI score above which a genome would be considered to have high metabolic independence. We tested several different thresholds by varying the average percent completeness of the 33 IBD-enriched metabolic modules that we expected from the

‘HMI’ genomes from ≥ 75% (corresponding to an HMI score of ≥ 24.75) to ≥ 85% (corresponding to an HMI score of ≥ 28.05). For each threshold, we computed the same statistics and ran the same statistical tests as those reported in our main manuscript to assess the impact of these thresholds on the results (Supplementary Table 3h). At the highest threshold we tested (HMI score ≥ 28.05), a small proportion of the reference genomes (7%, or n = 24) were classified as HMI, so we did not test higher thresholds.

We found that the results from comparing HMI genomes to non-HMI genomes are similar regardless of which HMI score threshold is used to classify genomes into either group. No matter which HMI score threshold was used, the mean genome size and mean number of genes were higher for HMI genomes than for non-HMI genomes. On average, the HMI genomes were about 1 Mb larger and had 1,032 more gene calls than non-HMI genomes. We ran two Wilcoxon Rank Sum statistical tests to assess the following null hypotheses: (1) HMI genomes do not have higher detection in IBD samples than non-HMI genomes, and (2) HMI genomes do not have higher detection in healthy samples than non-HMI genomes. For both tests, the p-values decreased (grew more significant) as the HMI score threshold decreased due to the inclusion of more genomes in the HMI bin. The first test for higher detection of HMI genomes than non-HMI genomes in IBD samples yielded p-values less than α = 0.05 at all HMI score thresholds. The second test for higher detection of HMI genomes than non-HMI genomes in healthy samples yielded p-values less than α = 0.05 for the three lowest HMI score thresholds (HMI score ≥ 24.75, ≥ 25.08, or ≥ 25.41). However, irrespective of significance threshold and HMI score threshold, there was always far stronger evidence to reject the first null hypothesis than the second, given that the p-value for the first test in IBD samples was 1 to 5 orders of magnitude lower (more significant) than the p-value for the second test in healthy samples.

IBD samples harbored a significantly higher fraction of genomes classified as HMI than healthy or non-IBD samples, regardless of HMI score threshold (p < 1e-15, Kruskal-Wallis Rank Sum test). The p-values for this test increased (grew less significant) as the HMI score threshold decreased. This suggests that, at higher thresholds, relatively more genomes drop out of the HMI fraction in healthy/non-IBD samples than in IBD samples, thereby leading to larger differences and more significant p-values. Consequently, the HMI scores of genomes detected in IBD samples must be higher than the HMI scores of genomes detected in the other sample groups – indeed, the average HMI score of genomes detected within at least one IBD sample is 24.75, while the average score of genomes detected within at least one healthy sample is 22.78. Within a given sample, the mean HMI score of genomes detected within that sample is higher for the IBD group than in the healthy group: the average per-sample mean HMI score is 25.14 across IBD samples compared to the average of 23.00 across healthy samples.”

Lines 357 and 454 - I would remove the discussion of the "gut environment" which isn't really addressed here. The observed trends could just as easily relate to microbial interactions or the effects of diet and pharmaceuticals. Perhaps the issue is the vague nature of this term, which I read to imply changes in the mammalian host. Given the level of evidence, I'd opt to keep the options open and discuss what additional data would help resolve these questions.

We are in complete agreement with the reviewer that microbial interactions are likely an important driver of our observations. In healthy communities, microbial cross-feeding enables microbes with lower metabolic independence to establish and increase microbial diversity. Which is exactly why we are stating that “Community-level signal translates to individual microbial populations and provides insights into the microbial ecology of stressed gut environments”.

Diet or usage of prescription drugs on the other hand, as discussed previously, likely varies substantially over the various cohorts investigated, and is thus not a driver of the observed trends. Instead, HMI works as a high level indicator that is not influenced by these variable host habits.

Lines 354-394 - Could remove or dramatically trim down this text. Too much discussion for a results section.

We kindly remind the reviewer that our manuscript is written following a “Results and Discussion” format. This section provides necessary context and justification for our classifier implementation, so we have left it as-is.

Lines 395-441 - This section raised a lot of issues and could be qualified or even removed. The model was trained on modules that were IBD-associated in the same dataset, so it's not surprising that it worked. An independent test set would be required to see if this model has any broader utility.

The point that we selected the IBD-enriched modules as features should not raise any concerns, as these modules would have emerged as the most important (ie, most highly weighted) features in our model even if we had included all modules in our training data. This is because machine learning classifiers by design pick out the features that best distinguish between classes, and the 33 IBD-associated modules are a selective subset of these (if they were not, they would not have been significantly enriched in the IBD sample group). That said, a carefully conducted feature selection process prior to model training is a standard best-practice in machine learning; thus, if anything, this should be interpreted as a point of confidence rather than a concern. Furthermore, we evaluated our model using cross-validation, a standard practice in the machine learning field that assesses the stability of model performance by training and testing the model on different subsets of the data. This effort established that the model is robust across different inputs as demonstrated by the per-fold confusion matrix and the ROC curve. These are all standard approaches in machine learning to quantify the model tradeoff between bias and variance. As for the independent test set, we went far and beyond, and applied our model to the antibiotic time-series dataset described later in this section, which, in our opinion, and likely also in the opinion of many experts, serves as one of the most convincing ways to test the utility of any model. Classification results here show that our hypothesis concerning the relevance of metabolic independence to microbial survival in stressed gut environments applies beyond the IBD case and includes antibiotic use, which is indeed a stronger validation for this hypothesis than any test we could have done on other IBD-related datasets. Regardless, we agree that any ‘broader’ utility of our model, such as its applications in clinical settings for diagnostic purposes, is something we certainly can not make strong claims about without more data. We have therefore qualified this section by adding the following sentence:

“Determining whether such a model has broader utility as a diagnostic tool requires further research and validation; however, these results demonstrate the potential of HMI as an accessible diagnostic marker of IBD.”

The application to the antibiotic intervention data raises additional concerns, as the model will predict IBD (labeled "stress" in Figure 5) where none exists.

We apologize for this misunderstanding. The label “stress” actually means stress, not IBD. The figure the reviewer is referring to demonstrates that metabolic modules enriched in the gut microbiome of IBD patients are also temporarily enriched in the gut microbiome of healthy individuals treated with antibiotics for the duration of the treatment. While the classifier uses PPCN values for 33 metabolic modules enriched in microbiomes of IBD patients, it does not mean that this enrichment is exclusive to IBD. The classifier will distinguish between metagenomes in which the PPCN values for those 33 metabolic modules is higher and metagenomes in which the PPCN values are lower. Hence, our analysis demonstrates that during antibiotic usage in healthy individuals, the PPCN values of these 33 metabolic modules spike in a similar fashion to how they would in the gut community of a person with IBD. This points to a more general trend of high metabolic independence as a factor supporting microbial survival in conditions of stress; that is, the increase in metabolic independence is not specific to the IBD condition but rather a more generic ecological response to perturbations in the gut microbial community. We have clarified this point with the following addition to the paragraph summarizing these results:

“All pre-treatment samples were classified as ‘healthy’ followed by a decline in the proportion of ‘healthy’ samples to a minimum 8 days post-treatment, and a gradual increase until 180 days post treatment, when over 90% of samples were classified as ‘healthy’ (Figure 5, Supplementary Table 4b). In other words, the increase in the HMI metric serves as an indicator of stress in the gut microbiome, regardless of whether that stress arises from the IBD condition or the application of antibiotics. These observations support the role of HMI as an ecological driver of microbial resilience during gut stress caused by a variety of environmental perturbations and demonstrate its diagnostic power in reflecting gut microbiome state.”

We’ve also added the following sentence to the end of the legend for Figure 5:

“Samples classified as ‘healthy’ by the model were considered to have ‘no stress’ (blue), while samples classified as ‘IBD’ were considered to be under ‘stress’ (red).”

Figure S5A - should probably split this into 2 graphs since different data is analyzed.

It is true that different sets of modules are used in either half of the figure; however, there is a significant amount of overlap between the sets (17 modules), which is why there are lines connecting the points for the same module as described in the figure legend. We are using this figure to make the point that the median PPCN value of each module increases, in both sets of modules, from the healthy sample group to the IBD sample group. Therefore, we believe the current presentation is appropriate.

Figure S6A – this shows a substantial study effect and raises concerns about reproducibility.

We examined potential batch effects in Supplementary Information File 1 (see section “Considerations of Batch Effect”), and found that any study effect was minor and overcome by the signal between groups:

“The similar distribution of the median normalized copy number for each of the 33 IBD-enriched metabolic modules (summarized across all samples within a given study), across all studies within a given sample group (Supplementary Figure 6b), confirms that the sample group explains more of the trend than the study of origin.”

Furthermore, within Supplementary Figure 6a, there is a clear increase between the non-IBD controls from Franzosa et al. 2018 and the IBD samples from the same study, as well as between the non-IBD controls from Schirmir et al. 2018 and the IBD samples from that study. As there is no study effect influencing those two comparisons, this reinforces the evidence that there is a true increase in the normalized copy numbers of these modules when comparing samples from more healthy individuals to those from less healthy individuals.

Figure S7B - check numbers, which I think should sum to 33.

The numbers should not sum to 33. In this test to determine whether the two largest studies had excessive influence on the identity of the IBD-enriched modules, we repeated our strategy to obtain 33 IBD-enriched modules (those with the 33 smallest p-values from the statistical test) from each set of samples – either (1) samples from Le Chatelier et al. 2013 and Vineis et al. 2016, or (2) samples that are not from those two studies. The 2 sets, containing 33 modules each, gives us a total of 66 IBD-enriched modules. By comparing those two sets, we found that 20 modules were present in both sets – hence the value of 20 in the center of the Venn Diagram. In each set, 13 modules were unique – hence the value of 13 on either side. 13 + 13 + 2*20 = 66 total modules.

We again thank our reviewers for their time and interest, and invaluable input.

Author response:

The following is the authors’ response to the original reviews.

Reveiwer#1 (Public Review):

Weaknesses:

While the novel compound showed a promising potency to the HER2-positive gastric cancer cells and xenograft model, it would be great to also to be evaluated with the HER2-positive breast cancer cell models. The author did not compare the current compounds with other therapeutic strategies targeting HER2 expression at the genetic level. It is unclear whether the EGFR inhibitors gefitinib and canertinib but not HER2-specific inhibitors (i.e. tucatinib) were used as a control in the manuscript.

We appreciate the reviewer’s insightful comments. Evaluating compound 10 on HER2-positive breast cancer cells is indeed crucial, especially given the established HER2-targeting therapies for breast cancer. In response to this concern, we conducted additional experiments to investigate the impact of compound 10 on HER2-positive breast cancer cell lines AU565 and BT474, specifically assessing its HER2 downregulating activity (Author response image 1).

Author response image 1.

HER2 downregulatory effect of compound 10 in HER2-positive breast cancer cell lines, AU565 and BT474.

The selection of gefitinib (an EGFR tyrosine kinase inhibitor) and canertinib (a pan-HER inhibitor) as positive controls in our manuscript is based on their demonstrated ability to inhibit the protein-protein interaction (PPI) between ELF3 and MED23, as previously reported (J Adv Res. 47, (2023) 173-87. 10.1016/j.jare.2022.08.003; Cancer letters. 325, (2012) 72-9. 10.1016/j.canlet.2012.06.004). In referenced studies, SEAP reporter gene assay was utilized to screen compounds for their capacity to disrupt the ELF3-MED23 PPI. This assay involves GAL4-ELF3 binding to a GAL4 binding site in the SEAP reporter gene, followed by interaction with MED23, leading to RNA polymerase II recruitment and SEAP expression in cells (J Am Chem Soc. 2004, 126(49), 15940. doi: 10.1021/ja0445140). Canertinib exhibited stronger inhibitory activity against ELF3-MED23 PPI compared to gefitinib, but also showed non-specific cytotoxicity. YK1 was subsequently developed based on structural analysis of the interfaces between gefitinib and MED23, and between ELF3 and MED23. Considering the previously validated inhibitory activities of gefitinib and canertinib, these drugs were selected as positive controls in the current study to compare the ELF3-MED23 inhibitory efficacy of novel compounds.

Reveiwer#1 (Recommendations For the Authors):

(1) It is unclear how compound 5 did not inhibit HER2 overexpression at mRNA but at protein levels as compounds 3 and 10. Could the author further explain the potential mechanism for compound 5?

While the exact mechanism remains unclear, the results indicated that compound 5 likely affects the protein level of HER2 through somewhat non-specific mechanisms rather than by inhibiting the ELF3-MED23 PPI. Based on this assessment, compound 5 was excluded from further investigation.

(2) The HER2 expression and its downstream signaling pathway assay are unclear about the approach. It needs to be included in the methods or supplementary.

We investigated the ELF3-MED23 PPI inhibitory activity and its subsequent effect on HER2 downregulation using a comprehensive approach involving multiple techniques to ensure precise and unbiased experimental results.

To assess PPI inhibition, we employed the following assays:

· SEAP reporter gene assay

· Fluorescence polarization (FP)

· Split-luciferase complementation assay

· GST-pulldown

· Immunoprecipiation (IP)

HER2 expression levels were evaluated through:

· SEAP reporter gene assay

· Luciferase promoter assay

· Quantification of HER2 mRNA using qPCR

· Measurement of HER2 protein levels via western blot analysis

To evaluate downstream signaling of HER2, we analyzed:

· Phosphorylation levels of MAPK (pMAPK) and AKT (pAKT)

These methods were systematically applied to elucidate the mechanism of action of compound 10 in inhibiting ELF3-MED23 interaction and subsequently downregulating HER2.

For clarity, we have revised the manuscript to provide a detailed description of the experimental methods to assess PPI, as described below.

“SEAP assay was performed as previously described to measure ELF3-MED23 PPI-dependent HER2 transcription [29]. In this assay, the GAL4-ELF3 fusion protein binds to one of the five GAL4 binding sites on the reporter gene (pG4IL2SX). The interaction between the GAL4-ELF3 fusion protein and endogenous MED23 induces the expression of the SEAP. Once expressed, SEAP acts as a phosphatase on the substrate 4-MUP (4-methyl umbelliferyl phosphate), resulting in increased fluorescence. The mammalian expression vector, …”

“FP assay was conducted following a previously described method to evaluate the molecular interaction between ELF3 and MED23 [29]. The FP assay operates on the principle of the molecular rotation dynamics. When a fluorescently labeled small molecule is excited by polarized light, the emitted fluorescence can be polarized or depolarized depending on the molecular status. Free small molecules rotate rapidly, altering the orientation of their fluorescence dipole and emitting depolarized light. However, when these small molecules bind to large molecules, such as proteins, the resulting complex rotates more slowly, and the emitted light retains much of its original polarization. In this study, different concentrations of (His)6-MED23391–582, as the large molecule, and 10 nM of FITC-labeled ELF3129–145 peptide, as the fluorescence-labeled small molecule, were combined in …”

(3) It is confusing to me about the order of the experiments, in which the SAR work came after the synthesis and a series of biochemical studies for the characterization of the synthetic compounds. What is the specific reason for this order?

We concluded that the current approach is appropriate because the analysis was not intended for structural modification and optimization through SAR (Structure-Activity Relationship) analysis. Instead, the primary objective was to elucidate the structural basis underlying the efficacy of PPI inhibition among compounds sharing the same scaffold. We believe this will provide valuable insights for future design and synthesis of new compounds.

(4) The yield for each step of the general synthesis needs to be included in the scheme 1.

Scheme 1 has been updated to include the yield of each step of the synthesis process.

(5) In line 532, the authors stated 28 compounds, should it be 26?

‘Twenty-eight compounds’ includes 26 newly synthesized compounds and 2 positive controls, gefitinib and canertinib.

(6) Introduction part, lines 74 to 75, "While HER2 gene amplification is the primary mechanism responsible for HER2 overexpression" may not be confirmed in lung cancers.

HER2 overexpression is usually a direct consequence of gene amplification, although overexpression can occur by other mechanisms [Nat Rev Cancer. 2009;9:463–475. doi: 10.1038/nrc2656.; Cell. 2007;129:1275–1286. doi: 10.1016/j.cell.2007.04.034.]. The levels of HER2 protein expression and gene amplification are linearly associated and highly concordant in breast cancer, colorectal cancer, ovarian cancer, and esophageal adenocarcinoma [World J Gastrointest Oncol. 2019, 11(4): 335–347. doi: 10.4251/wjgo.v11.i4.335; J Clin Oncol. 2002;20:719–26. doi.org/10.1200/JCO.2002.20.3.71; Oncology. 2001;61(Suppl 2):14–21. doi.org/10.1159/000055397; Science. 1989, 244(4905):707-12. doi: 10.1126/science.2470152; Cancer. 2014 Feb 1; 120(3): 415–424. doi: 10.1002/cncr.28435]. As reviewer mentioned, the linear association between of HER2 protein expression and gene amplification has not been fully established for NSCLC [ESMO Open. 2022, 100395. doi: 10.1016/j.esmoop.2022.100395].

Therefore, we change the sentence as describe below.

“While HER2 gene amplification is the primary mechanism responsible for HER2 overexpression in most HER2-positive cancers, except in lung cancer [16], high transcription rates of HER2 per gene copy have also been observed to contribute.”

(7) The abstract part, lines 31 and 32, the detailed experimental data for SEAP needs to be expressed in another way.

SEAP is a type of reporter gene assay. We revised the manuscript as follows and we additionally described it method part.

“Upon systematic analysis, candidate compound 10 was selected due to its potency in downregulating reporter gene activity of HER2 promoter confirmed by SEAP activity and its effect on HER2 protein and mRNA levels.”

(8) The author should combine the box for Chalcone, pyrazoline, Licochalcone E, and YK-1, Figures 1 and 2 into a new single Figure.

We revised the manuscript following the reviewer's comments.

(9) Provide the list of antibodies and sources for the cell-based and western blot assays.

Table S1 presents detailed information about the antibodies and dilution ratios used in the cell-based and western blot assays.

Reveiwer#2 (Public Reviews):

Weaknesses:

The rationale behind the proposed structural modifications for the three groups of compounds is not clear.

Reveiwer#2 (Recommendations For the Authors):

(1) Based on previous work experience, it would be interesting to evaluate the in silico mode of interaction of compound 10.

As suggested by the reviewers, we additionally performed in silico docking study to identify the mode of interaction of compound 10 (Author response image 2). As shown below, the results indicate that compound 10 shares a similar binding orientation with YK1, forming an H-bond with the H449 residue. Although it does not interact with the D400 residue, it was predicted to create an additional H-bond with S450, which is right next to H449, thereby reinforcing the overall binding of compound 10 to MED23. Moreover, compound 10 was additionally predicted to form a pi-pi interaction with F399, which has been previously identified as an important interaction for compounds to demonstrate outstanding PPI inhibitory effect against ELF3 and MED23.

Author response image 2.

Docking analysis of compound 10.

(2) The chalcones presented in this study are structurally similar to those previously presented by the group (ref 29). In said work, most of the compounds exhibited activities with IC50 values between 1.3 and 3 μM, with inhibition values at 10 μM ranging between 80 and 90% in the SEAP assay. These results are similar to those observed in this paper for the same assay. Can an explanation be found?

Chalcones are inherently flexible molecules, giving them a high chance of occupying critical hotspot residues within the binding interface of ELF3-MED23, irrespective of the side chains introduced to this moiety. However, depending on the type of side chains introduced, the overall drug-like properties of compounds can be significantly altered, while still maintaining their PPI inhibitory effect. The significance of this study lies in our effort to enhance metabolic stability through extensive introduction of methoxy groups and other hydrophobic side chains to the chalcone skeleton, while preserving high PPI inhibitory activity.

(3) Is the replacement of H and OH by OMe necessary? Does it improve any property (activity, selectivity, bioavailability, solubility, etc.)? Regarding the derivatives of group 2, why did they decide to replace the O-H, which in silico demonstrated favorable hydrogen bond interactions with Asp400? How do these molecules look in the binding site? Perhaps this is a point to discuss since the substitution of OH led to the obtaining of inactive molecules, or is the effect due to substitution with the terminal aromatic ring with 3 OMe?

We modified the hydroxyl group moiety of YK-1 into a methoxy group to reduce the polarity of the compound, thereby enhancing its cell membrane permeability (Author response image 3) and reducing the likelihood of rapid elimination through phase II metabolic pathways in vivo. Additionally, we considered the potential conversion of the methoxy group back to a hydroxyl group via phase I metabolism in vivo.

Author response image 3.

Impact of methoxy group introduction on TPSA (total polar surface area) of each molecule. TPSA of each molecule containing chalcone structure were calculated using the Molinspiration webserver.

(4) Lines 134 and 134: "Only compounds are in red."

We revised the manuscript following the reviewer's comments.

(5) Line 171: "Chalcone skeleton, shown in red."

We revised the manuscript following the reviewer's comments.

(6) Line 350: "N-1-acetyl-4,5-dihydropyrazoline."

We revised the manuscript following the reviewer's comments.

(7) Scheme 1. Replace "h" with "hr".

We revised the manuscript following the reviewer's comments. Scheme 1 has been replaced by a new version.

(8) Where is "Table S1" in SI?

Tables S1 and S2 are supposed to be included in SI. We will ensure that Tables S1 and S2 are properly uploaded to the SI section.

(9) In Figure 6, Graph D, to enhance comprehension, please incorporate red arrows indicating drug administration.

We revised Figure 6 (D) following the reviewer's comments. Red arrows indicating drug administration have been incorporated, along with a descriptive comment "Drug administration" next to each arrow. Additionally, the figure legend now includes a clear description of these additions.

Reveiwer#3 (Public review):

Weaknesses:

Compound 10 potency as PPI inhibitor has been shown in only one cell line NCI-N87.

Reveiwer#3 (Recommendations For the Authors):

(1) The authors should show this compound 10 is effective in other gastric cancer cells like KATOIII, SNU1.

We evaluated the HER2 downregulating activity of compound 10 in the gastric cancer cell line, SNU216, which is confirmed to express high level of HER2 protein (Author response image 4).

Author response image 4.

HER2 downregulatory effect of compound 10 in HER2-positive gastric cancer cell line, SNU216. (A) Expression levels of HER2 and ELF3 in various gastric cancer cell lines. (B) HER2 downregulation in the SNU216 cell line following treatment with compound 10.

Author response:

The following is the authors’ response to the original reviews.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

In this work by Wang et al., the authors use single-molecule super-resolution microscopy together with biochemical assays to quantify the organization of Nipah virus fusion protein F (NiV-F) on cell and viral membranes. They find that these proteins form nanoscale clusters which favors membrane fusion activation, and that the physical parameters of these clusters are unaffected by protein expression level and endosomal cleavage. Furthermore, they find that the cluster organization is affected by mutations in the trimer interface on the NiV-F ectodomain and the putative oligomerization motif on the transmembrane domain, and that the clusters are stabilized by interactions among NiV-F, the AP2-complex, and the clathrin coat assembly. This work improves our understanding of the NiV fusion machinery, which may have implications also for our understanding of the function of other viruses.

Strengths:

The conclusions of this paper are well-supported by the presented data. This study sheds light on the activation mechanisms underlying the NiV fusion machinery.

Weaknesses:

The authors provide limited details of the convolutional neural network they developed in this work. Even though custom-codes are made available, a description of the network and specifications of how it was used in this work would aid the readers in assessing its performance and applicability. The same holds for the custom-written OPTICS algorithm. Furthermore, limited details are provided for the imaging setup, oxygen scavenging buffer, and analysis for the single-molecule data, which limits reproducibility in other laboratories. The claim of 10 nm resolution is not backed up by data and seems low given the imaging conditions and fluorophores used. Fourier Ring Correlation analysis would have validated this claim. If the authors refer to localization precision rather than resolution, then this should be specified and appropriate data provided to support this claim.

We thank reviewer 1 for these suggestions. We described key steps in imaging setup, singlemolecule data reconstruction, the OPTICS algorithm in cluster identification, and 1D CNN in

classification of the OPTICS data in the Materials and Methods section. We also provided a recipe for the imaging buffer. We refer to 10 nm localization precision rather than resolution. The localization precision achieved by our SMLM system is shown in the Author response image 1.

Author response image 1.

The localization precision of the custom-built SMLM. Shows the distribution of localization error at the x (dX), y (dY), and z (dZ) direction in nanometer of blinks generated from Alexa Flour 647 labeled to NiV-F expressed on the plasma membrane of PK13 cells. The lateral precision is <10 nm and the axial precision is < 20 nm. 

Reviewer #2 (Public Review): 

Summary:

In this manuscript, Wang and co-workers employ single molecule light microscopy (SMLM) to detect NiV fusion protein (NiV-F) in the surface of cells. They corroborate that these glycoproteins form microclusters (previously seen and characterized together with the NiVG and Nipah Matrix protein by Liu and co-workers (2018) also with super-resolution light microscopy). Also seen by Liu and coworkers the authors show that the level of expression of NiV-F does not alter the identity of these microclusters nor endosomal cleavage. Moreover, mutations and the transmembrane domain or the hexamer-of-trimer interface seem to have a mild effect on the size of the clusters that the authors quantified.

Importantly, it has also been shown that these particles tend to cluster in Nipah VLPs.

We thank reviewer #2 for the comments and suggestions. This paper is built on Liu et al 1 to further characterize the nanoclusters formed by NiV-F and their role in membrane fusion activation. While Liu et al. studied the NiV glycoprotein distribution at the NiV assembly sites to inform mechanisms in NiV assembly and release, Wang et al. analyzed the nanoorganization and distribution of NiV-F at the prefusion conformation, providing insights into the membrane fusion activation mechanisms.  

Strengths:

The authors have tried to perform SMLM in single VLPs and have shown partially the importance of NiV-F clustering.

Weaknesses:

The labelling strategy for the NiV-F is not sufficiently explained. The use of a FLAG tag in the extracellular domain should be validated and compared with the unlabelled WT NiV-F when expressed in functional pseudoviruses (for example HIV-1 based particles decorated with NiV-F). This experiment should also be carried out for both infection and fusion (including BlaM-Vpr as a readout for fusion). I would also suggest to run a time-of-addition BlaM experiment to understand how this particular labelling strategy affects single virion fusion as compared to the the WT.  

We thank reviewer #2 for this suggestion. We have made various efforts to validate the expression and function of FLAG-tagged NiV-F. The NiV-F-FLAG shows comparable cell surface expression levels and induces similar cell-cell fusion levels in 293T cells as that of untagged NiV-F 1. The NiV-F-FLAG also showed similar levels of virus entry as untagged NiV-F when both were pseudotyped on a recombinant Vesicular Stomatitis Virus (VSV) with the VSV glycoprotein replaced by a Renilla luciferase reporter gene (VSV-ΔG-rLuc; Fig. S1D). We also performed a virus entry kinetics assay using NiV VLPs expressing NiV-M-βlactamase (NiV-M-Bla), NiV-G-HA, and NiV-F-FLAG, NiV-F-AU1 or untagged NiV-F. The intracellular AU1 tag is located at the C-terminus of NiV-F (Genbank accession no. AY816748.1). However, we detected different levels of NiV-M-Bla in equal volume of VLPs, suggesting that the tags in NiV-F affect the budding of the VLPs (Author response image 2A). Therefore, we performed fusion kinetics assay by using VLPs expressing the same levels of NiV-M-Bla. Among them, the NiV-F-FLAG on VLPs shows the most efficient fusion between VLP and HEK293T cell membranes (Author response image 2B), significantly more efficient than that of untagged NiV-F and NiV-FAU1. However, we cannot attribute the enhanced fusion activity to the FLAG tag, because the readout of this assay relies on both the levels of β-lactamase (introduced by NiV-M-Bla in VLPs) and the NiV-F constructs. The tags in NiV-F could affect both the budding of VLPs and the stoichiometry of F and M in individual VLPs. We did not use the HIV-based pseudovirus system because the incorporation of NiV-F into HIV pseudoviruses requires a C-terminal deletion 2,3.

In summary, the FLAG tag does not affect cell-cell fusion 1 and virus entry when pseudotyped to the recombinant VSV-ΔG-rLuc viruses (Fig. S1D). Given that we do not observe any difference in clustering between an HA- and FLAG-tagged NiV-F constructs on PK13 cell surface (Fig. S1A-C), we conclude that the FLAG tag has minimal effect on both the fusion activity and the nanoscale distribution of NiV-F. 

Author response image 2.

Viral entry is not affected by labeling of NiV-F. A) Western blot analysis of NiV-M-Bla in NiV-VLPs generated by HEK293T cells expressing NiV-M-Bla, NiV-G-HA and NiV-F-FLAG, untagged NiV-F, or NiV-F-AU1. Equal volume of VLPs were separated by a denaturing 10% SDS–PAGE and probed against β-lactamase (SANTA CRUZ, sc-66062). B) NiV-VLPs expressing NiV-M-BLa, NiV-G-HA, and NiV-F-FLAG, untagged NiV-F or NiV-F-AU1 expression plasmids were bond to the target HEK293T cells loaded with CCF2-AM dye at 4°C. The Blue/Green (B/G) ratio was measured at 37°C for 4 hrs at a 3-min interval. Results were normalized to the maximal B/G ratio of NiV-F-FLAG-NiV VLPs. Results from one representative experiment out of three independent experiments are shown. 

It would also be very important to compare the FLAG labelling approach with recent advances in the field (for instance incorporating noncanonical amino acids (ncAAs) into NiVF by amber stop-codon suppression, followed by click chemistry). 

We are greatly thankful for this comment from reviewer #2. Labeling noncanonical amino acids (ncAAs) with biorthogonal click chemistry is indeed a more precise labeling strategy compared to the traditional epitope labeling approach used in this paper. We will explore the applications of ncAAs labeling in single-molecule localization imaging and virus-host interactions in future projects. 

In this paper, the FLAG tag inserted in NiV-F protein seems to have minimal effect on the NiV-F-induced virus entry and cell-cell fusion 1 (Fig. S1). Although the FLAG tag labeling approach may increase the detectable size of NiV-F nanoclusters due to the use of the antibody complex, it should not affect our conclusions drawn from the relative comparisons between wt and mutant NiV-F or control and drug-treated cells. 

The correlation between the existence of microclusters of a particular size and their functionality is missing. Only cell-cell fusion assays are shown in supplementary figures and clearly, single virus entry and fusion cannot be compared with the biophysics of cell-cell fusion. Not only the environment is completely different, membrane curvature and the number of NiV-F drastically varies also. Therefore, specific fusion assays (either single virus tracking and/or time-of-addition BlaM kinetics with functional pseudoviruses) are needed to substantiate this claim.  

We thank Reviewer 2 for the suggestion. To support the link between F clustering and viruscell membrane fusion, we conducted pseudotyped virus entry and VLP fusion kinetics assays, as shown in revised Figure S4. The viral entry results (Fig. S4 E and F) corroborate that of the cell-cell fusion assay (Fig. S4A and B) and previously published data 4. The fusion kinetics confirmed that the real-time fusion kinetics was affected by mutations at the hexameric interface, with the hypo-fusogenic mutants L53D and V108D exhibited reduced entry efficiency while the hyper-fusogenic mutant Q393L showed increased efficiency (Fig. S4G and H). The results were described in detail in the revised manuscript. 

Additionally, we performed a pseudotyped virus entry assay on the LI4A (Fig. S6F and G) and YA (Fig. S7F and G) mutants to verify the function of these mutants on viruses in revised Supplemental Figures. Neither LI4A nor YA incorporated into the VSV/NiV pseudotyped viruses as shown by the Western blot analyses of the pseudovirions (Fig. S6F and S7F), and thus did not induce virus entry, consisting with the cell-cell fusion results (Fig. S6C, D and Fig. S7C, D). We did not perform the entry kinetic assay of these two mutants as they do not incorporate into VLPs or pseudovirions. 

The authors also claim they could not characterize the number of NiV-F particles per cluster. Another technique such as number and brightness (Digman et al., 2008) could support current SMLM data and identify the number of single molecules per cluster. Also, this technology does not require complex microscopy apparatus. I suggest they perform either confocal fluorescence fluctuation spectroscopy or TIRF-based nandb to validate the clusters and identify how many molecule are present in these clusters.  

We thank reviewer 2 for this suggestion. Determining the true copy number of NiV-F in individual clusters could verify whether the F clusters on the plasma membrane are hexamer-of-trimer assemblies. Regardless, it does not affect our conclusion that the organization of NiV-F into nanoclusters affects the membrane fusion triggering ability. The confocal fluorescence fluctuation spectroscopy (FFS) and TIRF-based analyses are accessible tools for quantifying fluorophore copy numbers and/or stoichiometry based on fluorescence fluctuation or photobleaching. However, these methods are unable to quantify the number of proteins in individual clusters because they analyze fluorophores either in the entire cell (as in wide-field epifluorescence microscopy coupled with FFS and TIRF-coupled photobleaching) 5–7 or within a large excitation volume (confocal laser scanning microscopycoupled FFS) 8. Both of these volumes are significantly larger than a single NiV-F cluster, which has an average diameter of 24-26 nm (Fig. 1F). 

The current SMLM setup is useful for characterizing the protein distribution and organization. However, quantifying the true protein copy number within a nanocluster is challenging because of the stochasticity of fluorophore blinking and the unknown labeling stoichiometry 9–11. To address the challenge in fluorophore blinking, quantitative DNA-PAINT (qDNA-PAINT) may be used because the on-off frequency of the fluorophores is tied to the well-defined kinetic constants of DNA binding and the influx rate of the imager strands, rather than the stochasticity of fluorophore blinking. Thus, the frequency of blinks can be translated to protein counting 12. To address the challenge in unknown labeling stoichiometry, DNA origami can be used as a calibration standard 11. DNA origami supports handles at a regular space with several to tens of nanometers apart, and the handles can be conjugated with a certain number of proteins of interest. The copy number of protein interest in the experimental group can be determined by comparing the SMLM localization distribution of the sample to that of the DNA origami calibration standard. Given the requirement of a more sophisticated SMLM setup and a high-precision calibration tool, we will explore the quantification of NiV-F copy numbers in nanoclusters in a future project. 

Also, it is not clear how many cells the authors employ for their statistics (at least 30-50 cells should be employed and not consider the number of events blinking events. I hope the authors are not considering only a single cell to run their stats... The differences between the mutants and the NiV-F is minor even if their statistical analyses give a difference (they should average the number and size of the clusters per cell for a total of 30-50 cells with experiments performed at least in three different cells following the same protocol). Overall, it seems that the authors have only evaluated a very low number of cells.

We disagree with this comment from Reviewer #2. The sample size for cluster analysis in SMLM images was chosen by considering the target of the study (cells and VLPs) and the data acquisition and analysis standards in the SMLM imaging field. We also noted the sample size (# of ROI and cells) in the figure legend. 

Below, we compared the sample sizes in our study to those in similar studies that used comparable imaging and cluster analysis methods from 2015 to 2024. The classical clustering analysis methods are categorized into global clustering (e.g. nearest neighbor analysis, Ripley’s K function, and pair correlation function) and complete clustering, such as density-based analysis (e.g. DBSCAN, Superstructure, FOCAL, ToMATo) and Tessellationbased analysis (e.g. Delaunay triangulation, Voronoii Tessellation). The global clustering analysis method provides spatial statistics for global protein clustering or organization (e.g. clustering extent), while the complete clustering approach extracts information from a single-cluster level, such as the morphology and localization density of individual clusters. We used the density-based analyses, DBSCAN and OPTICS, for cluster analysis on cell plasma membranes and VLP membranes. 

Author response table 1.

The comparison of imaging methods, analysis methods, and sample size in the current study to other studies conducted from 2015 to 2024.

They should also compare the level of expression (with the number of molecules per cell provided by number and brightness) with the total number of clusters. 

We thank reviewer 2 for this suggestion. We compared the level of expression with the total number of clusters for F-WT in Figure 1I in the main text.  

The same applies to the VLP assay. I assume the authors have only taken VLPs expressing both NiV-M and NiV-F (and NiV-G). But even if this is not clearly stated I would urge the authors to show how many viruses were compared per condition (normally I would expect 300 particles per condition coming from three independent experiments. As a negative control to evaluate the cluster effect I would mix the different conditions. Clearly you have clusters with all conditions and the differences in clustering depending on each condition are minimal. Therefore you need to increase the n for all experiments.

We thank reviewer 2 for this comment. We acquired and analyzed more images of NiV VLPs bearing F-WT, Q393L, L53D, and V108D. Results are shown in the revised Figure 4 and the number of VLPs (>300) used for analysis is specified in the figure legend. An increased number of VLP images does not affect the classification result in Figure 4C. 

As for the suggestion on “evaluating the cluster effect at different mixed conditions”, I assume that reviewer 2 would like to see how the presence of different viral structural proteins (F, M, and G) on VLPs could affect F clustering.  We showed that the organization of NiV envelope proteins on the VLP membrane is similar in the presence or absence of NiV-M by direct visualization 27, suggesting that the effect of NiV-M on F-WT clustering on VLPs is minimal. We also show comparable incorporation of NiV-F among the NiV-F hexamer-oftrimer mutants (Fig. 4A). Therefore, we did not test the F clustering at different F, M, and G combinations in this paper. However, this could be an interesting question to pursue in a paper focusing on NiV VLP production. 

Reviewer #3 (Public Review):

Summary:

The manuscript by Wang and colleagues describes single molecule localization microscopy to quantify the distribution and organization of Nipah virus F expressed on cells and on virus-like particles. Notably the crystal structure of F indicated hexameric assemblies of F trimers. The authors propose that F clustering favors membrane fusion.

Strengths:

The manuscript provides solid data on imaging of F clustering with the main findings of:

-  F clusters are independent of expression levels

-  Proteolytic cleavage does not affect F clustering

-  Mutations that have been reported to affect the hexamer interface reduce clustering on cells and its distribution on VLPs - - F nanoclusters are stabilized by AP

Weaknesses:

The relationship between F clustering and fusion is per se interesting, but looking at F clusters on the plasma membrane does not exclude that F clustering occurs for budding. Many viral glycoproteins cluster at the plasma membrane to generate micro domains for budding. 

This does not exclude that these clusters include hexamer assemblies or clustering requires hexamer assemblies. 

We thank reviewer #3 for this question. We did not focus on the role of NiV-F clusters for budding in the current manuscript, although this is an interesting topic to pursue. In this manuscript, we observed that NiV VLP budding is decreased for some cluster-disrupting mutants, such as F-YA, and F-LI4A. however, F-V108D showed increased budding compared to F-WT (Fig. 4A). We also observed that VLPs and VSV/NiV pseudoviruses expressing L53D have little NiV-G (Fig. 4A, Fig. S4F and S4H), although the incorporation level of L53D is comparable to that of wt F in both VLPs and pseudovirions (Fig. 4A and Fig. S4F). L53D is a hypofusogenic mutant with decreased clustering ability. Therefore, our current data do not show a clear link between F clustering and NiV VLP budding or glycoprotein incorporation. 

We reported that both NiV-F and -M form clusters at the plasma membrane although NiV-F clusters are not enriched at the NiV-M positive membrane domains 1. This result indicates that NiV-M is the major driving force for assembly and budding, while NiV-F is passively incorporated into the assembly sites. The central role of NiV-M in budding is also supported by a recent study showing that NiV-M induces membrane curvature by binding to PI(4,5)P2 in the inner leaflet of the plasma membrane 28. However, the expression of NiV-F alone induces the production of vesicles bearing NiV-F 29 and NiV-F recruits vesicular trafficking and actin cytoskeleton factors to VLPs either alone or in combination with NiV-G and -M, indicating a potential autonomous role in budding 30. Additionally, several electron microscopy studies show that the paramyxovirus F forms 2D lattice interspersed above the M lattice, suggesting the participation of F in virus assembly and budding. Nonetheless, the evidence above suggests that NiV-F may play a role in budding, but our data cannot correlate NiV-F clustering to budding. 

Assuming that the clusters are important for entry, hexameric clusters are not unique to Nipah virus F. Similar hexameric clusters have been described for the HEF on influenza virus C particles (Halldorsson et al 2021) and env organization on Foamy virus particles (Effantin et al 2016), both with specific interactions between trimers. What is the organization of F on Nipah virus particles? If F requires to be hexameric for entry, this should be easily imaged by EM on infectious or inactivated virus particles. 

We thank reviewer #3 for this suggestion. The hexamer-of-trimer NiV-F is observed on the VLP surface by electron tomography 4. The NiV-F hexamer-of-trimers are arranged into a soccer ball-like structure, with one trimer being part of multiple hexamer-of-trimers. The implication of NiV-F clusters in virus entry and the potential mechanism for NiV-F higherorder structure formation are discussed in the revised manuscripts. 

AP stabilization of the F clusters is curious if the clusters are solely required for entry? Virus entry does not recruit the clathrin machinery. Is it possible that F clusters are endocytosed in the absence of budding? 

We thank reviewer #3 for this question. The evidence from the current study does not exclude the role of NiV-F clustering in virus budding. NiV-F is known to be endocytosed in the virus-producing cells for cleavage by Cathepsin B or L at endocytic compartments at a pH-dependent manner31–33 in the absence of budding. However, given that all cleaved and uncleaved NiV-F have an endocytosis signal sequence at the cytoplasmic tail and are able to interact with AP-2 for endosome assembly and the cleaved and uncleaved F may have similar clustering patterns (Fig. 2), we do not think NiV-F clustering is specifically regulated for the cleavage of NiV-F. A plausible hypothesis is that NiV-F clusters are stabilized by multiple intrinsic factors (e.g. trimer interface) and host factors (e.g. AP-2) on cell membrane for cell-cell fusion and virus budding. We linked the clustering to the fusion ability of NiV-F in this study, but the NiV-F clustering may also be important in facilitating virus budding. Once in the viruses, the higher-order assembly of the clusters (e.g. lattice) may form due to protein enrichment, and the cell factors may not be the major maintenance force. 

Clusters are required for budding. 

Other points:

Fig. 3: Some of the V108D and L53D clusters look similar in size than wt clusters. It seems that the interaction is important but not absolutely essential. Would a double mutant abrogate clustering completely?

We thank Reviewer #3 for the suggestion. We generated a double mutant of NIV-F with L53D and V108D (NiV-F-LV) and assessed its expression and processing. Although the mutant retained processing capability, it exhibited minimal surface expression, making it unfeasible to analyze its nano-organization on the cell or viral membrane.

Author response image 4.

The expression and fusion activity of Flag-tagged NiV-F and NiV-F L53D-V108D (LV). (A) Representative western blot analysis of NiV-F-WT, LV in the cell lysate of 293T cells. 293T cells were transfected by NiV-F-WT or the LV mutant. The empty vector was used as a negative control. The cell lysates were analyzed on SDS-PAGE followed by western blotting after 28hrs post-transfection. F0 and F2 were probed by the M2 monoclonal mouse antiFLAG antibody. GAPDH was probed by monoclonal mouse anti-GAPDH. (B) Representative images of 293T cell-cell fusion induced by NiV-G and NiV-F-WT or NiV-F-LV. 293T cells were co-transfected with plasmids coding for NiV-G and empty vector (NC) or NiV-F constructs. Cells were fixed at 18 hrs post-transfection. Arrows point to syncytia. Scale bar: 10um. (C) Relative cell-cell fusion levels in 293T cells in (B). Five fields per experiment were counted from three independent experiments. Data are presented as mean ± SEM. (D) The cell surface expression levels of NiV-F-WT, NiV-F-LV in 293T cells measured by flow cytometry. Mean fluorescence Intensity (MFI) values were calculated by FlowJo and normalized to that of F-WT. Data are presented as mean ± SEM of three independent experiments. Statistical significance was determined by the unpaired t-test with Welch’s correction (*P<0.05, **P<0.01, ***P<0.001, ****P<0.0001). Values were compared to that of the NiV-F-WT.

Fig. 4: The distribution of F on VLPs should be confirmed by cryoEM analyses. This would also confirm the symmetry of the clusters. The manuscript by Chernomordik et al. JBC 2004 showed that influenza HA outside the direct contact zone affects fusion, which could be further elaborated in the context of F clusters and the fusion mechanism.

We thank reviewer 3 for this suggestion. The distribution of F on VLPs was resolved by electron tomogram which showed that the NiV-F hexamer-of-trimers are arranged into a soccer ball-like structure 4. The role of influenza HA outside of the contact zone in fusion activation is an interesting phenomenon. It may address the energy transmission within and among clusters. We will pursue this topic in a future project.  

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

•  Please define all used abbreviations throughout the manuscript and in the SI.

We defined the abbreviations at their first usage. 

•  The sentence starting with "Additionally, ..." on line 155 appears to be incomplete.

We corrected this sentence.  

•  The statement starting with "As reported, ..." on line 181 should be supported by a reference.

We added a reference. 

•  In Fig. 4C, it is unclear what the x and y axes represent.  

Fig. 4C is a t-SNE plot for visualizing high-dimensional data in a low-dimensional space. It maintains the local data structure but does not represent exact quantitative relationships. In other words, points that are close together in Fig. 4C are also close in the high-dimensional space, meaning the OPTICS plots, which reflect the clustering patterns, are similar for two points that are positioned near each other in Fig. 4C. Therefore, the x and y axes do not represent the original, quantitative data, and thus the axis titles are meaningless.  

•  The reference on line 306 appears to be unformatted.

We reformatted the reference.  

Reviewer #2 (Recommendations For The Authors):

The authors need to include the overall statistics for each experiment (at least 30 to 50 cells with three independent experiments are needed). 

We highlighted the sample size (number of ROI and number of cells) used for analysis in the figure legend. The determination of the sample size is justified in Table 1 in the response letter. 

The authors need to generate a functional pseudovirus system (for example HIVpp/NiV F) to run both infectivity and fusion experiments (including Apr-BlaM assay). 

We tested viral entry using a VSV/NiV pseudovirus system and the viral entry kinetics using VLPs expressing NiV-M-β-lactamase. The results are presented in Fig. S1, S4, S6, and S7.  

Reviewer #3 (Recommendations For The Authors):

Even low resolution EM data on VLPs or viruses would strengthen the conclusions.

We thank this reviewer for the suggestion. We cited the NiV VLP images acquired by electron tomography 4, but we currently have limited resources to perform cryoEM on NiV VLPs.  

References.

(1) Liu, Q., Chen, L., Aguilar, H. C. & Chou, K. C. A stochastic assembly model for Nipah virus revealed by super-resolution microscopy. Nature Communications 9, 3050 (2018).

(2) Khetawat, D. & Broder, C. C. A Functional Henipavirus Envelope Glycoprotein Pseudotyped Lentivirus Assay System. Virology Journal 7, 312 (2010).

(3) Palomares, K. et al. Nipah Virus Envelope-Pseudotyped Lentiviruses Efficiently Target ephrinB2Positive Stem Cell Populations In Vitro and Bypass the Liver Sink When Administered In Vivo. J Virol 87, 2094–2108 (2013).

(4) Xu, K. et al. Crystal Structure of the Pre-fusion Nipah Virus Fusion Glycoprotein Reveals a Novel Hexamer-of-Trimers Assembly. PLoS Pathog 11, e1005322 (2015).

(5)    Bakker, E. & Swain, P. S. Estimating numbers of intracellular molecules through analysing fluctuations in photobleaching. Sci Rep 9, 15238 (2019).

(6) Nayak, C. R. & Rutenberg, A. D. Quantification of Fluorophore Copy Number from Intrinsic

Fluctuations during Fluorescence Photobleaching. Biophys J 101, 2284–2293 (2011).

(7) Salavessa, L. & Sauvonnet, N. Stoichiometry of ReceptorsReceptors at the Plasma MembranePlasma membrane During Their EndocytosisEndocytosis Using Total Internal Reflection Fluorescent (TIRF) MicroscopyMicroscopy Live Imaging and Single-Molecule Tracking. in Exocytosis and Endocytosis: Methods and Protocols (eds. Niedergang, F., Vitale, N. & Gasman, S.) 3–17 (Springer US, New York, NY, 2021). doi:10.1007/978-1-0716-1044-2_1.

(8) Slenders, E. et al. Confocal-based fluorescence fluctuation spectroscopy with a SPAD array detector. Light Sci Appl 10, 31 (2021).

(9) Annibale, P., Vanni, S., Scarselli, M., Rothlisberger, U. & Radenovic, A. Identification of clustering artifacts in photoactivated localization microscopy. Nat Methods 8, 527–528 (2011).

(10) Baumgart, F. et al. Varying label density allows artifact-free analysis of membrane-protein nanoclusters. Nat Methods 13, 661–664 (2016).

(11) Zanacchi, F. C. et al. A DNA origami platform for quantifying protein copy number in super-resolution. Nat Methods 14, 789–792 (2017).

(12) Jungmann, R. et al. Multiplexed 3D cellular super-resolution imaging with DNA-PAINT and Exchange-PAINT. Nature Methods 11, 313–318 (2014).

(13) Rubin-Delanchy, P. et al. Bayesian cluster identification in single-molecule localization microscopy data. Nat Methods 12, 1072–1076 (2015).

(14) Griffié, J. et al. 3D Bayesian cluster analysis of super-resolution data reveals LAT recruitment to the T cell synapse. Sci Rep 7, 4077 (2017).

(15) Dynamic Bayesian Cluster Analysis of Live-Cell Single Molecule Localization Microscopy Datasets - Griffié - 2018 - Small Methods - Wiley Online Library. https://onlinelibrary.wiley.com/doi/full/10.1002/smtd.201800008.

(16) Caetano, F. A. et al. MIiSR: Molecular Interactions in Super-Resolution Imaging Enables the Analysis of Protein Interactions, Dynamics and Formation of Multi-protein Structures. PLOS Computational Biology 11, e1004634 (2015).

(17) Malkusch, S. & Heilemann, M. Extracting quantitative information from single-molecule superresolution imaging data with LAMA – LocAlization Microscopy Analyzer. Sci Rep 6, 34486 (2016).

(18) Zhang, Y., Lara-Tejero, M., Bewersdorf, J. & Galán, J. E. Visualization and characterization of individual type III protein secretion machines in live bacteria. Proceedings of the National Academy of Sciences 114, 6098–6103 (2017).

(19) Tobin, S. J. et al. Single molecule localization microscopy coupled with touch preparation for the quantification of trastuzumab-bound HER2. Sci Rep 8, 15154 (2018).

(20) Levet, F. et al. SR-Tesseler: a method to segment and quantify localization-based super-resolution microscopy data. Nature Methods 12, 1065–1071 (2015).

(21) Peters, R., Griffié, J., Burn, G. L., Williamson, D. J. & Owen, D. M. Quantitative fibre analysis of singlemolecule localization microscopy data. Sci Rep 8, 10418 (2018).

(22) Levet, F. et al. A tessellation-based colocalization analysis approach for single-molecule localization microscopy. Nat Commun 10, (2019).

(23) Banerjee, C. et al. ULK1 forms distinct oligomeric states and nanoscopic structures during autophagy initiation. Science Advances 9, eadh4094 (2023).

(24) Pageon, S. V. et al. Functional role of T-cell receptor nanoclusters in signal initiation and antigen discrimination. Proceedings of the National Academy of Sciences 113, E5454–E5463 (2016).

(25) Cresens, C. et al. Flat clathrin lattices are linked to metastatic potential in colorectal cancer. iScience 26, 107327 (2023).

(26) Seeling, M. et al. Immunoglobulin G-dependent inhibition of inflammatory bone remodeling requires pattern recognition receptor Dectin-1. Immunity 56, 1046-1063.e7 (2023).

(27) Liu, Q. T. et al. The nanoscale organization of Nipah virus matrix protein revealed by super-resolution microscopy. Biophysical Journal 121, 2290–2296 (2022).

(28) Norris, M. J. et al. Measles and Nipah virus assembly: Specific lipid binding drives matrix polymerization. Science Advances 8, eabn1440 (2022).

(29) Patch, J. R. et al. The YPLGVG sequence of the Nipah virus matrix protein is required for budding. Virol. J. 5, 137 (2008).

(30) Johnston, G. P. et al. Nipah Virus-Like Particle Egress Is Modulated by Cytoskeletal and Vesicular Trafficking Pathways: a Validated Particle Proteomics Analysis. mSystems 4, e00194-19 (2019).

(31) Diederich, S. et al. Activation of the Nipah Virus Fusion Protein in MDCK Cells Is Mediated by Cathepsin B within the Endosome-Recycling Compartment. J Virol 86, 3736–3745 (2012).

(32) Diederich, S., Thiel, L. & Maisner, A. Role of endocytosis and cathepsin-mediated activation in Nipah virus entry. Virology 375, 391–400 (2008).

(33) Pager, C. T., Craft, W. W., Patch, J. & Dutch, R. E. A mature and fusogenic form of the Nipah virus fusion protein requires proteolytic processing by cathepsin L. Virology 346, 251–257 (2006).

Author Response

The following is the authors’ response to the original reviews.

Public Reviews:

Reviewer #1 (Public Review):

The regulation of motor autoinhibition and activation is essential for efficient intracellular transport. This manuscript used biochemical approaches to explore two members in the kinesin-3 family. They found that releasing UNC-104 autoinhibition triggered its dimerization whereas unlocking KLP-6 autoinhibition is insufficient to activate its processive movement, which suggests that KLP-6 requires additional factors for activation, highlighting the common and diverse mechanisms underlying motor activation. They also identified a coiled-coil domain crucial for the dimerization and processive movement of UNC-104. Overall, these biochemical and single-molecule assays were well performed, and their data support their statements. The manuscript is also clearly written, and these results will be valuable to the field.

Thank you very much!

Ideally, the authors can add some in vivo studies to test the physiological relevance of their in vitro findings, given that the lab is very good at worm genetic manipulations. Otherwise, the authors should speculate the in vivo phenotypes in their Discussion, including E412K mutation in UNC-104, CC2 deletion of UNC-104, D458A in KLP-6.

1. We have shown the phenotypes unc-104(E412K) mutation in C. elegans (Niwa et al., Cell Rep, 2016) and described about it in discussion (p.14 line 3-4). The mutant worm showed overactivation of the UNC-104-dependent axonal transport, which is consistent with our biochemical data showing that UNC-104(1-653)(E412K) is prone to form a dimer and more active than wild type.

2. It has been shown that L640F mutation induces a loss of function phenotype in C. elegans (Cong et al., 2021). The amount of axonal transport is reduced in unc-104(L640F) mutant worms. L640 is located within the CC2 domain. To show the importance of CC2-dependent dimerization in the axonal transport in vivo, we biochemically investigated the impact of L640F mutation.

By introducing L640F into UNC-104(1-653)(E412K), we performed SEC analysis. The result shows that UNC-104(1-653)(E412K,L640F) failed to form stable dimers despite the release of their autoinhibition (new Figure S8). This result strongly suggests the importance of the CC2 domain in the axonal transport in vivo. Based on the result, we discussed it in the revised manuscript (p.13 line 6-8).

1. Regarding KLP-6(D458A), we need a genetic analysis using genome editing and we would like to reserve it for a future study. We speculate that the D458A mutation could lead to an increase in transport activity in vivo similar to unc-104(E412K). This is because the previous study have shown that wild-type KLP-6 was largely localized in the cell body, while KLP-6(D458A) was enriched at the cell periphery in the N2A cells (Wang et al., 2022). We described it in discussion (p.14 line 13-14).

While beyond the scope of this study, can the author speculate on the candidate for an additional regulator to activate KLP-6 in C. elegans?

The heterodimeric mechanoreceptor complex, comprising LOV-1 and PKD-2, stands as potential candidates for regulating KLP-6 dimerization. We speculate the heterodimerization property is suitable for the enhancement of KLP-6 dimerization. On the other hand, it's noteworthy that KLP-6 can undergo activation in Neuro 2a cells upon the release of autoinhibition (Wang et al., 2022). This observation implies the involvement of additional factors which are not present in sf9 cells may be able to induce dimerization. Post-translational modifications would be one of the candidates. We discussed it in p14 line 7-14.

The authors discussed the differences between their porcine brain MTs and chlamydonomas axonemes in UNC-104 assays. However, the authors did not really retest UNC-104 on axonemes after more than two decades, thereby not excluding other possibilities.

We thought that comparing different conditions used in different studies is essential for the advancement of the field of molecular motors. Therefore, we newly performed single-molecule assay using Chlamydomonas axonemes and compared the results with brain MTs (Fig. S6). Just as observed in the study by Tomoshige et al., we were also unable to observe the processive runs of UNC-104(1-653) on Chlamydomonas axonemes (Fig. S6A). Furthermore, we found that the landing rate of UNC-104(1-653) on Chlamydomonas axonemes was markedly lower in comparison to that on purified porcine microtubules (Fig. S6B).

Reviewer #1 (Recommendations For The Authors):

More discussion as suggested above would improve the manuscript.

We have improved our manuscript as described above.

Reviewer #2 (Public Review):

The Kinesin superfamily motors mediate the transport of a wide variety of cargos which are crucial for cells to develop into unique shapes and polarities. Kinesin-3 subfamily motors are among the most conserved and critical classes of kinesin motors which were shown to be self-inhibited in a monomeric state and dimerized to activate motility along microtubules. Recent studies have shown that different members of this family are uniquely activated to undergo a transition from monomers to dimers.

Niwa and colleagues study two well-described members of the kinesin-3 superfamily, unc104 and KLP6, to uncover the mechanism of monomer to dimer transition upon activation. Their studies reveal that although both Unc104 and KLP6 are both self-inhibited monomers, their propensities for forming dimers are quite different. The authors relate this difference to a region in the molecules called CC2 which has a higher propensity for forming homodimers. Unc104 readily forms homodimers if its self-inhibited state is disabled while KLP6 does not.

The work suggests that although mechanisms for self-inhibited monomeric states are similar, variations in the kinesin-3 dimerization may present a unique form of kinesin-3 motor regulation with implications on the forms of motility functions carried out by these unique kinesin-3 motors.

Thank you very much!

Reviewer #2 (Recommendations For The Authors):

The work is interesting but the process of making constructs and following the transition from monomers to dimers seems to be less than logical and haphazard. Recent crystallographic studies for kinesin-3 have shown the fold and interactions for all domains of the motor leading to the self-inhibited state. The mutations described in the manuscript leading to disabling of the monomeric self-inhibited state are referenced but not logically explained in relation to the structures. Many of the deletion constructs could also present other defects that are not presented in the mutations. The above issues prevent wide audience access to understanding the studies carried out by the authors.

We appreciate this comment. We improved it as described bellow.

Suggestions: Authors should present schematic, or structural models for the self-inhibited and dimerized states. The conclusions of the papers should be related to those models. The mutations should be explained with regard to these models and that would allow the readers easier access. Improving access to the readers in and outside the motor field would truly improve the impact of the manuscript on the field.

The structural models illustrating the autoinhibited state have been included in new Figure S4, accompanied by an explanation of the correlation between the mutations and these structures in the figure legend. Additionally, schematic models outlining the dimerization process of both UNC-104 and KLP-6 have been provided in Figure S9 to enhance reader comprehension of the process.

Reviewer #3 (Public Review):

In this work, Kita et al., aim to understand the activation mechanisms of the kinesin-3 motors KLP-6 and UNC-104 from C. elegans. As with many other motor proteins involved in intracellular transport processes, KLP-6 and UNC-104 motors suppress their ATPase activities in the absence of cargo molecules. Relieving the autoinhibition is thus a crucial step that initiates the directional transport of intracellular cargo. To investigate the activation mechanisms, the authors make use of mass photometry to determine the oligomeric states of the full-length KLP-6 and the truncated UNC-104(1-653) motors at sub-micromolar concentrations. While full-length KLP-6 remains monomeric, the truncated UNC-104(1-653) displays a sub-population of dimeric motors that is much more pronounced at high concentrations, suggesting a monomer-to-dimer conversion. The authors push this equilibrium towards dimeric UNC-104(1-653) motors solely by introducing a point mutation into the coiled-coil domain and ultimately unleashing a robust processivity of the UNC-104 dimer. The authors find that the same mechanistic concept does not apply to the KLP-6 kinesin-3 motor, suggesting an alternative activation mechanism of the KLP-6 that remains to be resolved. The present study encourages further dissection of the kinesin-3 motors with the goal of uncovering the main factors needed to overcome the 'self-inflicted' deactivation.

Thank you very much!

Reviewer #3 (Recommendations For The Authors):

126-128: It is surprising that surface-attachment does not really activate the full-length KLP6 motor (v=48 {plus minus} 42 nm/s). Can the authors provide an example movie of the gliding assay for the FL KLP6 construct? Gliding assays are done by attaching motors via their sfGFP to the surface using anti-GFP antibodies. Did the authors try to attach the full-length KLP-6 motor directly to the surface? If the KLP-6 motor sticks to the surface via its (inhibitory) C-terminus, this attachment would be expected to activate the motor in the gliding assay, ideally approaching the in vivo velocities of the activated motor.

We have included an example kymograph showing the gliding assay of KLP-6FL (Fig. S1A). When we directly attached KLP-6FL to the surface, the velocity was 0.15 ± 0.02 µm/sec (Fig. S1B), which is similar to the velocity of KLP-6(1-390). While the velocity observed in the direct-attachment condition is much better than those observed in GFP-mediated condition, the observed velocity remains considerably slower than in vivo velocities. Firstly, we think this is because dimerization of KLP-6 is not induced by the surface attachment. Previous studies have shown that monomeric proteins are generally slower than dimeric proteins in the gliding assay (Tomishige et al., 2002). These are consistent with our observation that KLP-6 remains to be monomeric even when autoinhibition is released. Secondly, in vitro velocity of motors is generally slower than in vivo velocity.

156-157: It seems that the GCN4-mediated dimerization induces aggregation of the KLP6 motor domains as seen in the fractions under the void volume in Figure 3B (not seen with the Sf9 expressed full-length constructs, see Figure 1B). Also, the artificially dimerized motor construct does not fully recapitulate the in vivo velocity of UNC-104. Did the authors analyze the KLP-6(1-390)LZ with mass photometry and is it the only construct that is expressed in E. coli?

KLP-6::LZ protein is not aggregating. We have noticed that DNA and RNA from E. coli exists in the void fraction and they occasionally trap recombinant kinesin-3 proteins in the void fraction. To effectively remove these nucleic acids from our protein samples, we employed streptomycin sulfate as a purification method (Liang et al., Electrophoresis, 2009). Please see Purification of recombinant proteins in Methods. In the size exclusion chromatography analysis, we observed that KLP-6(1-393)LZ predominantly eluted in the dimer fraction (New Figure 3). Subsequently, we reanalyzed the motor's motility using a total internal reflection fluorescence (TIRF) assay, as shown in the revised Figure 3. Even after these efforts, the velocity was not changed significantly. The velocity of KLP-6LZ is about 0.3 µm/sec while that of cellular KLP-6::GFP is 0.7 µm/sec (Morsci and Barr, 2011). Similar phenomena, "slower velocity in vitro", has been observed in other motor proteins.

169: In Wang et al., (2022) the microtubule-activated ATPase activities of the mutants were measured in vitro as well, with the relative activities of the motor domain and the D458A mutant being very similar. The D458A mutation is introduced into the full-length motor in Wang et al., while in the present work, the mutation is introduced into the truncated KLP-6(1-587) construct. Can the authors explain their reasoning for the latter?

(1) Kinesins are microtubule-stimulated ATPases. i.e. The ATPase activity is induced by the binding with a microtubule.

(2) Previous studies have shown that the one-dimensional movement of the monomeric motor domain of kinesin-3 depends on the ATPase activity even when the movement does not show clear plus-end directionality (Okada et al., Science, 1998).

(3) While KLP-6(1-587) does not bind to microtubules, both KLP-6(1-390) (= the monomeric motor domain) and KLP-6(1-587)(D458A) similarly bind to microtubules and show one dimensional diffusion on microtubules (Fig. 4E and S2B).

Therefore, the similar ATPase activities of the motor domain(= KLP-6(1-390)) and KLP-6(D458A) observed by Wang et al. is because both proteins similarly associate with and hydrolyze ATP on microtubules, which is consistent with our observation. On the other hand, because KLP-6(wild type) cannot efficiently bind to microtubules, the ATPase activity is low.

Can the authors compare the gliding velocities of the KLP-6(1-390)LZ vs KLP-6(1-587) vs KLP-6(1-587)(D458A) constructs to make sure that the motors are similarly active?

We conducted a comparative analysis of gliding velocities involving KLP-6(1-390), KLP-6(1-587), and KLP-6(1-587)(D458A) (Fig. S1C). We used KLP-6(1-390) instead of KLP-6(1-390)LZ, aligning with the protein used by Wang et al.. We demonstrated that both KLP-6(1-587) and KLP-6(1-587) (D458A) exhibited activity levels comparable to that of KLP-6(1-390). The data suggests that the motor of all recombinant proteins are similarly active.

Please note that, unlike full length condition (Fig. 1D and S1A and S1B), the attachment to the surface using the anti-GFP antibody can activates KLP-6(1-587). The data suggests that, due to the absence of coverage by the MBS and MATH domain (Wang et al., Nat. Commun., 2022), the motor domain of KLP-6(1-587) to some extent permits direct binding to microtubules under gliding assay conditions.

Are the monomeric and dimeric UNC-104(1-653) fractions in Figure 5B in equilibrium? Did the authors do a re-run of the second peak of UNC-104(1-653) (i.e. the monomeric fraction with ~100 kDa) to assess if the monomeric fraction re-equilibrates into a dimer-monomer distribution?

We conducted a re-run of the second peak of UNC-104(1-653) and verified its re-equilibration into a distribution of dimers and monomers after being incubated for 72 hours at 4°C (Fig. S5).

UNC-104 appears to have another predicted coiled-coiled region around ~800 aa (e.g. by NCoils) that would correspond to the CC3 in the mammalian homolog KIF1A. This raises the question if the elongated UNC-104(1-800) would dimerize more efficiently than UNC-104(1-653) (authors highlight the sub-population of dimerized UNC-104(1-653) at low concentrations in Figure 5C) and if this dimerization alone would suffice to 'match' the UNC-104(1-653)E412K mutant (Figure 5D). Did the authors explore this possibility? This would mean that dimerization does not necessarily require the release of autoinhibition.

We have tried to purify UNC-104(1-800) and full-length UNC-104 using the baculovirus system. However, unfortunately, the expression level of UNC-104(1-800) and full length UNC-104 was too low to perform in vitro assays even though codon optimized vectors were used. Instead, we have analyzed full-length human KIF1A. We found that full-length KIF1A is mostly monomeric, not dimeric (Please look at the Author response image 1). The property is similar to UNC-104(1-653) (Figure 5A-C). Therefore, we think CC3 does not strongly affect dimerization of KIF1A, and probably its ortholog UNC-104. Moreover, a recent study has shown that CC2 domain, but not other CC domains, form a stable dimer in the case of KIF1A (Hummel and Hoogenraad, JCB, 2021). Given the similarity in the sequence of KIF1A and UNC-104, we anticipate that the CC2 domain of UNC-104 significantly contributes to dimerization, potentially more than other CC domains. We explicitly describe it in the Discussion in the revised manuscript.

Author response image 1.

Upper left, A representative result of size exclusion chromatography obtained from the analysis of full-length human KIF1A fused with sfGFP. Upper right, A schematic drawing showing the structure of KIF1A fused with sfGFP and a result of SDS-PAGE recovered from SEC analysis. Presumable dimer and monomer peaks are indicated. Lower left, Presumable dimer fractions in SEC were collected and analyzed by mass photometry. The result confirms that the fraction contains considerable amount of dimer KIF1A. Lower right, Presumable monomer fractions were collected and analyzed by mass photometry. The result confirms that the fraction mainly consists of monomer KIF1A. Note that these results obtained from full-length KIF1A protein are similar to those of UNC-104(1-653) protein shown in Figure 5A-C.

Author response:

The following is the authors’ response to the original reviews.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

In this study, Bonnifet et al. profile the presence of L1 ORF1p in the mouse and human brain. They claim that ORF1p is expressed in the human and mouse brain at a steady state and that there is an age-dependent increase in expression. This is a timely report as two recent papers have extensively documented the presence of full-length L1 transcripts in the mouse and human brain (PMID: 38773348 & PMID: 37910626). Thus, the finding that L1 ORF1p is consistently expressed in the brain is not surprising, but important to document.

Thank you for recognizing the importance of this study. The two cited papers have indeed reported the presence of full-length transcripts in the mouse and human brain. However, the first (PMID: 38773348) report has shown evidence of full-length LINE-1 RNA and ORF1 protein expression in the mouse hippocampus (but not elsewhere) and the second (PMID: 37910626) shows full-length LINE-1 RNA expression and H3K4me3-ChIP data in the frontal and temporal lobe of the human brain, but not protein expression.

Strengths:

Several parts of this manuscript appear to be well done and include the necessary controls. In particular, the evidence for steady-state expression of ORF1p in the mouse brain appears robust.

Weaknesses:

Several parts of the manuscript appear to be more preliminary and need further experiments to validate their claims. In particular, the data suggesting expression of L1 ORF1p in the human brain and the data suggesting increased expression in the aged brain need further validation. Detailed comments:

(1) The expression of ORF1p in the human brain shown in Figure 1j is not convincing. Why are there two strong bands in the WB? How can the authors be sure that this signal represents ORF1p expression and not nonspecific labelling? Additional validations and controls are needed to verify the specificity of this signal.

We have validated the antibody against human ORF1p (Abcam 245249-> https://www.abcam.com/enus/products/primary-antibodies/line-1-orf1p-antibody-epr22227-6-ab245249), which we use for Western blotting experiments (please see Fig1J and new Suppl Fig.2A,B and C), by several means.

(1) We have done immunoprecipitations and co-immunoprecipitations followed by quantitative mass spectrometry (LC-MS/MS; data not shown as they are part of a different study). We efficiently detect ORF1p in IPs (Western blot now added in Suppl Fig2B) and by quantitative mass spectrometry (5 independent samples per IP-ORF1p and IP-IgG: ORF1p/IgG ratio: 40.86; adj p-value 8.7e-07; human neurons in culture; data not shown as they are part of a different study). We also did co-IPs followed by Western blot using two different antibodies, either the Millipore clone 4H1 (https://www.merckmillipore.com/CH/en/product/Anti-LINE-1-ORF1p-Antibody-clone-4H1,MM_NF-MABC1152?ReferrerURL=https%3A%2F%2Fwww.google.com%2F) or the Abcam antibody to immunoprecipitate and the Abcam antibody for Western blotting on human brain samples. Indeed, the Millipore antibody does not work well on Western Blots in our hands. We consistently revealed a double band indicating that both bands are ORF1p-derived. We have added an ORF1p IP-Western blot as Suppl Fig. 2B which clearly shows the immunoprecipitation of both bands by the Abcam antibody. Abcam also reports a double band, and they suspect that the lower band is a truncated form (see the link to their website above). ORF1p Western blots done by other labs with different antibodies have detected a second band in human samples

• Sato, S. et al. LINE-1 ORF1p as a candidate biomarker in high grade serous ovarian carcinoma. Sci Rep 13, 1537 (2023) in Figure 1D

• McKerrow, W. et al. LINE-1 expression in cancer correlates with p53 mutation, copy number alteration, and S phase checkpoint. Proc. Natl. Acad. Sci. U.S.A. 119, e2115999119 (2022)) showing a Western blot of an inducible LINE-1 (ORFeus) detected by the MABC1152 ORF1p antibody from Millipore Sigma in Figure 7 - Walter et al. eLife 2016;5:e11418. (DOI: 10.7554/eLife.11418) in mouse ES cells with an antibody made inhouse (gift from another lab; in Figure 2B)

The lower band might thus be a truncated form of ORF1p or a degradation product which appears to be shared by mouse and human ORF1p. We have now mentioned this in the revised version of the paper (lines 183-189).

(2) We have used the very well characterized antibody from Millipore ((https://www.merckmillipore.com/CH/en/product/Anti-LINE-1-ORF1p-Antibody-clone-4H1,MM_NFMABC1152?ReferrerURL=https%3A%2F%2Fwww.google.com%2F)) for immunostainings and detect ORF1p staining in human neurons in the very same brain regions (Fig 2H, new Suppl Fig. 2E) including the cerebellum in the human brain. We added a 2nd antibody-only control (Suppl Fig. 2E).

(3) We also did antibody validation by siRNA knock-down. However, it is important to note, that these experiments were done in LUHMES cells, a neuronal cell line which we differentiated into human dopaminergic neurons. In these cells, we only occasionally detect a double band on Western blots, but mostly only reveal the upper band at ≈ 40kD. The results of the knockdown are now added as Suppl Fig. 2C.

Altogether, based on our experimental validations and evidence from the literature, we are very confident that it is indeed ORF1p that we detect on the blots and by immmunostainings in the human brain.

(2) The data shown in Figure 2g are not convincing. How can the authors be sure that this signal controls are needed to verify the specificity of this signal. represents ORF1p expression and not non-specific labelling? Extensive additional validations and

In line 117-123 of the manuscript, we had specified “Importantly, the specificity of the ORF1p antibody, a widely used, commercially available antibody [18,34–38], was confirmed by blocking the ORF1p antibody with purified mouse ORF1p protein resulting in the complete absence of immunofluorescence staining (Suppl Fig. 1A), by using an inhouse antibody against mouse ORF1p[17] which colocalized with the anti-ORF1p antibody used (Suppl Fig. 1B, quantified in Suppl Fig. 1C), and by immunoprecipitation and mass spectrometry used in this study (see Author response image 1)”.

Figure 2G shows a Western blot using an extensively used and well characterized ORF1p antibody from abcam (mouse ORF1p, Rabbit Recombinant Monoclonal LINE-1 ORF1p antibody-> (https://www.abcam.com/enus/products/primary-antibodies/line-1-orf1p-antibody-epr21844-108-ab216324; cited in at least 11 publications) after FACS-sorting of neurons (NeuN+) of the mouse brain. We have validated this ORF1p antibody ourselves in IPs (please see Fig 6A) and co-IP followed by mass spectrometry (LC/MS-MS; see Fig 6, where we detect ORF1p exclusively in the 5 independent ORF1p-IP samples and not at all in 5 independent IgG-IP control samples, please also see Suppl Table 2). In this analysis, we detect ORF1p with a ratio and log2fold of ∞ , indicating that this proteins only found in IP-ORF1p samples (5/5) and not in the IP-control samples ((not allowing for the calculation of a ratio with p-value), please see Suppl Table 2)

Author response image 1.

In addition, we have added new data showing the entire membrane of the Western blot in Fig1H (now Suppl Fig.1E) and a knock-down experiment using siRNA against ORF1p or control siRNA in mouse dopaminergic neurons in culture (MN9D; new Suppl Fig.1D). This together makes us very confident that we are looking at a specific ORF1p signal. The band in Figure 2G is at the same height as the input and there are no other bands visible (except the heavy chain of the NeuN antibody, which at the same time is a control for the sorting). We added some explanatory text to the revised version of the manuscript in lines 120-124 and lines 253-256).

Please note that in the IP of ORF1p shown in Fig6A, there is a double band as well, strongly suggesting that the lower band might be a truncated or processed form of ORF1p. As stated above, this double band has been detected in other studies (Walter et al. eLife 2016;5:e11418. DOI: 10.7554/eLife.11418) in mouse ES cells using an in-house generated antibody against mouse ORF1p. Thus, with either commercial or in-house generated antibodies in some mouse and human samples, there is a double band corresponding to full-length ORF1p and a truncated or processed version of it.

We noticed that we have not added the references of the primary antibodies used in Western blot experiments in the manuscript, which was now corrected in the revised version.

(3) The data showing a reduction in ORF1p expression in the aged mouse brain is confusing and maybe even misleading. Although there is an increase in the intensity of the ORF1p signal in ORF1p+ cells, the data clearly shows that fewer cells express ORF1p in the aged brain. If these changes indicate an overall loss or gain of ORF1p, expression in the aged brain is not resolved. Thus, conclusions should be more carefully phrased in this section. It is important to show the quantification of NeuN+ and NeuN- cells in young vs aged (not only the proportions as shown in Figure 3b) to determine if the difference in the number of ORF1p+ cells is due to loss of neurons or perhaps a sampling issue. More so, it would be essential to perform WB and/or proteomics experiments to complement the IHC data for the aged mouse samples.

We thank the reviewer for this comment and we agree that the representation has been confusing, which is why we added data to Suppl Fig.5 (F-K) using a different representation. As suggested by the reviewer, in new Suppl Fig. 5F-K, we now show the number of ORF1p+, NeuN+ or NeuN- cells per mm2. These graphs indicate that the number per mm2 of ORF1p+ cells overall do not decrease significantly (with the dorsal striatum as an exception, but possibly due to technical limitations which we now discuss in the results section, line 332-335). Globally, there is thus no loss of ORF1p+ expressing cells. There is also no global nor region-specific decrease in the number of neuronal cells (NeuN+ per mm2) although proportions change (Suppl Fig 2E, confocal acquisitions), thus most likely due to a gain of non-neuronal cells in this region. Concerning Western blots on mouse brain tissues from young and aged individuals, we unfortunately ran into limits regarding tissue availability of aged mice.

(4) The transcriptomic data presented in Figure 4 and Figure 5 are not convincing. Quantification of transposon expression on short read sequencing has important limitations. Longer reads and complementary approaches are needed to study the expression of evolutionarily young L1s (see PMID: 38773348 & PMID: 37910626 for examples of the current state of the art). Given the read length and the unstranded sequencing approach, I would at least ask the authors to add genome browser tracks of the upregulated loci so that we can properly assess the clarity of the results. I would also suggest adding the mappability profile of the elements in question. In addition, since this manuscript focuses on ORF1p, it would be essential to document changes in protein levels (and not just transcripts) in the ageing human brain.

We agree that there are limitations to the analysis of TEs with short read sequencing and we have added more text on this aspect in the revised version (results section) and highlighted the problem of limited and disequilibrated sample size in the discussion (line 638-644). The approaches shown in PMID: 38773348 & PMID: 37910626 or even a combination of them, would be ideal of course. However, here we re-analyzed a unique preexisting dataset (Dong et al, Nature Neuroscience, 2018; http://dx.doi.org/10.1038/s41593-018-0223-0), which contains RNA-seq data of human post-mortem dopaminergic neurons in a relatively high number of brain-healthy individuals of a wide age range including some “young” individuals which is rare in post-mortem studies. Such data is unfortunately not available with long read sequencing or any other more appropriate approach yet. Limitations are evident, but all limitations will apply equally to both groups of individuals that we compare. The general mappability profile of the full-length LINE-1 “UIDs” was shown in old Suppl Fig 6A. We have colorhighlighted now in new Suppl Fig 8C the specific elements in this graph. Most importantly, we have now used, as a condensate of suggestions by all reviewers, a combination of mappability score, post-hoc power calculation, visualization and correlation with adjacent gene expression in order to retain a specific locus with confidence or not. Using these criteria, we retained UID-68 (Fig 5D) which has a relatively high mappability score (Suppl Fig.8C) plus an overlap of umap 50 mappability peaks and read mapping when visualizing the locus in IGV (new Fig. 5E), very high post-hoc power (96.6%; continuous endpoint, two independent samples, alpha 0.05) and no correlation with adjacent gene expression per individual (Fig. 5F, G). Based on these criteria, we had to exclude UID-129, UID-37, UID-127 and UID-137, reinforcing the notion that a combination of quality control criteria might be crucial to retain a specific locus with confidence. This is now mentioned in the manuscript in the discussion in line 427430).

We will not be able to document changes in protein levels in aged human dopaminergic neurons as we do not have access to this material. We have tried to obtain human substantia nigra tissues but were not able to get sufficient amounts to do laser-capture microdissection or FACS analyses, especially of young individuals. There are still important limitations to tissue availability, especially of young individuals, and even more so of specific regions of interest like the substantia nigra pars compacta affected in Parkinson disease.

(5) More information is needed on RNAseq of microdissections of dopaminergic neurons from 'healthy' postmortem samples of different ages. No further information on these samples is provided. I would suggest adding a table with the clinical information of these samples (especially age, sex, and cause of death). The authors should also discuss whether this experiment has sufficient power. The human ageing cohort seems very small to me.

This is a re-analysis of a published dataset (Dong et al, Nat Neurosci, 2018; doi:10.1038/s41593-018-0223-0), available through dbgap (phs001556.v1.p1). In this original article, the criteria for inclusion as a brain-healthy control were as follows:

“…Subjects… were without clinicopathological diagnosis of a neurodegenerative disease meeting the following stringent inclusion and exclusion criteria. Inclusion criteria: (i) absence of clinical or neuropathological diagnosis of a neurodegenerative disease, for example, PD according to the UKPDBB criteria[47], Alzheimer’s disease according to NIA-Reagan criteria[48], or dementia with Lewy bodies by revised consensus criteria[49]; for the purpose of this analysis incidental Lewy body cases (not meeting clinicopathological diagnostic criteria for PD or other neurodegenerative disease) were accepted for inclusion; (ii) PMI ≤ 48 h; (iii) RIN[50] ≥ 6.0 by Agilent Bioanalyzer (good RNA integrity); and (iv) visible ribosomal peaks on the electropherogram. Exclusion criteria were: (i) a primary intracerebral event as the cause of death; (2) brain tumor (except incidental meningiomas); (3) systemic disorders likely to cause chronic brain damage.”

We do not have access to the cause of death, but we have added available metadata as Suppl_Table 5 to the manuscript.

We have performed a post-hoc power analysis (using the “Post-hoc Power Calculator” https://clincalc.com/stats/Power.aspx, which evaluates the statistical power of an existing study and added the results to the revision. Due to this analysis, we have indeed taken out Suppl Fig 7 as a whole which had shown data of three full-length LINE-1 loci (UID-37, UID-127 and UID-137) with low power (between 17-66% power). The locus shown in Fig. 5D of the UID-68) had a post-hoc power score of 96.6% which increases our confidence in this full-length LINE-1 element being upregulated in aged dopaminergic neurons. UID-129 had a post-hoc power score of 97%. However, visualization and mappability analysis of the UID-129 locus led us to exclude this UID.

The post-hoc power analysis for L1HS and L1PA2 revealed a low power (28.4% and 32.8% respectively). We have added these results to the manuscript (line 359-362), but decided to keep the data in as this will hopefully be a motivation for future confirmation studies knowing that the availability of similar data from brain-healthy human dopaminergic neurons especially of young individuals will be low.

(6) The findings in this manuscript apply to both human and mouse brains. However, the landscape of the evolutionarily young L1 subfamilies between these two species is very different and should be part of the discussion. For example, the regulatory sequences that drive L1 expression are quite different in human and mouse L1s. This should be discussed.

Indeed, they are different. We have added a paragraph to the discussion (lines 539-548).

(7) On page 3 the authors write: "generally accepted that TE activation can be both, a cause and consequence of aging". This statement does not reflect the current state of the field. On the contrary, this is still an area of extensive investigation and many of the findings supporting this hypothesis need to be confirmed in independent studies. This statement should be revised to reflect this reality.

We agree, this is overstated, we have changed this sentence accordingly to:

“It is now, 31 years after the initial proposition of the “transposon theory of aging” by Driver and McKechnie [14], still a matter of debate whether TE activation can be both, a cause and a consequence of aging [15,16].”

Reviewer #2 (Public Review):

Summary:

Bonnifet et al. sought to characterize the expression pattern of L1 ORF1p expression across the entire mouse brain, in young and aged animals, and to corroborate their characterization with Western blotting for L1 ORF1p and L1 RNA expression data from human samples. They also queried L1 ORF1p interacting partners in the mouse brain by IP-MS.

Strengths:

A major strength of the study is the use of two approaches: a deep-learning detection method to distinguish neuronal vs. non-neuronal cells and ORF1p+ cells vs. ORF1p- cells across large-scale images encompassing multiple brain regions mapped by comparison to the Allen Brain Atlas, and confocal imaging to give higher resolution on specific brain regions. These results are also corroborated by Western blotting on six mouse brain regions. Extension of their analysis to post-mortem human samples, to the extent possible, is another strength of the paper. The identification of novel ORF1p interactors in the brain is also a strength in that it provides a novel dataset for future studies.

Thank you for highlighting the strength of our study.

Weaknesses:

The main weakness of the study is that cell type specificity of ORF1p expression was not examined beyond neuron (NeuN+) vs non-neuron (NeuN-). Indeed, a recent study (Bodea et al. 2024, Nature Neuroscience) found that ORF1p expression is characteristic of parvalbumin-positive interneurons, and it would be very interesting to query whether other neuronal subtypes in different brain regions are distinguished by ORF1p expression.

We agree that this point is important to address. We have mentioned in the manuscript our previous work, which showed that in the mouse ventral midbrain, dopaminergic neurons (TH+/NeuN+) express ORF1p and that these neurons express higher levels of ORF1p than adjacent non-dopaminergic neurons (TH-/NeuN+; Blaudin de Thé et al, EMBO J, 2018). Others have shown evidence of full-length L1 RNA expression in both excitatory and inhibitory neurons but much less expression in non-neuronal cells (Garza et al, SciAdv, 2023). Further, ORF1p expression was documented in excitatory (CamKIIa-positive) and CamKIIa-negative neurons in the mouse frontal cortex (Zhang et al, Cell Res, 2022, doi.org/10.1038/s41422-022-00719-6). We do detect ORF1p staining in mouse (Fig. 1B, panel 10) and human Purkinje cells (based on morphology and in accordance with data from Takahashi et al, Neuron, 2022; DOI: 10.1016/j.neuron.2022.08.011) and most probably basket cells (based on anatomical location in the molecular layer near Purkinje cells) of the cerebellum (Suppl Fig.4). Some Purkinje cells express PV in mice (https://doi.org/10.1016/j.mcn.2021.103650 and 10.1523/JNEUROSCI.22-1607055.2002), as do stellate and basket cells of the molecular layer (10.1523/JNEUROSCI.22-16-07055.2002). While ORF1p is expressed in PV cells of the hippocampus (Bodea et al, Nat Neurosci, 2024) and in the human and mouse cerebellum in PV-expressing neurons, it does not seem as if ORF1p expression is restricted to PV cells overall. To adress this question experimentally, we have now performed ORF1p stainings in different brain regions (hippocampus, cortex, hindbrain, thalamus, ventral midbrain and cerebellum) together with parvalbumin (PV) stainings and in some cases including the lectin WFA (Wisteria floribunda agglutinin, which specifically stains glycoproteins surrounding PV+ neurons). We have added this data to the manuscript as Suppl Fig.4. While PV-positive neurons often co-stain with ORF1p, not all ORF1p positive cells are PV-positive. We have also deepened the discussion of this aspect in the revised manuscript (line 579-599).

The data suggesting that ORF1p expression is increased in aged mouse brains is intriguing, although it seems to be based upon modestly (up to 27%, dependent on brain region) higher intensity of ORF1p staining rather than a higher proportion of ORF1+ neurons. Indeed, the proportion of NeuN+/Orf1p+ cells actually decreased in aged animals. It is difficult to interpret the significance and validity of the increase in intensity, as Hoechst staining of DNA, rather than immunostaining for a protein known to be stably expressed in young and aged neurons, was used as a control for staining intensity.

We have now separated the analysis of NeuN+, ORF1p+ and NeuN- cells (please see new Suppl Fig5F-K) which highlights the fact that there is indeed no change in the number of ORF1p+ cells in the young compared to the aged mouse brain. However, while neuronal cell numbers throughout the brain do not change significantly (new Suppl Fig.5F), while cell proportions in the ventral midbrain (confocal microscopy based quantifications) change, possibly due to a combination of a slight loss in neurons and a gain in non-neuronal cell numbers (Suppl Fig3E). Please also keep in mind that the ventral midbrain region on images taken on a confocal microscope are a much smaller region than the midbrain motor region as specified by ABBA on images taken by the slide scanner. A different marker than DNA as a control requires the use of a protein that is stably expressed throughout the brain and throughout age. We are not aware of a protein for which this has been established. To nevertheless try to address this issue, we used whole-brain imaging intensity data for the protein Rbfox3 (NeuN) which we originally used as a marker for cell identity. We have now added the quantifications of the protein Rbfox3 (NeuN) to Fig3 (new Fig3B). As shown in this figure, NeuN intensity is not stable from one individual to another, neither in control mice nor in the aged control group. Most importantly, NeuN staining intensity does not increase in aged mice. As we did not use NeuN intensity but presence or absence of NeuN as a marker for cell identity, the instability of NeuN intensity from one individual mouse to another does not have an influence on the data presented in this manuscript. It does indicate however, that the overall increase of ORF1p in aged mice is not a mere reflection of a general decrease in protein turnover. As stated above, the DNA staining with Hoechst controls for technical artefacts. Using Hoechst and NeuN as control, we have thus provided evidence for the fact that the increase in ORF1p intensity per cell is indeed specific for ORF1p. This is now added to the results section (line 299-301).

The main weakness of the IP-MS portion of the study is that none of the interactors were individually validated or subjected to follow-up analyses. The list of interactors was compared to previously published datasets, but not to ORF1p interactors in any other mouse tissue.

As stated in the manuscript, the list of previously published datasets does include a mouse dataset with ORF1p interacting proteins in mouse spermatocytes (please see line 479-480: “ORF1p interactors found in mouse spermatocytes were also present in our analysis including CNOT10, CNOT11, PRKRA and FXR2 among others (Suppl_Table4).”) -> De Luca, C., Gupta, A. & Bortvin, A. Retrotransposon LINE-1 bodies in the cytoplasm of piRNA-deficient mouse spermatocytes: Ribonucleoproteins overcoming the integrated stress response. PLoS Genet 19, e1010797 (2023)). We indeed did not validate any interactors for several reasons (economic reasons and time constraints (post-doc leaving)). However, we feel that the significant overlap with previously published interactors highlights the validity of our data and we anticipate that this list of ORF1p protein interactors in the mouse brain will be of further use for the community.

The authors achieved the goals of broadly characterizing ORF1p expression across different regions of the mouse brain, and identifying putative ORF1p interactors in the mouse brain. However, findings from both parts of the study are somewhat superficial in depth.

This provides a useful dataset to the field, which likely will be used to justify and support numerous future studies into L1 activity in the aging mammalian brain and in neurodegenerative disease. Similarly, the list of ORF1p interacting proteins in the brain will likely be taken up and studied in greater depth.

Reviewer #3 (Public Review):

The question about whether L1 exhibits normal/homeostatic expression in the brain (and in general) is interesting and important. L1 is thought to be repressed in most somatic cells (with the exception of some stem/progenitor compartments). However, to our knowledge, this has not been authoritatively / systematically examined and the literature is still developing with respect to this topic. The full gamut of biological and pathobiological roles of L1 remains to be shown and elucidated and this area has garnered rapidly increasing interest, year-by-year. With respect to the brain, L1 (and repeat sequences in general) have been linked with neurodegeneration, and this is thought to be an aging-related consequence or contributor (or both) of inflammation. This study provides an impressive and apparently comprehensive imaging analysis of differential L1 ORF1p expression in mouse brain (with some supporting analysis of the human brain), compatible with a narrative of non-pathological expression of retrotransposition-competent L1 sequences. We believe this will encourage and support further research into the functional roles of L1 in normal brain function and how this may give way to pathological consequences in concert with aging. However, we have concerns with conclusions drawn, in some cases regardless of the lack of statistical support from the data. We note a lack of clarity about how the 3rd party pre-trained machine learning models perform on the authors' imaging data (validation/monitoring tests are not reported), as well as issues (among others) with the particular implementation of co-immunoprecipitation (ORF1p is not among the highly enriched proteins and apparently does not reach statistical significance for the comparison) - neither of which may be sufficiently rigorous.

Thank you for your comments on our manuscript.

We have addressed the concerns about the machine learning paradigm (see Author response image 1). Concerning the co-IP-MS, we can confirm that ORF1p is among the highly enriched proteins as it was not found in the IgG control (in 5 independent samples), only in the ORF1p-IP (in 5 out of 5 independent samples). This is what the infinite sign in Suppl Table 2 indicates and this is why there is no p-value assigned as infinite/0 doesn’t allow to calculate a pvalue. We have made this clearer in the revised version of the manuscript and added a legend to Suppl Table 2.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

I would recommend the authors remove the human data and expand the analysis of the aged mice. This would most likely result in a much stronger manuscript.

We do think that the imaging data and the Western blots are convincing (please also see our detailed response above to the criticism concerning the antibody we used and the newly added data) and very much reflects what we find in the mouse brain, i.e. concerning the percentage of neurons expressing ORF1p and the percentage of ORF1p+ cells being neuronal. When it comes to the transcriptomic data on aged dopaminergic neurons, we have further discussed the limitations of this study in the revised manuscript and hope that the findings inspire others in the field to redo these types of analyses using the now state-of-the-art NGS technologies to address the question and validate what we have found.

Reviewer #2 (Recommendations For The Authors):

The characterization of ORF1p expression across the mouse brain would be vastly more informative if cell identity was established beyond NeuN+/NeuN---the neuronal predominance of L1 activity in the brain has long been observed. Indeed, even corroboration of the PV+ interneuron signature previously reported would both lend credence to the present study and provide valuable confirmation to the field.

We agree. Please see our response above as well as the new experimental data we added (Suppl Fig5.F-K).

The increased intensity (but not prevalence in terms of % of Orf1p positive cells) of Orf1p expression in aged mouse brains would be more convincing with further context and perhaps better controls. Is overall protein turnover in aged neurons simply slower than in neurons from younger brains? Immunostaining with another protein marker, rather than Hoescht staining of DNA, to demonstrate that increased staining intensity is unique to Orf1p, would make this result more compelling.

To address this question, we have now added the quantifications of the protein Rbfox3 (NeuN) to Fig3 (Fig. 3B). As shown in this figure, NeuN intensity is not stable from one individual to another, neither in control mice nor in the aged control group. As we did not use NeuN intensity but presence or absence of NeuN as a marker for cell identity, this does not have any influence on the data presented in this manuscript. It does indicate however, that the overall increase of ORF1p in aged mice is not a mere reflection of a general decrease in protein turnover. As stated above, the DNA staining with Hoechst controls for technical artefacts. Using Hoechst and NeuN as control, we have thus provided evidence for the fact that the increase in ORF1p intensity per cell is indeed specific for ORF1p.

Western blotting on cell lysates from aged vs young NeunN+ sorted cells would also strengthen this conclusion, although I appreciate the technical challenge of physically isolating whole mature neuronal cells.

Indeed, this would be feasible but only after FACS sorting, which is technically challenging on whole brain cells (less so on nuclei). We unfortunately do not have the possibility to embark on this right now.

Concerning data presentation, Figure 3A would be much more informative if the graph was broken down to show the proportion of ORF1p+ and ORF1p- cells, regardless of NeuN status, and the proportion of NeuN+ and NeuN- cells shown independently of Orf1p status. It is difficult to ascertain the relationship of either of these variables to age, as the graph is presented now.

We followed the suggestions of the reviewer agreeing that breaking down this figure into either ORF1p+ or NeuN+ or NeuN- cells without double attribution is easier to interpret. However, we also chose to use cell densities (cell numbers/ per mm2) to represent the data (new Suppl Fig.5F-K) which is even more precise while proportions are now shown in Suppl Fig.3A-E. Indeed, while it is important to realize that the variables ORF1p+/- or NeuN+/- are not completely independent of each other (as shown in proportions of old Fig4A and B, new Suppl Fig3A and B) as they form four categories (NeuN+/ORF1p+; NeuN+/ORF1p-. NeuN-/ORF1p+, NeuN-/ORF1p-), we can see from the data that there is no overall change in neuron number in the mouse brain between 3 month and 16 months of age. There isn’t an overall change of the density of ORF1p+ cells nor NeuN- cells in the mouse brain with the exception of a decrease in cell density of ORF1p-positive cells in the dorsal striatum accompanied by an increase in non-neuronal cell density (but as discussed above and in the manuscript (line 332-337), this might be due to technical limitations). Thus, while ORF1p intensities per cell increase significantly in older mice, here is no significant change in ORF1p+ cell number.

Reviewer #3 (Recommendations For The Authors):

(1) According to the description in Materials and Methods on the analysis of the confocal images (lines 731-743) the authors used Cell-Pose for both the nuclei and cell segmentation tasks, using model=cyto and diameter=30 for the first (nuclei) and model=cyto2 and diameter=40 for the second (cell). Description of analysis of sagittal brain regions (lines 746-764) indicates the pre-trained model DSB2018 from StarDist 2D was used for nuclei detection, and Cell-Pose using model cyto2 and diameter=30 for cell segmentation. Detected nuclei were then matched to segmented cell areas based on overlap criteria and each nucleus was labeled as 'positive' or 'negative' for either OFR1P or NEU-N.

As described in its three publications (1, 2, 3), Cell-Pose as a segmentation tool is trained in different datasets, with cyto2 being trained on a more varied dataset than cyto. In their library they also offer a model specific for nuclei2. Some description and explanation on the reasons two different models were used for nuclei detection and not choosing the offered specific pre-trained model by Cell-Pose in either case.

According to the cellpose library documentation "Changing the diameter will change the results that the algorithm outputs. When the diameter is set smaller than the true size then cellpose may over-split cells. Similarly, if the diameter is set too big then cellpose may over-merge cells.". It would be useful to offer the justification of the pixels chosen for the analysis (possibly average pixel counts in a subsample of Hoechst images).

Answers to questions 1-5:

Regarding ABBA, slices were first positioned and oriented manually along the Z-axis, without using DeepSlice. Automated affine registration was then applied in the XY plane, followed by manual refinement. 1 slice per mouse brain, 4 mouse brains per condition.

Regarding the gradient heatmap, as stated in the figure legend of Fig3F; Represented is the fold-change in percent (aged vs young) of the “mean of the mean” ORF1p expression per ORF1p+ cell quantified mapped onto the nine different regions analyzed. More precisely, the heatmap shows the percentage increase in the mean of all mean cell intensities in the aged condition, normalized to the mean of all mean cell intensities in the young condition. The pre-trained models and hyperparameters were selected based on their optimal performance across our image datasets. For slide scanner images, the StarDist DSB 2018 model was chosen over a Cellpose model because it more effectively avoided detecting out-of-focus nuclei, which were common in slide scanner images due to the lack of optical sectioning. This issue was not present in confocal images, where Cellpose cyto model was used instead. To assess the performance of each model and diameter setting, we computed the average precision (AP) metric, which is defined as AP = TP/(TP+FP+FN), where TP = true positives, FP = false positives, and FN = false negatives. The AP was calculated at the commonly used Intersection over Union (IoU) threshold of 0.5. For confocal images, Cellpose models and hyperparameters were evaluated on eight images per channel, capturing intensity variability across different mouse ages and brain regions. A total of approximately 2,000 nuclei and 1,000 NeuN and ORF1p cells were manually annotated. The AP values at an IoU threshold of 0.5 were: 0.995 for nuclei, 0.960 for NeuN, and 0.974 for ORF1p cells. These high AP values confirm that the selected models and diameter settings were well-suited for analyzing the entire dataset. For slide scanner images, nuclei and cell detection were evaluated on 14 images per channel, with approximately 800 nuclei and 400 NeuN and ORF1p cells manually annotated. The AP values were lower compared to confocal images, mainly due to a lower signal-to-noise ratio, which led to an increased number of false positives and false negatives: 0.806 for nuclei, 0.675 for NeuN, and 0.695 for ORF1p cells. This decline in performance was expected given the challenges posed by slide scanner images, including background noise and out-of-focus objects. Notably, the observed false positives primarily correspond to small-sized nuclei/cells or those with low intensity, which evade the stringent filters that were applied. While fine-tuning the models could further enhance detection robustness, we considered that the selected models and diameter settings were suitable for processing the entire dataset.

We added a paragraph to the materials & methods section with this new information; for confocal images (line 847-855), slide scanner images (line 878-885).

Author response table 1.

(2) Next to no information is offered regarding the brain segment registration and how the results were analyzed: The ABBA plug-in has two modules manual and automatic, via a DL pre-trained model called DeepSlice. The authors should report which mode of ABBA they used, how many slices per mouse brain, and how many brains. Moreover, there is no explanation of how the gradient heatmap of the brain regions (Figure 3G) was calculated.

Please see above

(3) Even the best algorithms produce some False predictions. In this application of the (3rd party) cellpose, StarDist, and ABBA pre-trained models, such cases of wrong predictions would have amplified downstream effects on the analysis e.g., wrongly characterizing certain cells as 'negative' (falsely not detected cell, falsely detected nucleus), or worse, biasing against certain cell subgroups (falsely not detected 'type' of nuclei). This is even more troubling with the variety of models used for the nuclei segmentation task, and the parameters in each. It is possible the authors performed optimizations and reported exactly such optimized values for their dataset, they should however still explicitly offer these detailed validation and optimization processes. The low statistical significance throughout the quantified results from these IF experiments (Figures 1-3) is also a cause for needing an explicit description of how these algorithms perform on the authors' data.

It is good practice that a pre-trained model when applied to a new dataset like the one that the authors produced for this work, would require basic monitoring for how it performs in the new, previously unseen dataset, even when the model's generalizability has been reported previously as great. It would be best if the authors had handannotated a few images as the validation set and produced some model performance metrics as a supplemental table for all pre-trained models they used, in the datasets they used them at. Alternatively, the authors are offered the ability by the cellpose team to fine-tune the model for their data, and this could be used to perform the experiments for this work instead if the performance metrics of the used cellpose (cyto and cyto2) models prove to be poor.

Please see above

(4) The legend for Figure 1A indicates that Cell-Pose was used for cell detection and StarDist for nuclei detection in the confocal images (line 960). This needs clarification and correction.

Please see above

(5) Some explanation of why the models used were changed when using confocal or the slide scanner microscope would be nice.

Please see above

(6) The legend title of Figure 3 (line 1040) "Fig. 3: ORF1p expression is increased throughout the whole mouse brain in the context of aging" is misleading as half the panels in the figure demonstrate a decrease in ORF1pexpressing cells. The two can be both true, but in a more nuanced relationship. A more modest representation of the data in the title is also warranted by the unimpressive statistical significance achieved (notably with no correction for multiple testing, which would further inflate them).

We have toned down the tile of Fig. 3 to “ORF1p expression is increased in some regions of the aged mouse brain” while leaving its meaning as globally. There is indeed no significant loss of ORF1p expressing cells (Suppl Fig. 5F; except in the dorsal striatum (Supl Fig. 5I, please see also discussion above), but there is a significant increase in ORF1p intensity per cell overall (Fig. 3A,C,F) and in several regions of the mouse brain (Fig E, G and H).

(7) Figure 4 suffers for significance. For example in panel A, the few genes with the highest -log10P value, ie above 1.3 (p-value of ~0.05) have a log2-fold change of 0.2-0.3 (fold change 1.14-1.23). There are no hits with even the modest log2-fold change of 0.5 (fold-change 1.4). The big imbalance between young/old samples for these RNA seq experiments (6 vs 36 mice) could be an issue here too.

The reviewer refers to mouse samples (“6 to 36 mice”), but this is data of human post-mortem dopaminergic neurons from brain-healthy individuals which were laser-captured and sequenced as reported by Dong et al, Nat Neurosci, 2018. There is indeed a big imbalance between young and old samples which are linked to the difficulties in availability of brain-healthy post-mortem tissues from young individuals which are obviously much rarer than from older people. We agree that the fold-enrichment are modest and p-values rather high, but we argue to keep this data in as it is based on rare post-mortem human brain tissues which were difficult to obtain and will be very difficult to obtain in sufficient number in future studies. We hope however, that these results will encourage such studies in the future and motivate researchers to further look into the expression of TEs in aging brain tissues with higher sample sizes and more suitable sequencing techniques. We have now in the revised version toned down some sentences (i.e. line 359: modest, but significant increase in several young…) and have now also added a post-hoc power analysis (results section line 359-362: “There was a modest but significant increase in several younger LINE-1 elements including L1HS and L1PA2 at the “name” level (Fig. 4A, B), an analysis which was however underpowered (post-hoc power calculation; L1HS: 28.4%; L1PA2: 32.8%) and thus awaits further confirmation in independent studies.”)

(8) Figure legend 4C (line 1088) should offer more explanation on what is compared for these correlations: the young vs old results, all intensities of all experiments, and intensities separately for each sample.

We have added the missing information to Figure legend 4C (line 1209-1215): “Correlation of the RNA expression levels of LINE-1 elements with known transposable element regulators in human dopaminergic neurons (all ages included). What was compared are the expression levels of LINE-1 elements with known regulators of TEs for each individual sample, all ages included.”

(9) Figure 5, panel D. The regressions are all driven by 1-2 outliers. Should be removed as they don't add anything.

We agree and therefore have performed an outlier test (ROUT (Q=1%) and identified outliers (1 in each graph) have been taken out from the analysis. We argue that the information of a non-correlation of UID-68 and adjacent gene expression is important as it rules out a dependency of expression of the full-length LINE-1 depending on neighboring gene expression (see new Fig5E-G).

(10) Figure 6 panel B. It is unexpected that the GO terms with the highest enrichment also show weak significance and vice-versa. Fold enrichment in the PANTHER tool is defined as the % of GO-term genes in the sample divided by the %GO-term genes in the background (organism).

This is not unexpected as GO terms contain different numbers of proteins. Indeed, the significance can be different if the GO term contains for example 3 or 300 proteins. A GO term containing only few proteins with a high fold change between the conditions (here: ORF1p-IP vs whole mouse genome) will lead to a rather low significance for example. If you look at the last 6 categories in Fig 6B, you can appreciate that they have very similar values for enrichment but very different significance levels (FDR).

(11) Many citations in the References sections are referred to by doi and "Published online" date. These should be corrected to include the citation in standard format (journal name, volume, issue, pages, etc).

We apologize for this and have corrected this in the revised version.

(12) (line 970) Legend of Figure 1 is missing label referencing panel C (ie (C) Bar plot showing the total....).

Thank you for pointing this out, this has been corrected.

(13) The bottom violin plot in Figure 1C lacks sufficient explanation (what are the M1-4 categories?). The same problem with panel G (same Figure 1).

This has now been better explained. The M1-M4 categories denominate individual mice numbered from 1 to 4 for (results are shown per individual).

-> specified in line 1098-1099 (Fig.1C) and new text (1117-1118: Fig.1G): Four three-month-old Swiss/ OF1 mice (labeled as M1 to M4) are represented each by a different color, the scattered line represents the median. ****p<0.0001, nested one-way ANOVA. Total cells analyzed = 4645

(14) Figure 1B; confocal image 2 (Hippocampus) does not seem to tell the same story as the main slide scanner image. Overall, more explicit phrasing regarding how the Images in Figure 1B are not blow-outs of the bigger one but different, confocal images of the same regions.

We have changed the sentence to: “Representative images acquired on a confocal microscope of immunostainings showing ORF1p expression (orange) in 10 different regions of the mouse brain.”, which hopefully helps to indicate that these images are indeed not blow-outs of the slide scanner image.

(15) Young are defined as 3 months and 'old' as 16 months mice. 16-month group name would be better as "adults". Example of age range considered 'old': "Young (3-6-month-old) and aged (18-27-month-old) male mice were age- and source-matched for each experiment." https://www.cell.com/cell-metabolism/fulltext/S1550-4131(23)00462X?_returnURL=https%3A%2F%2Flinkinghub.elsevier.com%2Fretrieve%2Fpii%2FS15504131230 0462X%3Fshowall%3Dtrue

This is true, but the 16-month age group does not have a designation when looking at Mouse Life history stages in C57Bl/6 mice from the Jackson laboratory (see https://www.jax.org/news-and-insights/jax-blog/2017/november/when-are-mice-considered-old#), they are neither middle-aged nor old. We therefore believe that the designation as “aged” still holds true.

(16) Lines 63-65 > To our understanding, both ORF1 and ORF2 proteins are thought to exhibit cis preference.

Yes, that is true, but the sentence as it is does not make a claim about ORF2p not having cis-preference.

(17) Figure 1I is only referred to as "Figure I". Twice. Page 8, line 173 & 176.

Thank you, has been corrected.

(18) Lines 178-182 >To investigate intra-individual expression patterns of ORF1p in the post-mortem human brain, we analyzed three brain regions of a neurologically healthy individual (Figure 1J) by Western blotting. ORF1p was expressed at different levels in the cingulate gyrus, the frontal cortex, and the cerebellum underscoring a widespread expression of human ORF1p across the human brain." > It is difficult for us to gauge how believable the blots are without knowing the amount of protein loaded.

We have loaded 10ug of tissue lysate per lane (tissue pulverized with a Covaris Cryoprep; amount now mentioned in the materials & methods section). We have added some more information on the antibody in the revised manuscript (line 183-194).

We say this from our experience conducting similar blots of anti-ORF1p IPs from human brain tissues using the same antibody (4H1) without successful detection of enriched protein by western blot (of course there can be many reasons for that, but knowing the amount of protein loaded is important for reproducibility). In addition, we find the "double" ORF1p bands they see in almost every blot atypical.

In our hands, the 4H1 antibody does not work well on Western blots, but it immunoprecipitates well and works very well on immunostainings. However, the abcam AB 245249 works well for Western blotting (and IPs) which is why we used this antibody for these applications, respectively. As described above, there is evidence that the double band is not atypical, but rather frequent, which we now also mention in the revised manuscript line 183191: “To investigate intra-individual expression patterns of ORF1p in the post-mortem human brain, we analyzed three brain regions of a neurologically-healthy individual (Fig. 1J, entire Western blot membrane in Suppl Fig. 2A) by Western blotting using a commercial and well characterized antibody which we further validated by several means. The double band pattern in Western blots has been observed in other studies for human ORF1p outside of the brain (Sato et al, SciRep, 2023, McKerrow et al, PNAS, 2022) as well as for mouse ORF1p (Walter et al, eLife, 2016). We also validated the antibody by immunoprecipitation and siRNA knock-down in human dopaminergic neurons in culture (differentiated LUHMES cells, Suppl Fig. 2B and 2C) where we detect however in most cases the upper band only. The nature of the lower band is unknown, but might be due to truncation, specific proteolysis or degradation. ORF1p was expressed at different levels in the human post-mortem cingulate gyrus, the frontal cortex and the cerebellum underscoring a widespread expression of human ORF1p across the human brain. This was in accordance with ORF1p immunostainings of the human post mortem cingulate gyrus (Fig. 2H and Suppl Fig. 2E) and frontal cortex (Suppl Fig. 2E), with an absence of ORF1p staining when using the secondary antibody only (Suppl Fig. 2E).”

In some images a band is labeled as IgG heavy chain (e.g. presumably from the FACS, Figure 2G, and IP, Figure 6A - which could contain residual antibody) - however, this is avoidable by using a different antibody for capture than detection - which also helps reduce false positive results.

Unfortunately, we have only an antibody raised in rabbit available to perform IPs and Western blots on mouse tissues and therefore cannot avoid the detection of the IgG heavy chain.

Aside from these, there seem to be persistent 'double bands' in the region of ORF1p. Generally, we are unaccustomed to seeing such 'double bands' in human anti-ORF1p western blots and IP-western blots, and since, in this study, this is seen in both mouse and human blots, it raises some doubts. Having the molecular mass ladder on each blot to at least allow for the assessment of migration consistency and would therefore be very helpful.

We have added the molecular weights on the Western blots (Fig.1H, Fig. 2G and Suppl Fig.1D and E). As discussed also above, there is accumulating evidence that in some tissues, there are persistent double bands detected using ORF1p antibodies in both, mouse and human tissues.

Human ORF1p detection:

We have validated the antibody against human ORF1p (Abcam 245249-> https://www.abcam.com/enus/products/primary-antibodies/line-1-orf1p-antibody-epr22227-6-ab245249), which we use for Western blotting experiments (please see Fig1J and new Suppl Fig.2A,B and C), by several means.

(1) We have done immunoprecipitations and co-immunoprecipitations followed by quantitative mass spectrometry (LC-MS/MS; data not shown as they are part of a different study). We efficiently detect ORF1p in IPs (Western blot now added in Suppl Fig2B) and by quantitative mass spectrometry (5 independent samples per IP-ORF1p and IP-IgG: ORF1p/IgG ratio: 40.86; adj p-value 8.7e-07; human neurons in culture; data not shown as they are part of a different study). We also did co-IPs followed by Western blot using two different antibodies, either the Millipore clone 4H1 (https://www.merckmillipore.com/CH/en/product/Anti-LINE-1-ORF1p-Antibody-clone- 4H1,MM_NF-MABC1152?ReferrerURL=https%3A%2F%2Fwww.google.com%2F) or the Abcam antibody to immunoprecipitate and the Abcam antibody for Western blotting on human brain samples. Indeed, the Millipore antibody does not work well on Western Blots in our hands. We consistently revealed a double band indicating that both bands are ORF1p-derived. We have added an ORF1p IP-Western blot as Suppl Fig. 2B which clearly shows the immunoprecipitation of both bands by the Abcam antibody. Abcam also reports a double band, and they suspect that the lower band is a truncated form (see the link to their website above). ORF1p Western blots done by other labs with different antibodies have detected a second band in human samples

• Sato, S. et al. LINE-1 ORF1p as a candidate biomarker in high grade serous ovarian carcinoma. Sci Rep 13, 1537 (2023) in Figure 1D

• McKerrow, W. et al. LINE-1 expression in cancer correlates with p53 mutation, copy number alteration, and S phase checkpoint. Proc. Natl. Acad. Sci. U.S.A. 119, e2115999119 (2022)) showing a Western blot of an inducible LINE-1 (ORFeus) detected by the MABC1152 ORF1p antibody from Millipore Sigma in Figure 7 - Walter et al. eLife 2016;5:e11418. (DOI: 10.7554/eLife.11418) in mouse ES cells with an antibody made inhouse (gift from another lab; in Figure 2B)

The lower band might thus be a truncated form of ORF1p or a degradation product which appears to be shared by mouse and human ORF1p. We have now mentioned this in the revised version of the paper (lines 183-189).

(2) We have used the very well characterized antibody from Millipore ((https://www.merckmillipore.com/CH/en/product/Anti-LINE-1-ORF1p-Antibody-clone-4H1,MM_NF-MABC1152?ReferrerURL=https%3A%2F%2Fwww.google.com%2F)) for immunostainings and detect ORF1p staining in human neurons in the very same brain regions (Fig 2H, new Suppl Fig. 2E) including the cerebellum in the human brain. We added a 2nd antibody-only control (Suppl Fig. 2E).

(3) We also did antibody validation by siRNA knock-down. However, it is important to note, that these experiments were done in LUHMES cells, a neuronal cell line which we differentiated into human dopaminergic neurons. In these cells, we only occasionally detect a double band on Western blots, but mostly only reveal the upper band at ≈ 40kD. The results of the knockdown are now added as Suppl Fig. 2C.

Altogether, based on our experimental validations and evidence from the literature, we are very confident that it is indeed ORF1p that we detect on the blots and by immmunostainings in the human brain.

Mouse ORF1p detection: In line 117-123 of the manuscript, we had specified “Importantly, the specificity of the ORF1p antibody, a widely used, commercially available antibody [18,34–38], was confirmed by blocking the ORF1p antibody with purified mouse ORF1p protein resulting in the complete absence of immunofluorescence staining (Suppl Fig. 1A), by using an inhouse antibody against mouse ORF1p[17] which colocalized with the anti-ORF1p antibody used (Suppl Fig. 1B, quantified in Suppl Fig. 1C), and by immunoprecipitation and mass spectrometry used in this study (see Author response image 1)”.

Figure 2G shows a Western blot using an extensively used and well characterized ORF1p antibody from abcam (mouse ORF1p, Rabbit Recombinant Monoclonal LINE-1 ORF1p antibody-> (https://www.abcam.com/enus/products/primary-antibodies/line-1-orf1p-antibody-epr21844-108-ab216324; cited in at least 11 publications) after FACS-sorting of neurons (NeuN+) of the mouse brain. We have validated this ORF1p antibody ourselves in IPs (please see Fig 6A) and co-IP followed by mass spectrometry (LC/MS-MS; see Fig 6, where we detect ORF1p exclusively in the 5 independent ORF1p-IP samples and not at all in 5 independent IgG-IP control samples, please also see Suppl Table 2). In this analysis, we detect ORF1p with a ratio and log2fold of ∞ , indicating that this proteins only found in IP-ORF1p samples (5/5) and not in the IP-control samples ((not allowing for the calculation of a ratio with p-value), please see Suppl Table 2)

In addition, we have added new data showing the entire membrane of the Western blot in Fig1H (now Suppl Fig.1E) and a knock-down experiment using siRNA against ORF1p or control siRNA in mouse dopaminergic neurons in culture (MN9D; new Suppl Fig.1D). This together makes us very confident that we are looking at a specific ORF1p signal. The band in Figure 2G is at the same height as the input and there are no other bands visible (except the heavy chain of the NeuN antibody, which at the same time is a control for the sorting). We added some explanatory text to the revised version of the manuscript in lines 120-124 and lines 253-256).

Please note that in the IP of ORF1p shown in Fig6A, there is a double band as well, strongly suggesting that the lower band might be a truncated or processed form of ORF1p. As stated above, this double band has been detected in other studies (Walter et al. eLife 2016;5:e11418. DOI: 10.7554/eLife.11418) in mouse ES cells using an in-house generated antibody against mouse ORF1p. Thus, with either commercial or in-house generated antibodies in some mouse and human samples, there is a double band corresponding to full-length ORF1p and a truncated or processed version of it.

We noticed that we have not added the references of the primary antibodies used in Western blot experiments in the manuscript, which was now corrected in the revised version.

(19) Figure 1H, 1J, 6A: Show/indicate molecular weight marker.

The molecular weight markers were added (please see Fig.1H, Fig. 2G and Suppl Fig.1D and E).

(20) Page 10, line 223. " ...expressing ORF1p and ORF1p"?

Thank you, this was corrected.

(21) Lines 279-280 "An increase of ORF1p expression was also observed in three other regions albeit not significant." > This means it is not distinguishable as a change under the assumptions and framework of the analysis; please remove this statement.

We agree, we removed this sentence.

(22) Page 13, line 301. Labeling the group with a mean age of 57.5 as "young" might be a bit misleading.

This is why we put the “young” in quotation marks.

(23) Lines 309-311 "however there was a significant increase in several younger LINE-1 elements including L1HS and L1PA2 at the "name" level (Figure 4A, B)". > Effect size is tiny; is this really viable as biologically significant? Maybe just remove the volcano plot? Does panel A add anything not covered by B?

We would like to keep the Volcano plot, even though effect sizes are small (which we acknowledge in the manuscript line 359-362: “There was a modest but significant increase in several younger LINE-1 elements including L1HS and L1PA2 at the “name” level (Fig. 4A, B), an analysis which was however underpowered (posthoc power calculation; L1HS: 28.4%; L1PA2: 32.8%) and thus awaits further confirmation in independent studies.” The reason for this decision is to illustrate a general increase in expression (even with a small effect size) of several LINE-1 elements at the name level with the youngest LINE-1 elements being amongst those with the highest effect.

(24) Lines 327-328 "The transcripts of these genes showed, although not statistically significant, a trend for decreased expression in the elderly (Supplementary Figure 5D-G). > I do not recommend doing this.

We agree and take it out.

(25) Lines 339-342 "While several tools using expectation maximization algorithms in assigning multi-mapping reads have been developed and successfully tested in simulations 48,54, we used a different approach in mapping unique reads to the L1Base annotation of full-length LINE-1" > Generally, this section is not clear - what is the rationale for the approach (compared to the stated norms)? Ideally, justify this analytical choice and provide a basic comparison to other more standard approaches (even if briefly in a supplement).

We thank the reviewer for his comment. Indeed, randomly assigning multi-mapping reads is usually a good strategy to quantify the expression of repeats at the family level (Teissandier et al. 2019) which we did in the first part of the analysis (class, family and name level). However, our main goal was to focus on specific single fulllength LINE elements which can encode ORF1p. We therefore decided to only use uniquely mapped reads, which is by definition the only way to be sure that a sequencing read really comes from a specific genomic location, and which will to not over-estimate their expression level. In this sense, we have added some explanatory text to this specific section. We also added a section to the discussion (line 638-644): This analysis has technical limitations inherent to transcriptomic analysis of repeat elements especially as it is based on short-read sequences and on a limited and disequilibrated number of individuals in both groups. Nevertheless, we tried to rule out several biases by demonstrating that mappability did not correlate with expression overall and used a combination of visualization, post-hoc power analysis and analysis of the mappability profile of each differentially expressed fulllength LINE-1 locus.

(26) Page 16, line 389. The age span covered is 59 years although the difference in mean age between the two groups is only 25.5 years - please indicate both metrics.

We have added this additional metric in line 432.

(27) Lines 394-397 "Further, correlation analyses suggest that L1HS expression might possibly be controlled by the homeoprotein EN1, a protein specifically expressed in dopaminergic neurons in the ventral midbrain 50, the heterochromatin binding protein HP1, two known regulators of LINE-1, and the DNA repair proteins XRCC5/6." > This reads like a drastic reach unless framed explicitly as a 'tempting speculation' (or similar). I don't think this claim should be made as it is without further validation.

We believe to have used careful language (“correlation analysis suggests”.“might possibly be controlled”) in the results section as well as in the discussion (line 660-671): “Matrix correlation analysis of several known LINE-1 regulators, both positive and negative, revealed possible regulators of young LINE-1 sequences in human dopaminergic neurons. Despite known and most probable cell-type unspecific regulatory factors like the heterochromatin binding protein CBX5/HP1 [51] or the DNA repair proteins XRCC5 and XRCC6 [49], we identified the homeoprotein EN1 as negatively correlated with young LINE-1 elements including L1HS and L1PA2. EN1 is an essential protein for mouse dopaminergic neuronal survival [50] and binds, in its properties as a transcription factor, to the promoter of LINE-1 in mouse dopaminergic neurons [17]. As EN1 is specifically expressed in dopaminergic neurons in the ventral midbrain, our findings suggests that EN1 controls LINE-1 expression in human dopaminergic neurons as well and serves as an example for a neuronal sub-type specific regulation of LINE-1.” To this we added: “Although these proteins are known regulators of LINE-1, this correlative relationship awaits experimental validation.”

(28) Mouse protein/gene names are all capital letters on page 17/18. Changes on page 18/19. This should be consistent.

Thank you, this has been corrected (all capital).

(29) Page 23, line 559. The estimated ORF1p/ORF2p ratio referenced is based on an overexpression of L1 from a plasmid (ref87). > It should be made clear to the reader that it is still unknown whether such a ratio is representative of native conditions.

OK, this is indeed true. Thank you for pointing this out. (line 621-622)

(30) Lines 613-616 "Further, GO term analysis contained expected categories like "P-body", mRNA metabolism related categories, and "ribonucleoprotein granule". We also identified NXF1 as a protein partner of ORF1p, a protein found to interact with LINE-1 RNA related to its nuclear export 89." > There is no reason to speculate that the proteins in the pulldown are specific to L1 RNAs.

We did not speculate that the proteins in the pulldown are specific to LINE-1 RNA. We just mentioned that NXF1 was an ORF1p protein partner and that it had been found previously as a LINE-1 RNA interactor.

ORF1p is present in large heterogeneous assemblies - not every protein should be assigned an L1-related function and many proteins will be participating in general RNA-granule functions (given L1 ORFs are known to accumulate in such structures). Moreover, the granules are not the same in every cell type. IP is done in low salt and overnight incubation (poorly controlled for non-specific accumulation).

We state that these key interactors are “probably” essential for completing or repressing the LINE-1 life cycle. It is true that we cannot affirm this. We therefore added a sentence to the discussion (line 679): “This supports the validity of the list of ORF1p partners identified, although we cannot rule out the possibility that unspecific protein partners might be pulled down due to colocalization in the same subcellular compartment.”

(31) Lines 629-631" These results complete the picture of the post-transcriptional and translational control of ORF1p and suggest that these mechanisms, despite a steady-state expression, are operational in neurons." > Stating that these results complete the picture, which is still very much open for completion (granted, these results add to the picture), is an unneeded over-reach.

We agree. We changed “complete” to “add to “ the picture.

(32) Lines 641-644 "Finally, we found components of RNA polymerase II and the SWI/SNF complex as partners of ORF1p. This further indicates that ORF1p has access to the nucleus in mouse brain neurons as described for other cells 95,96, implying that ORF1p potentially has access to chromatin." > There is no way to know if this is a post-lysis effect - we have no real specificity information. The mock IP control is insufficient for this conclusion without further validation.

We added: “however a bias due to a post-lysis effect cannot be excluded.” Line 711

(33) ab216324 for IF and ab245122 for IP - why? What is the difference? Both are rated equally for IF and IP - please provide a rationale for reagent selection and use.

These two antibodies are the same except their storage buffer. ab245122 is azide and BSA-free, while ab216324 contains the preservative sodium azide (0.01%) and the following constituents: PBS, 40% Glycerol (glycerin, glycerine), 0.05% BSA. As azide and BSA can affect coupling of antibodies to beads, antibodies which do not contain these components in their buffer are preferred for IPs (but can be stored less long).

(34) Page 35, line 862. "1.3 x 105" should be "1.3 x 105".

We added a regular x but we are not sure if this is what the reviewer was referring to ?

(35) MS comparison in Figure 6. Why is the comparison not being made between young vs. old brain/neurons? This would be more informative instead of just showing what they IP over a mock IgG control and the comparison would track better with other experiments in the rest of the paper.

Yes, that is true. However, we did not do this at the time as we did not have old mouse brain tissue available. Services from official animal providers in France have unfortunately only recently expanded their offer with regard to the availability of aged animals.

(36) Supplementary Table 2 (MS data) is lacking information. How many peptides (unique/total) were discovered for each protein? Why are all ratios and p-values not listed for every protein in the table? LFQ protein intensity values should also be listed. Each supplementary table should have a legend as a separate tab in the document.

As stated in the SupplTable2 and now made clearer in an independent tab file in SupplTable2 which contains a legend to the table, some proteins do not have associated p values and ratios as these proteins are found only in the ORF1p IP and not in the IgG control. This is why these proteins have an indefinite sign instead of a foldenrichment and no p-value assigned as we cannot calculate a ratio with X/0 which again makes it impossible to obtain a p-value. Concerning the absence of LFQ protein intensity values, as stated in the materials & methods section, we did not use these values (linear model) but instead the intensity values of the peptides: “The label free quantification was performed by peptide Extracted Ion Chromatograms (XICs), reextracted by conditions and computed with MassChroQ version 2.2.21 109. For protein quantification, XICs from proteotypic peptides shared between compared conditions (TopN matching) with missed cleavages were used. Median and scale normalization at peptide level was applied on the total signal to correct the XICs for each biological replicate (n=5). To estimate the significance of the change in protein abundance, a linear model (adjusted on peptides and biological replicates) was performed, and p-values were adjusted using the Benjamini–Hochberg FDR procedure.”

The number of peptides unique/total for each protein has been added to Suppl_Table2 along other available information.

(37) Poor overlap in 6C could in part be explained by the use of different sample/tissue types, but more likely the big difference could come from the very different conditions at which the IPs were performed (buffers and incubation times etc.).

The overlap seems poor, but nevertheless is bigger as by chance (representation factor 2.6, p<5.4e-08). We agree that this can be in part explained by different experimental conditions which we now added to the discussion (line 478: “However, differences in experimental conditions could also influence this overlap.”)

(38) Figure 6D is a very uninspiring representation of the data. What is the point of showing several binary interactions? Was the IgG control proteome also analyzed? Have proteins displayed in Figure 6 been corrected for that?

The point of showing these interactions is that OFR1p interacts with clustered proteins. ORF1p interacts with proteins that belong to specific GO terms (Fig6b), but these proteins are also interacting with each other more than expected (Fig6C). This is the benefit of showing a STRING (Search Tool for the Retrieval of Interacting Genes/Proteins) representation, which is a database of known and predicted protein–protein interactions. Indeed, proteins in Fig6 have been corrected for the IgG proteome. We only show proteins that were enriched or uniquely present in the ORF1p IP condition compared to the IgG control (please see Suppl_Table2).

Author Response

The following is the authors’ response to the original reviews.

Comment 1: The authors showed increased plasma IL-22 and its expression in the intestine. Are intestinal ILC3s the main source of plasma IL-22?

Reply: ILC3s are the main source of IL-22 as reported previously (PMID: 30700914). In the small intestine, ILC3s account for about 62% of IL22+ cells. Other IL22+ cells include γδ T, Foxp3+T and CD4+T cells.

Comment 2: The authors transplanted intestinal ILC3s from NCD mice to DIO mice and showed significant metabolic improvements. However, in Fig. 1, intermittent fasting increased IL-22positive ILC3s proportion rather than changing the total number. Please clarify whether this transplantation is due to increasing ILC3s number or introducing more IL-22 positive ILC3s (which are decreased in DIO). Are these transplanted ILC3s by default homing to the intestine rather than to other tissues?

Reply: We believe that the transplantation increases ILC3s number, leading to the increment in IL22 levels. The transplanted ILC3s by default are homing to the intestine rather than to other tissues because ILC3s express several homing receptors such as CCR7, CCR9, and α4β7, which modulate their capacity to migrate to the gut (PMID: 26141583; PMID: 26708278; PMID: 25575242; PMID: 34625492). Our observation that ILC3s in adipose tissue remained unchanged by ILC3 cell transplantation (Supplementary Figure 5F) also supports this concept.

Comment 3: Thermogenesis in this acute cold challenge is mainly by brown adipose tissue. Beiging is a chronic and adaptive response. Based on the data in WAT, there is a beiging phenotype, but the core body temperature in acute cold challenge is not an accurate readout. It would be a missed opportunity by not evaluating thermogenic activity in BAT. More browning genes should be included to strengthen the beiging phenotype of WAT. Moreover, inflammation in WAT can be examined to provide a whole picture of adipose tissue remodeling through this pathway.

Reply: Per suggestion, we performed additional experiments to measure levels of inflammation genes such as Il4, Il1b, Il6, Il22, Il23, Il17a. As shown in supplemental figure 2D, these inflammation relevant genes were not altered.

Comment 4: For the SVF beige adipocyte differentiation, 100 ng/mL IL-22 was used. This is highly above the physiological concentration at ~5 pg/mL. Please justify this high concentration used.

Reply: We agree with the reviewer that the dose of IL-22 used is high. However, the efficient dose at 100 ng/ml used in our studies is consistent with the literatures. Previous reports have shown that IL-22 directly activates Stat3 in adipose tissue and primary adipocytes, and promotes the expression of genes involved in triglyceride lipolysis (Lipe and Pnpla2) and fatty-acid β-oxidation (Acox1) at the dose of 100 ng/ml (Wang X, Ota N, et al. Nature. 2014). Consistently, other studies have reported that IL-22 at 100 ng/ml significantly reversed the enhanced expression of CCL2, CCL20 and IL1B mRNAs in granulosa cells in vitro (Qi X, et al. Nat Med. 2019).

Comment 5: The authors showed increased Ucp1 and Cidea expression by IL-22 treatment in SVFs. Please be aware that these increases are likely due to boosted adipogenesis as told by the morphology. Please examine more adipogenic markers to confirm. Is this higher adipogenesis caused by the high concentration of IL-22?

Reply: Per suggestion, we examined the expression of adipogenic marker genes such as Pparγand Fabp4. We found that IL-22 did not increase the levels of these adipogenic marker genes relevant to the PBS control as shown in supplemental figure 6F.

Author response image 1.

Comment 6: In line 201, the authors drew the conclusion that IL-22 increased SVF beige differentiation. To fully support this conclusion, the authors should assure adipogenesis at the same baseline and then compare beiging, or examine the effect of IL-22 on normal adipogenesis to compare with beige differentiation.

Reply: We examined the expression of adipogenic marker genes such as Pparγ and Fabp4 and found that IL-22 did not increase the expression of these adipogenic marker genes relevant to the PBS control.

Reviewer #2:

This study aims to investigate the mediatory role of intestinal ILC3-derived IL-22 in intermittent fasting-elicited metabolic benefits.

Strengths:

The observation of induction of IL-22 production by intestinal ILC3 is significant, and the scRNAseq provides new information into intestine-resident immune cell profiling in response to repeated fasting and refeeding.

Weaknesses:

The experimental design for some studies needs to be improved to enhance the rigor of the overall study. There is a lack of direct evidence showing that the metabolically beneficial effects of IF are mediated by intestinal ILC3 and their derived IL-22. The mechanism by which IL-22 induces a thermogenic program is unknown. The browning effect induced by IF may involve constitutive activation of lipolysis, which was not considered.

Comment 1: Lack of direct evidence showing that IL-22-expressing ILC3s in intestine is the key contributor to intermittent fasting (IF)-mediated elevation of circulating IL-22 levels. The fraction of IL-22-expressing cells was increased threefold by IF but the increase in circulating IL-22 is moderate (Figs. 1J and 1K).

Reply: IL-22 in circulation is subjected to clearance, degradation, and binding with plasma proteins, et al. Thus, circulating levels of IL-22 may be much lower than the amount secreted by the intestinal IL-22 positive ILC3s.

Comment 2: The loss of fat mass by IF suggests that the active lipolysis may explain the white fat browning which was not considered. This may apply to the observations in IL-22 treated mice as well as IL-22R KO mice.

Reply: We analyzed the expression of genes relate to lipolysis in NCD and NCD-IF mice and found that IF did not alter the levels of these genes in white adipose tissues (Supplementary figure 2D). We have addressed this concerns in lines 119, page 6.

Author response image 2.

Comment 3: IL-22 administration and adoptive transfer of ILC3 had no significant effect on body weight. Not clear how IL-22 improves insulin sensitivity in this case.

Reply: Our results are consistent with previous report showing that IL-22 administration improves insulin sensitivity without change in body weight (Qi X, et al. Nat Med. 2019). In addition, previous studies have demonstrated that IL-22 can increase Akt phosphorylation in muscle, liver and adipose tissues, leading to improvement in insulin sensitivity (Wang X, et al. Nature. 2014). We have addressed this potential mechanism in lines192-195, page 9.

Comment 4: The energy expenditure data look unusual given that there was little increase in oxygen consumption during dark cycle compared to light cycle (Fig.3).

Reply: The not so obvious difference in oxygen consumption between dark cycle and light cycle may be due to the technical problem of the system.

Comment 5: The thermogenic capacity for the whole fat pad needs to consider the expression of UCP1 in certain amount of tissue and the total mass for each individual animal because the mRNA level itself does not reflect the whole tissue capacity.

Reply: We used the whole subcutaneous adipose tissue from one side for qPCR to reflect the whole tissue capacity.

Comment 6: The design of studies for the adoptive transfer of ILC3 was concerned. The PBS is not a good control for the group with ILC3 cells (Figs. 2A-2H). Similar issue applies for the co-culture study in which beige only is not an ideal control for Beige+ILC3 (Figs. 2I-2J).

Reply: We agree with the reviewer that the PBS is not a good control. Because we cannot find a similar immune cell without any effect on adipocytes, we designed this experiment based on other studies in which saline or PBS are used as ILC transfer experiment controls (Sasaki T, et al. Cell Rep. 2019; Wang H, et al. Nat Commun. 2019)

Comment 7: The induction of thermogenesis by IL-22 treatment may be related to enhanced differentiation rather than direct activation of thermogenic genes (Figs. 4G and 4H).

Reply: Our observation that IL-22 did not alter the levels of genes related to adipogenesis (Supplemental figure 6F) indicates that IL-22 may not alter the differentiation of adipocytes. We addressed this concern in Lines 211-212, page 10.

Reviewer #3:

Chen et al. investigated how intermittent fasting causes metabolic benefits in obese mice and found that intestinal ILC3 and IL-22-IL-22R signaling contribute to the beiging of white adipose tissue (WAT) and consequent metabolic benefits including improved glucose and lipid metabolism in diet-induced obese mice. They demonstrate that intermittent fasting causes increased IL22+ILC3 in small intestines of mice. Adoptive transfer of purified intestinal ILC3 or administration of exogenous IL-22 can lead to increases in UCP1 gene expression and energy expenditure as well as improved glucose metabolism. Importantly, the above metabolic benefits caused by intermittent fasting are abolished in IL-22R-/- mice. Using an in vitro experiment, the authors show that ILC3derived IL-22 may directly act on adipocytes to promote SVF beige differentiation. Finally, by performing sc-RNA-seq analysis of intestinal immune cells from mice with different treatments, the authors indicate a possible way of intestinal ILC3 being activated by intermittent fasting. Overall, this study provides a new mechanistic explanation for the metabolic benefits of intermittent fasting and reveals the role of intestinal ILC3 in the enhancement of the whole-body energy expenditure and glucose metabolism likely via IL-22-induced beige adipogenesis.

Although this study presents some interesting findings, particularly IL-22 derived from intestinal ILC3 could induce beiging of WAT by directly acting on adipocytes, the experimental data are not sufficient to support the key claims in the manuscript.

Comment 1: Only increased UCP1 expression on mRNA level is not enough to support the beiging of WAT. More methods such as western blotting and immunostaining of UCP1 in WAT are needed to confirm the enhanced beige adipogenesis.

Reply: Additional experiments have been performed to measure the UCP1 protein by Western blot. The data is included in Figure 4I and Supplementary Figure 2E.

Comment 2: IL-22 is known to modulate metabolic pathways via multiple downstream functions. The use of whole-body knockout of IL-22R could not exclude the indirect effect on the promotion of beiging of WAT. Specific deletion of IL-22R in adipose tissues is therefore needed to confirm the direct effect of IL-22 on adipocytes which is suggested by the in vitro study.

Reply: We agreed with the reviewer that specific deletion of IL-22R in adipose tissues is critical to confirm the direct effect of IL-22 on adipocytes. We will generate the AdioQ-IL-22R-/- mice to test this concept further in vivo.

Comment 3: The authors failed to show the cellular distribution of IL-22R in adipose tissues. This is important because the mechanism that explains the increased beige adipogenesis could be different based on the expression of IL-22R in adipose progenitor cells or mature adipocytes. So it is not appropriate to conclude that "IL-22 then directly activates IL-22R on adipocytes, leading to subsequent induction of beiging of white adipose tissue" in line 407. Additionally, Oil red O staining is needed for Fig 4G and Fig 5J, and protein levels of UCP1 and adipogenesis-related markers are needed to evaluate beige fat differentiation and the whole adipogenesis.

Reply: Per suggestion, we have added the expression of IL-22R in adipose progenitor cells or mature adipocytes (Supplementary Figure 6E). In addition, protein levels of UCP1 and adipogenesis-related markers to evaluate the whole adipogenesis (Figure 4I, Supplementary figure 6F) are now included. We have also addressed this issue in lines 207-215, page 10.

Comment 4: Although the authors provided some hypothesis about how intermittent fasting increases IL-22+ILC3 in small intestines by sc-RNA-seq analysis, some functional assays are needed to identify the factors, for example, how about the levels of macrophage-derived IL-23 or AHR ligands in small intestines and whether they contribute to increased percentages of intestinal IL-22+ILC3 following intermittent fasting.

Reply: We used flow cytometry sorting of macrophages combined with qPCR experiments to preliminarily demonstrate that intermittent fasting increases the expression of molecules such as Cd44 and CCl4 (Supplementary Figure 10B), which may contribute to the increase in the proportion of IL-22+ ILC3s in the intestine under intermittent fasting. Our observation that IL-23 mRNA levels were not changed indicates that this molecule may not the major contributor for the communication between macrophage and ILC3s. Other potential molecules such as AHR ligands remain to be explored.

Comment 5: What are the differences between adipose ILC3 and intestinal ILC3? Why do transferred ILC3 only migrate to the small intestine but not WAT of recipient mice? It would be better to examine or at least discuss whether other factors from intestinal ILC3 may also contribute to beiging of WAT following intermittent fasting.

Reply: Intestinal ILC3s specifically express gut homing receptors CCR7, CCR9, and α4β7 (PMID: 26141583; PMID: 26708278; PMID: 25575242; PMID: 34625492). This may explain transplantation of intestinal ILC3s can migrate mainly to the intestine instead of adipose tissue (PMID: 34625492). The proportion of ILC3s in adipose tissue of mice is very small. Their functions have not been clarified yet. We have addressed this issue in lines 156-158, page 8.

There are some other factors from intestinal ILC3 which may also contribute to beiging of WAT following intermittent fasting. By secreting IL-22, ILC3 enhanced the intestinal mucosal barrier, leading to reduction of the influx of LPS and PGN into the bloodstream under high-fat diet conditions, and subsequent increase in the beiging of white adipose tissue (Chen H, et al. Acta Pharm Sin B. 2022). We have addressed this potential mechanism in lines 344-347, page 16.

Comment 6: The sensitivity of the IL-22 ELISA kit used in the manuscript was 8.2 pg/mL, according to the information from the methods, however, in Fig. 1J and Fig. 2B, the IL-22 levels in mouse plasma were lower than 6 pg/mL, which was below the sensitivity of the ELISA kit and also the assay range. Please explain.

Reply: We have double-checked the original data and found that we have made a mistake in calculating the concentration of IL-22. We have corrected this error (Fig. 1J, Fig. 2B).

Comment 7: In Fig 7A, the significance of the Hypothesis testing should be marked. In Fig 7F and 7G, the contrast between the two groups is not apparent, other comparing ways could be used to enhance the readability.

Reply: Per suggestion, we have marked the significance of the hypothesis testing between HFD vs NCD and HFD-IF vs HFD in Fig7A. Shown in Fig 7F and 7G are the top 20 enriched interacting proteins between different cell types. The dot plot displays the average expression level and significance of protein interactions in cell types.

Comment 8: The total food intake of fasting mice fed with NCD or HFD was less than those without fasting, and the food intake rate the author showed in Fig S1 represents the value that was normalized to body weight. So the author should describe it precisely In line 114.

Reply: We have revised the statement accordingly in line 114-115.

Comment 9: Western blotting analysis has been described in methods, however, there is no corresponding experimental data in the result part.

Reply: The Western blotting results are now included.

Author Response

The following is the authors’ response to the original reviews.

After thoroughly reviewing the comments and suggestions provided by the reviewers, we have revised our manuscript. We sincerely appreciate the reviewers' constructive approach and valuable feedback. We believe that the edited version of the manuscript is now more comprehensible and reader-friendly. Please find our responses to the comments below.

Reviewer #1 (Public Review):

This EEG study probes the prediction of a mechanistic account of P300 generation through the presence of underlying (alpha) oscillations with a non-zero mean. In this model, the P300 can be explained by a baseline shift mechanism. That is, the non-zero mean alpha oscillations induce asymmetries in the trial-averaged amplitudes of the EEG signal, and the associated baseline shifts can lead to apparent positive (or negative) deflections as alpha becomes desynchronized at around P300 latency. The present paper examines the predictions of this model in a substantial data set (using the typical P300-generating oddball paradigm and careful analyses). The results show that all predictions are fulfilled: the two electrophysiological events (P300, alpha desynchronization) share a common time course, anatomical sources (from inverse solutions), and covariations with behaviour; plus relate (negatively) in amplitude, while the direction of this relationship is determined by the non-zero-mean deviation of alpha oscillations pre-stimulus (baseline shift index, BSI). This is indicative of a tight link of the P300 with underlying alpha oscillations through a baseline shift account, at least in older adults, and hence that the P300 can be explained in large parts by non-zero mean brain oscillations as they undergo post-stimulus changes.

Specific comments

1) The baseline shift model predicts an inverse temporal similarity between alpha envelope changes and P300, confirmed over posterior regions (negative maxima over Pz, Fig 2B). It is therefore intriguing to see in this Figure a very high (positive) correlation in left frontal electrodes. I acknowledge that this is covered in the discussion, but given that this is somewhat unexpected at this point, I suggest providing the readers with a pointer in the Figure legend to this observation and the discussion. Also, I would recommend being more careful with the discussion of this left frontal positive correlation, where a "negative P300" over these areas is mentioned. Given the use of average-referenced sensor data (as opposed to source localized data) and the clear posterior localization of the P300 (Fig 4A), it is likely that what is picked up as "negative ERP potential" over left frontal sites is the posterior P300 forward-projected and inverted through the calculation of the average reference. Accordingly, the interpretation in terms of polarity (positive) of the correlation is likely misleading but what this observation seems to suggest is that other oscillatory processes (than posterior alpha) (e.g. of motor preparation during evidence accumulation) do substantially correlate with the posterior P300 build-up.

We agree that the name P300 should be used rather for positive potential over posterior sites. We edited the text, substituting mentions of “negative P300” for “negative ER”. Also, the following text has been added to the legend of Figure 2:

“Note the positive correlation between the low-frequency signal and the alpha amplitude envelope over central sites. Due to the negative polarity of ER over the fronto-central sites, such correlation may still indicate a temporal relationship between the P300 process and oscillatory amplitude envelope dynamics (due to the use of a common average reference). However, it cannot be entirely excluded that additional lateralized response-related activity contributes to this positive correlation (Salisbury et al., 2001).”

2) Parts of the conclusions are based on a relationship between alpha-amplitude modulation and size of P300-amplitude (amplitude-amplitude) using data binning (illustrated in Fig 3) and the bins seem to include different participants, rather than trials. As this is an analysis of EEG data, I wonder how much of this relationship can be explained by a confound of skull thickness (or other individual differences in anatomy picked up with the scalp measures such as gyral folding patterns and current source orientations etc). E.g. those with thicker/thinner skulls are expected to show less/more of a modulation in all signals. This could be ruled out by relating the bins in alpha modulation not to the P300 but to another event that does not coincide in time with the alpha changes (e.g. P100), where no changes across bins would be expected.

We are grateful for the suggestions on confound estimation. We repeated the analysis of binning of alpha rhythm amplitude normalised change in relation to early ER, which in our auditory paradigm was N100. The largest change in the alpha amplitude occurs later in the poststimulus window, but that does not necessarily mean that the activity in the window right after the stimulus onset is unaffected. As can be seen in Figure 3 (t-statistics between alpha bins), there is already a significant difference around 100 ms over the central regions of the scalp. For this plot, the broadband data was filtered from 0.1 to 3 Hz, thus assessing only changes in low-frequency signals. We repeated the same analysis for broadband data (0.1–45 Hz) and also observed a significant difference between two extreme bins around 100 ms over the central region (Figure S5A). However, if we filter the signal from 4 to 45 Hz, these significant differences almost completely disappear (only electrode TP9 was significant; Figure S5B). Importantly, this range (4–45 Hz) includes the frequency of N100, which is typically in the alpha range. It means that the differences in N100 are riding on top of the baseline shift created by an unfolding alpha amplitude decrease. When this low-frequency baseline shift was removed, significant differences were no longer visible. This is an indication that differences in P300 amplitude between alpha bins are restricted to the low-frequency range and are not propagated to other ERs with higher frequency content.

We added Figure S5 to the Supplementary material and introduced it in the main text, the Results section, as follows:

“The cluster within the earlier window (100–200 ms) over central regions (Figure 3C) possibly reflects the previously shown effect of prestimulus alpha amplitude on earlier ERs (Brandt et al., 1991, Babiloni et al., 2008) but may also be a manifestation of BSM. We tested this assumption for early ER, which in our auditory task was N100. We repeated the binning analysis for broadband data (0.1–45 Hz) and also observed a significant difference between two extreme bins around 100 ms over the central region (Figure S5A). However, if we filter the signal from 4 to 45 Hz (the range that includes the frequency of N100 but not low-frequency baseline shifts), these significant differences almost completely disappear (only electrode TP9 was significant; Figure S5B). It means that the difference in N100 amplitudes over frontal sites is driven by the baseline shift created by an unfolding alpha amplitude decrease. The significant difference at the TP9 electrode possibly reflects a genuine physiological effect of alpha rhythm amplitude on the excitability of a neuronal network and, as a consequence, on the amplitude of ER (as opposed to the baseline-shift mechanism, where the alpha rhythm doesn’t affect the amplitude of ER but creates an additional component of ER; Iemi et al. 2019).”

3) Related to the above: I assume it can be ruled out that the relationship between baseline-shift index and P300 amplitude (also determined through binning, Fig 6) could be influenced by the above-mentioned confounds, given the inverse relationship?

As in previous studies alpha rhythm power was found to depend on the size of the head (Candelaria-Cook et al., Cerebral Cortex, 2022), we agree that the contribution of this confounding factor should be estimated (and we did estimate it). However, we would like to point out that we looked into dependencies based on ratios, which eliminates absolute units potentially being affected by head size, skull thickness, etc. For instance, the baseline-shift index is estimated as the Pearson correlation coefficient between the alpha rhythm envelope and low-frequency signal during the resting state. Therefore, multiplying the alpha amplitude envelope by an arbitrary scale would not cause the correlation to change. Nonetheless, for a subset of participants (1034 participants, mean age 69.8 years, 496 female), we had MRI data, from which we extracted total intracranial volume. For each electrode, we computed the Pearson correlation between the variable of interest and total intracranial volume. Variables of interest were the peak amplitude of P300, the attenuation-peak amplitude of alpha rhythm, alpha rhythm normalised amplitude (computed as ), and the magnitude of the baseline shift index (BSI). The p-value was set at Bonferroni corrected 0.05. For P300, only one electrode, namely C4, demonstrated a significant correlation of –0.10. However,the C4 electrode is outside of the typical electrode range for P300. For alpha envelope amplitude, significant correlations were observed all over the head (19 out of 31 electrodes, maximum at Cz), and a larger total intracranial volume was related to a higher amplitude of alpha rhythm.

Candelaria-Cook et al. (Cerebral Cortex, 2022) showed a similar association in longitudinal data from children and adolescents, but the increase in alpha rhythm power in that study might have been due to additional factors beyond a growing head. Conversely, normalised alpha amplitude showed no significant correlations. Similarly, the absolute value of BSI did not correlate significantly with total intracranial volume at any electrode. Overall, only alpha amplitude shows a prominent correlation to total brain volume, thus reducing the concern that head size may be a confound.

4) This study is based on a sample of older participants. One wonders to what extent this is needed to reveal the alpha-P300 relationships (e.g. more variability in this population than in younger controls), and/or whether other mechanisms may be at play across the lifespan.

Our study is indeed based on a sample of older participants. However, in our previous study (Studenova et al., PLOS Comp Bio, 2022), we compared young and elderly participants using resting-state data. There, we measured the baseline-shift index (BSI) at rest, and BSI serves as a proxy for baseline shifts present in the task-based data (under the assumptions of the baseline-shift mechanism, ER is in essence a baseline shift). We found that BSIs for elderly participants were smaller in comparison to those for young participants. Yet, the distribution of BSI values across the scalp (as in Figure 6A) was similar between the two age groups.

Additionally, we observed that larger alpha rhythm power was positively correlated with the magnitude of BSI, but only for younger participants, which points out possible difficulties arising from the fact that elderly people have reduced alpha power. Therefore, we believe that for a sample of young participants, the results should not be different.

5) Legend to Figure 6: sentence under A: "A positive deflection of P300 at posterior sites coincides with a decrease in alpha amplitude, a case that corresponds to negative mean oscillations." I find this sentence at this place in the legend confusing, as Fig 6A seems to illustrate the BSI only (not yet any relationship?).

We expanded the text in the legend with this paragraph:

“BSI serves as a proxy for the relation between ER polarity and the direction of alpha amplitude change (Nikulin et al., 2010). Here, we observe predominantly negative BSIs (and thus negative mean oscillations) at posterior sites, which indicates the inverted relation between P300 and alpha amplitude change. Indeed, in the task data, a positive deflection of P300 at posterior sites coincides with a decrease in alpha amplitude.”

6) Page 4: repetition of "has been" "has been" one after each other in the text We are thankful for this catch. We removed the repetition.

Reviewer #2 (Public Review):

The authors attempt to show that event-related changes in the alpha band, namely a decrease in alpha power over parieto/occipital areas, explain the P300 during an auditory target detection task. The proposed mechanism by which this happens is a baseline-shift, where ongoing oscillations which have a non-zero mean undergo an event-related modulation in amplitude which then mimics a low frequency event-related potential. In this specific case, it is a negative-mean alpha-band oscillation that decreases in power post-stimulus and thus mimics a positivity over parieto-occipital areas, i.e. the P300. The authors lay out 4 criteria that should hold if indeed alpha modulation generates the P300, which they then go about providing evidence for.

Strengths:

• The authors do go about showing evidence for each prediction rigorously, which is very clearly laid out. In particular, I found the 3rd section connecting resting-state alpha BSI to the P300 quite compelling.

• The study is obviously very well-powered.

• Very well-written and clearly laid out. Also, the EEG analysis is thorough overall, with sensible analysis choices made.

• I also enjoyed the discussion of the literature, albeit with certain strands of P300 research missing.

Weaknesses:

In general, if one were to be trying to show the potential overlap and confound of alpha-related baseline shift and the P300, as something for future researchers to consider in their experimental design and analysis choices, the four predictions hold well enough. However, if one were to assert that the P300 is "generated" via alpha baseline shift, even partially, then the predictions either do not hold, or if they do, they are not sufficient to support that hypothesis. This general issue is to be found throughout the review. I will briefly go through each of the predictions in turn:

1) The matching temporal course of alpha and P300 is not as clear as it could be. Really, for such a strong statement as the P300 being generated by alpha modulation, one would need to show a very tight link between the signals temporally. There are many neural and ocular signals which occur over the course of target detection paradigms: P300, alpha decrease, motor-related beta decrease, the LRP, the CNV, microsaccade rate suppression etc. To specifically go above and beyond this general set of signals and show a tighter link between alpha and P300 requires a deeper comparison. To start, it would be a good idea to show the signals overlapping on the same plot to really get an idea of temporal similarity. Also, with the P300-alpha correlation, how much of this correlation is down to EEG-related issues such as skull thickness, cortical folding, or cognitive issues such as task engagement? One could perhaps find another slow wave ERP, e.g. the Lateralised Readiness Potential, and see if there is a similar strength correlation. If there is not, that would make the P300 relationship stand out.

Thank you for this comment. In our study, we outline the prerequisites for the baseline-shift mechanism (BSM) and show how they hold for the obtained data. Overall, for all the prerequisites, the evidence could be found in favour of BSM. However, as it is the case for all EEG/MEG data, the non-invasive nature of the data puts constraints on the interpretation of the results. In order to specifically address the points raised by the reviewer about the results, we provide additional information about the overlap (Figure 2) and non-specific anatomical parameters.

The baseline-shift mechanism makes a general prediction about the generation of some ERs (those that coincide with a change in oscillatory amplitudes). The fact that neuronal oscillations (especially alpha oscillations) are modulated in almost any task indicates that other ERs can also contain a contribution from the baseline-shift mechanism. In our study, it is plausible that several sources of alpha oscillations orchestrated several ER components that appeared on the scalp after the presentation of a target stimulus. Due to the substantial spatial mixing and temporal overlap, it is difficult to disentangle the processes indexing perceptual, memory, or motor functions. However, currently, we are working on showing that the readiness potential (movement related potential) in the classical Libet’s paradigm also complies with the baseline-shift mechanism.

Concerns about confounds such as skull thickness are valid; therefore, we performed additional analysis. For a subset of participants (1034 participants, mean age 69.8 years, 496 female), we had MRI data, from which we extracted total intracranial volume. We tested the correlation between total intracranial volume and several variables of interest: the peak amplitude of P300, the attenuation-peak amplitude of alpha rhythm, alpha rhythm normalised change, and the magnitude of the baseline shift index (BSI). For P300 amplitude, only the C4 electrode showed a significant correlation of –0.10. For alpha envelope amplitude, there were significant correlations all over the head (19 out of 31 electrodes, maximum at Cz). The correlations showed that a larger total intracranial volume was related to a higher amplitude of alpha rhythm. For a normalised change in alpha amplitude, we observed no significant correlations. Similarly, the absolute value of BSI did not correlate significantly with total intracranial volume at any electrode. Overall, alpha amplitude indeed shows a prominent correlation to total brain volume, but none of the relational variables (normalised amplitude change, BSI) show any correlation.

In Figure 3, it is clear that alpha binning does not account for even 50% of the variance of P300 amplitude. Again, if there is such a tight link between the two signals, one would expect the majority of P300 variance to be accounted for by alpha binning. As an aside, the alpha binning clearly creates the discrepancy in the baseline period, with all alpha hitting an amplitude baseline at approx. 500ms. I wonder if could you NOT, in fact, baseline your slow wave ERP signal, instead using an appropriate high pass filter (see "EEG is better left alone", Arnaud Delorme, 2023) and show that the alpha binning creates the difference in ERP at the baseline which then is reinterpreted as a P300 peak difference after baselining.

The difference in the baseline window for alpha rhythm amplitude is indeed prominent (Figure R1A,B), so we proceed with the suggested analysis. Before anything else, we would like to reiterate that the baseline correction per se does not generate ER; it just moves the whole curve (in the pre- and poststimulus intervals) up and down. Firstly, we repeated the analysis without baseline correction (filter 0.1–3 Hz) and still observed the difference in P300 amplitude across bins (Figure R1D). Moreover, based on cluster-based permutation testing, ERs in the two most extreme bins were not significantly different in the prestimulus window. However, when we opt for no baseline correction, there will still be a baseline, namely, the average of the signal will be zero within a filtering window (e.g., 10 sec for a high-pass filter at 0.1 Hz). Thus, secondly, we computed an ER but with the baseline in the poststimulus window (400–600 ms; Figure R1E). In this case, the difference between bin 1 and bin 5 (for the prestimulus interval) in the window before 0 ms was significant in the posterior regions. The differences in the baseline are perceived as being smaller than the differences in alpha amplitude. This can be attributed to the fact that there are other low-frequency processes in the EEG signal that are different from alpha baseline shifts. Additionally, P300 in bin 1 in comparison with P300 in bin 5 is significantly different in shape (Figure R1C). This can be an indication of overlapping components; namely, for bin 5 (where alpha amplitude change is the highest), associated baseline shift dominates, and for bin 1 (where alpha amplitude change is the smallest), associated baseline shift is hidden behind other components. We believe that this proposed analysis demonstrates the intuition behind the baseline-shift mechanism: the baseline shift is generated due to a change in the oscillatory amplitude; and the change is simply the difference between two time points.

Author response image 1.

The difference in the strength of alpha amplitude modulation correlates with the difference in P300 amplitude. A. The alpha rhythm amplitude was binned according to the percentage of change. The bins were the following: (66, –25), (–25, –37), (–37, –47), (–47, –58), (–58,–89) % change. A is identical to Figure 3A, main text. B. The alpha rhythm amplitude is multiplied by –1 and evened within the prestimulus window. This may be an approximation for baseline shifts in the low-frequency signal. C. P300 responses are sorted into the corresponding bins. The C is identical to Figure 3B, main text. D. P300 are obtained without applying a baseline correction and are sorted into the corresponding bins. The difference in peak amplitude of P300 remains visible and significant. E. P300 is baselined at 400–600 ms. As a consequence, there are significant differences in the prestimulus window.

2) The topographies are somewhat similar in Figure 4, but not overwhelmingly so. There is a parieto-occipital focus in both, but to support the main thesis, I feel one would want to show an exact focus on the same electrode. Showing a general overlap in spatial distribution is not enough for the main thesis of the paper, referring to the point I make in the first paragraph re Weaknesses. Obviously, the low density montage here is a limitation. Nevertheless, one could use a CSD transform to get more focused topographies (see https://psychophysiology.cpmc.columbia.edu/software/csdtoolbox/), which apparently does still work for lower-density electrode setups (see Kayser and Tenke, 2006).

As we mentioned in our provisional response, we believe that we would not benefit from using CSD. First, the CSD transform is a spatial high-pass filter, and, hence, it is commonly used for spatially localised activities. In our case, we have two activities—P300 and alpha amplitude decrease—that are widespread with low spatial frequency, and we believe that applying CSD is not helpful. Second, CSD is more sensitive to surface sources that emanate from the crowns of gyri. For activity in the P300 window, there is a possibility that sources are localised within the longitudinal fissure. Third, as we completely agree that low density montage is a limitation, we used source reconstruction with eLoreta (Figure 5) to clarify the spatial localisation of the potential source of P300 and alpha amplitude change, which indeed shows a considerable spatial overlap.

3) Very nice analysis in Figure 6, probably the most convincing result comparing BSI in steady state to P300, thus at least eliminating task-related confounds.

4) Also a good analysis here, wherein there seem to be similar correlation profiles across P300 and alpha modulation. One analysis that would really nail this down would be a mediation analysis (Baron and Kenny, 1986; https://davidakenny.net/cm/mediate.htm), where one could investigate if e.g. the relationship between P300 amplitude and CERAD score is either entirely or partially mediated by alpha amplitude. One could do this for each of the relationships. To show complete mediation of P300 relationship with a cog task via alpha would be quite strong.

We agree that mediation analysis better suits the purpose of our claim. We added this analysis to the edited version of the manuscript. Additionally, we became concerned that the total alpha power effect may be driving the correlation. Therefore, we used alpha amplitude change in percentage instead of the absolute values of the amplitude. Significant mediation was present only for attention and executive scores.

In the updated version of the manuscript, the Methods section reads as follows:

“The correlation between cognitive scores (see Methods/Cognitive tests) and the amplitude and latency of P300 and alpha oscillations was calculated with linear regression using age as a covariate (R lme4, Bates et al., 2015). To estimate what proportion of the correlation between P300 and cognitive score is mediated by alpha oscillations, we used mediation analysis (Baron et al., 1986; R mediation, Tingley et al, 2014). First, we estimated the effect of P300 on the cognitive variable of interest (total effect, cogscore ~ P300+age). Second, we computed the association between P300 and alpha oscillations (the effect on the mediator, alpha ~ P300). Third, we run the full model (the effect of the mediator on the variable of interest, cogscore ~ P300+alpha+age). Lastly, we estimated the proportion mediated.”

The Results section reads as follows:

“Stimulus-based changes in brain signals are thought to reflect cognitive processes that are involved in the task. A simultaneous and congruent correlation of P300 and alpha rhythm to a particular cognitive score would be another evidence in favour of the relation between P300 and alpha oscillations. Moreover, if thus found, the correlation directions should correspond to the predictions according to BSM. Along with the EEG data, in the LIFE data set, a variety of cognitive tests were collected, including the Trail-making Test (TMT) A&B, Stroop test, and CERADplus neuropsychological test battery (Loeffler et al., 2015). From the cognitive tests, we extracted composite scores for attention, memory, and executive functions (Liem et al., 2017, see Methods/Cognitive tests) and tested the correlation between composite cognitive scores vs. P300 and vs. alpha amplitude modulation. The scores were available for a subset of 1549 participants (out of 2230), age range 60.03–80.01 years old. Cognitive scores correlated significantly with age (age and attention: −0.25, age and memory: −0.20, age and executive function: −0.23). Therefore, correlations between cognitive scores and electrophysiological variables were evaluated, regressing out the effect of age. To rule out the possibility of a absolute alpha power association with cognitive scores, for this analysis, we used alpha amplitude normalised change computed as , where 𝐴 𝑝𝑜𝑠𝑡 is at the latency of strongest amplitude decsease. Computed this way, negative alpha amplitude change would correspond to a more pronounced decrease, i.e., stronger oscillatory response.

To increase the signal-to-noise ratio of both P300 and alpha rhythm, we performed spatial filtering (see Methods/Spatial filtering, Figures 7B,C). Following this procedure, both P300 and alpha latency, but not amplitude, significantly correlated with attention scores (Figure 7A, left column). Larger latencies were related to lower attentional scores, which corresponded to a longer time-to-complete of TMT and Stroop tests and hence poorer performance. The proportion of correlation between P300 latency and attention, mediated by alpha attenuation peak latency, is 0.12. Memory scores were positively related to P300 amplitude and negatively to P300 latency (Figure 7A, middle column). The direction of correlation is such that higher memory scores, which reflected more recalled items, corresponded to a higher P300 amplitude and an earlier P300 peak. The association between alpha rhythm parameters and memory scores is not significant, but it goes in the same direction as the association for P300. Executive function (Figure 7A, right column) were related significantly to both P300 and alpha amplitude latencies. The proportion of correlation between P300 latency and attention, mediated by alpha attenuation peak latency, is 0.14. Overall, the direction of correlation is similar for P300 and alpha oscillations, as expected for BSM. Moreover, the direction of correlation is consistent across cognitive functions.

And an additional paragraph in the Discussion:

“The mediation analysis showed that the modulation of alpha oscillations only partially explained the correlation between P300 and cognitive variables. This, in general, corresponds to the idea that not the whole P300 but only its fraction can be explained by the changes in the alpha amplitudes. Figure 5 shows that alpha oscillations change not only in the cortical areas where P300 is generated; therefore, we cannot expect a complete correspondence between the two processes. Moreover, since cognitive tests and EEG recordings were performed at different time points, the associations between the cognitive variables and EEG markers are expected to be rather weak and to reflect only some neuronal processes common to P300, alpha rhythm, and tasks. For these reasons, a complete mediation of one EEG variable through another EEG variable in the context of a separate cognitive assessment cannot be expected.”

One last point, from the methods it appears that the task was done with eyes closed? That is an extremely important point when considering the potential impact of alpha amplitude modulation on any other EEG component due to the well-known substantial increase in alpha amplitude with eyes closed versus open. I wonder, would we see any of these effects with eyes opened?

The task was auditory and was indeed conducted in an eyes-closed state. In an eyes-closed state, alpha rhythm amplitude in the occipital regions shows a prominent increase. However, we believe that in our case, it was neither an advantage nor a disadvantage. First, occipital sources of alpha rhythm that demonstrate an increase in amplitude are not likely to be those sources that attenuate as a reaction to a target tone. The source reconstruction of alpha rhythm amplitude change (although with a limited number of channels) displayed widespread regions with a prominent decrease on the posterior midline, including the precuneus and posterior cingulate cortex (which contain polymodal association areas; Leech et al., Brain, 2014; Al-Ramadhani et al., Epileptic Disord, 2021). Second, in our previous study, we tested resting-state data with both eyes-closed and eyes-open conditions. There, we computed the baseline-shift index (BSI), which serves as an approximation for estimating if oscillations have a non-zero mean. We found no significant difference between the eyes-open and eyes-closed states in terms of the absolute value of the BSI. Moreover, the average distribution of BSIs on the scalp was the same for both conditions.

Overall, there is a mix here of strengths of claims throughout the paper. For example, the first paragraph of the discussion starts out with "In the current study, we provided comprehensive evidence for the hypothesis that the baseline-shift mechanism (BSM) is accountable for the generation of P300 via the modulation of alpha oscillations." and ends with "Therefore, P300, at least to a certain extent, is generated as a consequence of stimulus-triggered modulation of alpha oscillations with a non-zero mean." In the limitations section, it says the current study speaks for a partial rather than exhausting explanation of the P300's origin. I would agree with the first part of that statement, that it is only partial. I do not agree, however, that it speaks to the ORIGIN of the P300, unless by origin one simply means the set of signals that go to make up the ERP component at the scalp-level (as opposed to neural origin).

We have edited parts of the manuscript that have overly exuberant claims. However, we would argue further that alpha rhythm amplitude change does partially explain P300 origin. When a stimulus is being processed by the neuronal network, some part of this network presumably breaks from synchronous oscillation mode. Hence, on the scalp, we observe a decrease in oscillatory amplitude. According to the baseline-shift mechanism (BSM), this stimulus-related decrease in the amplitude generates the baseline shift in the frequency range of modulation (under 3 Hz for alpha rhythm). The P300 component that is explained by alpha rhythm amplitude modulation is, in essence, a baseline shift. Therefore, the origin of a part of P300 is the oscillating network that was pushed out of its synchronous oscillating regime.

Again, I can only make these hopefully helpful criticisms and suggestions because the paper is very clearly written and well analysed. Also, the fact that alpha amplitude modulation potentially confounds with P300 amplitude via baseline shift is a valuable finding.

Specific comments:

Perhaps give a brief overview of the task involved at the start. I know it is not particularly relevant, but I think necessary for those unfamiliar with cog tasks.

We added a short description of a task in the Introduction section.

“In this data set, the experimental task was an auditory oddball paradigm. Participants would hear tones, one type of which—the target tone—would occur in only 12% of trials. Target tones elicit both P300 and the modulation of the alpha amplitude. ”

Author response:

The following is the authors’ response to the original reviews.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

The authors tested whether learning to suppress (ignore) salient distractors (e.g., a lone colored nontarget item) via statistical regularities (e.g., the distractor is more likely to appear in one location than any other) was proactive (prior to paying attention to the distractor) or reactive (only after first attending the distractor) in nature. To test between proactive and reactive suppression the authors relied on a recently developed and novel technique designed to "ping" the brain's hidden priority map using EEG inverted encoding models. Essentially, a neutral stimulus is presented to stimulate the brain, resulting in activity on a priority map which can be decoded and used to argue when this stimulation occurred (prior to or after attending to a distracting item). The authors found evidence that despite learning to suppress the high probability distractor location, the suppression was reactive, not proactive in nature.

Overall, the manuscript is well-written, tests a timely question, and provides novel insight into a long-standing debate concerning distractor suppression.

Strengths (in no particular order):

(1) The manuscript is well-written, clear, and concise (especially given the complexities of the method and analyses).

(2) The presentation of the logic and results is mostly clear and relatively easy to digest.

(3) This question concerning whether location-based distractor suppression is proactive or reactive in nature is a timely question.

(4) The use of the novel "pinging" technique is interesting and provides new insight into this particularly thorny debate over the mechanisms of distractor suppression.

Weaknesses (in no particular order):

(1) The authors tend to make overly bold claims without either A) mentioning the opposing claim(s) or B) citing the opposing theoretical positions. Further, the authors have neglected relevant findings regarding this specific debate between proactive and reactive suppression.

(2) The authors should be more careful in setting up the debate by clearly defining the terms, especially proactive and reactive suppression which have recently been defined and were more ambiguously defined here.

(3) There were some methodological choices that should be further justified, such as the choice of stimuli (e.g., sizes, colors, etc.).

(4) The figures are often difficult to process. For example, the time courses are so far zoomed out (i.e., 0, 500, 100 ms with no other tick marks) that it makes it difficult to assess the timing of many of the patterns of data. Also, there is a lot of baseline period noise which complicates the interpretations of the data of interest.

(5) Sometimes the authors fail to connect to the extant literature (e.g., by connecting to the ERP components, such as the N2pc and PD components, used to argue for or against proactive suppression) or when they do, overreach with claims (e.g., arguing suppression is reactive or feature-blind more generally).

We thank the reviewer for their insightful feedback and have made several adjustments to address the concerns raised. To provide a balanced discussion, we tempered our claims about suppression mechanisms and incorporated additional references to opposing theoretical positions, including the signal suppression hypothesis, while clarifying the definitions of proactive and reactive suppression based on recent terminology (Liesefeld et al., 2024). We justified methodological choices, such as the slight size differences between stimuli to achieve perceptual equivalence and the randomization of target and distractor colors to mitigate potential luminance biases. We have revised our figure to enhance figure clarity. Lastly, while our counterbalanced design precluded reliable ERP assessments (e.g., N2pc, PD), we discussed their potential relevance for future research and ensured consistency with the broader literature on suppression mechanisms.

Reviewer #2 (Public Review):

Summary:

The authors investigate the mechanisms supporting learning to suppress distractors at predictable locations, focusing on proactive suppression mechanisms manifesting before the onset of a distractor. They used EEG and inverted encoding models (IEM). The experimental paradigm alternates between a visual search task and a spatial memory task, followed by a placeholder screen acting as a 'ping' stimulus -i.e., a stimulus to reveal how learned distractor suppression affects hidden priority maps. Behaviorally, their results align with the effects of statistical learning on distractor suppression. Contrary to the proactive suppression hypothesis, which predicts reduced memory-specific tuning of neural representations at the expected distractor location, their IEM results indicate increased tuning at the high-probability distractor location following the placeholder and prior to the onset of the search display.

Strengths:

Overall, the manuscript is well-written and clear, and the research question is relevant and timely, given the ongoing debate on the roles of proactive and reactive components in distractor processing. The use of a secondary task and EEG/IEM to provide a direct assessment of hidden priority maps in anticipation of a distractor is, in principle, a clever approach. The study also provides behavioral results supporting prior literature on distractor suppression at high-probability locations.

Weaknesses:

(1) At a conceptual level, I understand the debate and opposing views, but I wonder whether it might be more comprehensive to present also the possibility that both proactive and reactive stages contribute to distractor suppression. For instance, anticipatory mechanisms (proactive) may involve expectations and signals that anticipate the expected distractor features, whereas reactive mechanisms contribute to the suppression and disengagement of attention.

This is an excellent point. Indeed, while many studies, including our own, have tried to dissociate between proactive and reactive mechanisms, as if it is one or the other, the overall picture is arguably more nuanced. We have added a paragraph to the discussion on page 19 to address this. At the same time, (for more details see our responses to your comments 3 and 5), we have added a paragraph where we provide an alternative explanation of the current data in the light of the dual-task nature of our experiment.

(2) The authors focus on hidden priority maps in pre-distractor time windows, arguing that the results challenge a simple proactive view of distractor suppression. However, they do not provide evidence that reactive mechanisms are at play or related to the pinging effects found in the present paradigm. Is there a relationship between the tuning strength of CTF at the high-probability distractor location and the actual ability to suppress the distractor (e.g., behavioral performance)? Is there a relationship between CTF tuning and post-distractor ERP measures of distractor processing? While these may not be the original research questions, they emerge naturally and I believe should be discussed or noted as limitations.

Thank you for raising these important points. While CTF slopes have been shown to provide spatially and temporally resolved tracking of covert spatial attention and memory representations at the group level, to the best of our knowledge, no study to date has found a reliable correlation between CTFs and behavior. Moreover, the predictive value of the learned suppression effect, while also highly reliable at the group level, has been proven to be limited when it comes to individual-level performance (Ivanov et al. 2024; Hedge et al., 2018). Nevertheless, based on your suggestion, we explored whether there was a correlation between the averaged gradient slope within the time window where the placeholder revived the memory representation and the average distance slope in reaction times for the learned suppression effect. This correlation was not significant (r = .236, p = 0.267), which, considering our sample size and the reasons mentioned earlier, is not particularly surprising. Given that our sample size was chosen to measure group level effects, we decided not to include individual differences analysis it in the manuscript.

Regarding the potential link between the CTF tuning profile and post-distractor ERP measures like N2pc and Pd, our experimental design presented a specific challenge. To reliably assess lateralized ERP components like N2pc or Pd the high probability location must be restricted to static lateralized positions (e.g., on the horizontal midline). Our counterbalanced design (see also our response to comment 9 by reviewer 1), which was crucial to avoid bias in spatial encoding models, precluded such a targeted ERP analysis.

(3) How do the authors ensure that the increased tuning (which appears more as a half-split or hemifield effect rather than gradual fine-grained tuning, as shown in Figure 5) is not a byproduct of the dual-task paradigm used, rather than a general characteristic of learned attentional suppression? For example, the additional memory task and the repeated experience with the high-probability distractor at the specific location might have led to longer-lasting and more finely-tuned traces for memory items at that location compared to others.

Thank you for raising these important points. Indeed, a unique aspect of our study that sets it apart from other studies, is that the effects of learned suppression were not measured directly via an index of distractor processing, but rather inferred indirectly via tuning towards a location in memory. The critical assumption here, that we now make explicit on page 18, is that various sources of attentional control jointly determine the priority landscape, and this priority landscape can be read out by neutral ping displays. An alternative however, as suggested by the reviewer, is that memory representations may have been sharper when they remembered location was at the high probability distractor location. We believe this is unlikely for various reasons. First, at the behavioral level there was no evidence that memory performance differed for positions overlapping high and low probability distractor locations (also see our response to reviewer 3 minor comment 4). Second, there was no hint whatsoever that the memory representation already differed during encoding or maintenance (This is now explicitly indicated in the revised manuscript on page 14), which would have been expected if the spatial distractor imbalance modulated the spatial memory representations.

Nevertheless, as discussed in more detail in response to comment 5, there is an alternative explanation for the observed gradient modulation that may be specific to the dual nature of our experiment.

(4) It is unclear how IEM was performed on total vs. evoked power, compared to typical approaches of running it on single trials or pseudo-trials.

Thank you for pointing out that our methods were not clear. We did not run our analysis on single trials because we were interested in separately examining the spatial selectivity of both evoked alpha power (phase locked activity aligned with stimulus onset) and total alpha power (all activity regardless of signal phase). It is only possible to calculate evoked and total power when averaging across trials. Thus, when we partitioned the data into sets for the IEM analysis, we averaged trials for each condition/stimulus location to obtain a measurement of evoked and total power each condition for each set. This is the same approach used in previous work (e.g. Foster et al., 2016; van Moorselaar et al., 2018).

We reviewed our method section and can see why this was unclear. In places, we had incorrectly described the dimensions of training and test data as electrodes x trials. To address this, we’ve rewritten the “Time frequency analysis”, “Inverted encoding model” sections, and added a new “Training and test data” section. We hope that these sections are easier to follow.

(5) Following on point 1. What is the rationale for relating decreased (but not increased) tuning of CTF to proactive suppression? Could it be that proactive suppression requires anticipatory tuning towards the expected feature to implement suppression? In other terms, better 'tuning' does not necessarily imply a higher signal amplitude and could be observable even under signal suppression. The authors should comment on this and clarify.

We appreciate your highlighting of these highly relevant alternative explanations. In response, we have revised a paragraph in the General Discussion on page 18 to explicitly outline our rationale for associating decreased tuning with proactive suppression. However, in doing so, we now also consider the alternative perspective that proactive suppression might actually require enhanced tuning towards the expected feature to implement suppression effectively.

It's important to note that both of these interpretations – decreased tuning as a sign of suppression and increased tuning as a preparatory mechanism for suppression – diverge significantly from the commonly held model (including our own initial assumptions) wherein weights at the to-be-suppressed location are simply downregulated.

Minor:

(1) In the Word file I reviewed, there are minor formatting issues, such as missing spaces, which should be double-checked.

Thank you! We have now reviewed the text thoroughly and tried our best to avoid formatting issues.

(2) Would the authors predict that proactive mechanisms are not involved in other forms of attention learning involving distractor suppression, such as habituation?

Habituation is a form of non-associative learning where the response to a repetitive stimulus decreases over time. As such, we would not characterize these changes as “proactive”, as it only occurs following the (repeated) exposure to the stimulus. 

(3) A clear description in the Methods section of how individual CTFs for each location were derived would help in understanding the procedure.

Thank you. We have now added several sentences on page 27 to clarify how individual CTFs in Figure 3 and distance CTFs in Figure 5 are calculated.

“The derived channel responses (8 channels × 8 location bins) were then used for the following analyses: (a) calculating individual Channel Tuning Functions (CTFs) based on each of the eight physical location bins (e.g., Figure 3C and 3D); (b) grouping responses according to the distance between each physical location and the high-probability distractor location to calculate distance CTFs (e.g., Figure 5); and (c) averaging across location bins to represent the general strength of spatial selectivity in tracking the memory cue, irrespective of its specific location (e.g., Figure 3A and 3B).”

(4) Why specifically 1024 resampling iterations?

Thank you for your question. The statistical analysis was conducted using the permutation_cluster_1samp_test function within the MNE package in Python. We have clarified this on page 25. The choice of 1024 permutations reflects the default setting of the function, which is generally considered sufficient for robust non-parametric statistical testing. This number provides a balance between computational efficiency and the precision of p-value estimation in the context of our analyses.

Reviewer #3 (Public Review):

Summary:

In this experiment, the authors use a probe method along with time-frequency analyses to ascertain the attentional priority map prior to a visual search display in which one location is more likely to contain a salient distractor.  The main finding is that neural responses to the probe indicate that the high probability location is attended, rather than suppressed, prior to the search display onset.  The authors conclude that suppression of distractors at high-probability locations is a result of reactive, rather than proactive, suppression.

Strengths:

This was a creative approach to a difficult and important question about attention.  The use of this "pinging" method to assess the attentional priority map has a lot of potential value for a number of questions related to attention and visual search. Here as well, the authors have used it to address a question about distractor suppression that has been the subject of competing theories for many years in the field. The paper is well-written, and the authors have done a good job placing their data in the larger context of recent findings in the field.

Weaknesses:

The link between the memory task and the search task could be explored in greater detail. For example, how might attentional priority maps change because of the need to hold a location in working memory? This might limit the generalizability of these findings. There could be more analysis of behavioral data to address this question. In addition, the authors could explore the role that intertrial repetition plays in the attentional priority map as these factors necessarily differ between conditions in the current design. Finally, the explanation of the CTF analyses in the results could be written more clearly for readers who are less familiar with this specific approach (which has not been used in this field much previously).

We appreciate the reviewer's valuable feedback and have made significant revisions to address the concerns raised. To clarify the connection between the memory and search tasks, we conducted additional analyses to explore the effects of spatial distance between the memory cue location and the high-probability distractor location on behavioral performance. We also investigated the potential influence of intertrial repetition effects on the observed results by removing trials with location repetitions. To enhance clarity, we revised the explanation of the CTF analyses in the Results section and improved figure annotations to ensure accessibility for readers unfamiliar with this approach. Collectively, these updates further discuss how the pattern of CTF slopes reflect the interplay between memory and search tasks while addressing key methodological and interpretative considerations.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Suggestions/Critiques (in no particular order)

(1) The authors discuss the tripartite model (bottom-up, top-down, and selection history) but neglect recent and important discussions of why this trichotomy might be unnecessarily complicated (e.g., Anderson, 2024: Trichotomy revisited: A monolithic theory of attentional control). Simply put, one of the 3 pillars (i.e., selection history) likely does not fall into a unitary construct or "box"; instead, it likely contains many subcomponents (e.g., reward associations, stimulus-response habit learning, statistical learning, etc.). Since the focus of the current study is learned distractor suppression based on the statistical regularities of the distractor, the authors should comment on which aspects of selection history are relevant, perhaps by using this monolithic framework.

We appreciate the reviewer's insightful suggestion regarding theoretical frameworks of attentional control. While Anderson (2024) proposes a monolithic theory that challenges the traditional tripartite model, our study deliberately maintains a pragmatic approach. The main purpose of our experiment is empirically investigating the mechanisms of learned distractor suppression, rather than adjudicating between competing theoretical models.

We agree that selection history is not a unitary construct but comprises multiple subcomponents, including reward associations, stimulus-response habit learning, and statistical learning. In this context, our study specifically focuses on statistical learning as a key mechanism of distractor suppression. By explicitly acknowledging the multifaceted nature of selection history and referencing Anderson's monolithic perspective, we invite readers to consider the theoretical implications while maintaining our research's primary focus on empirical investigation. To this end, we have modified the manuscript to read (see page 3):

"The present study investigates the mechanisms underlying statistical learning, specifically learned distractor suppression, which represents one critical subcomponent of selection history. While theoretical models like the tripartite framework and the recent monolithic theory (Anderson, 2024) offer complementary perspectives on attentional control, our investigation focuses on empirically characterizing the statistical learning mechanisms underlying learned distractor suppression."

(2) The authors discuss previous demonstrations of location-based and feature-based learned distractor suppression. The authors admit that there have been a large number of studies but seem to mainly cite those that were conducted by the authors themselves (with the exception being Vatterott & Vecera, 2012). For example, there are other studies investigating location-based suppression (Feldmann-Wüstefeld et al., 2021; Sauter et al., 2021), feature-based suppression (Gaspelin & Luck, 2018a; Stilwell et al., 2022; Stilwell & Gaspelin, 2021; Vatterott et al., 2018), or both (Stilwell et al., 2019). The authors do not cite Gaspelin and colleagues at all in the manuscript, despite claiming that singleton-based suppression is not proactive.

We appreciate your pointing out the need for a more comprehensive citation of the literature on learned distractor suppression, particularly with respect to location-based and feature-based suppression. In response to your comment, we have now expanded the reference list on page 4 to include relevant studies that further support our discussion of both location-based and feature-based suppression mechanisms.

(3) The authors use the terms "proactive" and "reactive" suppression without taking into consideration the recent terminology paper, which one of the current authors, Theeuwes, helped to write (Liesefeld et al., 2024, see Figure 8). The terms proactive and reactive suppression need to be defined relative to a time point. The authors need to be careful in defining proactive suppression as prior to the first shift of attention, but after the stimuli appear and reactive suppression as after the first shift of attention and after the stimuli appear. Thus, the critical time point is the first shift of attention. Does suppression occur before or after the first shift of attention? The authors could alleviate this by using the term "stimulus-triggered suppression" to refer to "suppression that occurs after the distractor appears and before it captures attention" (Liesefeld et al., 2024).

Thank you for pointing out that this was insufficiently clear in the previous version. In the revised version we specifically refer to the recent terminology paper on page 5 to make clear that suppression could theoretically occur at three distinct moments in time, and that the present paper was designed to dissociate between suppression before or after the first shift of attention.

(4) Could the authors justify why the circle stimulus (2° in diameter) was smaller than the diamonds (2.3° x 2.3°)? Are the stimuli equated for the area? Or, for width and height? Doesn't this create a size singleton target on half of all trials (whenever the target is a circle) in addition to the lone circle being a shape singleton? Along these lines, could the authors justify why the colors were used and not equiluminant? This version of red is much brighter than this version of green if assessed by a spectrophotometer. Thus, there are sensory imbalances between the colors. Further, the grey used as the ping is likely not equiluminant to both colors. Thus, the grey "ping" is likely dimmer for red items but brighter for green items. Is this a fair "ping"?

Thank you for raising these important points. We chose, as is customary in this experimental paradigm (e.g., Huang et al., 2023; Duncan et al., 2023), to make the diamond slightly larger (2.3° x 2.3°) than the circle (2° in diameter) to ensure a better visual match in overall size appearance. If the circle and diamond stimuli were equated strictly in terms of size (both at 2°), the diamond would appear visually smaller due to the differences in geometric shape. By adjusting the dimensions slightly, we aimed to minimize any unintentional differences in perceptual salience.

As for the colors used in the experiment, the reviewer is right that there might be sensory imbalances between the red and green stimuli, with red appearing brighter than green based on measurements such as spectrophotometry. To ensure that any effects couldn’t be explained by sensory imbalance in the displays, we randomized target and distractor colors across trials, meaning that roughly half the trials had a red distractor and half had a green distractor. This randomization should have mitigated any systematic biases caused by color differences.

We appreciate your feedback and have clarified these points in method section in the revised manuscript on page 22:

"Please note that although the colors were not equiluminant, the target and distractor colors were randomized across trials such that roughly half the trials had a red distractor, and half had a green distractor. This randomization process should help mitigate any systematic biases this may cause."

(5) For the eye movement artifact rejection, the authors use a relatively liberal rejection routine (i.e., allowing for eye movements up to 1.2° visual angle and a threshold of 15 μV). Given that every 3.2 μV deviation in HEOG corresponds to ~ ± 0.1° of visual angle (Lins, et al., 1993), the current oculomotor rejection allows for eye movements between 0.5° and 1.2° visual angle to remain which might allow for microsaccades (e.g., Poletti, 2023) to contaminate the EEG signal (e.g., Woodman & Luck, 2003).

The reviewer correctly points out that our eye rejection procedure, which is the same as in our previous work (e.g., Duncan et al., 2023), still allows for small, but systematic biases in eye position towards the remembered location and potentially towards or away from the high probability distractor location. While we cannot indefinitely exclude this possibility, we believe this is unlikely for the following reasons. First, although there is a link between microsaccades and covert attention, it has been demonstrated that subtle biases in eye position cannot explain the link between alpha activity and the content of spatial WM (Foster et al., 2016, 2017). Specifically, Foster et al. (2017) found no evidence for a gaze-position-related CTF, while an analysis on that same data yielded clear target related CTFs. Similarly, within the present data set there was no evidence that the observed revival induced by the ping display could be attributed to systematic changes in gaze position, as a multivariate cross-session decoding analysis with x,y positions from the tracker did not yield reliable above-chance decoding of the location in memory.

Author response image 1.

(6) The authors claim that "If the statistically learned suppression was spatial-based and feature-blind, one would also expect impaired target processing at the high-probability location." (p. 7, lines 194-195). Why is it important that suppression is feature-blind here? Further, is this a fair test of whether suppression is feature-blind? What about inter-trial priming of the previous trial? If the previous trial's singleton color repeated RTs might be faster than if it switched. In other words, the more catastrophic the interference (the target shape, target color, distractor shape, distractor color) change between trials, the more RTs might slow (compared with consistencies between trials, such that the target and distractor shapes repeat and the target and distractor colors repeat). Lastly, given the variability across both the shape and color dimensions, the claim that this type of suppression is feature-blind might be an artifact of the design promoting location-based instead of feature-based suppression.

Thank you for raising this point. In the past we have used the finding that learned suppression was not specific to distractors, but also generalized to targets to argue in favor of proactive (or stimulus triggered) suppression. However, we agree that given the current experimental parameters it may be an oversimplification to conclude that the effect was feature-blind based on the impaired target processing as observed here. As this argument is also not relevant to our main findings, we have removed this interpretation and simply report that the effect was observed for both distractor and targets. Nevertheless, we would like to point out that while inter-trial priming could influence reaction times, the features of both target and distractors (shape and color) were randomly assigned on each trial. This should mitigate consistent feature repetitions effects. Additionally, previous research has demonstrated that suppression effects persist even when immediate feature repetitions are controlled for or statistically accounted for (e.g., Wang & Theeuwes 2018 JEP:HPP; Huang et al., 2021 PB&R).

(7) The authors should temper claims such as "suppression occurs only following attentional enhancement, indicating a reactive suppression mechanism rather than proactive suppression." (p. 15, lines 353-353). Perhaps this claim may be true in the current context, but this claim is too generalized and not supported, at least yet. Further, "Within the realm of learned distractor suppression, an ongoing debate centers around the question of whether, and precisely when, visual distractors can be proactively suppressed. As noted, the idea that learned spatial distractor suppression is applied proactively is largely based on the finding that the behavioral benefit observed when distractors appear with a higher probability at a given location is accompanied by a probe detection cost (measured via dot offset detection) at the high probability distractor location (Huang et al., 2022, 2023; Huang, Vilotijević, et al., 2021)." (p. 15, lines 355-361). Again, the authors should either cite more of the opposing side of the debate (e.g., the signal suppression hypothesis, Gaspelin & Luck, 2019 or Luck et al., 2021) and the many lines of converging evidence of proactive suppression) or temper the claims.

Thank you for your constructive feedback regarding our statements on suppression mechanisms. We acknowledge that our original claim was intended to reflect our specific findings within the context of this study and was not meant to generalize across all research in the field. To prevent any misunderstanding, we have tempered our claims to avoid overgeneralization by clarifying that our findings suggest a tendency toward reactive suppression within the specific experimental conditions we investigated (see page 17).

Furthermore, learned distractor suppression is multifaceted, encompassing both feature-based suppression (as proposed by the signal suppression hypothesis) and spatial-based suppression (as examined in the current study). The signal suppression hypothesis provides proactive evidence related to the suppression of specific feature values (Gaspelin et al., 2019; Gaspelin & Luck, 2018b; Stilwell et al., 2019). We have incorporated references to these studies to offer a more comprehensive perspective on the ongoing debate at a broader level (see page 17).

(8) "These studies however, mainly failed to find evidence in support of active preparatory inhibition (van Moorselaar et al., 2020, 2021; van Moorselaar & Slagter, 2019), with only one study observing increased preparatory alpha contralateral to the high probability distractor location (Wang et al., 2019)." (p. 15, lines 367-370). This is an odd phrasing to say "many studies" have shown one pattern (citing 3 studies) and "only" one showing the opposite, especially given these were all from the current authors' labs.

Agreed. We have rewritten this text on page 17.

“These studies however, failed to find evidence in support of active preparatory inhibition as indexed via increased alpha power contralateral to the high probability distractor location  (van Moorselaar et al., 2020, 2021; van Moorselaar & Slagter, 2019; but see Wang et al., 2019).”

(9) Could the authors comment on why total power was significantly above baseline immediately (without clearer timing marks, ~10-50 ms) after the onset of the cue (Figure 3)? Is this an artifact of smearing? Further, it appears that there is significant activity (as strong as the evoked power of interest) in the baseline period of the evoked power when the memory item is presented on the vertical midline in the upper visual field (this is also true, albeit weaker, for the memory cue item presented on the horizontal midline to the right). This concern again appears in Figure 4 where the Alpha CTF slope was significantly below or above the baseline prior to the onset of the memory cue. Evoked Alpha was already significantly higher than baseline in the baseline period. In Figure 5, evoked power is already higher and different for the hpl than the lpls even at the memory cue (and before the memory cue onsets). There are often periods of differential overlap during the baseline period, or significant activity in the baseline period or at the onset of the critical, time-locked stimulus array. The authors should explain why this might be (e.g., smearing).

Thank you for pointing this out. As suggested by the reviewer, this ‘unexpected’ pre-stimulus decoding is indeed the result of temporal smearing induced by our 5th order Butterworth filter. The immediate onset of reliable tuning (sometimes even before stimulus onset) is then also a typical aspect of studies that track tuning profiles across time in the lower frequency bands such as alpha (van Moorselaar & Slagter 2019; van Moorselaar et al., 2020; Foster et al., 2016).

Indeed, visual inspection also suggests that evoked activity tracked items at the top of the screen, an effect that is unlikely to result from temporal smearing as it is temporally interrupted around display onset. However, it is important to note that CTFs by location are based on far fewer trials, making them inherently noisier. The by-location plots primarily serve to show that the observed pattern is generally consistent across locations. In any case, given that the high probability distractor location was counterbalanced across participants it did not systematically influence our results.

(10) Given that EEG was measured, perhaps the authors could show data to connect with the extant literature. For example, by showing the ERP N2pc and PD components. A strong prediction here is that there should be an N2pc component followed by a PD component if there is the first selection of the singleton before it is suppressed.

Thank you for your great suggestion regarding the analysis of ERP components such as N2pc and Pd. To reliably assess lateralized ERP components like N2pc or Pd the high probability location must be restricted to static lateralized positions (e.g., on the horizontal midline such as Wang et al., 2019). In contrast, our study was designed to utilize an inverted encoding model to investigate the mechanisms underlying spatial suppression. To avoid bias in training the spatial model toward specific spatial locations (see also the previous comment), we counterbalanced the high-probability location across participants, ensuring an equal distribution of high-probability locations within the sample. Given this counterbalanced design, it was not feasible to reliably assess these components within the scope of the current study. Yet, we agreed with the reviewer that it would be of theoretical interest to examine Pd and N2pc evoked by the search display, particularly in this scenario where suppression has been triggered prior to search onset.

(11) Figure 2 (behavioral results) is difficult to see (especially the light grey and white bars). A simple fix might be to outline all the bars in black.

Thank you! We have incorporated your suggestion by outlining all the bars on page 10.

Reviewer #3 (Recommendations For The Authors):<br /> (1) I'm wondering about the link between the memory task and the search task.  I think the interpretation of the data should include more discussion of the fact that much of the search literature doesn't involve simultaneously holding an unrelated location in memory.  How might that change the results?

For example - what happens behaviorally on the subset of trials in which the location to be held in memory is near the high probability distractor location?  All the behavioral data is more or less compartmentalized, but I think some behavioral analysis of this and related questions might be quite useful.  I know there are comparisons of behavior in single vs. dual-task cases (for the memory task at least), but I think the analyses could go deeper.

Thank you for your great suggestion. To investigate the potential interactions between the spatial memory task and the visual search task, we conducted additional analyses on the behavioral data. First, we examined whether memory recall was influenced by the spatial distance (dist0 to dist4) between the memory cue location and the high-probability distractor location. As shown in the figure below, memory recall is not systematically biased either toward or away from the high-probability distractor location (p = .562, ηp² = .011).

We also assessed how the memory task might affect search performance. Specifically, we plotted reaction times as a function of the spatial overlap between the memory cue location and any of the search items, separating trials by distractor-present (match-target, match-distractor, match-neutral) and distractor-absent (match-target, match-neutral) conditions. Although visually the result pattern seems to suggest that search performance was facilitated when the memory cue spatially overlapped with the target and interfered with when it overlapped with the distractor, this pattern did not reach statistical significance (distractor-present: p = .249, ηp² = .002; distractor-absent: p = .335, ηp² = .002). We have now included these analyses in our supplemental material.

Beyond additional data analyses, there are also theoretical questions to be asked.  For example, one could argue that in order to maintain a location near or at the high probability distractor location in working memory, the priority map would have to shift substantially. This doesn't necessarily mean that proactive suppression always occurs in search when there is a high probability location. Instead, one could argue that when you need to maintain a high probability location in memory but also know that this location might contain a distractor, the representation necessarily looks quite different than if there were no memory tasks.  Maybe there are reasons against this kind of interpretation but more discussion could be devoted to it in the manuscript. I guess another way to think of this question is - how much is the ping showing us about attentional priority for search vs. attentional priority for memory, or is it simply a combination of those things, and if so, how might that change if we could ping the attentional priority map without a simultaneous memory task?

Thank you for this valuable suggestion. The aim of our study was to explore how the CTFs elicited by the memory cue were influenced by the search task. We employed a simultaneous memory task because directly measuring CTFs in relation to the search task was not feasible, as the HPL typically does not vary within individual participants. Consequently, CTFs locked to placeholder onsets could reflect arbitrary differences between (subgroups of) participants rather than true differences in the HPL. To address this, we combined the search task with a VWM task, leveraging the fact that location-specific CTFs can reliably be elicited by a memory cue and that the location of this cue relative to the HPL can be systematically varied within participants (Foster et al., 2016, 2017; van Moorselaar et al., 2018). This approach allowed us to examine the CTFs elicited by the memory cue and how these were modulated by their distance from the HPL.

While it is theoretically possible that the observed changes resulted from alterations in how the memory cue was maintained in memory only, this explanation seems unlikely, for memory performance (recall) did not vary as a function of the cue's distance from the HPL, suggesting that the distance-related changes in the CTFs are reflections of both tasks. Moreover, distractor learning typically occurs without awareness (Gao & Theeuwes 2022; Wang & Theeuwes 2018). It is difficult to understand how such unconscious processes could lead to anticipations in the memory task and subsequently modulate the representation of the consciously remembered memory cue only. We therefore believe that if we would have pinged the attentional priority map without a simultaneous memory task, the results would have been similar to those obtained in the present experiment, indicating stronger tuning at the HPL. Yet, this work still needs to be done.

To address this comment, we have added a paragraph on p. 18:

“However, two alternative explanations warrant consideration. First, one could argue that observed modulations in the revived CTFs do not provide insight into the mechanisms underlying distractor suppression but instead reflect changes in the memory representation itself, potentially triggered by the anticipation of the HPL in the search task. According to this view, the changes in the revived CTFs would be unrelated to how search performance (in particular distractor suppression) was achieved. While this is theoretically possible, we believe it to be unlikely. Memory performance (recall) did not vary as a function of the cue's distance from the HPL, whereas the revived CTFs did, indicating that these changes likely reflect contributions from both tasks. Additionally, distractor learning typically occurs without conscious awareness (Gao & Theeuwes 2022; Wang & Theeuwes 2018). It is difficult to conceive how such unconscious processes could produce anticipatory effects in the memory task and selectively modulate the representation of the consciously remembered memory cue. Second, the apparent lack of suppression and the presence of a pronounced tuning at the high-probability distractor location could actually reflect a proactive mechanism that manifests in a way that seems reactive due to the dual-task nature of our experiment.”

(2) When the distractor appears at a particular location with a high probability it necessarily means that intertrial effects differ between high and low probability distractor locations.  Consecutive trials with a distractor at the same location are far more frequent in the high probability condition.  You may not have enough power to look at this, and I know this group has analyzed this behaviorally in the past, but I do wonder how much that influences the EEG data reported here.  Are CTFs also sensitive to distractors/targets from the most recent trial?  And does that contribute to the overall patterns observed here?

Thank you for your thoughtful comment. Indeed, Statistical distractor learning studies naturally involve a higher proportion of intertrial effects for high-probability distractors compared to low-probability ones. Previous research, including the present study, has demonstrated that while distractor location improves performance—shown by faster response times (t(23) = 6.32, p < .001, d = 0.33) and increased accuracy (t(23) = 4.21, p < .001, d = 0.86)—intertrial effects alone cannot fully account for the learned suppression effects induced by spatial distractor imbalances. This analysis in now reflected in the revised manuscript on page 9.

However, as noted by the reviewer, this leaves uncertain to what extent the neural indices of statistical learning, in this case the modulation of channel tuning functions, capture the effects of interest beyond the contributions of intertrial priming. To address this issue, one possible approach is to rerun the CTF analysis after excluding trials with location repetitions. Since the distractor location is unknown to participants at the time the CTF is revived by the placeholder, we removed trials where the memory cue location repeated the distractor location from the preceding trial, rather than trials with distractor location repetitions between consecutive trials. Our analyses indicate that after trials removal (~ 9% of overall trials), the spatial gradient pattern in the CTF slopes remains similar. However, the cluster-based permutation analysis fails to reveal any significant findings, and a one-sample t-test on the slopes averaged within the 100 ms time window of interest yields a p-value of 0.106. While this could suggest that the current pattern is influenced by distractor-cue repetition, it is more likely that the trial removal resulted in an underpowered analysis. To investigate this, we randomly removed an equivalent number of trials (9%), which similarly resulted in insignificant findings, although the overall result pattern remained comparable (p = 0.066 for the one-sample t-test on the slopes average within the interested time window of 100 ms).

Author response image 2.

Also, in our previous pinging study we observed that, despite the trial imbalance, decoding was approximately equal between high probability trailing (i.e., location intertrial priming) and non-trailing trials, suggesting that the ping is able to retrieve the priority landscape that build up across longer timescales.

(3) Maybe there is too much noise in the data for this, but one could look at individual differences in the magnitude of the high probability distractor suppression and the magnitude of the alpha CTF slope.  If there were a correlation here it would bolster the argument about the relationship between priority to the distractor location and subsequent behavior reduction of interference from that distractor.  

Thank you for this valuable suggestion. We investigated whether there was a correlation between the average gradient slope during the time window in which the placeholder revived the memory representation and the average distance slope in reaction times for the learned suppression effect. This correlation was not significant (r = .236, p = 0.267), which is perhaps expected given the potential noise levels, as noted by the reviewer. Furthermore, while the learned suppression effect is robust at the group level, its predictive value for individual-level performance has been shown to be limited (Ivanov et al., 2024; Hedge et al., 2018). Consequently, we chose not to include this analysis in the manuscript (see also our response to comment 2 by reviewer 2).

(4) The results sections are a bit dense in places, especially starting at the bottom of page 11.  For readers who are familiar with the general questions being asked but less so with the particular time-frequency analyses and CTF approaches being used (like myself), I think a bit more time could be spent setting up these analyses within the results section to make extra clear what's going on.

Thank you for your feedback regarding the clarity of our Results section. We have revised this section to make it more understandable and easier to follow, especially for readers who may be less familiar with the specific time-frequency analyses and modeling approaches used in our study. Specifically, we have provided additional interpretations alongside the reported results from page 10 to page 13 to aid comprehension and ensure that the methodology and findings are accessible to a broader audience. Additionally, we have revised the figure notes to further enhance clarity and understanding.

Other comments:

Abstract: "a neutral placeholder display was presented to probe how hidden priority map is reconfigured..."  i think the word "the" is missing before "priority map"

Thank you. We have added the word “the” before “hidden priority map”.

p. 4, Müller's group also has a number of papers that demonstrate how learned distractor regularities impact search (From the ~2008-2012 range, probably others as well), it might be worth citing a few here.

Thank you for your suggestion. In the revised manuscript, we have added citations to several key papers from Muller’s group on page 4 as well as other research groups.

p.5 - Chang et al. (2023) seems highly relevant to the current study (and consistent with its results) - depending on word limits, it might make sense to expand the description of this in the introduction to make clear how the present study builds upon it

Thank you! We have expanded the discussion of Chang et al. (2023) on page 5 to provide more detailed elaboration of their study and its relevance to our work.

p. 7 - maybe not for the current study, but I do wonder whether the distortion of spatial memory by the presence of the search task occurs only when there is a relevant regularity in the search task. In other words, if the additional singleton task had completely unpredictable target and distractor locations, would there be memory distortions?  Possibly for the current dataset, the authors could explore whether the behavioral distortion is systematically towards or away from the high probability distractor location.

Thank you for your insightful suggestion. Following your recommendation, we conducted an additional analysis to examine memory recall as a function of the distance between the memory cue location and the high-probability distractor location. Figure S1A illustrates the results, depicting memory recall deviation across various distances (dist0 to dist4) from the high-probability distractor location.

Our statistical analysis indicates that memory recall is not systematically biased either towards or away from the high-probability distractor location (p = .562, η_p² = .011). This finding suggests that spatial memory recall remains relatively stable and is not heavily influenced by the presence of regularities in the distractor locations.

p. 7 - in addition to stats it would be helpful to report descriptive statistics for the high probability vs. other distractor location comparisons

Thank you! We have added descriptive statistics on page 8 and page 9.

p. 19, "64%" repeated unnecessarily - also, shouldn't it be 65% if it's 5% at each of the other seven locations?

Thank you. This is now corrected in the revised manuscript.

p. 20 "This process continued until participants demonstrated a thorough understanding of the assigned tasks" Were there objective criteria to measure this?

Thank you for pointing out this issue. To clarify, objective criteria were indeed used to assess participants’ readiness to proceed. Specifically:

For the training phase practice trials, participants were required to achieve an average memory recall deviation of less than 13°.

For the test phase practice trials, participants needed to demonstrate a minimum of 65% accuracy in the search task. In addition, participants were asked to verbally confirm their understanding of the task goals with the experimenter before proceeding.

We have revised the manuscript to clearly indicate these criteria on p. 23.

p. 21 "P-values were Greenhouse-Geiser corrected in case where the..." I think "case" should be "cases"

Thank you. We have corrected this in the revised manuscript.

Author Response

The following is the authors’ response to the original reviews.

Reviewer 1

We thank the reviewer for their thoughtful comments. We have addressed them below, and we believe that have significantly strengthened the clarity of the manuscript.

Main Comments:

In Fig. 2C-D, I am not sure I understand why ≈ 100 mutations fix with β = 0. In the absence of epistasis, and since the coefficients hi are sampled from a symmetric distribution centered at zero, it is to be expected that roughly half of the mutations will have positive fitness effects and thus will eventually fix in the population. With L = 250, I would have expected to see the number of fixed mutations approach ≈ 125 for β = 0. Perhaps I am missing something?

• In our simulations, we initialize all populations from a state where there are only 100 available beneficial mutations (i.e., the initial rank is always 100). Without epistasis, these initial beneficial mutations are the only beneficial mutations that will be present throughout the entire trajectory. Hence, for β = 0, only 100 beneficial mutations can fix. Previously, this information could be found in the “Materials and methods” section of the SI. To make this aspect of our simulation more clear in the revision, we have added a discussion of the initial rank to the “Landscape structure” subsection of the model definition section. In addition, we have merged “Materials and methods” with “Further simulation details” in the SI into one section, and have listed the values for the simulation parameters in the model definition section.

Along these lines, the authors show that increasing β leads to a higher number of fixed mutations. I am not sure I understand their explanation for this. In line 209 they write that as β increases, “mutations are needed to cease adaptation”. The way I see it, in the absence of epistasis the fitness peak should correspond to a genotype with ≈ L/2 mutations (the genotype carrying all mutations with hi > 0). Increasing the magnitude of microscopic epistasis (i.e., increasing β ), and assuming that there is no bias towards positive epistasis (which there shouldn’t be based on the model formulation, i.e., section "Disorder statistics" on page 4), can change the “location” of the fitness peak, such that it now corresponds to a different genotype. Statistically speaking, however, there are more genotypes with L/2 mutations than with any other number of mutations, so I would have expected that, on average, the number of mutations fixed in the population would still have been ≈ L/2 (naturally with somewhat large variation across replicates, as seems to be the case).

• With epistasis, the situation becomes more complex. The structure of our model imposes significant sign epistasis in general (i.e. mutations can be beneficial on one background genotype and deleterious on another). This means that in the presence of epistasis, more than 100 mutations can be required to reach a local optimum even when the initial rank was 100. Intuitively, this occurs because mutations that were deleterious on the ancestral background genotype can become beneficial on future genotypes. We find that this occurs consistently throughout adaptation, leading to the accumulation of more mutations with increasing epistasis.

• Please note that we use the value L = 1000 in our simulations. We have also made the fact that we use L = 1000 more clear by moving the description of the simulation parameters to the main text.

I do see how, in the clonal interference regime, there can be multiple genotypes in the population at a given time (each with a different mutational load), thus making the number of fixed mutations larger than L/2 when aggregating over all genotypes in the population. But this observation makes less intuitive sense to me in the SSWM regime. In lines 207-208, the authors state that “as beta increases, a greater number of new available beneficial mutations are generated per each typical fixation event”. While this is true, it is also the case that a greater number of mutations that would have been beneficial in the absence of epistasis are now deleterious due to negative epistasis (if I am understanding what the authors mean correctly).

• The reviewer is correct to note that in the strong clonal interference regime, there will be more accumulated mutations across the entire population than in any single strain. However, we report the number mutations that have fixed, i.e., become present in the entire population.

• We find that the typical decrease in rank (per fixation event) of the population decreases with increasing epistasis — i.e., the number of available beneficial mutations that are “consumed” when a mutation fixes is typically lower in systems with stronger epistasis.

Similarly, I am not sure I understand how one goes from equation (6) to equation (7). In particular, it would seem to me that the term 4αiαj Ji j in equation (6) should be equally likely to be positive or negative (again assuming no bias towards positive Ji j). I thus do not see why ηi j in equation (7) is sampled from a normal distribution with mean µβ instead of just mean zero.

• The reviewer is correct that, for a uniformly random initial state, αi , αj , and Ji j will be uncorrelated so that the distribution of 4αiαj Ji j can be computed exactly (and has mean zero). However, we initialize from a state with rank 100, so that we need to compute the distribution of the random variable E[αiαj Ji j|αiαj Ji j > 0, R = 100]. This is mathematically very challenging, because there are nontrivial correlations between spins even at initialization. For these reasons, we found the uniformly random approximation insufficient. This is described in the paragraph following Equation (7) in the resubmission.

Minor Comments:

The authors use a model including terms up to second-order epistasis. To be clear, I think this choice is entirely justified: as they mention in their manuscript, this structure allows to approximate any fitness model defined on a Boolean hypercube. As I understand it, the reason for not incorporating higher-order terms (as in e.g. Reddy and Desai, eLife 2021) has to do with computational efficiency, i.e., accommodating higher-order terms in equation (10) may lead to a substantial increase in computation time. Is this the case?

• The author is correct that the incorporation of higher-order terms leads to significantly more expensive computation. It’s an interesting direction of future inquiry to see if our adaptive fast fitness computation method can be extended to higher-order interactions.

Reviewer 2

We would like to thank the reviewer for their careful reading and their useful comments connecting our work to spin glass physics. We believe the resulting additions to the paper have made our contributions stronger, and that they reveal some novel connections between the substitution trajectory and correlation functions in spin glasses. A summary of our investigation is provided below, and we have added two paragraphs to the discussion section under the heading “Connections to spin glass physics”.

Main Comments:

In spin glasses, slowdown of dynamics could have contributions from stretched exponential relaxation of spin correlations as well as aging, each of which are associated with their own exponents. In the present model, these processes could be quantified by computing two-point correlations associated with genomic overlap, as a function of lag time as well as waiting time (generation number). The population dynamics of competing strains makes the analysis more complicated. But it should be possible to define these correlations by separately averaging over lineages starting from a single parent genome, and over distinct parent genomes. It would be interesting to see how exponents associated with these correlations relate to the exponent c associated with asymptotic fitness growth.

• To investigate this point, we first considered the two-point correlation function 〈αi (tw)αi (tw+ ∆t)〉 for waiting time tw and lag time ∆t. Because all spins are statistically identical, it is natural to average this over the spin index i, leading to the quantity

Viewed as a function of ∆t for any fixed tw, it is clear that . If m mutations with respect to α(tw) have fixed at time tw + ∆t, a similar calculation shows that . Surprisingly, this simple derivation reveals that the two-spin correlation function commonly studied in spin glass physics is an affine transformation of the substitution trajectory commonly studied in population genetics. Moreover, it shows that the effect of tw is to change the definition of the ancestral strain, so that we may set tw = 0 without loss of generality and study the correlation function χ2(t) = 1 − 2m(t) where m(t) is the mean substitution trajectory of the population. Much of our analysis proceeds by analyzing the effect of epistasis on the accumulation of mutations. This relation provides a novel connection between this analysis and the analysis of correlation functions in the spin glass literature.

• It is well known that in the SSWM limit without epistasis, the substitution trajectory follows a power law similar to the fitness trajectory with relaxation exponent 1.0 [1]. Informed by this identity, we performed simulations in the SSWM limit and fit power laws to the correlation function χ2 as a function of time. We have verified that χ2(t) obeys a power- law relaxation with exponent roughly 1.0 for β = 0; moreover, as anticipated by the reviewer, the corresponding exponent decreases with increasing β . Nevertheless, we find that these relaxation exponents are distinct from those found for the fitness trajectory, despite following the same qualitative trend. This point is particularly interesting, as it highlights that the dynamics of fixation induce a distinct functional form at the level of the correlation functions when compared to, for example, the Glauber dynamics in statistical physics.

The strength of dynamic correlations in spin glasses can be characterized by the four-point susceptibility, which contains information about correlated spin flips. These correlations are maximized over characteristic timescales. In the context of evolution, such analysis may provide insights on the correlated accumulation of mutations on different sets of loci over different timescales. It would be interesting to see how these correlations change as a function of the mutation rate as well as the strength of epistasis.

• To study this point, we considered the four-point correlation function

Because spins are statistically identical, we found numerically that the genotype average is roughly equivalent to the angular average over trajectories. Inter-changing the order of the summation and the angular averaging, we then find that

so that the information contained in the four-point correlation function is the same as the information contained in the two-point correlation function.

Fig. 2E and Fig. 5 together suggests an intriguing possibility when interpreted in the spin glass context. It is clear that in the absence of epistasis, clonal interference accelerates fitness growth. Fig. 2E additionally suggests that this scenario will continue to hold even in the presence of weak, but finite epistasis, but disappears for sufficiently strong epistasis. I wonder if the two regimes are separated by a phase transition at some non-trivial strength of epistasis. Indeed, the qualitative behavior appears to change from that of a random field Ising spin glass for small β , to that of a zero field Sherrington-Kirkpatrick spin glass for sufficiently large β . While the foregoing comments are somewhat speculative, perhaps a discussion along these lines, and what it means in the context of evolution could be a useful addition to the discussion section of the paper.

• We thank the reviewer for this interesting suggestion, and we have added a discussion of this point to the text in the future directions section, lines 483–489.

Minor Comments:

1. In the abstract (line 17-18), I recommend use of the phrase "a simulated evolving population" to avoid a possible misinterpretation of the work as experimental as opposed to numerical.

• We have added the word “simulated”.

1. In line 70, the word "the" before "statistical physics" is redundant.

• We have removed “the”.

1. To make the message in lines 294-295 visually clear, I recommend keeping the Y-axis scale bars constant across Fig. 4A and Fig. 4B.

• We appreciate the suggestion. However, we found that when putting the two figures on the same scale, because the agreement is only qualitative and not quantitative (as emphasized in the text), it becomes difficult to view the trend in both systems. For this reason, we have chosen to keep the figure as-is.

1. Fig. 6 caption states: "Without epistasis, the rank decreases with increasing µ". It should be "rank increases".

• We have fixed this.

1. In the last sentence in the caption to Fig. 8, the labels "(A, β =0)" and "(B, β =0.25)" need to be swapped.

• We have fixed this.

Editor Comments

We thank the editor for pointing our attention towards these three interesting references, in particular the second, which appears most relevant to our work. We have added a discussion of reference 2 in the future directions section (lines 471–482), commenting on how to determine the contribution of within-path clonal interference to the fitness dynamics in our model. We have also added a reference to article 3 in the model description, commenting on the importance of sign epistasis and the prevalence of sign epistasis in our model with β > 0.

References:

1. Good BH, Desai MM. The impact of macroscopic epistasis on long-term evolutionary dynamics. Genetics. 2015.

Author Response

The following is the authors’ response to the original reviews.

Reviewer #1 (Public Review):

The enteroviruses comprise a medically important genus in the large and diverse picornavirus family, and are known to be released without lysis from infected cells in large vesicles containing numerous RNA genome-containing capsids - a feature allowing for en bloc transmission of multiple viral genomes to newly infected cells that engulf these vesicles. SIRT-1 is an NAD-dependent protein deacetylase that has numerous and wide ranging effects on cellular physiology and homeostasis, and it is known to be engaged in cellular responses to stress and autophagy.

Jassey et al. show that RNAi depletion of SIRT-1 impairs the release of enterovirus D-68 (EVD68) in EVs recovered from the supernatant fluids of infected cells using a commercial exosome isolation kit. The many functions attributed to SIRT-1 in the literature reflect its capacity to deacetylate various cell proteins engaged in transcription, DNA repair, and regulation of metabolism, apoptosis and autophagy. However, Jassey et al. make the surprising claim that the proviral role of SIRT-1 in promoting enterovirus release is not dependent on its deacetylase activity. Fig. S1C is crucial to this suggestion, as it is said to show that reconstituting expression with a catalytically-inactive mutant can rescue virus release from SIRT-1 depleted cells. However, no information is provided concerning the levels of endogenous and ectopicallyexpressed SIRT-1 proteins in this experiment, making it very difficult to interpret the results. Is the mutant SIRT-1 protein expressed at a higher level than the non-mutant protein? Is there a 'sponging' effect with these transfections that lessens the siRNA efficiency and reduces knockdown of the endogenous protein? Fig. S1B and Fig. 4C convincingly show that EX527, a small molecule inhibitor of the deacetylase activity of SIRT-1, inhibits extracellular release of the virus. This suggests that the deacetylase activity of SIRT-1 is in fact required for the proviral effect of SIRT-1. This is a fundamentally important question that will require more investigation.

We have included western blot data (Fig. S1D), which shows comparable levels of expression between the wild-type and mutant SIRT-1 constructs as well as the endogenous SIRT-1. While both constructs partially rescued EV-D68 titers in SIRT-1 knockdown cells, only the wild-type construct rescued SERCA2A protein levels, indicating that SIRT-1 deacetylase activity is required for SERCA2A expression but not for EV-D68 infection.

Fig. 6 shows how SIRT-I knockdown impacts the release of enterovirus D68 in EVs recovered from cell culture supernatant using a commercial 'Total Exosome Isolation Kit'. The authors should describe the principle this kit exploits to isolate 'exosomes' (affinity isolation?) and specify which antibodies it involves (anti-phosphatidylserine, anti-CD63, others?) This could impact the outcome of these experiments, and moreover is important to include in the longterm scientific record. The authors are appropriately cautious in describing the vesicles they presume to be isolated by the kit as simply 'extracellular vesicles', since there are multiple types of EVs with very different mechanisms of biogenesis, of which 'exosomes' are but one specific type. It would have been more elegant had the authors shown that SIRT-1 is required for EVD68 release in detergent-sensitive vesicles with low buoyant density in isopycnic gradients, and to characterize the size and number of viral capsids in these vesicles by electron microscopy.

We have added a description of the Total Exosome Isolation Kit principle to the materials and methods. The reagent, in brief, ties up water molecules and forces less soluble components, such as vesicles, out of the culture media, which can then be pelleted by centrifugation. The purity and size distribution of exosomes isolated with this kit is comparable to ultracentrifugation.

Fig. 6 shows that SIRT-1 depletion upregulates CD63 expression, but has no apparent impact on the release of CD63-positive 'EVs' from uninfected cells. EV-D68 infection also upregulates CD63 expression in SIRT-1 replete cells, and in this case, increases the release of CD63-positive EVs. The combination of infection and SIRT-1 depletion massively upregulates CD63 expression, but appears to eliminate the enhanced release of CD63-positive EVs resulting from infection alone. These are interesting results, from which the authors infer CD63 is associated with EVs containing EV-D68. But, do we know this? Can a CD63 pulldown immunoprecipitate EV-D68 capsid proteins or viral RNA? CD63 is strongly associated with exosomes released from cells through the multi-vesicular body pathway, which are distinct from the LC3-positive EVs released by secretory autophagy that have previously been associated with enteroviruses. The authors suggest that 'knockdown of SIRT-1 may prevent the exocytosis of CD63-positive EVs", but this is a very broad claim (and not really demonstrated by Fig. 6): it requires a clearer definition of what the authors mean by 'exocytosis' and a much more detailed analysis of the size and buoyant density of EVs released in a SIRT-1-dependent process.

We have toned down this suggestion, which sets up our logic for what is now Figure 7 but we agree does not prove the specific nature of these vesicles.

The authors suggest that almost all EV-D68 released from infected cells is released without cell lysis in EVs. However, they generally show data from only a single time point following infection (5 or 6 hrs post-infection). It would have been interesting to see a more complete temporal analysis, and to know whether a high proportion of virus continues to be released in EVs, or if it is swamped out ultimately by lytic release of nonenveloped virus.

In these cells, very little virus is released at earlier timepoints, and after 6hpi it is difficult to analyze virus release because of cell detachment and lysis. In a future publication we will use less susceptible cells to analyze a time course of release.

Fig. 1D indicates that a small fraction of SIRT-1 leaks from the nucleus in EV-D68 infected cells. The authors suggest this is due to targeted nuclear export, rather than simply leaky nuclear pores which are well known to exist in enterovirus-infected cells. The authors present similar fluorescent microscopy data showing inhibition of TFEB export in leptomycin-B treated cells in Fig. S2A in support of their claim that this is specific SIRT-1 export, but these data are far from convincing - there is equivalent residual TFEB and SIRT-1 in the cytoplasm of the treated cells. Quantitative immunoblots of cytoplasmic and nuclear cell fractions might prove more compelling.

We have changed the text to remove the word “block” and instead suggest that there is inhibition, given the difference we observe with and without leptomycin-B.

Finally, the authors should be more specific in describing the viruses they have studied (EV-D68 and PV). It would be preferable to describe these as 'enteroviruses' (including in the title of the manuscript), rather than more broadly as 'picornaviruses'. There is no certainty that the requirement for SIRT-1 in non-lytic release of virus extends to hepatoviruses or other picornaviral genera, for which mechanisms of nonlytic release may be quite different.

We have made this change and thank the reviewer for pointing this out.

Reviewer #2 (Public Review):

The authors aimed to connect SIRT-1 to EV-D68 virus release through mediating ER stress. They are successful in robustly connecting these pathways experimentally and show a new role for SIRT-1 in EV-D68 infection. These results extend to additional viruses, suggesting role(s) for SIRT-1 in diverse virus infection.

The authors note that EV-D68 does not significantly impact SIRT-1 protein levels (Fig 1E and F), though this has been described for other picornaviruses (Xander et al., J Immunol 2019; Han et al., J Cell Sci 2016; Kanda et al Biochem Biophys Res Commun 2015). This may be of interest to note in the manuscript.

We have cited the above papers in the manuscript and thank the reviewer for these suggestions.

The data regarding CVB3 (Fig S4) are especially interesting because they show no discernable impact on infection. The manuscript should describe this further and perhaps speculate on potential reasons. Could it be due to inefficient knockdown?

We have shown that both genetic and pharmacological inhibition of SIRT-1 does not significantly alter CVB3 titers. We do not think this is due to inefficient knockdown since the CVB3 and PV experiments were done concurrently. We are currently investigating why CVB3 responds differently from EV-D68 and PV.

SIRT-1 (and other sirtuins) have been linked to an innate interferon response. Are any of the phenotypes observed here due to IFN responses? The use of H1HeLa cells would suggest this is not the case.

We think this is unlikely because H1HeLas are not IFN-competent and the knockdown of SIRT1 did not significantly alter viral RNA replication

Reviewer #1 (Recommendations For The Authors):

In Fig. 1, it would be informative to show an immunoblot of the protein in knockdown vs control cells (this is shown in different experiments in Fig. 2A and 3C, with variable degrees of knockdown efficiency, but ideally should be shown here also).

The knockdown efficiency of SIRT-1 is now shown in Fig. S1D. We thank the reviewer for this suggestion.

Why is the extracellular virus titer in the control cells in Fig. 1C so much lower (over a 1.5 logs) than in Fig. 1B? Has the plasmid transfection induced an innate immune response, and could this be confounding the experiment?

We think this is due to stress induced by transfection and not an innate immune response, since H1Hela are not interferon competent.

SIRT-1 is recognized to have a regulatory role in autophagy, but the author's claim that it is "essential for stress induced and basal autophagy" would be strengthened by including in Fig. 2B control images of starved and CCCP-treated cells.

LC3 lipidation and p62 degradation are the hallmarks of autophagy initiation and flux, which are shown in Fig. 2A. The goal of Fig. 2B was to verify the impact of SIRT-1 knockdown in restricting basal autophagic degradation. We will examine the effect of starvation and CCCP treatment in future studies. We thank the reviewer for understanding.

The BiP immunoblot shown in Fig. 4B does not support the claim that 'TG [thapsigargin] treatment induced BiP protein levels' whereas 'EV-D68 infection reduced BiP levels...suggesting that EV-D68 blocks ER stress.' The apparent differences in BiP expression are minimal and of questionable biological significance.

We have consistently observed a reduction in BiP levels during EV-D68 infection in both hSABCi-NS1.1 as indicated in Fig. 4B and H1HeLa (see Author response image 1), which is consistent with an ER stress blockade during EV-D68 infection.

Author response image 1.

Minor comments:

1) The variable and wide-ranging scale of the y-axis in Figs. 1A-C and S1 is distracting, exaggerates small differences, and makes it difficult to assess the magnitude of differences in virus titers. The scale should be standardized and held constant in graphs showing results from similar types of experiments.

Our graphs are plotted based on the viral titers from experiments, mostly done on different days. We are confident that the variabilities in the y-axis do not affect the statistical analyses.

2) The number and types of (technical or biological?) of experimental replicates should be indicated in the figure legends. Ideally, each replicate should be individually plotted in graphs.

All experiments are repeated at least three times unless otherwise indicated. We have added this information to the figure legends.

3) Fig. S5C - how many replicates were done, and is there a statistically significant difference in viral RNA abundance at the last time point?

The experiment was done three times, twice with a low MOI (0.1) and once with a high MOI (30). There is no statistical difference at the last time point as shown in the graphs in Author response image 2.

Author response image 2.

Reviewer #2 (Recommendations For The Authors):

Figure 1D would benefit from staining for viral replication compartments (J2, for instance) to correlate the amount of viral dsRNA with nuclear egress of SIRT-1. Similar data would benefit Figure 5A. The data in Figure S5 suggests that most, but not all cells, are infected, so having this control seems important for their IFA experiments.

SIRT-1 dsRNA staining for EV-D68 infection is shown in Fig. S5A and all cells appear to be infected. The IFA data (Author response image 3) shows dsRNA staining of CVB3-infected cells.

Author response image 3.

Are EVs not released as efficiently with SIRT-1 knockdown? The authors show that knockdown reduces CD63 levels in purified EVs, but this could be explained if exosomes are not generated as robustly with SIRT-1 knockdown.

We don’t want to use the word “exosomes” since their definition is very specific, and only use it once in our manuscript, to describe known membrane associations of CD63. We do not think SIRT-1 knockdown affects the intracellular generation of EVs, since depleting SIRT-1 leads to the buildup of CD63 positive signals in the whole cell lysates compared to the scramble control (Fig. 7B and C). Instead, our data suggest that SIRT-1 regulates the release of EVs during EV-D68 infection.

Labels of graphs for "Infection" versus treatment ("TG" or "EX527") is unclear. All samples are presumably infected, so perhaps the authors meant to label these diagrams as untreated.

We have made the changes in the labels and thank the reviewer for helping make these graphs more clear.

The induction of ER stress with TG and repression of stress with EV-D68 infection is clear from BiP western blots. Are BiP levels reduced in SIRT-1 knockdown cells? Their data with TG treatment and knockdown suggests this may be possible.

We have not examined the impact of SIRT-1 knockdown on BiP protein levels. But since SIRT1 KD increases ER stress, as evidenced by a reduction in SERCA2A levels (Fig. 3C and E), we would expect an increase in BiP levels in SIRT-1 depleted cells.

Would the authors expect TG to reduce EVs with EV-D68 as well? Presumably, combination of TG with SIRT-1 would reduce EVs similar to the results shown in Figure 6C. They mention in the discussion that TG and SIRT-1 "share common cellular targets" so it would be interesting to determine if TG acts similar to SIRT-1 knockdown with regard to EVs.

We think TG will similarly reduce EVs in EV-D68-infected cells, and we are currently testing this hypothesis.

Because of the inclusion of the SARS-CoV-2 data and mention in the abstract, it may be appropriate to include that data (Fig S7) in the main figures. The authors mention SIRT-1 as important to MERS-CoV infection in the introduction, but SIRT-1 has been implicated in RNA virus infection, including picornaviruses (noted above). The expansion of this section to provide additional context would benefit the introduction and discussion.

We have moved the former Fig. S7 to the main manuscript as Fig. 6.

Author response:

The following is the authors’ response to the current reviews.

eLife assessment

This study presents an important finding on the influence of visual uncertainty and Bayesian cue combination on implicit motor adaptation in young healthy participants, hereby linking perception and action during implicit adaptation. The evidence supporting the claims of the authors is convincing. The normative approach of the proposed PEA model, which combines ideas from separate lines of research, including vision research and motor learning, opens avenues for future developments. This work will be of interest to researchers in sensory cue integration and motor learning.

Thank you for the updated assessment. We are also grateful for the insightful and constructive comments from the reviewers, which have helped us improve the manuscript again. We made necessary changes following their comments (trimmed tests, new analysis results, etc) and responded to the comments in a point-by-point fashion below. We hope to publish these responses alongside the public review. Thank you again for fostering the fruitful discussion here.

Public Reviews:

Reviewer #1 (Public Review):

I appreciate the normative approach of the PEA model and am eager to examine this model in the future. However, two minor issues remain:

(1) Clarification on the PReMo Model:

The authors state, "The PReMo model proposes that this drift comprises two phases: initial proprioceptive recalibration and subsequent visual recalibration." This description could misinterpret the intent of PReMo. According to PReMo, the time course of the reported hand position is merely a read-out of the *perceived hand position* (x_hat in your paper). Early in adaptation, the perceived hand position is biased by the visual cursor (x_hat in the direction of the cursor); towards the end, due to implicit adaptation, x_hat reduces to zero. This is the same as PEA. I recommend that the authors clarify PReMo's intent to avoid confusion.

Note, however, the observed overshoot of 1 degree in the reported hand position. In the PReMo paper, we hypothesized that this effect is due to the recalibration of the perceived visual target location (inspired by studies showing that vision is also recalibrated by proprioception, but in the opposite direction). If the goal of implicit adaptation is to align the perceived hand position (x_hat) with the perceived target position (t_hat), then there would be an overshoot of x_hat over the actual target position.

PEA posits a different account for the overshoot. It currently suggests that the reported hand position combines x_hat (which takes x_p as input) with x_p itself. What is reasoning underlying the *double occurrence* of x_p?

There seem to be three alternatives that seem more plausible (and could lead to the same overshooting): 1) increasing x_p's contribution (assuming visual uncertainty increases when the visual cursor is absent during the hand report phase), 2) decreasing sigma_p (assuming that participants pay more attention to the hand during the report phase), 3) it could be that the perceived target position undergoes recalibration in the opposite direction to proprioceptive recalibration. All these options, at least to me, seem equally plausible and testable in the future.

For clarification of the PReMo model’s take on Fig4A, we now write:

“The PReMo model proposes that the initial negative drift reflects a misperceived hand location, which gradually reduces to zero, and the late positive drift reflects the influence of visual calibration of the target (Tsay, Kim, Saxena, et al., 2022). ”

However, we would like to point out that the PEA model does not predict a zero (perceived hand location) even at the late phase of adaptation: it remains negative, though not as large as during initial adaptation (see Figure 4A, red line). Furthermore, we have not seen any plausible way to use a visually biased target to explain the overshoot of the judged hand location (see below when we address the three alternative hypotheses the reviewer raised).

We don’t think the “double” use of xp is a problem, simply because there are TWO tasks under investigation when the proprioceptive changes are measured along with adaptation. The first is the reaching adaptation task itself: moving under the influence of the clamped cursor. This task is accompanied by a covert estimation of hand location after the movement (). Given the robustness of implicit adaptation, this estimation appears mandatory and automatic. The second task is the hand localization task, during which the subject is explicitly asked to judge where the hand is. Here, the perceived hand is based on the two available cues, one is the actual hand location xp, and the other is the influence from the just finished reaching movement (i.e., ). For Bayesian modeling from a normative perspective, sensory integration is based on the available cues to fulfill the task. For the second task of reporting the hand location, the two cues are xp and (with a possible effect of the visual target, which is unbiased since it is defined as 0 in model simulation; thus, its presence does not induce any shift effect). xp is used sequentially in this sense. Thus, its dual use is well justified.

Our hypothesis is that the reported hand position results from a combination of from the previous movement and the current hand position xp. However, specifically for the overshoot of the judged hand location in the late part of the adaptation (Fig4A), the reviewer raised three alternative explanations by assuming that the PReMo model is correct. Under the PReMo model, the estimated hand location is only determined by , and xp is not used in the hand location report phase. In addition, (with xp used once) and a visual recalibration of the target can explain away the gradual shift from negative to positive (overshoot).

We don’t think any of them can parsimoniously explain our findings here, and we go through these three hypotheses one by one:

(1) increasing xp's contribution (assuming visual uncertainty increases when the visual cursor is absent during the hand report phase)

(2) decreasing σp (assuming that participants pay more attention to the hand during the report phase)

The first two alternative explanations basically assume that xp has a larger contribution (weighting in Bayesian terms) in the hand location report phase than in the adaptation movement phase, no matter due to an increase in visual uncertainty (alternative explanation 1) or a reduction in proprioceptive uncertainty (alternative explanation 2). Thus, we assume that the reviewer suggests that a larger weight for xp can explain why the perceived hand location changes gradually from negative to positive. However, per the PReMo model, a larger weight for the xp will only affect , which is already assumed to change from negative to zero. More weight in  in the hand report phase (compared to the adaptation movement phase) would not explain away the reported hand location from negative to positive. This is because no matter how much weight the xp has, the PReMo model assumes a saturation for the influence of xp on . Thus would not exceed zero in the late adaptation. Then, the PReMo model would rely on the so-called visual shift of the target to explain the overshoot. This leads us to the third alternative the reviewer raised:

(3) it could be that the perceived target position undergoes recalibration in the opposite direction to proprioceptive recalibration.

The PReMo model originally assumed that the perceived target location was biased in order to explain away the positive overshoot of the reported hand location. We assume that the reviewer suggests that the perceived target position, which is shifted to the positive direction, also “biases” the perceived hand position. We also assume that the reviewer suggests that the perceived hand location after a clamp trial () is zero, and somehow the shifted perceived target position “biases” the reported hand location after a clamp trial. Unfortunately, we did not see any mathematical formulation of this biasing effect in the original paper (Tsay, Kim, Haith, et al., 2022). We are not able to come up with any formulation of this hypothesized biasing effect based on Bayesian cue integration principles. Target and hand are two separate perceived items; how one relates to another needs justification from a normative perspective when discussing Bayesian models. Note this is not a problem for our PEA models, in which both cues used are about hand localization, one is and the other is xp.

We believe that mathematically formulating the biasing effect (Figure 4A) is non-trivial since the reported hand location changes continuously from negative to positive. Thus, quantitative model predictions, like the ones our PEA model presents here, are needed.

To rigorously test the possible effect of visual recalibration of the target, there are two things to do: 1) use the psychometric method to measure the biased perception of the target, and 2) re-do Tsay et al. 2020 experiment without the target. For 2), compared to the case with the target, the PEA model would predict a larger overshoot, while the PReMo would predict a smaller overshoot or even zero overshoot. This can be left for future studies.

(2) Effect of Visual Uncertainty on Error Size:

I appreciate the authors' response about methodological differences between the cursor cloud used in previous studies and the Gaussian blob used in the current study. However, it is still not clear to me how the authors reconcile previous studies showing that visual uncertainty reduced implicit adaptation for small but not large errors (Tsay et al, 2021; Makino, et al 2023) with the current findings, where visual uncertainty reduced implicit adaptation for large but not small errors.

Could the authors connect the dots here: I could see that the cursor cloud increases potential overlap with the visual target when the visual error is small, resulting in intrinsic reward-like mechanisms (Kim et al, 2019), which could potentially explain attenuated implicit adaptation for small visual errors. However, why would implicit adaptation in response to large visual errors remain unaffected by the cursor cloud? Note that we did verify that sigma_v is increased in (Tsay et al. 2021), so it is unlikely due to the cloud simply failing as a manipulation of visual uncertainty.

In addition, we also reasoned that testing individuals with low vision could offer a different test of visual uncertainty (Tsay et al, 2023). The advantage here is that both control and patients with low vision are provided with the same visual input-a single cursor. Our findings suggest that uncertainty due to low vision also shows reduced implicit adaptation in response to small but not large errors, contrary to the findings in the current paper. Missing in the manuscript is a discussion related to why the authors' current findings contradict those of previous results.

For connecting the dots for two previous studies (Tsay et al., 2021, 2023); Note Makino et al., 2023 is not in this discussion since it investigated the weights of multiple cursors, as opposed to visual uncertainty associated with a cursor cloud):

First, we want to re-emphasize that using the cursor cloud to manipulate visual uncertainty brings some confounds, making it not ideal for studying visuomotor adaptation. For example, in the error clamp paradigm, the error is defined as angular deviation. The cursor cloud consists of multiple cursors spanning over a range of angles, which affects both the sensory uncertainty (the intended outcome) and the sensory estimate of angles (the error estimate, the undesired outcome). In Bayesian terms, the cursor cloud aims to modulate the sigma of a distribution (σv) in our model), but it additionally affects the mean of the distribution (µ). This unnecessary confound is neatly avoided by using cursor blurring, which is still a cursor with its center (µ) unchanged from a single cursor. Furthermore, as correctly pointed out in the original paper by Tsay et al., 2020, the cursor cloud often overlaps with the visual target; this "target hit" would affect adaptation, possibly via a reward learning mechanism (Kim et al., 2019). This is a second confound that accompanies the cursor cloud. Yes, the cursor cloud was verified as associated with high visual uncertainty (Tsay et al., 2021); this verification was done with a psychophysics method with a clean background, not in the context of a hand reaching a target that is needed. Thus, despite the cursor cloud having a sizeable visual uncertainty, our criticisms for it still hold when used in error-clamp adaptation.

Second, bearing these confounds of the cursor cloud in mind, we postulate one important factor that has not been considered in any models thus far that might underlie the lack of difference between the single-cursor clamp and the cloud-cursor clamp when the clamp size is large: the cursor cloud might be harder to ignore than a single cursor. For Bayesian sensory integration, the naive model is to consider the relative reliability of cues only. Yes, the cloud is more uncertain in terms of indicating the movement direction than a single cursor. However, given its large spread, it is probably harder to ignore during error-clamp movements. Note that ignoring the clamped cursor is the task instruction, but the large scatter of the cursor cloud is more salient and thus plausible and harder to ignore. This might increase the weighting of the visual cue despite its higher visual uncertainty. This extra confound is arguably minimized by using the blurred cursor as in our Exp4 since the blurred cursor did not increase the visual angle much (Figure 5D; blurred vs single cursor: 3.4mm vs 2.5mm in radius, 3.90o vs  2.87o in spread). In contrast, the visual angle of the dot cloud is at least a magnitude larger (cursor cloud vs. single cursor: at least 25o vs. 2.15o in the spread, given a 10o standard deviation of random sampling).

Third, for the low-vision study (Tsay et al., 2023), the patients indeed show reduced implicit adaptation for a 3 o clamp (consistent with our PEA model) but an intact adaptation for 30-degree clamp (not consistent). Though this pattern appears similar to what happens for normal people whose visual uncertainty is upregulated by cursor cloud (Tsay et al., 2021), we are not completely convinced that the same underlying mechanism governs these two datasets. Low-vision patients indeed have higher visual uncertainty about color, brightness, and object location, but their visual uncertainty about visual motion is still unknown. Due to the difference in impairment among low vision people (e.g., peripheral or central affected) and the different roles of peripheral and central vision in movement planning and control (Sivak & Mackenzie, 1992), it is unclear about the overall effect of visual uncertainty in low vision people. The direction of cursor movement that matters for visuomotor rotation here is likely related to visual motion perception. Unfortunately, the original study did not measure this uncertainty in low-vision patients. We believe our Exp1 offers a valid method for this purpose for future studies. More importantly, we should not expect low-vision patients to integrate visual cues in the same way as normal people, given their long-term adaptation to their vision difficulties. Thus, we are conservative about interpreting the seemingly similar findings across the two studies (Tsay et al., 2021, 2023) as revealing the same mechanism.

A side note: these two previous studies proposed a so-called mis-localization hypothesis, i.e., the cursor cloud was mislocated for small clamp size (given its overlapping with the target) but not for large clamp size. They suggested that the lack of uncertainty effect at small clamp sizes is due to mislocalization, while the lack of uncertainty effect at large clamp sizes is because implicit adaptation is not sensitive to uncertainty at large angles. Thus, these two studies admit that cursor cloud not only upregulates uncertainty but also generates an unwanted effect of so-called “mis-localization” (overlapping with the target). Interestingly, their hypothesis about less sensitivity to visual uncertainty for large clamps is not supported by a model or theory but merely a re-wording of the experiment results.

In sum, our current study cannot offer an easy answer to "connect the dots" in the aforementioned two studies due to methodology issues and the specialty of the population. However, for resolving conflicting findings, our study suggests solutions include using a psychometric test to quantify visual uncertainty for cursor motion (Exp1), a better uncertainty-manipulation method to avoid a couple of confounds (Exp4, blurred cursor), and a falsifiable model. Future endeavors can solve the difference between studies based on the new insights from the current.

Reviewer #2 (Public Review):

Summary:

The authors present the Perceptual Error Adaptation (PEA) model, a computational approach offering a unified explanation for behavioral results that are inconsistent with standard state-space models. Beginning with the conventional state-space framework, the paper introduces two innovative concepts. Firstly, errors are calculated based on the perceived hand position, determined through Bayesian integration of visual, proprioceptive, and predictive cues. Secondly, the model accounts for the eccentricity of vision, proposing that the uncertainty of cursor position increases with distance from the fixation point. This elegantly simple model, with minimal free parameters, effectively explains the observed plateau in motor adaptation under the implicit motor adaptation paradigm using the error-clamp method. Furthermore, the authors experimentally manipulate visual cursor uncertainty, a method established in visuomotor studies, to provide causal evidence. Their results show that the adaptation rate correlates with perturbation sizes and visual noise, uniquely explained by the PEA model and not by previous models. Therefore, the study convincingly demonstrates that implicit motor adaptation is a process of Bayesian cue integration

Strengths:

In the past decade, numerous perplexing results in visuomotor rotation tasks have questioned their underlying mechanisms. Prior models have individually addressed aspects like aiming strategies, motor adaptation plateaus, and sensory recalibration effects. However, a unified model encapsulating these phenomena with a simple computational principle was lacking. This paper addresses this gap with a robust Bayesian integration-based model. Its strength lies in two fundamental assumptions: motor adaptation's influence by visual eccentricity, a well-established vision science concept, and sensory estimation through Bayesian integration. By merging these well-founded principles, the authors elucidate previously incongruent and diverse results with an error-based update model. The incorporation of cursor feedback noise manipulation provides causal evidence for their model. The use of eye-tracking in their experimental design, and the analysis of adaptation studies based on estimated eccentricity, are particularly elegant. This paper makes a significant contribution to visuomotor learning research.

The authors discussed in the revised version that the proposed model can capture the general implicit motor learning process in addition to the visuomotor rotation task. In the discussion, they emphasize two main principles: the automatic tracking of effector position and the combination of movement cues using Bayesian integration. These principles are suggested as key to understanding and modeling various motor adaptations and skill learning. The proposed model could potentially become a basis for creating new computational models for skill acquisition, especially where current models fall short.

Weaknesses:

The proposed model is described as elegant. In this paper, the authors test the model within a limited example condition, demonstrating its relevance to the sensorimotor adaptation mechanisms of the human brain. However, the scope of the model's applicability remains unclear. It has shown the capacity to explain prior data, thereby surpassing previous models that rely on elementary mathematics. To solidify its credibility in the field, the authors must gather more supporting evidence.

Indeed, our model here is based on one particular experimental paradigm, i.e., the error-clamp adaptation. We used it simply because 1) this paradigm is one rare example that implicit motor learning can be isolated in a clean way, and 2) there are a few conflicting findings in the literature for us to explain away by using a unified model.

For our model’s broad impact, we believe that as long as people need to locate their effectors during motor learning, the general principle laid out here will be applicable. In other words, repetitive movements with a Bayesian cue combination of movement-related cues can underlie the implicit process of various motor learning. To showcase its broad impact, in upcoming studies, we will extend this model to other motor learning paradigms, starting from motor adaptation paradigms that involve both explicit and implicit processes.

Reviewer #3 (Public Review):

(2.1) Summary

In this paper, the authors model motor adaptation as a Bayesian process that combines visual uncertainty about the error feedback, uncertainty about proprioceptive sense of hand position, and uncertainty of predicted (=planned) hand movement with a learning and retention rate as used in state space models. The model is built with results from several experiments presented in the paper and is compared with the PReMo model (Tsay, Kim et al., 2022) as well as a cue combination model (Wei & Körding, 2009). The model and experiments demonstrate the role of visual uncertainty about error feedback in implicit adaptation.

In the introduction, the authors notice that implicit adaptation (as measured in error-clamp based paradigms) does not saturate at larger perturbations, but decreases again (e.g. Moorehead et al., 2017 shows no adaptation at 135{degree sign} and 175{degree sign} perturbations). They hypothesized that visual uncertainty about cursor position increases with larger perturbations since the cursor is further from the fixated target. This could decrease importance assigned to visual feedback which could explain lower asymptotes.

The authors characterize visual uncertainty for 3 rotation sizes in a first experiment, and while this experiment could be improved, it is probably sufficient for the current purposes. Then the authors present a second experiment where adaptation to 7 clamped errors are tested in different groups of participants. The models' visual uncertainty is set using a linear fit to the results from experiment 1, and the remaining 4 parameters are then fit to this second data set. The 4 parameters are 1) proprioceptive uncertainty, 2) uncertainty about the predicted hand position, 3) a learning rate and 4) a retention rate. The authors' Perceptual Error Adaptation model ("PEA") predicts asymptotic levels of implicit adaptation much better than both the PReMo model (Tsay, Kim et al., 2022), which predicts saturated asymptotes, or a causal inference model (Wei & Körding, 2007) which predicts no adaptation for larger rotations. In a third experiment, the authors test their model's predictions about proprioceptive recalibration, but unfortunately compare their data with an unsuitable other data set (Tsay et al. 2020, instead of Tsay et al. 2021). Finally, the authors conduct a fourth experiment where they put their model to the test. They measure implicit adaptation with increased visual uncertainty, by adding blur to the cursor, and the results are again better in line with their model (predicting overall lower adaptation), than with the PReMo model (predicting equal saturation but at larger perturbations) or a causal inference model (predicting equal peak adaptation, but shifted to larger rotations). In particular the model fits for experiment 2 and the results from experiment 4 show that the core idea of the model has merit: increased visual uncertainty about errors dampens implicit adaptation.

(2.2) Strengths

In this study the authors propose a Perceptual Error Adaptation model ("PEA") and the work combines various ideas from the field of cue combination, Bayesian methods and new data sets, collected in four experiments using various techniques that test very different components of the model. The central component of visual uncertainty is assessed in a first experiment. The model uses 4 other parameters to explain implicit adaptation. These parameters are: 1) a learning and 2) a retention rate, as used in popular state space models and the uncertainty (variance) of 3) predicted and 4) proprioceptive hand position. In particular, the authors observe that asymptotes for implicit learning do not saturate, as claimed before, but decrease again when rotations are very large and that this may have to do with visual uncertainty (e.g. Tsay et al., 2021, J Neurophysiol 125, 12-22). The final experiment confirms predictions of the fitted model about what happens when visual uncertainty is increased (overall decrease of adaptation). By incorporating visual uncertainty depending on retinal eccentricity, the predictions of the PEA model for very large perturbations are notably different from, and better than, the predictions of the two other models it is compared to. That is, the paper provides strong support for the idea that visual uncertainty of errors matters for implicit adaptation.

(2.3) Weaknesses

Although the authors don't say this, the "concave" function that shows that adaptation does not saturate for larger rotations has been shown before, including in papers cited in this manuscript.

For a proper citation of the “concave” adaptation function: we assume the reviewer is referring to the study by Morehead, 2017 which tested large clamp sizes up to 135 o and 175 o. Unsurprisingly, the 135 o and 175 o conditions lead to nearly zero adaptation, possibly due to the trivial fact that people cannot even see the moving cursor. We have quoted this seminar study from the very beginning. All other error-clamp studies with a block design emphasized an invariant or saturated implicit adaptation with large rotations (e.g., Kim, et al., 2019).

The first experiment, measuring visual uncertainty for several rotation sizes in error-clamped paradigms has several shortcomings, but these might not be so large as to invalidate the model or the findings in the rest of the manuscript. There are two main issues we highlight here. First, the data is not presented in units that allow comparison with vision science literature. Second, the 1 second delay between movement endpoint and disappearance of the cursor, and the presentation of the reference marker, may have led to substantial degradation of the visual memory of the cursor endpoint. That is, the experiment could be overestimating the visual uncertainty during implicit adaptation.

For the issues related to visual uncertainty measurement in Exp1:

First, our visual uncertainty is about cursor motion direction in the display plane, and the measurement in Exp1 has never been done before. Thus, we do not think our data is comparable to any findings in visual science about fovea/peripheral comparison. We quoted Klein and others’ work (Klein & Levi, 1987; Levi et al., 1987) in vision science since their studies showed that the deviation from the fixation is associated with an increase in visual uncertainty. Their study thus inspired us to conduct Exp1 to probe how our concerned visual uncertainty (specifically for visual motion direction) changes with an increasing deviation from the fixation. Any model and its model parameters should be specifically tailored to the task or context it tries to emulate. In our case, motion direction in a center-out-reaching setting is the modeled context, and all the relevant model parameters should be specified in movement angles. This is particularly important since we need to estimate parameters from one experiment to predict behaviors in another experiment.

Second, the 1s delay of the reference cursor has minimal impact on the estimate of visual uncertainty based on previous vision studies. Our Exp1 used a similar visual paradigm by (White et al., 1992), which shows that delay does not lead to an increase in visual uncertainty over a broad range of values (from 0.2s to >1s, see their Figure 5-6).

These two problems have been addressed in the revised manuscript, with proper citations listed.

The paper's third experiment relies to a large degree on reproducing patterns found in one particular paper, where the reported hand positions - as a measure of proprioceptive sense of hand position - are given and plotted relative to an ever present visual target, rather than relative to the actual hand position. That is, 1) since participants actively move to a visual target, the reported hand positions do not reflect proprioception, but mostly the remembered position of the target participants were trying to move to, and 2) if the reports are converted to a difference between the real and reported hand position (rather than the difference between the target and the report), those would be on the order of ~20° which is roughly two times larger than any previously reported proprioceptive recalibration, and an order of magnitude larger than what the authors themselves find (1-2°) and what their model predicts. Experiment 3 is perhaps not crucial to the paper, but it nicely provides support for the idea that proprioceptive recalibration can occur with error-clamped feedback.

Reviewer 3 thinks Tsay 2020 dataset is not appropriate for our theorization, but we respectfully disagree. For the three points raised here, we would like to elaborate:

(1) As we addressed in the previous response, the reported hand location in Figure 4A (Tsay et al., 2020) is not from a test of proprioceptive recalibration as conventionally defined. In the revision, we explicitly state that this dataset is not about proprioceptive recalibration and also delete texts that might mislead people to think so (see Results section). Instead, proprioceptive recalibration is measured by passive movement, as in our Exp3 (Figure 4E). For error-clamp adaptation here, "the remembered position of the target" is the target. Clearly, the participants did not report the target position, which is ever-present. Instead, their reported hand location shows an interestingly continuous change with ongoing adaptation.

(2) Since the Tsay 2020 dataset is not a so-called proprioceptive recalibration, we need not take the difference between the reported location and the actual hand location. Indeed, the difference would be ~20 degrees, but comparing it to the previously reported proprioceptive recalibration is like comparing apples to oranges. In fact, throughout the paper, we refer to the results in Fig 4A as “reported hand location”, not proprioceptive recalibration. The target direction is defined as zero degree thus its presence will not bias the reported hand in the Bayesian cue combination (as this visual cue has a mean value of 0). Using the target as the reference also simplifies our modeling.

(3) Exp3 is crucial for our study since it shows our model and its simple Bayesian cue combination principle are applicable not only to implicit adaptation but also to proprioceptive measures during adaptation. Furthermore, it reproduced the so-called proprioceptive recalibration and explained it away with the same Bayesian cue combination as the adaptation. We noticed that this field has accumulated an array of findings on proprioceptive changes induced by visuomotor adaptation. However, currently, there is a lack of a computational model to quantitatively explain them. Our study at least made an initial endeavor to model these changes.

Perhaps the largest caveat to the study is that it assumes that people do not look at the only error feedback available to them (and can explicitly suppress learning from it). This was probably true in the experiments used in the manuscript, but unlikely to be the case in most of the cited literature. Ignoring errors and suppressing adaptation would also be a disastrous strategy to use in the real world, such that our brains may not be very good at this. So the question remains to what degree - if any - the ideas behind the model generalize to experiments without fixation control, and more importantly, to real life situations.

The largest caveat raised by the reviewer appears to be directed to the error-clamp paradigm in general, not only to our particular study. In essence, this paradigm indeed requires participants to ignore the clamped error; thus, its induced adaptive response can be attributed to implicit adaptation. The original paper that proposed this paradigm (Morehead et al., 2017) has been cited 220 times (According to Google Scholar, at the time of this writing, 06/2024), indicating that the field has viewed this paradigm in a favorable way.

Furthermore, we agree that this kind of instruction and feedback (invariant clamp) differ from daily life experience, but it does not prevent us from gaining theoretical insights by studying human behaviors under this kind of "artificial" task setting. Thinking of the saccadic adaptation (Deubel, 1987; Kojima et al., 2004): jumping the target while the eye moves towards it, and this somewhat artificial manipulation again makes people adapt implicitly, and the adaptation itself is a "disastrous" strategy for real-life situations. However, scientists have gained an enormous understanding of motor adaptation using this seemingly counterproductive adaptation in real life. Also, think of perceptual learning of task-irrelevant stimuli (Seitz & Watanabe, 2005, 2009): when participants are required to learn to discriminate one type of visual stimuli, the background shows another type of stimuli, which people gradually learn even though they do not even notice its presence. This "implicit" learning can be detrimental to our real life, too, but the paradigm itself has advanced our understanding of the inner workings of the cognitive system.

Recommendations for the authors:

Reviewer #2 (Recommendations For The Authors):

L101: There is a typo: (Tsay et al., 2020), 2020) should be corrected to (Tsay et al., 2020).

Thanks for pointing it out, we corrected this typo.

L224-228: It would be beneficial to evaluate the validity of the estimated sigma_u and sigma_p based on previous reports.

We can roughly estimate σu by evaluating the variability of reaching angles during the baseline phase when no perturbation is applied. The standard deviation of the reaching angle in Exp 2 is 5.128o±0.190o, which is close to the σu estimated by the model (5.048o). We also used a separate perceptual experiment to test the proprioceptive uncertainty (n = 13, See Figure S6), σp from this experiment is 9.737o±5.598o, also close to the σp extracted by the model (11.119o). We added these new analysis results to the final version of the paper.

L289-298: I found it difficult to understand the update equations of the proprioceptive calibration based on the PEA model. Providing references to the equations or better explanations would be helpful.

We expanded the process of proprioceptive calibration in Supplementary Text 1 with step-by-step equations and more explanations. 

Reviewer #3 (Recommendations For The Authors):

Suggestions (or clarification of previous suggestions) for revisions

The authors persist on using the Tsay et al 2020 paper despite its many drawbacks which the authors attempt to address in their reply. But the main drawback is that the results in the 2020 paper is NOT relative to the unseen hand but to the visual target the participants were supposed to move their hand to. If the results were converted so to be relative to the unseen hand, the localization biases would be over 20 deg in magnitude.

The PEA simulations are plotted relative to the unseen hand which makes sense. If the authors want to persist using the Tsay 2020 dataset despite any issues, they at least need to make sure that the simulations are mimicking the same change. That is, the data from Tsay 2020 needs to be converted to the same variable used in the current paper.

If the main objection for using the Tsay 2021 is that the design would lead to forgetting, we found that active localization (or any intervening active movements like no-cursor reach) does lead to some interference or forgetting (a small reduction in overall magnitude of adaptation) this is not the case for passive localization, see Ruttle et al, 2021 (data on osf). This was also just a suggestion, there may of course also be other, more suitable data sets.

As stated above, changing the reference system is not necessary, nor does it affect our results. Tsay et al 2020 dataset is unique since it shows the gradual change of reported hand location along with error-clamp adaptation. The forgetting (or reduction in proprioceptive bias), even if it exists, would not affect the fitting quality of our model for the Tsay 2020 dataset: if we assume that forgetting is invariant over the adaptation process, the forgetting would only reduce the proprioceptive bias uniformly across trials. This can be accounted for by a smaller weight on . The critical fact is that the model can explain the gradual drift of the proprioceptive judgment of the hand location.

By the way, Ruttle et al.'s 2021 dataset is not for error-clamp adaptation, and thus we will leave it to test our model extension in the future (after incorporating an explicit process in the model).

References

Deubel, H. (1987). Adaptivity of gain and direction in oblique saccades. Eye Movements from Physiology to Cognition. https://www.sciencedirect.com/science/article/pii/B9780444701138500308

Kim, H. E., Parvin, D. E., & Ivry, R. B. (2019). The influence of task outcome on implicit motor learning. ELife, 8. https://doi.org/10.7554/eLife.39882

Klein, S. A., & Levi, D. M. (1987). Position sense of the peripheral retina. JOSA A, 4(8), 1543–1553.

Kojima, Y., Iwamoto, Y., & Yoshida, K. (2004). Memory of learning facilitates saccadic adaptation in the monkey. The Journal of Neuroscience: The Official Journal of the Society for Neuroscience, 24(34), 7531–7539.

Levi, D. M., Klein, S. A., & Yap, Y. L. (1987). Positional uncertainty in peripheral and amblyopic vision. Vision Research, 27(4), 581–597.

Morehead, J. R., Taylor, J. A., Parvin, D. E., & Ivry, R. B. (2017). Characteristics of implicit sensorimotor adaptation revealed by task-irrelevant clamped feedback. Journal of Cognitive Neuroscience, 29(6), 1061–1074.

Seitz, & Watanabe. (2005). A unified model for perceptual learning. Trends in Cognitive Sciences, 9(7), 329–334.

Seitz, & Watanabe. (2009). The phenomenon of task-irrelevant perceptual learning. Vision Research, 49(21), 2604–2610.

Sivak, B., & Mackenzie, C. L. (1992). Chapter 10 The Contributions of Peripheral Vision and Central Vision to Prehension. In L. Proteau & D. Elliott (Eds.), Advances in Psychology (Vol. 85, pp. 233–259). North-Holland.

Tsay, J. S., Avraham, G., Kim, H. E., Parvin, D. E., Wang, Z., & Ivry, R. B. (2021). The effect of visual uncertainty on implicit motor adaptation. Journal of Neurophysiology, 125(1), 12–22.

Tsay, J. S., Kim, H. E., Saxena, A., Parvin, D. E., Verstynen, T., & Ivry, R. B. (2022). Dissociable use-dependent processes for volitional goal-directed reaching. Proceedings. Biological Sciences / The Royal Society, 289(1973), 20220415.

Tsay, J. S., Kim, H., Haith, A. M., & Ivry, R. B. (2022). Understanding implicit sensorimotor adaptation as a process of proprioceptive re-alignment. ELife, 11, e76639.

Tsay, J. S., Parvin, D. E., & Ivry, R. B. (2020). Continuous reports of sensed hand position during sensorimotor adaptation. Journal of Neurophysiology, 124(4), 1122–1130.

Tsay, J. S., Tan, S., Chu, M. A., Ivry, R. B., & Cooper, E. A. (2023). Low Vision Impairs Implicit Sensorimotor Adaptation in Response to Small Errors, But Not Large Errors. Journal of Cognitive Neuroscience, 35(4), 736–748.

White, J. M., Levi, D. M., & Aitsebaomo, A. P. (1992). Spatial localization without visual references. Vision Research, 32(3), 513–526.

The following is the authors’ response to the original reviews.

eLife assessment

This study presents a valuable finding on the influence of visual uncertainty and Bayesian cue combination on implicit motor adaptation in young healthy participants. The evidence supporting the claims of the authors is solid, although a better discussion of the link between the model variables and the outcomes of related behavioral experiments would strengthen the conclusions. The work will be of interest to researchers in sensory cue integration and motor learning.

Public Reviews:

Reviewer #1 (Public Review):

This valuable study demonstrates a novel mechanism by which implicit motor adaptation saturates for large visual errors in a principled normative Bayesian manner. Additionally, the study revealed two notable empirical findings: visual uncertainty increases for larger visual errors in the periphery, and proprioceptive shifts/implicit motor adaptation are non-monotonic, rather than ramp-like. This study is highly relevant for researchers in sensory cue integration and motor learning. However, I find some areas where statistical quantification is incomplete, and the contextualization of previous studies to be puzzling.

Thank you for your feedback and the positive highlights of our study. We appreciate your insights and will address the concerns in our revisions.

Issue #1: Contextualization of past studies.

While I agree that previous studies have focused on how sensory errors drive motor adaptation (e.g., Burge et al., 2008; Wei and Kording, 2009), I don't think the PReMo model was contextualized properly. Indeed, while PReMo should have adopted clearer language - given that proprioception (sensory) and kinaesthesia (perception) have been used interchangeably, something we now make clear in our new study (Tsay, Chandy, et al. 2023) - PReMo's central contribution is that a perceptual error drives implicit adaptation (see Abstract): the mismatch between the felt (perceived) and desired hand position. The current paper overlooks this contribution. I encourage the authors to contextualize PReMo's contribution more clearly throughout. Not mentioned in the current study, for example, PReMo accounts for the continuous changes in perceived hand position in Figure 4 (Figure 7 in the PReMo study).

There is no doubt that the current study provides important additional constraints on what determines perceived hand position: Firstly, it offers a normative Bayesian perspective in determining perceived hand position. PReMo suggests that perceived hand position is determined by integrating motor predictions with proprioception, then adding a proprioceptive shift; PEA formulates this as the optimal integration of these three inputs. Secondly, PReMo assumed visual uncertainty to remain constant for different visual errors; PEA suggests that visual uncertainty ought to increase (but see Issue #2).

Thank you for the comments and suggestions. We have now incorporated the citation for (Tsay et al., 2024), to acknowledge their clarification on the terms of perceptual error. We also agree that our model differs in two fundamental ways. One is to ditch the concept of proprioceptive shift and its contribution to the perceived hand location; instead, we resort to a “one-shot” integration of three types of cues with Bayesian rules. This is a more elegant and probably more ecological way of processing hand location per Occam's Razor. The second essential change is to incorporate the dependency of visual uncertainty on perturbation size into the model, as opposed to resorting to a ramp function of proprioceptive changes relative to perturbation size. The ramp function is not well grounded in perception studies. Yes, we acknowledged that PReMo is the first to recognize the importance of perceptual error, but highlighted the model differences in our Discussion.

We also think the PReMo model has the potential to explain Fig 4A. But the Tsay et al., 2022 paper assumes that “a generic shift in visual space” explains the gradual proprioceptive changes from negative to positive (see page 17 in Tsay et al., 2022). We do not think that evoking this visual mechanism is necessary to explain Fig 4A; instead, the proprioceptive change is a natural result of hand deviations during implicit adaptation. As the hand moves away from the target (in the positive direction) during adaptation, the estimated hand location goes alone with it. We believe this is the correct way of explaining Fig4A results. As we played around with the PReMo model, we found it is hard to use visual shift to explain this part of data without additional assumptions (at least not with the ones published in Tsay et al., 2022). Furthermore, our PEA model also parsimoniously explains away the proprioceptive shift observed in a completely different setting, i,e., the proprioceptive changes measured by the passive method as a function of perturbation size in Exp 3.

We expanded the discussion about the comparison between the two models, especially about their different views for explaining Fig4A.

Issue #2: Failed replication of previous results on the effect of visual uncertainty.

(2a) A key finding of this paper is that visual uncertainty linearly increases in the periphery; a constraint crucial for explaining the non-monotonicity in implicit adaptation. One notable methodological deviation from previous studies is the requirement to fixate on the target: Notably, in the current experiments, participants were asked to fixate on the target, a constraint not imposed in previous studies. In a free-viewing environment, visual uncertainty may not attenuate as fast, and hence, implicit adaptation does not attenuate as quickly as that revealed in the current design with larger visual errors. Seems like this current fixation design, while important, needs to be properly contextualized considering how it may not represent most implicit adaptation experiments.

First, we don’t think there is any previous study that examined visual uncertainty as a function of perturbation size. Thus, we do not have a replication problem here. Secondly, our data indicate that even without asking people to fixate on the target, people still predominantly fixate on the target during error-clamp adaptation (when they are “free” viewing). For our Exp 1, the fixation on the straight line between the starting position and the target is 86%-95% (as shown in Figure S1 now， also see below). We also collected eye-tracking data in Exp 4, which is a typical error-clamp experiment. More than 95% fall with +/- 50 pixels around the center of the screen, even slightly higher than Exp 1. This is well understandable: the typical error-clamp adaptation requires people to ignore the cursor and move the hand towards the target. To minimize the interference of the concurrently moving cursor, people depend on the fixation on the target, the sole task-relevant visual marker in the workspace, to achieve the task goal.

In sum, forcing the participants to fixate on the target is not because we aimed to make up the linear dependency of visual uncertainty; we required them to do so to mimic the eye-tracking pattern in typical error-clamp learning, which has been revealed in our pilot experiment. The visual uncertainty effect is sound, our study is the first to clearly demonstrate it.

Author response image 1.

On a side note (but an important one), the high percentage of fixation on the aiming target is also true for conventional visuomotor rotation, which involves strategic re-aiming (shown in Bromberg et al., 2019; de Brouwer et al., 2018, we have an upcoming paper to show this). This is one reason that our new theory would also be applicable to other types of motor adaptation.

(2b) Moreover, the current results - visual uncertainty attenuates implicit adaptation in response to large, but not small, visual errors - deviates from several past studies that have shown that visual uncertainty attenuates implicit adaptation to small, but not large, visual errors (Tsay, Avraham, et al. 2021; Makino, Hayashi, and Nozaki, n.d.; Shyr and Joshi 2023). What do the authors attribute this empirical difference to? Would this free-viewing environment also result in the opposite pattern in the effect of visual uncertainty on implicit adaptation for small and large visual errors?

We don’t think all the mentioned previous studies manipulated the visual uncertainty in a parametric way, and none of them provided quantitative measures of visual uncertainty. As we detailed in our Exp4 and in our Discussion, we don’t think Tsay et al., 2021 paper’s manipulation of visual uncertainty is appropriate (see below for 2d). Makino et al., 2023 study used multiple clamped cursors to perturb people, and its effect is not easily accountable since additional processes might be invoked given this kind of complex visual feedback. More importantly, we do not think this is a direct way of modulating visual uncertainty, nor did they provide any evidence.

(2c) In the current study, the measure of visual uncertainty might be inflated by brief presentation times of comparison and referent visual stimuli (only 150 ms; our previous study allowed for a 500 ms viewing time to make sure participants see the comparison stimuli). Relatedly, there are some individuals whose visual uncertainty is greater than 20 degrees standard deviation. This seems very large, and less likely in a free-viewing environment.

For our 2AFC, the reference stimulus is the actual clamped cursor, which lasts for 800 ms. The comparison stimulus is a 150-ms dot representation appearing near the reference. For measuring perception of visual motion, this duration is sufficient as previous studies used similar durations (Egly & Homa, 1984; Owsley et al., 1995). We think the 20-degree standard deviation is reasonable given that people fixate on the target, with only peripheral vision to process the fast moving cursor. The steep linear increase in visual uncertainty about visual motion is well documented. The last author of this paper has shown that the uncertainty of visual motion speed (though not about angels) follows the same steep trend (Wei et al., 2010). It is noteworthy that without using our measured visual uncertainty in Exp1, if we fit the adaptation data in Exp2 to “estimate” the visual uncertainty, they are in fact well aligned with each other (see Figure S7 and Supplementary Text 2). This is a strong support that our estimation is valid and accurate. We think this high visual uncertainty is an important message to the field. Thus we now highlighted its magnitude in our Discussion.

(2d) One important confound between clear and uncertain (blurred) visual conditions is the number of cursors on the screen. The number of cursors may have an attenuating effect on implicit adaptation simply due to task-irrelevant attentional demands (Parvin et al. 2022), rather than that of visual uncertainty. Could the authors provide a figure showing these blurred stimuli (gaussian clouds) in the context of the experimental paradigm? Note that we addressed this confound in the past by comparing participants with and without low vision, where only one visual cursor is provided for both groups (Tsay, Tan, et al. 2023).

Thank you for raising this important point about types of visual stimuli for manipulating uncertainty. We used Gaussian blur of a single cursor (similar to Burge et al., 2008) instead of a cloud of dots. We now added a figure inset to show how this blur looks.

Using a cursor cloud Makino et al., 2023; Tsay et al., 2021 to modulate visual uncertainty has inherent drawbacks that make it unsuitable for visuomotor adaptation. For the error clamp paradigm, the error is defined as angular deviation. The cursor cloud consists of multiple cursors spanning over a range of angles, which affects both the sensory uncertainty (the intended outcome) and the sensory estimate of angles (the error estimate, the undesired outcome). In Bayesian terms, the cursor cloud aims to modulate the sigma of a distribution (sigma_v       in         our       model), but it additionally affects the mean of the distribution (mu). This unnecessary confound is avoided by using cursor blurring, which is still a cursor with its center (mu) unchanged from a single cursor. Furthermore, as correctly pointed out in the original paper by Tsay et al., 2021, the cursor cloud often overlaps with the visual target, this “target hit” would affect adaptation, possibly via a reward learning mechanism (See Kim et al., 2019). This is a second confound that accompanies the cursor cloud.

Issue #3: More methodological details are needed.

(3a) It's unclear why, in Figure 4, PEA predicts an overshoot in terms of perceived hand position from the target. In PReMo, we specified a visual shift in the perceived target position, shifted towards the adapted hand position, which may result in overshooting of the perceived hand position with this target position. This visual shift phenomenon has been discovered in previous studies (e.g., (Simani, McGuire, and Sabes 2007)).

Visual shift, as it is called in Simani et al., 2007, is irrelevant for our task here. The data we are modeling are motor adaptation (hand position changes) and so-called proprioceptive changes (hand localization changes), both are measured and referenced in the extrinsic coordinate, not referenced to a visual target. For instance, the proprioceptive changes are either relative to the actual hand location (Exp 3) or relative to the goal (Fig 4A). We also don’t think visual shift is necessary in explaining the perceptual judgment of an unseen hand (the target shown during the judgment indeed has an effect of reducing the biasing effect of PE, see below for responses to reviewer 3).

In the PEA model, the reported hand angle is the result of integrating cues from the actual hand position and the estimated hand position (x_hand_hat) from previous movements. This integration process leads to the combined reported hand position potentially overshooting or undershooting, depending on the degree of adaptation. It is the changed proprioceptive cue (because the actively moved hand slowly adapted to the error clamp) leading to the overshoot of the perceived hand position.

In Results, we now explain these value changes with parentheses. Model details about the mechanisms of cue combination and model predictions can be found in Supplementary Text 1. We believe these detailed explanations can make this apparent.

(3b) The extent of implicit adaptation in Experiment 2, especially with smaller errors, is unclear. The implicit adaptation function seems to be still increasing, at least by visual inspection. Can the authors comment on this trend, and relatedly, show individual data points that help the reader appreciate the variability inherent to these data?

Indeed, the adaptation for small errors appears not completely saturated with our designated number of trials. However, this will not affect our model analysis. Our model fitting for PEA and other competing models is done on the time-series of adaptation, not on the saturated adaptation extent (see Fig 3A). Thus, despite that some conditions might not produce the full range of adaptation, the data is sufficient to constrain the models. We now mention this concern in Results; we also emphasize that the model not only explains the adaptation magnitude (operationally defined as adaptation extent measured at the same time, i.e., the end of the adaptation phase) but also the full learning process.

In response, we have included individual data points in the revised Figure 3B-D to provide a clear illustration of the extent of implicit adaptation, particularly for small perturbations.

(3c) The same participants were asked to return for multiple days/experiments. Given that the authors acknowledge potential session effects, with attenuation upon re-exposure to the same rotation (Avraham et al. 2021), how does re-exposure affect the current results? Could the authors provide clarity, perhaps a table, to show shared participants between experiments and provide evidence showing how session order may not be impacting results?

Thank you for raising the issue of session and re-exposure effects. First, we don’t think Exp1 has an effect on Exp4. Exp1 is a perceptual task and Exp4 is a motor adaptation task. Furthermore, Exp1 used random visual stimuli on both sides, thus it did not lead to any adaptation effect on its own. Second, Exp4 indeed had three sessions performed on three days, but the session effect does not change our main conclusion about the visual uncertainty. We used a 3-way repeated-measures anova (3 day x 3 perturbation x 2 visual uncertainty) revealed a significant main effect of day (F(2,36) = 17.693, p<0.001), indicating changes in performance across sessions (see Figure below). Importantly, the effects of perturbation and visual uncertainty (including their interactions) remain the same. The day factor did not interact with them. The main effect of day shows that the overall adaptation effect is reduced across days. Post-hoc pairwise comparisons elucidated that single-trial learning (STL) performance on Day 1 was significantly higher than on Day 2 (p = 0.004) and Day 3 (p < 0.001), with no significant difference between Day 2 and Day 3 (p = 0.106). Other ANOVA details: significant main effects for perturbation (F(1,36) = 8.872, p<0.001) and visual uncertainty (F(1,18) = 49.164, p<0.001), as well as a significant interaction between perturbation size and visual uncertainty (F(2,36) = 5.160, p = 0.013). There were no significant interactions involving the day factor with any other factors (all p > 0.182). Thus, the overall adaptation decreases over the days, but the day does not affect our concerned interaction effect of visual uncertainty and perturbation. The fact that their interaction preserved over different sessions strengthened our conclusion about how visual uncertainty systematically affects implicit adaptation.

Author response image 2.

(3d) The number of trials per experiment should be detailed more clearly in the Methods section (e.g., Exp 4). Moreover, could the authors please provide relevant code on how they implemented their computational models? This would aid in future implementation of these models in future work. I, for one, am enthusiastic to build on PEA.

We have clarified the number of trials conducted in each experiment, with detailed information now readily available in the Methods section of the main text. In addition, we have made the code for data analysis and modeling publicly accessible. These resources can be found in the updated "Data Availability" section of our paper.

(3f) In addition to predicting a correlation between proprioceptive shift and implicit adaptation on a group level, both PReMo and PEA (but not causal inference) predict a correlation between individual differences in proprioceptive shift and proprioceptive uncertainty with the extent of implicit adaptation (Tsay, Kim, et al. 2021). Interestingly, shift and uncertainty are independent (see Figures 4F and 6C in Tsay et al, 2021). Does PEA also predict independence between shift and uncertainty? It seems like PEA does predict a correlation.

Thank you for addressing this insightful question. Our PEA model indeed predicts a positive correlation (although not linear) between the proprioceptive uncertainty and the amplitude of the estimated hand position (x_hand_hat). This prediction is consistent with the simulations conducted, using the same parameters that were applied to generate the results depicted in

Figure 4B of our manuscript (there is a sign flip as x_hand_hat is negative).

Author response image 3.

Regarding the absence of a correlation observed in Tsay et al., 2021, we offer several potential explanations for this discrepancy. First, the variability observed in passive hand localization during motor adaptation (as in Tsay et al., 2021) does not directly equal proprioceptive uncertainty, which typically requires psychophysical testing to accurately assess. Second, our study showed that the proprioceptive bias attenuates during the repetitive measurements; in our Exp3, it decreased within a block of three trials. We noticed that Tsay et al., 2021 study used 36 measurements in a row without interleaving adaptation trials. Thus, the “averaged” proprioceptive bias in Tsay’s study might not reflect the actual bias during adaptation. We also noticed that that study showed large individual differences in both proprioceptive bias and proprioceptive variability (not uncertainty), thus getting a positive result, if it were really there, would require a large number of participants, probably larger than their n=30ish sample size. These putative explanations are not put in the revision, which already has a long discussion and has no space for discussing about a null result.

Reviewer #2 (Public Review):

Summary:

The authors present the Perceptual Error Adaptation (PEA) model, a computational approach offering a unified explanation for behavioral results that are inconsistent with standard state-space models. Beginning with the conventional state-space framework, the paper introduces two innovative concepts. Firstly, errors are calculated based on the perceived hand position, determined through Bayesian integration of visual, proprioceptive, and predictive cues. Secondly, the model accounts for the eccentricity of vision, proposing that the uncertainty of cursor position increases with distance from the fixation point. This elegantly simple model, with minimal free parameters, effectively explains the observed plateau in motor adaptation under the implicit motor adaptation paradigm using the error-clamp method. Furthermore, the authors experimentally manipulate visual cursor uncertainty, a method established in visuomotor studies, to provide causal evidence. Their results show that the adaptation rate correlates with perturbation sizes and visual noise, uniquely explained by the PEA model and not by previous models. Therefore, the study convincingly demonstrates that implicit motor adaptation is a process of Bayesian cue integration

Strengths:

In the past decade, numerous perplexing results in visuomotor rotation tasks have questioned their underlying mechanisms. Prior models have individually addressed aspects like aiming strategies, motor adaptation plateaus, and sensory recalibration effects. However, a unified model encapsulating these phenomena with a simple computational principle was lacking. This paper addresses this gap with a robust Bayesian integration-based model. Its strength lies in two fundamental assumptions: motor adaptation's influenced by visual eccentricity, a well-established vision science concept, and sensory estimation through Bayesian integration. By merging these well-founded principles, the authors elucidate previously incongruent and diverse results with an error-based update model. The incorporation of cursor feedback noise manipulation provides causal evidence for their model. The use of eye-tracking in their experimental design, and the analysis of adaptation studies based on estimated eccentricity, are particularly elegant. This paper makes a significant contribution to visuomotor learning research.

Weaknesses:

The paper provides a comprehensive account of visuomotor rotation paradigms, addressing incongruent behavioral results with a solid Bayesian integration model. However, its focus is narrowly confined to visuomotor rotation, leaving its applicability to broader motor learning paradigms, such as force field adaptation, saccadic adaptation, and de novo learning paradigms, uncertain. The paper's impact on the broader fields of neuroscience and cognitive science may be limited due to this specificity. While the paper excellently demonstrates that specific behavioral results in visuomotor rotation can be explained by Bayesian integration, a general computational principle, its contributions to other motor learning paradigms remain to be explored. The paper would benefit from a discussion on the model's generality and its limitations, particularly in relation to the undercompensating effects in other motor learning paradigms.

Thank you for your thoughtful review and recognition of the contributions our work makes towards understanding implicit motor adaptation through the Perceptual Error Adaptation (PEA) model. We appreciate your suggestion to broaden the discussion about the model's applicability beyond the visuomotor rotation paradigm, a point we acknowledge was not sufficiently explored in our initial discussion.

Our model is not limited to the error-clamp adaptation, where the participants were explicitly told to ignore the rotated cursor. The error-clamp paradigm is one rare example that implicit motor learning can be isolated in a nearly idealistic way. Our findings thus imply two key aspects of implicit adaptation: 1) localizing one’s effector is implicitly processed and continuously used to update the motor plan; 2) Bayesian cue combination is at the core of integrating movement feedback and motor-related cues (motor prediction cue in our model) when forming procedural knowledge for action control.

We will propose that the same two principles should be applied to various kinds of motor adaptation and motor skill learning, which constitutes motor learning in general. Most of our knowledge about motor adaptation is from visuomotor rotation, prism adaptation, force field adaptation, and saccadic adaptation. The first three types all involve localizing one’s effector under the influence of perturbed sensory feedback, and they also have implicit learning. We believe they can be modeled by variants of our model, or at least should consider using the two principles we laid out above to think of their computational nature. For skill learning, especially for de novo learning, the area still lacks a fundamental computational model that accounts for skill acquisition process on the level of relevant movement cues. Our model suggests a promising route, i.e., repetitive movements with a Bayesian cue combination of movement-related cues might underlie the implicit process of motor skills.

We added more discussion on the possible broad implications of our model in the revision.

Reviewer #3 (Public Review):

Summary

In this paper, the authors model motor adaptation as a Bayesian process that combines visual uncertainty about the error feedback, uncertainty about proprioceptive sense of hand position, and uncertainty of predicted (=planned) hand movement with a learning and retention rate as used in state space models. The model is built with results from several experiments presented in the paper and is compared with the PReMo model (Tsay, Kim, et al., 2022) as well as a cue combination model (Wei & Körding, 2009). The model and experiments demonstrate the role of visual uncertainty about error feedback in implicit adaptation.

In the introduction, the authors notice that implicit adaptation (as measured in error-clamp-based paradigms) does not saturate at larger perturbations, but decreases again (e.g. Moorehead et al., 2017 shows no adaptation at 135{degree sign} and 175{degree sign} perturbations). They hypothesized that visual uncertainty about cursor position increases with larger perturbations since the cursor is further from the fixated target. This could decrease the importance assigned to visual feedback which could explain lower asymptotes.

The authors characterize visual uncertainty for 3 rotation sizes in the first experiment, and while this experiment could be improved, it is probably sufficient for the current purposes. Then the authors present a second experiment where adaptation to 7 clamped errors is tested in different groups of participants. The models' visual uncertainty is set using a linear fit to the results from experiment 1, and the remaining 4 parameters are then fit to this second data set. The 4 parameters are 1) proprioceptive uncertainty, 2) uncertainty about the predicted hand position, 3) a learning rate, and 4) a retention rate. The authors' Perceptual Error Adaptation model ("PEA") predicts asymptotic levels of implicit adaptation much better than both the PReMo model (Tsay, Kim et al., 2022), which predicts saturated asymptotes, or a causal inference model (Wei & Körding, 2007) which predicts no adaptation for larger rotations. In a third experiment, the authors test their model's predictions about proprioceptive recalibration, but unfortunately, compare their data with an unsuitable other data set. Finally, the authors conduct a fourth experiment where they put their model to the test. They measure implicit adaptation with increased visual uncertainty, by adding blur to the cursor, and the results are again better in line with their model (predicting overall lower adaptation) than with the PReMo model (predicting equal saturation but at larger perturbations) or a causal inference model (predicting equal peak adaptation, but shifted to larger rotations). In particular, the model fits experiment 2 and the results from experiment 4 show that the core idea of the model has merit: increased visual uncertainty about errors dampens implicit adaptation.

Strengths

In this study, the authors propose a Perceptual Error Adaptation model ("PEA") and the work combines various ideas from the field of cue combination, Bayesian methods, and new data sets, collected in four experiments using various techniques that test very different components of the model. The central component of visual uncertainty is assessed in the first experiment. The model uses 4 other parameters to explain implicit adaptation. These parameters are 1) learning and 2) retention rate, as used in popular state space models, and the uncertainty (variance) of 3) predicted and 4) proprioceptive hand position. In particular, the authors observe that asymptotes for implicit learning do not saturate, as claimed before, but decrease again when rotations are very large and that this may have to do with visual uncertainty (e.g. Tsay et al., 2021, J Neurophysiol 125, 12-22). The final experiment confirms predictions of the fitted model about what happens when visual uncertainty is increased (overall decrease of adaptation). By incorporating visual uncertainty depending on retinal eccentricity, the predictions of the PEA model for very large perturbations are notably different from and better than, the predictions of the two other models it is compared to. That is, the paper provides strong support for the idea that visual uncertainty of errors matters for implicit adaptation.

Weaknesses

Although the authors don't say this, the "concave" function that shows that adaptation does not saturate for larger rotations has been shown before, including in papers cited in this manuscript.

The first experiment, measuring visual uncertainty for several rotation sizes in error-clamped paradigms has several shortcomings, but these might not be so large as to invalidate the model or the findings in the rest of the manuscript. There are two main issues we highlight here. First, the data is not presented in units that allow comparison with vision science literature. Second, the 1 second delay between the movement endpoint and the disappearance of the cursor, and the presentation of the reference marker, may have led to substantial degradation of the visual memory of the cursor endpoint. That is, the experiment could be overestimating the visual uncertainty during implicit adaptation.

The paper's third experiment relies to a large degree on reproducing patterns found in one particular paper, where the reported hand positions - as a measure of proprioceptive sense of hand position - are given and plotted relative to an ever-present visual target, rather than relative to the actual hand position. That is, 1) since participants actively move to a visual target, the reported hand positions do not reflect proprioception, but mostly the remembered position of the target participants were trying to move to, and 2) if the reports are converted to a difference between the real and reported hand position (rather than the difference between the target and the report), those would be on the order of ~20{degree sign} which is roughly two times larger than any previously reported proprioceptive recalibration, and an order of magnitude larger than what the authors themselves find (1-2{degree sign}) and what their model predicts. Experiment 3 is perhaps not crucial to the paper, but it nicely provides support for the idea that proprioceptive recalibration can occur with error-clamped feedback.

Perhaps the largest caveat to the study is that it assumes that people do not look at the only error feedback available to them (and can explicitly suppress learning from it). This was probably true in the experiments used in the manuscript, but unlikely to be the case in most of the cited literature. Ignoring errors and suppressing adaptation would also be a disastrous strategy to use in the real world, such that our brains may not be very good at this. So the question remains to what degree - if any - the ideas behind the model generalize to experiments without fixation control, and more importantly, to real-life situations.

Specific comments:

A small part of the manuscript relies on replicating or modeling the proprioceptive recalibration in a study we think does NOT measure proprioceptive recalibration (Tsay, Parvin & Ivry, JNP, 2020). In this study, participants reached for a visual target with a clamped cursor, and at the end of the reach were asked to indicate where they thought their hand was. The responses fell very close to the visual target both before and after the perturbation was introduced. This means that the difference between the actual hand position, and the reported/felt hand position gets very large as soon as the perturbation is introduced. That is, proprioceptive recalibration would necessarily have roughly the same magnitude as the adaptation displayed by participants. That would be several times larger than those found in studies where proprioceptive recalibration is measured without a visual anchor. The data is plotted in a way that makes it seem like the proprioceptive recalibration is very small, as they plot the responses relative to the visual target, and not the discrepancy between the actual and reported hand position. It seems to us that this study mostly measures short-term visual memory (of the target location). What is astounding about this study is that the responses change over time to begin with, even if only by a tiny amount. Perhaps this indicates some malleability of the visual system, but it is hard to say for sure.

Regardless, the results of that study do not form a solid basis for the current work and they should be removed. We would recommend making use of the dataset from the same authors, who improved their methods for measuring proprioception shifts just a year later (Tsay, Kim, Parvin, Stover, and Ivry, JNP, 2021). Although here the proprioceptive shifts during error-clamp adaptation (Exp 2) were tiny, and not quite significant (p<0.08), the reports are relative to the actual location of the passively placed unseen hand, measured in trials separate from those with reach adaptation and therefore there is no visual target to anchor their estimates to.

Experiment 1 measures visual uncertainty with increased rotation size. The authors cite relevant work on this topic (Levi & Klein etc) which has found a linear increase in uncertainty of the position of more and more eccentrically displayed stimuli.

First, this is a question where the reported stimuli and effects could greatly benefit from comparisons with the literature in vision science, and the results might even inform it. In order for that to happen, the units for the reported stimuli and effects should (also) be degrees of visual angle (dva).

As far as we know, all previous work has investigated static stimuli, where with moving stimuli, position information from several parts of the visual field are likely integrated over time in a final estimate of position at the end of the trajectory (a Kalman filter type process perhaps). As far as we know, there are no studies in vision science on the uncertainty of the endpoint of moving stimuli. So we think that the experiment is necessary for this study, but there are some areas where it could be improved.

Then, the linear fit is done in the space of the rotation size, but not in the space of eccentricity relative to fixation, and these do not necessarily map onto each other linearly. If we assume that the eye-tracker and the screen were at the closest distance the manufacturer reports it to work accurately at (45 cm), we would get the largest distances the endpoints are away from fixation in dva. Based on that assumed distance between the participant and monitor, we converted the rotation angles to distances between fixation and the cursor endpoint in degrees visual angle: 0.88, 3.5, and 13.25 dva (ignoring screen curvature, or the absence of it). The ratio between the perturbation angle and retinal distance to the endpoint is roughly 0.221, 0.221, and 0.207 if the minimum distance is indeed used - which is probably fine in this case. But still, it would be better to do fit in the relevant perceptual coordinate system.

The first distance (4 deg rotation; 0.88 dva offset between fixation and stimulus) is so close to fixation (even at the assumed shortest distance between eye and screen) that it can be considered foveal and falls within the range of noise of eye-trackers + that of the eye for fixating. There should be no uncertainty on or that close to the fovea. The variability in the data is likely just measurement noise. This also means that a linear fit will almost always go through this point, somewhat skewing the results toward linearity. The advantage is that the estimate of the intercept (measurement noise) is going to be very good. Unfortunately, there are only 2 other points measured, which (if used without the closest point) will always support a linear fit. Therefore, the experiment does not seem suitable to test linearity, only to characterize it, which might be sufficient for the current purposes. We'd understand if the effort to do a test of linearity using many more rotations requires too much effort. But then it should be made much clearer that the experiment assumes linearity and only serves to characterize the assumed linearity.

Final comment after the consultation session:

There were a lot of discussions about the actual interpretation of the behavioral data from this paper with regards to past papers (Tsay et al. 2020 or 2021), and how it matches the different variables of the model. The data from Tsay 2020 combined both proprioceptive information (Xp) and prediction about hand position (Xu) because it involves active movements. On the other hand, Tsay et al. 2021 is based on passive movements and could provide a better measure of Xp alone. We would encourage you to clarify how each of the variables used in the model is mapped onto the outcomes of the cited behavioral experiments.

The reviewers discussed this point extensively during the consultation process. The results reported in the Tsay 2020 study reflect both proprioception and prediction. However, having a visual target contributes more than just prediction, it is likely an anchor in the workspace that draws the response to it. Such that the report is dominated by short-term visual memory of the target (which is not part of the model). However, in the current Exp 3, as in most other work investigating proprioception, this is calculated relative to the actual direction.

The solution is fairly simple. In Experiment 3 in the current study, Xp is measured relative to the hand without any visual anchors drawing responses, and this is also consistent with the reference used in the Tsay et al 2021 study and from many studies in the lab of D. Henriques (none of which also have any visual reach target when measuring proprioceptive estimates). So we suggest using a different data set that also measures Xp without any other influences, such as the data from Tsay et al 2021 instead.

These issues with the data are not superficial and can not be solved within the model. Data with correctly measured biases (relative to the hand) that are not dominated by irrelevant visual attractors would actually be informative about the validity of the PEA model. Dr. Tsay has so much other that we recommend using a more to-the-point data set that could actually validate the PEA model.

As the comments are repetitive at some places, we summarize them into three questions and address it one by one below:

(1) Methodological Concerns about visual uncertainty estimation in Experiment 1: a) the visual uncertainty is measured in movement angles (degrees), while the unit in vision science is in visual angles (vda). This mismatch of unit hinders direct comparison between the found visual uncertainty and those reported in the literature, and b) a 1-second delay between movement endpoint and the reference marker presentation causes an overestimate of visual uncertainty due to potential degradation of visual memory. c) The linear function of visual uncertainty is a result of having only three perturbation sizes.

a) As noted by the reviewer, our visual uncertainty is about cursor motion direction in the display plane, which has never been measured before. We do not think our data is comparable to any findings in visual science about fovea/peripheral comparison. We quoted Klein and others’ work Klein & Levi, 1987; Levi et al., 1987 in vision science since their studies showed that the deviation from the fixation is associated with the increase in visual uncertainty. Their study thus inspired our Exp1 to probe how our concerned visual uncertainty (specifically for visual motion direction) changes with an increasing deviation from the fixation. We believe that any model and its model parameters should be specifically tailored to the task or context it tries to emulate. In our case, motion direction in a center-out reaching setting is the modeled context, and all the relevant model parameters should be specified in movement angles.

b) The 1s delay of the reference cursor appears to have minimum impact on the estimate of visual uncertainty, based on previous vision studies. Our Exp1 used a similar visual paradigm by White et al., 1992, which shows that delay does not lead to an increase in visual uncertainty over a broad range of values (from 0.2s to >1s, see their Figure 5-6). We will add more methodology justifications in our revision.

c) We agree that if more angles are tested we can be more confident about the linearity of visual uncertainty. However, the linear function is a good approximation of visual uncertainty (as shown in Figure 2C). More importantly, our model performance does not hinge on a strict linear function. Say, if it is a power function with an increasing slope, our model will still predict the major findings presented in the paper, as correctly pointed out by the reviewer. It is the increasing trend of visual uncertainty, which is completely overlooked by previous studies, that lead to various seemingly puzzling findings in implicit adaptation. Lastly, without assuming a linear function, we fitted the large dataset of motor adaptation from Exp2 to numerically estimate the visual uncertainty. This estimated visual uncertainty has a strong linear relationship with perturbation size (R = 0.991, p<0.001). In fact, the model-fitted visual uncertainty is very close to the values we obtained in Exp1. We now included this analysis in the revision. See details in Supplementary text 2 and Figure S7.

(2) Experiment 3's: the reviewer argues that the Tsay et al., 2020 data does not accurately measure proprioceptive recalibration, thus it is not suitable for showing our model’s capacity in explaining proprioceptive changes during adaptation.

Response: We agree that the data from Tsay et al., 2020 is not from passive localization, which is regarded as the widely-accepted method to measure proprioceptive recalibration, a recalibration effect in the sensory domain. The active localization, as used in Tsay et al., 2020, is hypothesized as closely related to people’s forward prediction (where people want to go as the reviewer put it in the comments). However, we want to emphasize that we never equated Tsay’s findings as proprioceptive recalibration: throughout the paper we call them “reported hand location”. We reserved “proprioceptive recalibration” to our own Exp3, which used a passive localization method. Thus, we are not guilty of using this term. Secondly, as far as we know, localization bias or changes, no matter measured by passive or active methods, have not been formally modeled quantitatively. We believe our model can explain both, at least in the error-clamp adaptation setting here. Exp3 is for passive localization, the proprioceptive bias is caused by the biasing effect from the just-perceived hand location (X_hand_hat) from the adaptation trial. Tsay et al. 2020 data is for active localization, whose bias shows a characteristic change from negative to positive. This can be explained by just-perceived hand location (X_hand_hat again) and a gradually-adapting hand (X_p). We think this is a significant advance in the realm of proprioceptive changes in adaptation. Of course, our idea can be further tested in other task conditions, e.g., conventional visuomotor rotation or even gain adaptation, which should be left for future studies.

For technical concerns, Tsay et al., 2020 data set is not ideal: when reporting hand location, the participants view the reporting wheel as well as the original target. As correctly pointed out by the reviewer, the presence of the target might provide an anchoring cue for perceptual judgment, which acts as an attractor for localization. If it were the case, our cue combination would predict that this extra attractor effect would lead to a smaller proprioceptive effect than that is currently reported in their paper. The initial negative bias will be closer to the target (zero), and the later positive bias will be closer to the target too. However, the main trend will remain, i.e. the reported hand location would still show the characteristic negative-to-positive change. The attractor effect of the target can be readily modeled by giving less weight to the just-perceived hand location (X_hand_hat). Thus, we would like to keep Tsay et al., 2020 data in our paper but add some explanations of the limitations of this dataset as well as how the model would fare with these limitations.

That being said, our model can explain away both passive and active localization during implicit adaptation elicited by error clamp. The dataset from Tsay et al., 2021 paper is not a good substitute for their 2020 paper in terms of modeling, since that study interleaved some blocks of passive localization trials with adaptation trials. This kind of block design would lead to forgetting of both adaptation (Xp in our model) and the perceived hand (X_hand_hat in our model), the latter is still not considered in our model yet. As our Exp3, which also used passive localization, shows, the influence of the perceived hand on proprioceptive bias is short-lived, up to three trials without adaptation trials. Of course, it would be of great interest to design future studies to study how the proprioceptive bias changes over time, and how its temporal changes relate to the perceptual error. Our model provides a testbed to move forward in this direction.

(3) The reviewer raises concerns about the study's assumption that participants ignore error feedback, questioning the model's applicability to broader contexts and real-world scenarios where ignoring errors might not be viable or common.

Reviewer 2 raised the same question above. We moved our responses here. “We appreciate your suggestion to broaden the discussion about the model's applicability beyond the visuomotor rotation paradigm, a point we acknowledge was not sufficiently explored in our initial discussion.

Our model is not limited to the error-clamp adaptation, where the participants were explicitly told to ignore the rotated cursor. The error-clamp paradigm is one rare example that implicit motor learning can be isolated in a nearly idealistic way. Our findings thus imply two key aspects of implicit adaptation: 1) localizing one’s effector is implicitly processed and continuously used to update the motor plan; 2) Bayesian cue combination is at the core of integrating movement feedback and motor-related cues (motor prediction cue in our model) when forming procedural knowledge for action control.

We will propose that the same two principles should be applied to various kinds of motor adaptation and motor skill learning, which constitutes motor learning in general. Most of our knowledge about motor adaptation is from visuomotor rotation, prism adaptation, force field adaptation, and saccadic adaptation. The first three types all involve localizing one’s effector under the influence of perturbed sensory feedback, and they also have implicit learning. We believe they can be modeled by variants of our model, or at least should consider using the two principles we laid out above to think of their computational nature. For skill learning, especially for de novo learning, the area still lacks a fundamental computational model that accounts for skill acquisition process on the level of relevant movement cues. Our model suggests a promising route, i.e., repetitive movements with a Bayesian cue combination of movement-related cues might underlie the implicit process of motor skills.”

We also add one more important implication of our model: as stated above, our model also explains that the proprioceptive changes, revealed by active or passive localization methods, are brought by (mis)perceived hand localization via Bayesian cue combination. This new insight, though only tested here using the error-clamp paradigm, can be further utilized in other domains, e.g., conventional visuomotor rotation or force field adaptation. We hope this serves as an initial endeavor in developing some computational models for proprioception studies. Please see the extended discussion on this matter in the revision.

Recommendations for the authors:

Revisions:

All three reviewers were positive about the work and have provided a set of concrete and well-aligned suggestions, which the authors should address in a revised version of the article. These are listed below.

A few points of particular note:

(1) There are a lot of discussions about the actual interpretation of behavioral data from this paper or past papers (Tsay et al. 2020 or 2021) and how it matches the different variables of the model.

(2) There are some discussions on the results of the first experiment, both in terms of how it is reported (providing degrees of visual angle) and how it is different than previous results (importance of the point of fixation). We suggest also discussing a few papers on eye movements during motor adaptation from the last years (work of Anouk de Brouwer and Opher Donchin). Could the authors also discuss why they found opposite results to that of previous visual uncertainty studies (i.e., visual uncertainty attenuates learning with large, but not small, visual errors); rather than the other way around as in Burge et al and Tsay et al 2021 and Makino Nozaki 2023 (where visual uncertainty attenuates small, but not large, visual errors).

(3) It is recommended by several reviewers to discuss the applicability of the model to other areas/perturbations.

(4) Several reviewers and I believe that the impact of the paper would be much higher if the code to reproduce all the simulations of the model is made available to the readers. In addition, while I am very positive about the fact that the authors shared the data of their experiments, metadata seems to be missing while they are highly important because these data are otherwise useless.

Thank you for the concise summary of the reviewers’ comments. We have addressed their concerns point by point.

Reviewer #2 (Recommendations For The Authors):

L142: The linear increase in visual uncertainty should be substantiated by previous research in vision science. Please cite relevant papers and discuss why the linear model is considered reasonable.

We cited relevant studies in vision science. Their focus is more about eccentricity inflate visual uncertainty, similar to our findings that deviations from the fixation direction inflate visual uncertainty about motion direction.

We also want to add that our model performance does not hinge on a strict linear function of visual uncertainty. Say, if it is a power function with an increasing slope, our model will still predict the major findings presented in the paper. It is the increasing trend of visual uncertainty, which is completely overlooked by previous studies, that lead to various seemingly puzzling findings in implicit adaptation. Furthermore, without assuming a linear function, we fitted the large dataset of motor adaptation from Exp2 to numerically estimate the visual uncertainty. This estimated visual uncertainty has a strong linear relationship with perturbation size (R = 0.991, p<0.001). In fact, the model-fitted visual uncertainty is very close to the values we obtained in Exp1. We now included this new analysis in the revision. See details in Supplementary text 2 and Figure S7.

L300: I found it challenging to understand the basis for this conclusion. Additional explanatory support is required.

We unpacked this concluding sentence as follows:

“The observed proprioceptive bias is formally modeled as a result of the biasing effect of the perceived hand estimate x_hand_hat. In our mini-block of passive localization, the participants neither actively moved nor received any cursor perturbations for three trials in a row. Thus, the fact that the measured proprioceptive bias is reduced to nearly zero at the third trial suggests that the effect of perceived hand estimate x_hand_hat decays rather rapidly.”

L331: For the general reader, a visual representation of what the blurring mask looks like would be beneficial.

Thanks for the nice suggestion. We added pictures of a clear and a blurred cursor in Figure 5D.

L390: This speculation is intriguing. It would be helpful if the authors explained why they consider causal inference to operate at an explicit process level, as the reasoning is not clear here, although the idea seems plausible.

Indeed, our tentative conclusion here is only based on the model comparison results here. It is still possible that causal inference also work for implicit adaptation besides explicit adaptation. We make a more modest conclusion in the revision:

“The casual inference model is also based on Bayesian principle, then why does it fail to account for the implicit adaptation? We postulate that the failure of the causal inference model is due to its neglect of visual uncertainty as a function of perturbation size, as we revealed in Experiment 1. In fact, previous studies that advocating the Bayesian principle in motor adaptation have largely focused on experimentally manipulating sensory cue uncertainty to observe its effects on adaptation (Burge et al., 2008; He et al., 2016; Körding & Wolpert, 2004; Wei & Körding, 2010), similar to our Experiment 4. Our findings suggest that causal inference of perturbation alone, without incorporating visual uncertainty, cannot fully account for the diverse findings in implicit adaptation. The increase in visual uncertainty by perturbation size is substantial: our Experiment 1 yielded an approximate seven-fold increase from a 4° perturbation to a 64° perturbation. We have attributed this to the fact that people fixate in the desired movement direction during movements. Interestingly, even for conventional visuomotor rotation paradigm where people are required to “control” the perturbed cursor, their fixation is also on the desired direction, not on the cursor itself (de Brouwer, Albaghdadi, et al., 2018; de Brouwer, Gallivan, et al., 2018). Thus, we postulate that a similar hike in visual uncertainty in other “free-viewing” perturbation paradigms. Future studies are warranted to extend our PEA model to account for implicit adaptation in other perturbation paradigms.”

L789: The method of estimating Sigma_hand in the brain was unclear. Since Bayesian computation relies on the magnitude of noise, the cognitive system must have estimates of this noise. While vision and proprioception noise might be directly inferred from signals, the noise of the hand could be deduced from the integration of these observations or an internal model estimate. This process of estimating noise magnitude is theorized in recursive Bayesian integration models (or Kalman filtering), where the size estimate of the state noise (sigma_hand) is updated concurrently with the state estimate (x_hand hat). The equation in L789 and the subsequent explanation appear to assume a static model of noise estimation. However, in practice, the noise parameters, including Sigma_hand, are likely dynamic and updated with each new observation. A more detailed explanation of how Sigma_hand is estimated and its role in the cognitive process.

This is a great comment. In fact, if a Kalman filter is used, the learning rate and the state noise all should be dynamically updated on each trial, under the influence of the observed (x_v). In fact, most adaptation models assume a constant learning rate, including our model here. But a dynamic learning rate (B in our model) is something worth trying. However, in our error-clamp setting, x_v is a constant, thus this observation variable cannot dynamically update the Kalman filter; that’s why we opt to use a “static” Bayesian model to explain our datasets. Thus, Sigma_hand can be estimated by using Bayesian principles as a function of three cues available, i.e., the proprioceptive cue, the visual cue, and the motor prediction cue. We added a

detailed derivation of sigma_hand in the revision in Supplementary text 1.

Reviewer #3 (Recommendations For The Authors):

We observed values in Fig 2C for the 64-degree perturbation that seem to be outliers, i.e., greater than 50 degrees. It is unclear how a psychometric curve could have a "slope" or JNP of over 60, especially considering that the tested range was only 60. Since the data plotted in panel C is a collapse of the signed data in panel B, it is perplexing how such large data points were derived, particularly when the signed uncertainty values do not appear to exceed 30.

Related to the previous point, we would also recommend connecting individual data points: if the uncertainty increases (linearly or otherwise), then people with low uncertainty at the middle distance should also have low uncertainty at the high distance, and people with high uncertainty at one point, should also have that at other distances. Or perhaps the best way to go about this is to use the uncertainty at the two smaller perturbations to predict uncertainty at the largest perturbation for each participant individually?

Thank you for your suggestion to examine the consistency of individual levels of visual uncertainty across perturbation sizes. First, a sigma_v of 60 degrees is well possible, naturally falling out of the experimental data. It shows some individuals indeed have large visual uncertainty. Given these potential outliers (which should not be readily removed as we don’t have any reason to do so), we estimated the linear function of sigma_v with a robust method, i.e., the GLM with a gamma distribution, which favors right-skewed distribution that can well capture positive outliers. Furthermore, we added in our revision a verification test of our estimates of sigma_v: we used Exp2’s adaptation data to estimate sigma_v without assuming its linear dependency. As shown, the model-fitted sigma_v closely matched the estimated ones from Exp1 (see Supplementary text 2 and Figure S7).

We re-plotted the sigma_v with connected data points provided, and the data clearly indicate that individuals exhibit consistent levels of visual uncertainty across different perturbation sizes, i.e. those with relatively lower uncertainty at middle distances (in fact, angles) tend to exhibit relatively lower uncertainty at higher distances too, and similarly, those with higher uncertainty at one distance maintain that level of uncertainty at other distances. This is confirmed by spearman correlation analysis to assess the consistency of uncertainties across different degrees of perturbation among individuals. Again, we observed significant correlations between perturbation angles, indicating good individual consistency (4 and 16 degrees, rho = 0.759, p<0.001; 16 and 64 degrees, rho = 0.527, p = 0.026).

Author response image 4.

The illustration in Fig 2A does not seem to show a stimulus that is actually used in the experiment (looks like about -30{degree sign} perturbation). It would be good to show all possible endpoints with all other visual elements to scale - including the start-points of the PEST procedure.

Thanks for the suggestion. We updated Fig 2A to show a stimulus of +16 degree, as well as added an additional panel to show all the possible endpoints.

Finally (related to the previous point), in lines 589-591 it says the target is a blue cross. Then in lines 614-616, it says participants are to fixate the blue cross or the start position. The start position was supposed to have disappeared, so perhaps the blue plus moved to the start position (which could be the case, when looking at the bottom panel in Fig 2A, although in the illustration the plus did not move fully to the start position, just toward it to some degree). Perhaps the descriptions need to be clarified, or it should be explained why people had to make an eye movement before giving their judgments. And if people could have made either 1) no eye movement, but stayed at fixation, 2) moved to the blue plus as shown in the last panel in Fig 2A, or 3) fixated on the home position, we'd be curious to know if this affected participants' judgments.

Thanks for pointing that out. The blue cross serves as the target in the movement task, then disappears with the cursor after 800ms of frozen time. The blue cross then appeared in the discrimination task at the center of the screen, i.e. the start location. Subjects were asked to fixate at the blue cross during the visual discrimination task. Note this return the fixation to the home position is exactly what we will see in typical error-clamp adaptation: once the movement is over, people guided their hand back to the home position. We performed a pilot study to record the typical fixation pattern during error-clamp adaptation, and Exp1 was intentionally designed to mimic its fixation sequence. We have now updated the description of Figure 2A, emphasizing the stimulus sequence. .

In Figure 4A, the label "bias" is confusing as that is used for recalibrated proprioceptive sense of hand position as well as other kinds of biases elsewhere in the paper. What seems to be meant is the integrated hand position (x-hat_hand?) where all three signals are apparently combined. The label should be changed and/or it should be clarified in the caption.

Thanks for pointing that out, it should be x_hand_hat, and we have corrected this in the revised version of Figure 4.

In the introduction, it is claimed that larger perturbations have not been tested with "implicit adaptation" paradigms, but in the same sentence, a paper is cited (Moorehead et al., 2017) that tests a rotation on the same order of magnitude as the largest one tested here (95{degree sign}), as well as much larger rotations (135{degree sign} and 175{degree sign}). With error-clamps. Interestingly, there is no adaptation in those conditions, which seems more in line with the sensory cue integration model. Can the PEA model explain these results as well? If so, this should be included in the paper, and if not, it should be discussed as a limitation.

First, we double checked our manuscript and found that we never claimed that larger perturbations had not been tested.

We agree that it is always good to have as many conditions as possible. However, the 135 and 175 degree conditions would lead to minimum adaptation, which would not help much in terms of model testing. We postulated that this lack of adaptation is simply due to the fact that people cannot see the moving cursor, or some other unknown reasons. Our simple model is not designed to cover those kinds of extreme cases.

Specify the size of the arc used for the proprioceptive tests in Exp 3 and describe the starting location of the indicator (controlled by the left hand). Ideally, the starting location should have varied across trials to avoid systematic bias.

Thank you for the comments. The size of the arc used during these tests, as detailed in the methods section of our paper, features a ring with a 10 cm radius centered at the start position. This setup is visually represented as a red arc in Figure 7B.

After completing each proprioceptive test trial, participants were instructed to position the indicator at approximately -180° on the arc and then relax their left arm. Although the starting location for the subsequent trial remained at-180°, it was not identical for every trial, thereby introducing slight variability.

Please confirm that the proprioceptive biases plotted in Fig 4E are relative to the baseline.

Thank you for bringing this to our attention. Yes, the proprioceptive biases illustrated in Figure 4E are indeed calculated relative to the baseline measurements. We have added this in the method part.

Data availability: the data are available online, but there are some ways this can be improved. First, it would be better to use an open data format, instead of the closed, proprietary format currently used. Second, there is no explanation for what's in the data, other than the labels. (What are the units? What preprocessing was done?) Third, no code is made available, which would be useful for a computational model. Although rewriting the analyses in a non-proprietary language (to increase accessibility) is not a reasonable request at this point in the project, I'd encourage it for future projects. But perhaps Python, R, or Julia code that implements the model could be made available as a notebook of sorts so that other labs could look at (build on) the model starting with correct code - increasing the potential impact of this work.

Great suggestions. We are also fully supportive of open data and open science. We now:

(1) Updated our data and code repository to include the experimental data in an open data format (.csv) for broader accessibility.

(2) The data are now accompanied by detailed descriptions to clarify their contents.

(3) We have made the original MATLAB (.m) codes for data analysis, model fitting and simulation available online.

(4) We also provide the codes in Jupyter Notebook (.ipynb) formats.

These updates can be found in the revised “Data Availability” section of our manuscript.

References

Bromberg, Z., Donchin, O., & Haar, S. (2019). Eye Movements during Visuomotor Adaptation Represent Only Part of the Explicit Learning. eNeuro, 6(6). https://doi.org/10.1523/ENEURO.0308-19.2019

Burge, J., Ernst, M. O., & Banks, M. S. (2008). The statistical determinants of adaptation rate in human reaching. Journal of Vision, 8(4), 1–19.

de Brouwer, A. J., Gallivan, J. P., & Flanagan, J. R. (2018). Visuomotor feedback gains are modulated by gaze position. Journal of Neurophysiology, 120(5), 2522–2531.

Egly, R., & Homa, D. (1984). Sensitization of the visual field. Journal of Experimental Psychology. Human Perception and Performance, 10(6), 778–793.

Kim, H. E., Parvin, D. E., & Ivry, R. B. (2019). The influence of task outcome on implicit motor learning. eLife, 8. https://doi.org/10.7554/eLife.39882

Klein, S. A., & Levi, D. M. (1987). Position sense of the peripheral retina. JOSA A, 4(8), 1543–1553.

Levi, D. M., Klein, S. A., & Yap, Y. L. (1987). Positional uncertainty in peripheral and amblyopic vision. Vision Research, 27(4), 581–597.

Makino, Y., Hayashi, T., & Nozaki, D. (2023). Divisively normalized neuronal processing of uncertain visual feedback for visuomotor learning. Communications Biology, 6(1), 1286.

Owsley, C., Ball, K., & Keeton, D. M. (1995). Relationship between visual sensitivity and target localization in older adults. Vision Research, 35(4), 579–587.

Simani, M. C., McGuire, L. M. M., & Sabes, P. N. (2007). Visual-shift adaptation is composed of separable sensory and task-dependent effects. Journal of Neurophysiology, 98(5), 2827–2841.

Tsay, J. S., Avraham, G., Kim, H. E., Parvin, D. E., Wang, Z., & Ivry, R. B. (2021). The effect of visual uncertainty on implicit motor adaptation. Journal of Neurophysiology, 125(1), 12–22.

Tsay, J. S., Chandy, A. M., Chua, R., Miall, R. C., Cole, J., Farnè, A., Ivry, R. B., & Sarlegna, F. R. (2024). Minimal impact of proprioceptive loss on implicit sensorimotor adaptation and perceived movement outcome. bioRxiv : The Preprint Server for Biology. https://doi.org/10.1101/2023.01.19.524726

Tsay, J. S., Kim, H., Haith, A. M., & Ivry, R. B. (2022). Understanding implicit sensorimotor adaptation as a process of proprioceptive re-alignment. eLife, 11, e76639.

Wei, K., Stevenson, I. H., & Körding, K. P. (2010). The uncertainty associated with visual flow fields and their influence on postural sway: Weber’s law suffices to explain the nonlinearity of vection. Journal of Vision, 10(14), 4.

White, J. M., Levi, D. M., & Aitsebaomo, A. P. (1992). Spatial localization without visual references. Vision Research, 32(3), 513–526.

Author response:

The following is the authors’ response to the original reviews.

eLife Assessment

The authors identify new mechanisms that link a PIK3R1 mutant to cellular signaling and division in Activated PI3 Kinase Delta Syndrome 1 and 2 (APDS1/2). The conclusion that this mutant serves as a dominant negative form of the protein, impacting PI3K complex assembly and IRS/AKT signaling, is important, and the evidence from constitutive and inducible systems in cultured cells is convincing. Nevertheless, there are several limitations relating to differences between cell lines and expression systems, as well as more global characterization of the protein interaction landscape, which would further enhance the work.

We are pleased by this fair assessment, while noting that this work relates to APDS2 (PIK3R1-related) rather than APDS1 (PIK3CD-related). Our findings we believe are clear, but the observation that studies including more global proteomics/phosphoproteomics in cells expressing mutants at endogenous levels would add further insight is well made. We hope that this report may motivate such studies by laboratories with wider access to primary cells from patients and knock-in mice.

Public Reviews

Reviewer #1 (Public Review):

Summary:

This study provides convincing data showing that expression of the PIK3R1(delta Exon11) dominant negative mutation in Activated PI3K Delta Syndrome 1/2 (APDS1/2) patient-derived cells reduces AKT activation and p110δ protein levels. Using a 3T3-L1 model cell system, the authors show that overexpressed p85α delta Exon 11) displays reduced association with the p110α catalytic subunit but strongly interacts with Irs1/2. Overexpression of PIK3R1 dominant negative mutants inhibits AKT phosphorylation and reduces cellular differentiation of preadipocytes. The strength of this article is the clear results derived from Western blots analysis of cell signaling markers (e.g. pAKT1), and co-immunoprecipitation of PI3K holoenzyme complexes and associated regulatory factors (e.g. Irs1/2). The experimental design, interpretation, and quantification broadly support the authors' conclusions.

Strengths:

The authors analyze a variety of PIK3R1 mutants (i.e. delta Exon11, E489K, R649W, and Y657X), which reveals a range of phenotypes that support the proposed model for dominant negative activity. The use of clonal cell lines with doxycycline-induced expression of the PIK3R1 mutants (DExon 11, R649W, and Y657X) provides convincing experimental data concerning the relationship between p85α mutant expression and AKT phosphorylation in vivo. The authors convincingly show that p85α delta Exon11, R649W, or Y657X) is unable to associate with p110α but instead more strongly associates with Irs1/2 compared to wild type p85α. This helps explain why the authors were unable to purify the recombinant p110α/p85α delta Exon 11) heterodimeric complex from insect cells.

Weaknesses:

Future experimentation will be needed to reconcile the cell type specific differences (e.g. APDS2 patient-derived cells vs. the 3T3-L1 cell model system) in PIK3R1 mutant behavior reported by the authors.

This is a fair comment. It has been established for many years that relative protein levels even of wild type PIK3CA and PIK3R1 gene products influence sensitivity of PI3K to growth factor stimulation. Such issues of stoichiometry become exponentially more complicated when the numerous potential interactions among the full repertoire of Class 1 PI3K regulatory subunits (3 splice variants of PIK3R1, and also PIK3R2 and PIK3R3) and corresponding catalytic subunits (PIK3CA, PIK3CB, PIK3CD) are considered, and when different activities and stabilities of PIK3R1 mutants are added to the mix. It thus seems obvious to us that different levels of expression of different mutants in different cellular contexts will have different signalling consequences. We establish a paradigm in this paper using an overexpression system, and we strongly agree that this merits further investigation in a wider variety of primary cells (or cells with knock in at the endogenous locus), where available.

An unbiased proteomic study that broadly evaluates the cell signaling landscape could provide a more holistic understanding of the APDS2 and SHORT mutants compared to a candidate-based approach.

We agree. This would be highly informative, but we think would best be carried out in both “metabolic” and “immune” cells with endogenous levels of expression of SHORT or APDS2 PIK3R1 mutants. These are not all currently available to us, and require follow up studies.

Additional biochemical analysis of p110α/p85α delta Exon 11 complex is needed to explain why this mutant regulatory subunit does not strongly associate with the p110 catalytic subunit.

We agree. We present this observation in our overexpression system, which is clear and reproducible, even though somewhat surprising. The failure to bind p110a is likely not absolute, as sufficient p110a-p85a^DEx11 was synthesised in vitro in a prior study to permit structural and biochemical studies, although a series of technical workarounds were required to generate enough heterodimeric PI3K to study in vitro given the manifest instability of the complex, particularly when concentrated (PMID 28167755). We already note in discussion that p85a can homodimerize and bind PTEN, likely among other partners, and it may be that the APDS2 deletion strongly favours binding to proteins that effectively compete with p110a. However this requires further study of the wider interactome of the mutant PIK3R1, which, as noted above, are beyond the scope of the current study.

It remains unclear why p85α delta Exon 11 expression reduces p110δ protein levels in APDS2 patient-derived dermal fibroblasts.

We caution that we only had the opportunity to study dermal fibroblasts cultured from a single APDS2 patient, as noted in the paper, and so replication of this finding in future will be of interest. Nevertheless the observation is robust and reproducible in these cells, and we agree that this apparently selective effect on p110d  is not fully explained. Having said that, it has been observed previously that heterodimers of the DEx11 p85a variant with either p110a or p110d are unstable, and when the unstable complexes were eventually synthesised, p110a and p110d were demonstrated to show differences in engagement with the mutant p85, with greater disruption of inhibitory interactions observed for p110d (PMID 28167755). It is thus not a great stretch to imagine that as well as disinhibiting p110d more, the DEx11 p85a variant also destabilises the p85a-p110d complex more, potentially explaining its near disappearance in cells with low baseline p110d expression. Following on from the preceding question and response, however, is an alternative explanation, based on the 3T3-L1 overexpression studies in this paper, wherein we were unable to demonstrate binding of p110a by DEx11 p85a. If, in any given cellular context, the mutant p85 could bind p110d but not p110a, then the destabilising effect would be observed only for p110d. So in summary, we believe the selective effect on p110d is explained by differences in binding kinetics and heterodimer stability for different DEx11 p85a-containing complexes. The net effect of these differences may vary among cell types depending on relative levels of subunit expression.

This study would benefit from a more comprehensive biochemical analysis of the described p110α/p85α, p110β/p85α, and p110δ/p85α mutant protein complexes. The current limitation of this study to the use of a single endpoint assay to measure PI3K lipid kinase activity in the presence of a single regulatory input (i.e. RTK-derived pY peptide). A broader biochemical analysis of the mutant PI3K complexes across the canonical signaling landscape will be important for establishing how competition between wild-type and mutant regulatory subunits is regulated in different cell signaling pathways.

We agree that a wider analysis of upstream inputs and downstream network would be of interest, though as noted above the ultimate functional consequences of mutants will be an amalgam of any differential signalling effects of complexes that are stable enough to function, and differential effects of mutant p85a on the kinetics of distinct heterodimer assembly and stability. In this paper we seek to suggest a paradigm worthy of further, deeper assessment. We note that the search space here is large indeed (A. different cell types with differing profiles of PI3K subunit expression B. Multiple upstream stimuli and C. Multiple downstream outputs, with timecourse of responses an additional important factor to consider). These studies are realistically beyond the scope of the current work, but we hope that further studies, as suggested by the reviewer, follow.

Reviewer #2 (Public Review)

Summary:

Patsy R. Tomlinson et al; investigated the impact of different p85alpha variants associated with SHORT syndrome or APDS2 on insulin-mediated signaling in dermal fibroblasts and preadipocytes. They find no evidence of hyperactive PI3K signalling monitored by pAKT in APDS2 patient-derived dermal fibroblast cells. In these cells p110alpha protein levels were comparable to levels in control cells, however, the p110delta protein levels were strongly reduced. Remarkably, the truncated APDS2-causal p85alpha variant was less abundant in these cells than p85alpha wildtype. Afterwards, they studied the impact of ectopically expressed p85alpha variants on insulin-mediated PI3K signaling in 3T3-L1 preadipocytes. Interestingly they found that the truncated APDS2-causal p85alpha variant impaired insulin-induced signaling. Using immunoprecipitation of p110alpha they did not find truncated APDS2-causal p85alpha variant in p110alpha precipitates. Furthermore, by immunoprecipitating IRS1 and IRS2, they observed that the truncated APDS2-causal p85alpha variant was very abundant in IRS1 and IRS2 precipitates, even in the absence of insulin stimulation. These important findings add in an interesting way possible mechanistic explanation for the growing number of APDS2 patients described with features of SHORT syndrome.

Strengths:

Based on state-of-the-art functional investigation the authors propose indicating a loss-of-function activity of the APDS2-disease causing p85alpha variant in preadipocytes providing a possible mechanistic explanation for the growing number of APDS2 patients described with features of SHORT syndrome.

Weaknesses:

Related to Figure 1: PIK3R1 expression not only by Western blotting but also by quantifying the RNA transcripts, e.g. mutant and wildtype transcripts, was not performed. RNA expression analysis would further strengthen the suggested impaired stabilization/binding.

It is not completely clear to us how further PIK3R1 mRNA analysis would enhance the points we seek to make. Perhaps the reviewer’s point is that changes in protein expression could be explained by reduced transcription rather than having anything to do with altered protein turnover? As shown in Figure 1 supplemental figure 1, sequencing cDNA from each of the primary cell lines studied indicates that both mutant and WT alleles are expressed at or close to 50% of the total mRNA for PIK3CA or PIK3R1 as relevant. While this is not strictly quantitative, allied to prior evidence that these are dominant alleles which require to be expressed to exert their effect, with no evidence for altered mRNA expression of these variants in prior studies, we don’t believe any further quantification of mRNA expression would add value.

Related to Figure 2

As mentioned by the authors in the manuscript the expression of p110delta but also p110beta in 3T3-L1 preadipocytes ectopically expressing p85alpha variants has not been analyzed.

We agree that such determination would have been a useful addition to the study, but regretfully it was not undertaken in these modified 3T3-L1 cells at the time of study. However independent bulk RNAseq studies of the founder 3T3-L1 cells from which the stably transduced cells were generated, undertaken as part of an unrelated study, revealed the following relative levels of endogenous expression of PI3K subunit mRNA:

Author response table 1.

We have not determined endogenous protein expression, and so have left the text of the discussion unchanged, simply noting that we have not formally assessed protein expression of p110d/p110b. However these transcriptomic findings suggest that p110d protein is likely either undetectable, or else present at extremely low levels compared to endogenous p110a. p110b also appears to be expressed at a much lower level than p110a. In our studies overexpressing mutant PIK3R1 and assessing insulin action, we believe we are largely or perhaps entirely assaying the effect of the mutants on p110a, in keeping with the fact that genetic and pharmacological studies have firmly established that it is p110a that is responsible for mediating the metabolic actions of insulin in adipose tissue and preadipocytes including 3T3-L1 (e.g. PMID 16647110). Indeed, to quote from this study, in 3T3-L1 “… inhibitors of p110b (TGX-115 and TGX-286) and p110d (IC87114 and PIK-23) had no effect on the insulin-stimulated phosphorylation of any protein in the PI3-K pathway.”

We have added the following sentence to the discussion:

“The current study has limitations. We have studied primary cells from only a single APDS2 patient, and in the 3T3-L1 cell model, we did not determine whether p110d protein could be detected. If not, this could explain the lack of detectable AKT phosphorylation with induction of Pik3r1 DEx11.  Indeed, previous pharmacological studies in 3T3-L1 adipocytes has shown that selective inhibition of p110d or p110b does not alter insulin-induced phosphorylation of any protein studied in the PI3-K pathway, attesting to the dominance of p110a in insulin action in this cell model (Knight et al, 2006).” 

Furthermore, a direct comparison of the truncated APDS2-causal p85alpha variant with SHORT syndrome-causal p85alpha variants in regard to pAKT level, and p85alpha expression level has not been performed.

These investigations would further strengthen the data.

The cell lines conditionally expressing SHORT syndrome variants have been reported already, as cited (PMID: 27766312). Remarkably, the degree of inhibition of insulin-stimulated signalling is actually less pronounced for the SHORT syndrome variants than for the overexpressed APDS2 variant, as seen in the excerpt from the prior paper below. In this prior paper the maximum insulin concentration used, 100nM, was the concentration used in the current study. While overexpression of the APDS2 p85a variant ablated the response to insulin entirely, it is still seen in the prior study, albeit at a clearly reduced level.

Related to Figure 3

The E489K and Y657X p85alpha variants should be also tested in combination with p110delta in the PI3K activity in vitro assay. This would help to further decipher the overall impact, especially of the E489K variant.

We agree that this would make our data more complete, but for logistical reasons (primarily available personnel) we were compelled to constrain the number of p85-p110 combinations we studied. We elected to prioritise the PIK3R1 R649W variant as by far the most common causal SHORT syndrome variant, and as the variant showing the “cleanest” functional perturbation, namely severely impaired or absent ability to dock to phosphotyrosines in cognate proteins.  The paradox that we sought to explain in this paper, namely the phenotypic combination of gain-of-function APDS2 with loss-of-function SHORT syndrome features holds only for APDS2 PIK3R1 variants, and so while it is interesting to document that the canonical SHORT syndrome variant also inhibits PI3Kb and PI3Kd activation in vitro, this was not the main purpose of our study.

Reviewer #1 (Recommendations For The Authors):

Points of clarification and suggestions for improving the manuscript:

(1) Explain whether there are any PIK3R1-independent genetic alterations in the APDS2 and PROS-derived cell lines. For example, are there differences in the karyotype of mutant cell lines compared to wild-type cells?

Karyotypic abnormalities are not an established feature of either PROS or APDS2, and the patients from whom cells were derived were documented to be of normal karyotype. Karyotypic abnormalities acquired during cell culture would not be unprecedented, but confirming normal karyotypes in primary cell lines where there is no specific reason to suppose any alteration exceeds normal expectations for primary cell studies, and so this has not been undertaken.

(2) When introducing the APDS2-associated PIK3R1 mutation (lines 126-128), the authors describe both the exon 11 skipping and in-frame deletions. I recommend rewording this sentence to say exon 11 skipping results in an in-frame deletion of PIK3R1. The current wording makes it seem like APDS2-derived cells contain two genetic perturbations: (1) exon 11 skipping and (2) in-frame deletion. Include a diagram in Figure 1 to help explain the location of the mutations being studied in relationship to the PIK3R1 gene sequence and domains (i.e. nSH2, iSH2, cSH2). The description of the exon 11 skipping and in-frame deletions (lines 126-128) would benefit from having a complementary figure that diagrams the location of these mutations in the PIK3R1 gene.

On review we agree that clarity of description could be enhanced. We have now edited these lines as follows:

“We began by assessing dermal fibroblasts cultured from a previously described woman with APDS2 due to the common causal PIK3R1 mutation. This affects a splice donor site and causes skipping of exon 11, leading to an in-frame deletion of 42 amino acids (434-475 inclusive) in the inter-SH2 domain, which is shared by all PIK3R1 isoforms (Patient A.1 in (Lucas et al., 2014b))(Figure 1 figure supplement 1).”

We have moreover introduced a further figure element including a schematic of all PIK3R1 mutations reported in the current study (new Figure 1 figure supplement 1)

(3) For Figure 2, I recommend including a cartoon that illustrates the experimental design showing the induced expression of PIK3R1 mutants, R649W and Y657X, in the background of the wild-type endogenous gene expression.

Such a figure element has now been generated and included as Figure 2 figure supplement 1, duly called out in the results section where appropriate.

(4) For the data plotted in Figure 1B-1C, please clarify whether the experiments represent a single patient or all 3-4 patients shown in Figure 1A.

Each datapoint shown represents one of the patients in the immunoblots, with all patients included. Each point in turn is the mean from 3 independent experiments. We have added the following to the Figure legend:

“(B)-(E) quantification of immunoblot bands from 3 independent experiments shown for phosphoAKT-S473, phosphoAKT-T308, p110d and p110a respectively. Each point represents data from one of the patient cell lines in the immunoblots. Paired datapoints +/- insulin are shown in (B) and (C), and dotted lines mark means.”

(5) I recommend rewording the following sentence: "Given this evidence that APDS2-associated PIK3R1 delta Exon 11 potently inhibits PI3Kα when overexpressed in 3T3-L1 preadipocytes," to say "... potently inhibits PI3Kα signaling when overexpressed in 3T3-L1 preadipocytes." The data shown in Figures 1 and 2 do not support a direct biochemical inhibition of PI3Kα lipid kinase activity by p85α (delta Exon 11).

This edit has been made.

(6) Provide more discussion concerning the percentage of humans with APDS2 or SHORT syndrome that contain the mutations discussed in this paper. How strong is the genotype-phenotype link for these diseases? Are these diseases inherited or acquired through environmental stresses?

Both APDS2 and SHORT syndrome are very well established, highly penetrant and stereotyped monogenic disease. APDS is defined by the presence of activating PIK3R1 mutations such as the one studied here (by far the commonest causal mutation).  SHORT syndrome clinically has some superficial resemblance to other human genetic syndrome including short stature, but when careful attention is paid to characteristic features it is nearly universally attributable to loss-of-function PIK3R1 mutations with the single exception of one case in which a putatively pathogenic PKCE mutation was described (PMID: 28934384). Although both syndromes are monogenic it is often not accurate to refer to them as inherited, as, particularly in SHORT syndrome, de novo mutations (i.e. not found in either parent) are common. Environmental modifiers of phenotypes have not been described. To the introduction has now been added the comment that both conditions are highly penetrant and monogenic.

(7) The data presented in Figure 5 would benefit from additional discussion and citations that describe the molecular basis of the interaction between PI3K and Irs1/2. What studies have previously established this is a direct protein-protein interactions? Are there PI3K mutants that don't interact with Irs1/2 that can be included as a negative control? Alternatively, the authors can simply reference other papers to support the mechanism of interaction.

There is a voluminous literature dating back to the early 1990s documenting the mode of interaction of PI3K with Irs1/2. Relevant papers have now been cited as requested:

p85-Irs1 binding: PMID 1332046 (White lab, PNAS 1992)

p85-Irs2 binding: PMID 7675087 (White lab, Nature 1995)

“This may be important, as p85a mediates recruitment of PI3K to activated tyrosine kinase receptors and their tyrosine phosphorylated substrates, including the insulin-receptor substrate proteins Irs1 (PMID 1332046) and Irs2 (PMID 7675087).”

Regarding PI3K mutants that don't interact with Irs1/2, the SHORT syndrome mutant R649W which we include in this study is perhaps the best example of this, so it is both disease-causing and functions as such a negative control.

(8) To see the effect of the dominant negative delta Exon 11, the truncated p85α needs to be super stoichiometric to the full-length p85α (Figure 2 - Supplemental Figure 2). This is distinct from the results in Figure 1 showing the ADPS2-derived dermal fibroblast express 5-10x lower levels of p85α delta Exon 11 compared to full-length p85α (Figure 1A), but still strongly inhibits pAKT S473 and T308 (Figure 1B-1C). The manuscript would benefit from more discussion concerning the cell type specific differences in phenotypes. Alternatively, do the APDS2-derived dermal fibroblasts have other genetic perturbations that are not accounted for that potentially modulate cell signaling differently compared to 3T3-L1 preadipocytes?

The reviewer is astute to point out this apparent contrast. First of all, we have no reason to suppose there is any specific, PI3K-modifying genetic perturbation present in the primary dermal fibroblasts studied, although of course the genetic background of these cells is very distinct to that of 3T3-L1 mouse embryo fibroblasts. Related to such background differences, however, substantial variability is usually apparent in insulin-responsiveness even of healthy control dermal fibroblasts. This means that caution should be exercised in extrapolating from studies of the primary cells of a single individual. To illustrate this, we point the reviewer to our 2016 study in which we extensively studied the dermal fibroblasts of a proband with SHORT syndrome due to PIK3R1 Y657X:

From this study we conclude that A. WT controls show quite substantial variation in insulin-stimulated AKT phosphorylation and B. even the SHORT syndrome p85a Y657X variant, expressed at higher levels that WT p85a in dermal fibroblasts, does not produce an obvious decrease in insulin-stimulated AKT phosphorylation, despite extensive evidence from other human cell studies and knock-in mice that it does indeed impaired insulin action in metabolic tissues. For both these reasons we are not convinced that the lower insulin-induced AKT phosphorylation we described in Figure 1 should be overinterpreted until reproduced in other studies with primary cells from further APDS2 patients. This is why we did not comment more extensively on this. We now add the following qualifier in results:

“Despite this, no increase in basal or insulin-stimulated AKT phosphorylation was seen in APDS2 cells compared to cells from wild-type volunteers or from people with PROS and activating PIK3CA mutations H1047L or H1047R (Fig 1A-C, Fig 1 figure supplement 3A,B). Although insulin-induced AKT phosphorylation was lower in fibroblasts from the one APDS2 patient studied compared to controls, we have previously reported extensive variability in insulin-responsiveness of primary dermal fibroblasts from WT controls. Moreover even primary cells from a patient expressing high levels of the SHORT syndrome-associated p85a Y657X did not show attenuated insulin action, so we do not believe the reduced insulin action in APDS2 cells in the current study should be overinterpreted until reproduced in further APDS2 cells.”

Nevertheless we remind the reviewer that the main purpose of our primary cell experiment was to determine if there were any INCREASE in basal PI3K activity, or any difference in p110a or p110d protein levels, and we regard our findings in these regards to be clear.

The manuscript would benefit from additional explanation concerning why the E489K, R649W, and Y657X are equivalent substitutes for the characterization of p110α/p85α delta Exon 11). Perhaps a more explicit description of these mutations in relationship to the location of p85α delta Exon 11) mutation would help. I recommend including a diagram in Figure 3 showing the position of the delta Exon 11, E489K, R649W, and Y657X mutations in the PIK3R1 coding sequence. B. Also, please clarify whether all three holoenzyme complexes were biochemically unstable (i.e. p110α/p85α, p110β/p85α, p110δ/p85α) when p85α delta Exon 11) was expressed in insect cells.

A. Whether or not E489K, R649W and Y657X are “equivalent” to the APDS2 mutant is not really a meaningful issue here. These mutants are being studied because they cause SHORT syndrome without immunodeficiency, while the APDS2 mutant causes APDS2 often with features of SHORT syndrome. That is, it is naturally occurring mutations and the associated genotype-phenotype correlation that we seek to understand. Of the 3 SHORT syndrome causal mutations chosen, R649W is by far the commonest, effectively preventing phosphotyrosine binding, Y657X has the interesting attribute that it can be discriminated from full length p85 on immunoblots due to its truncation, and is moreover a variant that we have studied in cells and mice before, while the rarer E489K is an interesting SHORT syndrome variant as it is positioned more proximally in the p85a protein than most SHORT syndrome causal variants. All variants studied are now illustrated in the new Figure 1 figure supplement 1. B. Regarding stability of PI3K heterodimers containing the APDS2 p85a variant, we tried extensively to purify p110a and p110d complexes without success despite several approaches to optimise production. We did not try to synthesise the p110b-containing complex.

(10) I recommend presenting the results in Figure 4 before Figure 3 because it provides a good rationale for why it's difficult to purify the p110α/p85α delta Exon 11) holoenzyme from insect cells.

This would be true of p110d were studied in Figure 4 but it is not. Figure 4 looks instead at effects on p110a of heterologous overexpression of mutant p85, is a natural lead in to the ensuing figures 5 and 6, and we do not agree it would add value or enhance flow to swap Figures 3 and 4.

(11) The authors show that overexpression of the p85α delta Exon 11) did not result in p110α/p85α delta Exon 11) complex formation based on co-immunoprecipitation. Do the authors get the same result when they co-immunoprecipitation p110α/p85α and p110δ/p85α in the APDS2-derived dermal fibroblasts used in Figure 1A?

This is an interesting question but not an experiment we have done. It is not unfeasible, but generating enough cells to undertake IP experiments of this nature in dermal fibroblasts is a significant undertaking, and with finite resources available and only one primary cell line to study we elected not to pursue this.

Details in Methods section:

(1) Include catalog numbers and vendors for reagents (e.g. lipids, PhosSTOP, G-Dynabeads, etc.). There is not enough information provided to reproduce this work.

We have now added all vendors and catalogue numbers where relevant.

(2) Concerning the stated lipid composition (5/10/15/45/20/5 %) in the liposome preparation protocol. Please clarify whether these numbers represent molar percentages or mg/mL percentages.

We have now added that this is expressed as “(wt/vol)”

(3) What is the amino acid sequence of the PDGFR (pY2) peptide used for the PI3K activity assay?

This assay has been published and references with detailed methods are cited. For clarity, however we now say:

“PI(3,4,5)P3 production was measured by modified PI3-Kinase activity fluorescence polarisation assay (Echelon Biosciences, Salt Lake City, UT, USA). 10μL reactions in 384-well black microtitre plates used 1mM liposomes containing 50μM PI(4,5)P2, optimised concentrations of purified PI3K proteins, 100μM ATP, 2mM MgCl2, with or without 1μM tyrosine bisphosphorylated 33-mer peptide derived from mouse PDGFRβ residues 735-767, including phosphotyrosine at positions 740 and 751 (“pY2”; 735-ESDGGYMDMSKDESIDYVPMLDMKGDIKYADIE-767;  Cambridge peptides).”

(4) Include a Supplemental file containing a comprehensive description of the plasmids and coding sequencing used in this study.

Such a supplemental file has been created and is included as Table 2

Minor points of clarification, citations, and typos:

(1) Clarify why Activated PI3K Delta Syndrome 1 (APDS1) is thus named APDS2. See lines 71-72 of the introduction. Also see line 89: "...is common in APDS2, but not in APDS1." Briefly describe the difference between APDS1 and APDS2?

This is described in the introduction, but we apologise if our wording was not sufficiently clear. We have tried now to remove any ambiguity:

“Some PIK3R1 mutations reduce basal inhibition of catalytic subunits, usually due to disruption of the inhibitory inter-SH2 domain, and are found in cancers (Philp et al, 2001) and vascular malformations with overgrowth(Cottrell et al, 2021). In both diseases, hyperactivated PI3Ka, composed of heterodimers of PIK3R1 products and p110a, drives pathological growth. Distinct inter-SH2 domain PIK3R1 mutations, mostly causing skipping of exon 11 and deletion of residues 434-475, hyperactivate PI3Kd in immune cells, causing highly penetrant monogenic immunodeficiency (Deau et al, 2014; Lucas et al, 2014b). This phenocopies the immunodeficiency caused by genetic activation of p110d itself, which is named Activated PI3K Delta Syndrome 1 (APDS1) (Angulo et al, 2013; Lucas et al, 2014a). The PIK3R1-related syndrome, discovered shortly afterwards, is thus named APDS2.”

(2) Figure legend 1. Clarify reference to "Figure EV2".

(3) Figure legend 2. Clarify reference to "Figure EV3".

(4) Figure legend 3. Clarify reference to "Figure EV5".

Thank you for pointing out this oversight, arising from failure to update nomenclature fully between versions. “EV” figures actually are the figure supplements in the submission. All labels have now been updated.

(5) For Figure 1 - supplemental figure 1C, indicate experimental conditions on the blot (e.g. -/+ insulin).

This is now added

(6) Figure 4B, y-axis. Clarify how data was quantified. Perhaps reword "(% WT without DOX)" for clarity.

We have left the Y axis label as it is, but have added the following to the figure legend:

“(B) Quantification of immunoblot bands from immunoprecipitates from 3 independent experiments, expressed as a percentage relative to the intensity of the band in WT cells without doxycycline exposure.”

(7) In the results section (lines 117-124), please explicitly state whether the described mutations are homo- or heterozygous.

All mutations are heterozygous, as now explicitly stated

(8) I recommend spelling out the SHORT and APDS2 acronyms in the abstract to make this study more accessible.

We respectfully disagree that such spelling out in the abstract would improve accessibility. Both acronyms are clunky and wordy and are more likely to obscure meaning by squeezing out other words in the abstract. APDS is already spelled out in the introduction, and we now add the following for SHORT syndrome:

“More surprisingly, phenotypic overlap is reported between APDS2 and SHORT syndrome. SHORT syndrome, named for the characteristic developmental features (Short stature, Hyperextensibility, Hernia, Ocular depression, Rieger anomaly, and Teething delay) is caused by loss of PI3Ka function due to disruption of the phosphotyrosine-binding C-terminal SH2 domain (Chudasama et al, 2013; Dyment et al, 2013; Thauvin-Robinet et al, 2013).”

(9) I recommend explaining in more detail or rewording the following jargon/terms to make the writing more accessible to a broad audience: "reduced linear growth" (line 83) and "larger series" (line 86). I assume "reduced linear growth" is height.

Edited as follows:

“It  features short stature, insulin resistance, and dysmorphic features (Avila et al, 2016). In recent years, both individual case reports (Bravo Garcia-Morato et al, 2017; Petrovski et al, 2016; Ramirez et al, 2020; Szczawinska-Poplonyk et al, 2022) and larger case series (Elkaim et al, 2016; Jamee et al, 2020; Maccari et al, 2023; Nguyen et al, 2023; Olbrich et al, 2016; Petrovski et al., 2016) have established that many people with APDS2 have overt features of SHORT syndrome, while, more generally, linear growth impairment is common in APDS2, but not in APDS1.”

(10) For clarity, reword lines 214-215 to read, "No increase in p110α levels was seen on conditional overexpression of wild-type or R649W p85α."

Change made, thank you

(11) Figure 6A - Western blot label says, "657X" instead of "Y657X."

Now corrected

(12) Lines 214-215: For clarity, reword the sentence to say, "No increase in p110α was seen on conditional overexpression...".

REPEAT OF POINT 10 ABOVE

(13) Clarify what interactions are being competed for in the following statement: "... delta Ex11 may exert its inhibitory action by competing with PI3K holoenzyme" (lines 237-238). Are you referring to the interaction between p110α and p85α or the interaction between p110α/p85α and another protein?

We have endeavoured to clarify by editing as follows:

“As APDS2 p85a DEx11 does not appear to displace wild-type p85a from p110a despite strong overexpression, it is likely that there are high levels of truncated p85a unbound to p110a in the cell. This may be important, as p85a mediates recruitment of PI3K to activated tyrosine kinase receptors and their tyrosine phosphorylated substrates, including the insulin-receptor substrate proteins Irs1 and Irs2. Excess free regulatory subunits compete with heterodimeric PI3K holoenzyme for binding to these phosphotyrosines (Ueki et al., 2002), raising the possibility that excess free, truncated APDS2 p85a DEx11 may exert its inhibitory action similarly by outcompeting PI3K holoenzyme for phosphotyrosine binding.”

(14) Provide more information about the following statement and how it relates to the mutations in this study: "Homozygous truncating PIK3R1 mutations abolishing p85α expression while preserving p55α and p50α produce agammaglobulinaemia" (lines 271-272). The manuscript would benefit from a more explicit description of the nature of these mutations.

This wording seems to us to be explicit, however we agree that a schematic of PIK3R1 genotype-phenotype correlation, as requested elsewhere, would help readers. Such a schematic is now included as Figure 1 figure supplement 1.

(15) Typo on line 299: "unclike".

Corrected.

(16) The data presented in this study support a model in which p85α (DExon 11) expression functions as a dominant negative. Please clarify why in the discussion section you explain that p85α (DExon 11) activates PI3K. For example, "...skipping of exon 11, were shown in 2014 to activate PI3K..." (lines 290-291), "...activate PI3Kδ on one hand..." (line 309); "...APDS2 mutations in PIK3R1 has mixed consequences, producing greater hyperactivation of p110δ than p110α" (lines 354-355).

We do not entirely understand the reviewer’s question and thus request here. p85α (DExon 11) activates PI3Kd in immune cells and in vitro, and this is accepted, based on numerous reports, to be the mechanism underlying immunodeficiency. We do not challenge this, and cite evidence for any such claims in our report. The dominant negative activity we describe here towards PI3Ka activation is based not on inhibition of mutant-containing heterodimer, but rather on destabilisation of and/or competition with heterodimeric WT holoenzyme. This is the basis of the model we present; that is, a finely balanced competition between enzymic activation and mutant holoenzyme destabilisation and competition of mutant free p85a with WT holoenzyme, whose net effect likely differs among cells and tissues, most likely based on the repertoire and proportions of PI3K subunit expression. If the reviewer has specific suggestions for us that will make this point clearer still we should be happy to consider them.

(17) Provide references for the statements in lines 349-353 of the discussion.

This brief closing paragraph is a succinct recap and summary of the key points made throughout the manuscript and thoroughly referenced therein. We prefer to keep this section clean to maximise clarity, but are happy to copy references from the various other places in the manuscript to back up these assertions if this is preferred by the editorial team. Current text:

“In summary, it is already established that: A. genetic activation of PIK3CD causes immunodeficiency without disordered growth, while B. inhibition of PIK3R1 recruitment to RTKs and their substrates impairs growth and insulin action, without immunodeficiency, despite all catalytic subunits being affected and C. loss of p85 alone causes immunodeficiency.”

Reviewer #2 (Recommendations For The Authors):

In the abstract line 42 I would rather talk from SHORT syndrome like features.

Some patients do indeed meet the criteria for SHORT syndrome, but there is a spectrum. We have thus added this qualification and removed “short stature” to maintain the word count, as this is itself a SHORT syndrome-like feature.

Line 74 It would be helpful for the reader to give the amino-acid exchange and affected position of this single case.

We agree. Now added.

Furthermore, an illustration indicating the location of the different PIK3R1 variants on the p85 alpha level would be helpful for the reader.

As noted above such a figure element is now included as Figure 1 figure supplement 1 and duly called out in the text

The sentence in lines 298-300 makes no sense to me. Do you mean, unlike APDS1 murine models?

We agree, on review, that this paragraph is convoluted and makes a simple observation complex. We have rewritten now in what we hope is a more accessible style:

“Thus, study of distinct PIK3R1-related syndromes shows that established loss-of-function PIK3R1 mutations produce phenotypes attributable selectively to impaired PI3Ka hypofunction, while activating mutations produce phenotypes attributable to selectively increased PI3Kd signalling. Indeed, not only do such activating mutations not produce phenotypes attributable to PI3Ka activation, but they surprisingly have features characteristic of impaired PI3Ka function.”

Line 321 I propose including the notion of different cells: “The balance between expression and signalling in different cells may be a fine one ...”

This change has been made

Line 352 C. loss replace with complete loss.

“C.” actually denotes the last in a list after “A.” and “B.”. We have now used bold to emphasise this, but we imagine house style may dictate how we approach this.

Author response:

The following is the authors’ response to the original reviews.

Public Reviews:

Reviewer #1 (Public Review):

The study describes a new computational method for unsupervised (i.e., non-artificial intelligence) segmentation of objects in grayscale images that contain substantial noise, to differentiate object, no object, and noise. Such a problem is essential in biology because they are commonly confronted in the analysis of microscope images of biological samples and recently have been resolved by artificial intelligence, especially by deep neural networks. However, training artificial intelligence for specific sample images is a difficult task and not every biological laboratory can handle it. Therefore, the proposed method is particularly appealing to laboratories with little computational background. The method was shown to achieve better performance than a threshold-based method for artificial and natural test images. To demonstrate the usability, the authors applied the method to high-power confocal images of the thalamus for the identification and quantification of immunostained potassium ion channel clusters formed in the proximity of large axons in the thalamic neuropil and verified the results in comparison to electron micrographs.

Strengths:

The authors claim that the proposed method has higher pixel-wise accuracy than the threshold-based method when applied to gray-scale images with substantial noises.

Since the method does not use artificial intelligence, training and testing are not necessary, which would be appealing to biologists who are not familiar with machine learning technology.

The method does not require extensive tuning of adjustable parameters (trying different values of "Moran's order") given that the size of the object in question can be estimated in advance.

We appreciate the positive assessment of our approach.

Weaknesses:

It is understood that the strength of the method is that it does not depend on artificial intelligence and therefore the authors wanted to compare the performance with another non-AI method (i.e. the threshold-based method; TBM). However, the TBM used in this work seems too naive to be fairly compared to the expensive computation of "Moran's I" used for the proposed method. To provide convincing evidence that the proposed method advances object segmentation technology and can be used practically in various fields, it should be compared to other advanced methods, including AI-based ones, as well.

Protein localization studies revealed that protein distributions are frequently inhomogeneous in a cell. This is very common in neurons which are highly polarized cell types with distinct axo-somato-dendritic functions. Moreover, due to the nature of the cell-to-cell interactions among neurons (e.g. electrical and chemical synapses) the cell membrane includes highly variable microdomains with unique protein assemblies (i.e. clusters). Protein clusters are defined as membrane segments with higher protein densities compared to neighboring membrane regions. However, protein density can continuously change between “clusters” and “non-clusters”. As a consequence, differentiating proteins involved vs not involved in clusters is a challenging task.  Indeed, our analysis showed that the boundaries of protein clusters varied remarkably when 23 human experts delineated them.

Despite the fact the protein clusters can only be vaguely defined numerous studies have demonstrated the functional relevance of inhomogeneous protein distribution. Thus, there is a high relevance and need for an observer independent, “operative” segmentation method that can be accomplished and compared among different conditions and specimens. The strength of the Moran’s I analysis we propose here, as pointed out by our reviewers and editors, is that it can extract the relevant signals from an image generated in different, often noisy condition using a simple algorithm that allows quantitative characterization and identification of changes in many biological and non-biological samples.

In AI based analysis the ground truth is known by an observer and using a large training set AI learns to extract the relevant information for image segmentation. As outlined above the “ground truth”, however, cannot be unequivocally defined for protein clusters. There is no doubt, that with sufficient resource investment there would be an AI based analysis of the same problem. In our view, however, in an average laboratory setting generating a training set using hundreds of images examined by many experts may not be plausible. Moreover, generalization of one training set to another set of cluster, resistance to noise or different levels of background could also not be guaranteed.

This method was claimed to be better than the TBM when the noise level was high. Related to the above, TBMs can be used in association with various denoising methods as a preprocess. It is questionable whether the claim is still valid when compared to the methods with adequate complexity used together with denoising. Consider for example, Weigert et al. (2018) https://doi.org/10.1038/s41592-018-0216-7; or Lehtinen et al (2018) https://doi.org/10.48550/arXiv.1803.04189.

In Weigert et al. AI was trained with high-quality images of the same object obtained with extreme photon exposure in confocal microscope. As delineated above without training AI systems cannot be used for such purposes. The Lehtinen paper is unfortunately no longer available at this doi.

We must emphasize that in our work we did not intend to compare the image segmentation method based on local Moran’s I with all other available segmentation techniques. Rather we wanted to demonstrate a straightforward method of grouping pixels with similar intensities and in spatial proximity which does not require a priori knowledge of the objects. We used TBM to benchmark the method. We agree that with more advanced TBM methods the difference between Moran’s and TBM might have been smaller. The critical component here is, however, that even with most advanced TBM an artificial threshold is needed to be defined. The optimal threshold may change from sample to sample depending on the experimental conditions which makes quantification questionable. Moran’s method overcomes this problem and allows more objective segmentation of images even if the exact conditions (background labeling, noise, intensity etc) are not identical among the samples.

The computational complexity of the method, determined by the convolution matrix size (Moran's order), linearly increases as the object size increases (Fig. S2b). Given that the convolution must be run separately for each pixel, the computation seems quite demanding for scale-up, e.g. when the method is applied for 3D image volumes. It will be helpful if the requirement for computer resources and time is provided.

Here we provide the required data concerning the hardware and the computational time:

Hardware used for performing the analysis:

Intel(R) Xeon(R) Silver 4112 CPU @ 2.60GHz, 2594 Mhz, 4 kernel CPU, 64GB RAM, NVIDIA GeForce GTX 1080 graphic card.

MATLAB R2021b software was used for implementation.

Author response table 1.

Computation times:

Reviewer #2 (Public Review):

Summary:

The manuscript by David et al. describes a novel image segmentation method, implementing Local Moran's method, which determines whether the value of a datapoint or a pixel is randomly distributed among all values, in differentiating pixel clusters from the background noise. The study includes several proof-of-concept analyses to validate the power of the new approach, revealing that implementation of Local Moran's method in image segmentation is superior to threshold-based segmentation methods commonly used in analyzing confocal images in neuroanatomical studies.

Strengths:

Several proof-of-concept experiments are performed to confirm the sensitivity and validity of the proposed method. Using composed images with varying levels of background noise and analyzing them in parallel with the Local Moran's or a Threshold-Based Method (TBM), the study is able to compare these approaches directly and reveal their relative power in isolating clustered pixels.     

Similarly, dual immuno-electron microscopy was used to test the biological relevance of a colocalization that was revealed by Local Moran's segmentation approach on dual-fluorescent labeled tissue using immuno-markers of the axon terminal and a membrane-protein (Figure 5). The EM revealed that the two markers were present in terminals and their post-synaptic partners, respectively. This is a strong approach to verify the validity of the new approach for determining object-based colocalization in fluorescent microscopy. 

The methods section is clear in explaining the rationale and the steps of the new method (however, see the weaknesses section). Figures are appropriate and effective in illustrating the methods and the results of the study. The writing is clear; the references are appropriate and useful.

We are grateful for the constructive assessment of our results.

Weaknesses:

While the steps of the mathematical calculations to implement Local Moran's principles for analyzing high-resolution images are clearly written, the manuscript currently does not provide a computation tool that could facilitate easy implementation of the method by other researchers. Without a user-friendly tool, such as an ImageJ plugin or a code, the use of the method developed by David et al by other investigators may remain limited.

The code for the analysis is now available online as a user-friendly MATLAB script at: https://github.com/dcsabaCD225/Moran_Matlab/blob/main/moran_local.m

Recommendations for the authors:

Summary of reviews:

Both reviewers acknowledge the potential significance and practicality of the newly proposed image segmentation method. This method uses Local Moran's principles, offering an advantage over traditional intensity thresholding approaches by providing more sensitivity, particularly in reducing background noise and preserving biologically relevant pixels.

Strengths Highlighted:

• The proposed method can provide more accurate results, especially for grayscale images with significant noise.

• The method is not dependent on artificial intelligence, making it appealing for researchers with minimal computational background.    

• The approach can operate without the need for extensive tuning, given that the size of the object is known.

• Several proof-of-concept experiments were carried out, revealing the effectiveness of the method in comparison with the threshold-based segmentation methods.

• The manuscript is clear in terms of methodology, and the results are supported by effective illustrations and references.

Weaknesses Noted:

• The study lacked a comparative analysis with advanced segmentation methods, especially those that employ artificial intelligence.

See our response above to the same question of Reviewer 1.

• There are concerns about computational complexity, especially when dealing with larger data sets or 3D image volumes.

See our response about the calculations of computation times above to the similar question of Reviewer 1.

• Both reviewers noted the absence of a data/code availability statement in the manuscript, which might restrict the method's adoption by other researchers.

The code availability is provided now.

• Reviewer 2 suggested that some results, particularly related to Kv4.2 in the thalamus, might be better presented in a separate study due to their significance.

We thank our reviewers for this suggestion. We carefully evaluated the pros and cons of publishing the Kv4.2 data separately. We finally decided to keep the segmentation and experimental data together due to the following reason. We believe that the ultrastructural localization provides strong experimental proof for the relevance of our novel segmentation method. In order to make the potassium channel data more visible we added a subsentence to the title. In this manner we think scientist interested in the imaging method as well as the neurobiology will be both find and cite the paper. The novel title reads now:

“An image segmentation method based on the spatial correlation coefficient of Local Moran’s I - identification of A-type potassium channel clusters in the thalamus.”

Reviewer Recommendations:

(1) Provide details about the data and program code availability.

See our response above

(2) Offer practical recommendations and provide clarity on software packages and coding for the proposed method to enhance its adoption.

Done.

(3) Consider presenting the findings about Kv4.2 in the thalamus separately as they hold significant importance on their own.

See our response above

Given the reviews, the proposed image segmentation method presents a promising advancement in the domain of image analysis. The technique offers tangible benefits, especially for researchers dealing with biological microscopy data. However, for this method to see a broader application, it's imperative to provide clearer practical guidance and make data or code easily accessible. Additionally, while the findings regarding Kv4.2 in the thalamus are intriguing, they might achieve more impact if detailed in a dedicated paper.

Reviewer #1 (Recommendations For The Authors):

The availability of data or program code was not stated in the manuscript.

Reviewer #2 (Recommendations For The Authors):

(1) While the principles of the method are explained clearly in a step-by-step fashion in the Methods section, the practical aspects of running sequential computations over a large matrix of pixel values are not well described. It would be very useful if the authors could provide recommendations on how to set the data structure and clarify which software and programming package for Local Moran's analysis they used. In addition, providing the code for the sequential implementation described in the Methods section would facilitate the adoption of the method by other researchers, and thus, the impact of the study. Currently, there is no data or code availability statement included in the manuscript.

See our response above.

(2) Figure 4 illustrates an experiment in which transmission electron microscopy and freeze-fracture replica labeling approaches were used to demonstrate that a potassium channel marker, Kv4.2 was selective to synapses forming on larger caliber dendrites in the thalamus. As impressive as the EM approaches utilized in this figure are, the results of this experiment have a somewhat tangential bearing on the segmentation method that is the focus of this study. In fact, the experiments illustrated in Figure 5, dual immuno-EM, are more than sufficient to confirm what the dual-confocal imaging coupled with Local Moran's segmentation analysis reveals. Furthermore, the author's findings about the localization and selectivity of Kv4.2 in the thalamus are too important and exciting to bury in a paper focusing on the methodology. Those results may have a wider impact if they are presented and discussed in a separate experimental paper.

See our response above

Author Response

The following is the authors’ response to the original reviews.

Reviewer #1

The study provides a complete comparative interactome analysis of α-arrestin in both humans and drosophila. The authors have presented interactomes of six humans and twelve Drosophila α-arrestins using affinity purification/mass spectrometry (AP/MS). The constructed interactomes helped to find α-arrestins binding partners through common protein motifs. The authors have used bioinformatic tools and experimental data in human cells to identify the roles of TXNIP and ARRDC5: TXNIP-HADC2 interaction and ARRDC5-V-type ATPase interaction. The study reveals the PPI network for α-arrestins and examines the functions of α-arrestins in both humans and Drosophila.

Comments

I will like to congratulate the authors and the corresponding authors of this manuscript for bringing together such an elaborate study on α-arrestin and conducting a comparative study in drosophila and humans.

Introduction:

The introduction provides a rationale behind why the comparison between humans and Drosophila is carried out.

• Even though this is a research manuscript, including existing literature on similar comparison of α-arrestin from other articles will invite a wide readership.

Results:

The results cover all the necessary points concluded from the experiments and computational analysis.

1) The authors could point out the similarity of the α-arrestin in both humans and Drosophila. While comparing α-arrestin in both humans and Drosophila If percentage homology between α-arrestin of both Drosophila and humans needs to be calculated.

Thank you for your insightful feedback. As suggested by reviewer, we determined percentage homology of α-arrestin protein sequences from human and Drosophila using Clustal Omega. This homology is now illustrated as a heatmap in revised Figure S5. Please note that only the values with percentage homology of 40% or higher are selectively labeled.

• Citing the direct connecting genes from the network in the text will invite citations and a wider readership.

Figures:

The images are elaborate and well-made.

2) The authors could use a direct connected gene-gene network that pointing interactions. This can be used by other readers working on the same topic and ensure reproducibility and citations.

We appreciate your valuable comment. Based on the reviewer’s suggestion, we have developed a new website in which one can navigate the gene-gene networks of α-arrestins. These direct connected gene-gene networks are housed in the network data exchange (NDEx) project. Additionally, we have included gene ontology and protein class details for α-arrestins’ interactors in these set of networks, offering a more comprehensive view of α-arrestins’ interactomes.

On page 24 lines 15-18, we have revised the manuscript to introduce the newly developed website, as follows.

“Lastly, to assist the research community, we have made comprehensive α-arrestin interactome maps on our website (big.hanyang.ac.kr/alphaArrestin_PPIN). Researchers can search and download their interactomes of interest as well as access information on potential cellular functions and protein class associated with these interactomes.”  

3-1) The co-expression interactions represented as figures should reveal interaction among the α-arrestin and other genes. Which are the sub-network genes does the α- arrestin interact to/ with from the sub-network? The arrows are only pointing at the sub-networks. The figures do not reveal their interaction. Kindly reveal the interaction in the figure with the proper nodes in the figure.

3-2) Figure 2: the network attached in both human and drosophila is well represented. The green lines from α-arrestin indicate the strength of the interaction. Several smaller expression networks are seen. But "α-arrestin" in both organisms seems highly disconnected from all the genes. Connected genes have edges, not arrows. If α-arrestin can be shown connected to these gene-gene networks will help in identifying which genes connect with which gene through α-arrestin. This can be used by other readers working on the same topic and ensure reproducibility and citations.

Thank you for your valuable comment. In response to the reviewer’s recommendation, we’ve added supplementary figure, Figure S4, which illustrates direct interaction between α-arrestin and protein components of clustered complexes (or sub-networks) in addition to the associations shown between α-arrestins and the clustered complexes in Figure 2. We believe that this newly incorporated information regarding direct protein interactions will invite citations and wider readership as the reviewer pointed out.

On page 12 line 27 to page 13 line 5, we have revised the manuscript to cite the direction interactions between ARRDC3 and proteins involved in ubiquitination-dependent proteolysis, as follows.

“While the association of ARRDC3 with these ubiquitination-dependent proteolysis complexes is statistically insignificant, ARRDC3 does interact with individual components of these complexes such as NEDD4, NEDD4L, WWP1, and ITCH (Figure S4A). This suggest their functional relevance in this context, as previously reported in both literatures and databases (Nabhan et al., 2010; Shea et al., 2012; Szklarczyk et al., 2015; Warde-Farley et al., 2010) (Puca & Brou, 2014; Xiao et al., 2018).”

Direct interaction between α-arrestins and protein components of clustered complexes are illustrated in the newly added figure, Figure S4.

4-1) Figure 4. The Protein blot image was blurred. Kindly provide a higher-resolution image.

4-2) Figure 5. B. - The authors can provide images with higher resolution blot images. The bands were not visible.

We appreciate for valuable comment. Unfortunately, the protein blot image was scanned from the original film and the images we provided in the figure represent the highest resolution that we have obtained to date. Raw, uncropped images are shown in Author response image 1 and 2.

Author response image 1.

Raw image of Figure 4B

Author response image 2.

Raw image of Figure 5B

5) Figure: 5. A. - I see non-specific amplifications in the gel images. Are these blotting images? or the gel images that were changed to "Grayscale"? Non-specific amplification may imply that the experiment was not repeated and standardized. Was it gel images or blot images?

We appreciate your insightful comment. The images in Figure 5A represent western blot bands from co-immunoprecipitation assay for analysis of the interaction between TXNIP and HDAC2 proteins. Since immunoblotting using immunoprecipitates can usually detect some non-specific bands from heavy (~ 50 kDa) and light (~25 kDa) chains of the target antibody or from multiple co-immunoprecipitated proteins, we assume that the vague non-specific bands in Figure 5A might be a heavy chain of TXNIP or HDAC2 antibody or an unclear non-specific band. Because target bands showed strong intensity and very clear pattern compared to the non-specific bands in the co-immunoprecipitation assay, we believe that this data is sufficient to support the interaction of TXNIP with HDAC2. Finally, In the revised Figure 5A, we’ve modified the labeling for different experimental conditions, namely siCon and siTXNIP treatments, and added expected size of proteins (kDa), as shown below.

6) Figure 5. A. RT-PCR analysis: What was your expected size of the amplifications? the ladder indicated is in KDa. Is that right?

We appreciate your insightful questions. As mentioned above, Figure 5A shows the blotting images of co-immunoprecipitation analysis, and the ladder indicates the molecular weight (kDa) of protein markers. For clearer interpretation, the expected size of target proteins has been added in Figure 5A in the revised manuscript.

7) How were the band intensities determined?

Thank you for your question. For quantification of immunoblot results, the densities of target protein bands were analyzed with Image J, as we described in the Materials and Methods.

Discussion:

The authors have utilized and discussed the conclusion they draw from their study. But could highlight more on ARRDCs and why it was selected out of the other arrestins. The authors have provided future work directions associated with their work.

8) Why were only ARRDCs presented amongst all the arrestin in the main part of the manuscript?

We’re grateful for your valuable feedback. The reason we focused on α-arrestins was that α-arrestins have been discovered relatively recently, especially when compared to more established visual/ β-arrestin proteins in the same arrestin family but the biological functions of many α-arrestins remain largely unexplored, with notable exceptions in the budding yeast model and a few α-arrestins in mammals and invertebrate species. Most importantly, comparative study highlighting the shared or unique features of α-arrestins is yet to be undertaken. To gain a more comprehensive understanding of these unexplored α-arrestins across multiple species, we’ve centered our research on the ARRDCs within the arrestin protein family.

On page 21 lines 8-17, we’ve edited the manuscript to emphasize the importance of a comparative study on α-arrestins, as detailed below.

“According to a phylogenetic analysis of arrestin family proteins, α-arrestins were shown to be ubiquitously conserved from yeast to human (Alvarez, 2008). However, compared to the more established visual/ β-arrestin proteins, α-arrestins have been discovered more recently and much of their molecular mechanisms and functions remain mostly unexplored except for budding yeast model (Zbieralski & Wawrzycka, 2022). Based on the high-confidence interactomes of α-arrestins from human and Drosophila, we identified conserved and specific functions of these α-arrestins. Furthermore, we uncovered molecular functions of newly discovered function of human specific α-arrestins, TXNIP and ARRDC5. We anticipate that the discovery made here will enhance current understanding of α-arrestins.”

9) The discussion could be elaborated more by utilizing the data.

We appreciate your insightful feedback. Based on the reviewer’s suggestion, we’ve enhanced the discussion in the manuscript to provide a clearer interpretation of our results. First, we’ve added description of conserved protein complexes significantly associated with α-arrestins, stated on page 22 lines 5-12 and lines 23-26.

Page 22 lines 5-12: “The integrative map of protein complexes also highlighted both conserved and unique relationships between α-arrestins and diverse functional protein complexes. For instance, protein complexes involved in ubiquitination-dependent proteolysis, proteasome, RNA splicing, and intracellular transport (motor proteins) were prevalently linked with α-arrestins in both human and Drosophila. To more precisely identify conserved PPIs associated with α-arrestins, we undertook ortholog predictions within the α-arrestins’ interactomes. This revealed 58 orthologous interaction groups that were observed to be conserved between human and Drosophila (Figure 3).”

Page 22 lines 23-26: “Additionally, interaction between α-arrestins and entities like motor proteins, small GTPase, ATP binding proteins, and endosomal trafficking components were identified to be conserved. Further validation of these interactions could unveil molecular mechanisms consistently associated with these cellular functions.”

Secondly, we’ve added description of role of ARRDC5 in osteoclast maturation, as stated on page 23 lines 22-24.

“Conversely, depletion of ARRDC5 reduces osteoclast maturation, underscoring the pivotal role of ARRDC5 in osteoclast development and function (Figure S9A and B).”

Lastly, we examined the association between α-arrestins’ interactomes and human diseases, incorporating our findings into the discussion. The newly introduced figure based on the result is Figure S10.

On page 24 lines 10-14, we’ve added discussion on Figure S10 as follows.

“We further explored association between α-arrestins’ interactomes and disease pathways (Figure S10). Notably, the interactomes of α-arrestins in human showed clear links to specific diseases. For instance, ARRDC5 is closely associated with disease resulting from viral infection and cardiovascular conditions. ARRDC2, ARRDC4, and TXNIP share common association with certain neurodegenerative diseases, while ARRDC1 is implicated in cancer.”

Supplementary figures:

The authors have a rigorous amount of work added together for the success of this manuscript.

10) The reference section needs editing before publication. Maybe the arrangement was disturbed during compiling.

Thank you for your valuable comment. Based on the reviewer’s suggestion, we have rearranged the reference section to enhance its clarity. Below are excerpts from the update reference section in the manuscript.

“Adenuga, D., & Rahman, I. (2010). Protein kinase CK2-mediated phosphorylation of HDAC2 regulates co-repressor formation, deacetylase activity and acetylation of HDAC2 by cigarette smoke and aldehydes. Arch Biochem Biophys, 498(1), 62-73. doi:10.1016/j.abb.2010.04.002

Adenuga, D., Yao, H., March, T. H., Seagrave, J., & Rahman, I. (2009). Histone Deacetylase 2 Is Phosphorylated, Ubiquitinated, and Degraded by Cigarette Smoke. American Journal of Respiratory Cell and Molecular Biology, 40(4), 464-473. doi:10.1165/rcmb.2008-0255OC

Akalin, A., Franke, V., Vlahovicek, K., Mason, C. E., & Schubeler, D. (2015). Genomation: a toolkit to summarize, annotate and visualize genomic intervals. Bioinformatics, 31(7), 1127-1129. doi:10.1093/bioinformatics/btu775

Alvarez, C. E. (2008). On the origins of arrestin and rhodopsin. BMC Evol Biol, 8, 222. doi:10.1186/1471-2148-8-222”

11) many important references were missing.

We appreciate and agree with the reviewer’s comment. In response to the reviewer’s recommendation, we’ve thoroughly reviewed the manuscript and below are sections of the manuscript where around 20 new references have been added.

On page 8 lines 12-14:

“Utilizing the known affinities between short linear motifs in α-arrestins and protein domains in interactomes(El-Gebali et al., 2019; UniProt Consortium, 2018) “

On page 8 lines 19-22:

“One of the most well-known short-linear motifs in α-arrestin is PPxY, which is reported to bind with high affinity to the WW domain found in various proteins, including ubiquitin ligases (Ingham, Gish, & Pawson, 2004; Macias et al., 1996; Sudol, Chen, Bougeret, Einbond, & Bork, 1995)”

On page 9 lines 3-6:

“Next, we conducted enrichment analyses of Pfam proteins domains (El-Gebali et al., 2019; Huang da, Sherman, & Lempicki, 2009b) among interactome of each α-arrestin to investigate known and novel protein domains commonly or specifically associated (Figure S3A; Table S5).”

On page 9 lines 7-10:

“HECT and C2 domains are well known to be embedded in the E3 ubiquitin ligases such as NEDD4, HECW2, and ITCH along with WW domains (Ingham et al., 2004; Melino et al., 2008; Rotin & Kumar, 2009; Scheffner, Nuber, & Huibregtse, 1995; Weber, Polo, & Maspero, 2019)”

On page 10 lines 12-16:

“In fact, the known binding partners, NEDD4, WWP2, WWP1, and ITCH in human and CG42797, Su(dx), Nedd4, Yki, Smurf, and HERC2 in Drosophila, that were detected in our data are related to ubiquitin ligases and protein degradation (C. Chen & Matesic, 2007; Ingham et al., 2004; Y. Kwon et al., 2013; Marin, 2010; Melino et al., 2008; Rotin & Kumar, 2009) (Figure 1E; Figure S2F).”

On page 13 lines 20-21:

“Given that α-arrestins are widely conserved in metazoans (Alvarez, 2008; DeWire, Ahn, Lefkowitz, & Shenoy, 2007), “

On page 14 lines 12-17:

“The most prominent functional modules shared across both species were the ubiquitin-dependent proteolysis, endosomal trafficking, and small GTPase binding modules, which are in agreement with the well-described functions of α-arrestins in membrane receptor degradation through ubiquitination and vesicle trafficking (Dores et al., 2015; S. O. Han et al., 2013; Y. Kwon et al., 2013; Nabhan et al., 2012; Puca & Brou, 2014; Puca et al., 2013; Shea et al., 2012; Xiao et al., 2018; Zbieralski & Wawrzycka, 2022) (Figure 3).”  

Reviewer #2

In this manuscript, the authors present a novel interactome focused on human and fly alpha-arrestin family proteins and demonstrate its application in understanding the functions of these proteins. Initially, the authors employed AP/MS analysis, a popular method for mapping protein-protein interactions (PPIs) by isolating protein complexes. Through rigorous statistical and manual quality control procedures, they established two robust interactomes, consisting of 6 baits and 307 prey proteins for humans, and 12 baits and 467 prey proteins for flies. To gain insights into the gene function, the authors investigated the interactors of alpha-arrestin proteins through various functional analyses, such as gene set enrichment. Furthermore, by comparing the interactors between humans and flies, the authors described both conserved and species-specific functions of the alpha-arrestin proteins. To validate their findings, the authors performed several experimental validations for TXNIP and ARRDC5 using ATAC-seq, siRNA knockdown, and tissue staining assays. The experimental results strongly support the predicted functions of the alpha-arrestin proteins and underscore their importance. `

I would like to suggest the following analyses to further enhance the study:

1) It would be valuable if the authors could present a side-by-side comparison of the interactomes of alpha-arrestin proteins, both before and after this study. This visual summary network would demonstrate the extent to which this work expanded the existing interactome, emphasizing the overall contribution of this study to the investigation of the alpha-arrestin protein family.

We greatly appreciate your insightful feedback. In response to the reviewer’s suggestion, we’ve depicted a network of known PPIs associated with α-arrestins (Figure S2C and D). Furthermore, by comparing our high-confidence PPIs to these known sets, we found that the overlaps are statistically significant and the high-confidence PPIs of α-arrestins broaden the existing interactome (Figure S2E).

From page 7 line 26 to page 8 line 8, we’ve detailed this side-by-side comparisons of existing interactome and newly discovered high-confidence PPIs of α-arrestins, as outline below.

“As a result, we successfully identified many known interaction partners of α-arrestins such as NEDD4, WWP2, WWP1, ITCH and TSG101, previously documented in both literatures and PPI databases (Figure S2C-F) (Colland et al., 2004; Dotimas et al., 2016; Draheim et al., 2010; Mellacheruvu et al., 2013; Nabhan et al., 2012; Nishinaka et al., 2004; Puca & Brou, 2014; Szklarczyk et al., 2015; Warde-Farley et al., 2010; Wu et al., 2013). Additionally, we greatly expanded repertoire of PPIs associated with α-arrestins in human and Drosophila, resulting in 390 PPIs between six α-arrestins and 307 prey proteins in human, and 740 PPIs between twelve α-arrestins and 467 prey proteins in Drosophila (Figure S2E). These are subsequently referred to as ‘high-confidence PPIs’ (Table S3).”

2) While the authors conducted several analyses exploring protein function, there is a need to further explore the implications of the interactome in human diseases. For instance, it would be beneficial to investigate the association of the newly identified interactome members with specific human diseases. Including such investigations would strengthen the link between the interactome and human disease contexts.

Thank you for your valuable comment. As suggested by the reviewer, we examined the association between α-arrestins’ interactomes and human diseases, incorporating our findings into the discussion. The newly introduced figure based on the result is Figure S10.

On page 24 lines 10-14, we’ve added discussion on Figure S10 as follows.

“We further explored association between α-arrestins’ interactomes and disease pathways (Figure S10). Notably, the interactomes of α-arrestins in human showed clear links to specific diseases. For instance, ARRDC5 is closely associated with disease resulting from viral infection and cardiovascular conditions. ARRDC2, ARRDC4, and TXNIP share common association with certain neurodegenerative diseases, while ARRDC1 is implicated in cancer.”

Reviewer #3:

Lee, Kyungtae and colleagues have discovered and mapped out alpha-arrestin interactomes in both human and Drosophila through the affinity purification/mass spectrometry and the SAINTexpress method. They found the high confident interactomes, consisting of 390 protein-protein interactions (PPIs) between six human alpha-arrestins and 307 preproteins, as well as 740 PPIs between twelve Drosophila alpha-arrestins and 467 prey proteins. To define and characterize these identified alpha-arrestin interactomes, the team employed a variety of widely recognized bioinformatics tools. These included protein domain enrichment analysis, PANTHER for protein class enrichment, DAVID for subcellular localization analysis, COMPLEAT for the identification of functional complexes, and DIOPT to identify evolutionary conserved interactomes. Through these analyses, they confirmed known alpha-arrestin interactors' role and associated functions such as ubiquitin ligase and protease. Furthermore, they found unexpected biological functions in the newly discovered interactomes, including RNA splicing and helicase, GTPase-activating proteins, ATP synthase. The authors carried out further study into the role of human TXNIP in transcription and epigenetic regulation, as well as the role of ARRDC5 in osteoclast differentiation. This study holds important value as the newly identified alpha-arrestin interactomes are likely aiding functional studies of this group of proteins. Despite the overall support from data for the paper's conclusions, certain elements related to data quantification, interpretation, and presentation demand more detailed explanation and clarification.

1) In Figure 1B, it is shown that human alpha-arrestins were N-GFP tagged (N-terminal) and Drosophila alpha-arrestins were C-GFP (C-terminal). However, the rationale of why the authors used different tags for human and fly proteins was not explained in the main text and methods.

We appreciate your valuable comment. Both N- and C-terminally tagged α-arrestins have been used previously. Given that our study aims to increase the repertoire of α-arrestin interacting proteins, where GFP is added might not be a concern. We note that GFP is a relatively bulky tag, and tagging a protein with GFP can potentially abolish the interaction with some of the binding proteins. Follow-up studies utilizing different approaches for detecting protein-protein interactions, such as BioID and yeast two-hybrid, will allow us to build more comprehensive α-arrestin interactomes.

2) In Figure 2A, there seems to be an error for labeling the GAL4p/GAL80p complex that includes NOTCH2, NOTCH1 and TSC2.

Thank you for comment. We double-checked COMPLEAT (protein COMPLex Enrichment Analysis Tool) database for the name of protein complex consisting of NOTCH1, NOTCH2, AND TSC2. The database indeed labeled this complex as the “GAL4p/GAL80p complex”. However, given the potential for mis-annotation (since we could not ascertain the relevance of these proteins to the “GAL4p/GAL80p complex”), we chose to exclude this protein complex from the network. The update protein complex network is illustrated in the revised Figure 2A.

3) In Figure 5, given that knockdown of TXNIP did not affect the levels and nuclear localization of HDAC2, the authors suggest that TXNIP might modulate HDAC2 activity. However, the ChiP assay suggest a different model - TXNIP-HDAC2 interaction might inhibit the chromatin occupancy of HDAC2, reducing histone deacetylation and increasing global chromatin accessibly. The authors need to propose a model consistent with these sets of all data.

We greatly appreciate your detailed feedback. Our data indicates a global decrease in chromatin accessibility (Figure 4C-G) and a diminished interaction between TXNIP and HDAC2 under depletion of TXNIP (Figure 5A). Additionally, we observed an increased occupancy of HDAC2 and subsequent histone deacetylation at TXNIP-target promoter regions (Figure 5C) without any changes in the HDAC2 expression level (Figure 5A) in TXNIP- knockdown cells. From these observations, we infer that the interaction between TXNIP-HDAC2 might suppress the function of HDAC2, a major gene silencer affecting the formation of condensed or accessible chromatin by deacetylating activity. Although we checked whether TXNIP could induce cytosolic retention of HDAC2 to inhibit nuclear function of HDAC2, TNXIP knockdown did not alter its subcellular localization (Figure 5B).

To elucidate the mechanism by which TXNIP inhibits the function of HDAC2, we further investigated the effect of TXNIP on the levels of HDAC2 phosphorylation, which is known to be crucial for its deacetylase activity and the formation of transcriptional repressive complex. However, as shown in the Figure S8C and D, the knockdown of TXNIP did not affect the HDAC2 phosphorylation status, as well as the interaction between HDAC2 and other components in NuRD complex in the immunoblotting and co-IP assays, respectively. The results suggest that TXNIP may inhibit the function of HDAC2 independently of these factors.

Following the reviewer’s suggestion, we carefully provided a proposed model describing the possible role of TXNIP in transcriptional regulation through interaction with HDAC2 and co-repressor complex in Figure S8E.

Description of these newly added figures can be found in the revised manuscript from page 18 line 7 to 27, as outlined below.

“HDAC2 typically operates within the mammalian nucleus as part of co-repressor complexes as it lacks ability to bind to DNA directly (Hassig, Fleischer, Billin, Schreiber, & Ayer, 1997). The nucleosome remodeling and deacetylation (NuRD) complex is one of the well-recognized co-repressor complexes that contains HDAC2 (Kelly & Cowley, 2013; Seto & Yoshida, 2014) and we sought to determine if depletion of TXNIP affects interaction between HDAC2 and other components in this NuRD complex. While HDAC2 interacted with MBD3 and MTA1 under normal condition, the interaction between HDAC2 and MBD3 or MTA1 was not affected upon TXNIP depletion (Figure S8C). Next, given that HDAC2 phosphorylation is known to influence its enzymatic activity and stability (Adenuga & Rahman, 2010; Adenuga, Yao, March, Seagrave, & Rahman, 2009; Bahl & Seto, 2021; Tsai & Seto, 2002), we tested if TXNIP depletion alters phosphorylation status of HDAC2. The result indicated, however, that phosphorylation status of HDAC2 does not change upon TXNIP depletion (Figure S8D). In summary, our findings suggest a model where TXNIP plays a role in transcriptional regulation independent of these factors (Figure S8E). When TXNIP is present, it directly interacts with HDAC2, a key component of transcriptional co-repressor complex. This interaction suppresses the HDAC2 ‘s recruitment to target genomic regions, leading to the histone acetylation of target loci possibly through active complex including histone acetyltransferase (HAT). As a result, transcriptional activation of target gene occurs. In contrast, when TXNIP expression is diminished, the interaction between TXNIP and HDAC2 weakens. This restores histone deacetylating activity of HDAC2 in the co-repressor complex, leading to subsequent repression of target gene transcription.”

4) The authors showed that ectopic expression of ARRDC5 increased osteoclast differentiation and function. Does loss of ARDDC5 lead to defects in osteoclast function and fate determination?

We appreciate your valuable comment. We have confirmed the endogenous expression of ARRDC5 in osteoclasts and conducted a loss-of-function study using shARRDC5. As determined by qPCR, ARRDC5 was endogenously expressed very low in osteoclasts. Even during RANKL-induced osteoclast differentiation, the CT value (29-31) for ARRDC5 expression was high in osteoclasts compared to the CT value (17-24) for the expression of marker genes Cathepsin K, TRAP, and NFATc1. Even though its endogenous expression was very low, we generated ARRDC5 knockdown cells by infecting BMMs with lentivirus expressing shRNA of ARRDC5 and subsequently differentiated the cells into mature osteoclasts. After five days of differentiation, we observed a significant decrease in the total number of TRAP-positive multinucleated cells (No. of TRAP+ MNCs) in shARRDC5 cells compared to that in the control cells. This result indicates that the loss of ARRDC5 leads to defects in osteoclast differentiation. Result of this loss-of-function study using shARRDC5 is depicted in Figure S9A and B.

In the revised manuscript, following sentence explaining Figure S9A and B was added on page 19 lines 15-17 as follows.

“Depletion of ARRDC5 using short hairpin RNA (shRNA) impaired osteoclast differentiation, further affirming its crucial role in this differentiation process (Figure S9A and B).”

5) From Figure 6D, the authors argued that ARRDC5 overexpression resulted in more V-ATPase signals: however, there is no quantification. Quantification of the confocal images will foster the conclusion. Also, western blots for V-ATPase proteins will provide an alternative way to determine the effects of ARRDC5.

We appreciate your insightful feedback. As suggested by the reviewer, we quantified V-type ATPase signals using confocal images, which were shown in Figure 6D. The ImageJ program was employed for integrated density measurements, and the integrated density of GFP-GFP overexpressing osteoclasts was set to 1 for relative comparison. The result in the revised Figure 6D revealed a significant increase in V-type ATPase signals in GFP-ARRDC5 overexpressing osteoclasts compared to that in GFP-GFP overexpressing osteoclasts, as outlined below.

We also agree with the reviewer’s comment that Western blot for V-ATPase proteins will be an alternative way to determine the effects of ARRDC5 in osteoclast differentiation. We have confirmed no different expression of V-type ATPase between GFP-GFP and GFP-ARRDC5 overexpressing osteoclasts using qPCR and western blot analysis. The corresponding western blot result is shown in the revised Figure S9C.

In addition, the corresponding qPCR that measures the expression level of V-type ATPase between GFP-GFP and GFP-ARRDC5 overexpressing osteoclasts is shown in Author response image 3.

Author response image 3.

Moreover, based on the references, the V-type ATPase is localized at the plasma membrane during osteoclast differentiation (Toyomura et al., 2003). Although mRNA and protein expression levels were similar in both cells, localization of V-ATPase in plasma membrane was significantly increased in GFP-ARRDC5 overexpressing osteoclasts compared to that in GFP-GFP osteoclasts, as shown in the revised Figure 6D above.

6) The results from Figure 6D did not support the authors' argument that ARRDC5 might control the membrane localization of the V-ATPase, as bafilomycin is the V-ATPase inhibitor. ARRDC5 knockdown experiments will help to determine whether ARRDC5 can control the membrane localization of the V-ATPase in osteoclast.

Thank you for your insightful comment. V-type ATPase has been reported to play an important role in the differentiation and function of osteoclasts (Feng et al., 2009; Qin et al., 2012). Given that various subunits of the V-type ATPase interact with ARRDC5 (Figure 6A), we speculated that ARRDC5 might be involved in the function of this complex and play a role in osteoclast differentiation and function. As answered above, GFP-ARRDC5 overexpressing osteoclasts showed a similar expression level of V-type ATPase to GFP-GFP cells but exhibited increased V-type ATPase signals at the cell membrane compared to those in GFP-GFP cells (Figure 6D). Additionally, co-localization of ARRDC5 and V-type ATPase was observed in the osteoclast membrane (Figure 6D), as predicted by the human ARRDC5-centric PPI network. On the other side, bafilomycin A1, a V-type ATPase inhibitor, not only blocked localization of V-type ATPase to plasma membrane in GFP-ARRDC5 overexpressing osteoclasts, but also reduced ARRDC5 signals (Figure 6D). These results indicate that ARRDC5 plays a role in osteoclast differentiation and function by interacting with V-type ATPase and promoting the localization of V-type ATPase to plasma membrane in osteoclasts.

V-type ATPase present in osteoclast membrane is important to cell fusion, maturation, and function during osteoclast differentiation (Feng et al., 2009; Qin et al., 2012). GFP-ARRDC5 overexpressing osteoclasts showed a significant increase of V-type ATPase signals in the cell membrane compared to GFP-GFP cells (Figure 6D), and also significantly increased cell fusion (No. of TRAP+ MNCs in Figure 6B) and resorption activity (resorption pit formation in Figure 6C). However, ARRDC5 knockdown in osteoclasts (shARRDC5 cells) showed a significant decrease in No. of TRAP+ MNCs compared to that in the control cells, indicating that the loss of ARRDC5 leads to defects in cell fusion during osteoclast differentiation (Figure S9A and B). As described above, the endogenous expression of ARRDC5 was very low in osteoclasts and could be specifically expressed in a certain timepoint during the differentiation. Therefore, to better understand the interaction with V-type ATPase of ARRDC5 in osteoclasts, ARRDC5 overexpression is more suitable than its knockdown.

Part of the manuscript on page 19 line 21 to page 20 line 6 was edited to support our statement, as outlined below.

“The V-type ATPase is localized at the osteoclast plasma membrane (Toyomura et al., 2003) and its localization is important for cell fusion, maturation, and function during osteoclast differentiation (Feng et al., 2009; Qin et al., 2012). Furthermore, its localization is disrupted by bafilomycin A1, which is shown to attenuate the transport of the V-type ATPase to the membrane (Matsumoto & Nakanishi-Matsui, 2019). We analyzed changes in the expression level and localization of V-type ATPase, especially V-type ATPase V1 domain subunit (ATP6V1), in GFP-GFP and GFP-ARRDC5 overexpressing osteoclasts. The level of V-type ATPase expression did not change in osteoclasts regardless of ARRDC5 expression levels (Figure S9C). GFP signals were detected at the cell membrane when GFP-ARRDC5 was overexpressed, indicating that ARRDC5 might also localize to the osteoclast plasma membrane (Figure 6D; Figure S9D). In addition, we detected more V-type ATPase signals at the cell membrane in the GFP-ARRDC5 overexpressing osteoclasts, and ARRDC5 and V-type ATPase were co-localized at the osteoclast membrane (Figure 6D; Figure S9D).”

7) The tables (excel files) do not have proper names for each table S numbers. Please correct the name of excel files for readers.

We appreciate your valuable comments. In response to the reviewer’s suggestion, we’ve renamed excel files to more appropriate titles for easier readability. List of renamed tables (excel files) are shown below.

Table S1. List of α-arrestins from human and Drosophila Table S2. Evaluation sets of α-arrestins PPIs Table S3. Summary tables of SAINTexpress results Table S4. Protein domains and short linear motifs in the α-arrestin interactomes Table S5. Enriched Pfam domains in the α-arrestin interactomes Table S6. Subcellular localizations of α-arrestin interactomes Table S7. Summary of protein complexes and cellular components associated with α-arrestin Table S8. Orthologous relationship of α-arrestin interactomes between human and Drosophila Table S9. Summary of ATAC- and RNA-seq read counts before and after processing Table S10. Differential accessibility of ACRs and gene expression Table S11. Summary of ATAC-seq peaks located in promoters and gene expression level Table S12. List of primer sequences used in this study

8) http://big.hanyang.ac.kr/alphaArrestin_Fly link does not work. Please fix the link.

We appreciate your comment. In response to the reviewer’s comment, we have made comprehensive α-arrestin interactome maps on our new website (big.hanyang.ac.kr/alphaArrestin_PPIN) and confirmed that users can be re-directed to networks housed in NDEx.

Author response image 4.

Screen shot of the first page of the newly developed website.

Website address: big.hanyang.ac.kr/‌‌‌‌‌‍‍‍‌‌alphaArrestin_PPIN

Author response image 5.

Screen shot of the gene-gene network involving α-arrestin in human.

Author Response

The following is the authors’ response to the original reviews.

Public review

Reviewer 1

Zhang et al. tackle the important topic of primate-specific structural features of the brain and the link with functional specialization. The authors explore and compare gyral peaks of the human and macaque cortex through non-invasive neuroimagery, using convincing techniques that have been previously validated elsewhere. They show that nearly 60% of the macaque peaks are shared with humans, and use a multi-modal parcellation scheme to describe the spatial distribution of shared and unique gyral peaks in both species.

We thank the reviewer for his/her summary and affirmation of our work.

The claim is made that shared peaks are mainly located in lower-order cortical areas whereas unique peaks are located in higher-order regions, however, no systematic comparison is made. The authors then show that shared peaks are more consistently found across individuals than unique peaks, and show a positive but small and non-significant correlation between cross-individual counts of the shared peaks of the human and the macaque i.e. the authors show a non-significant trend for shared peaks that are more consistently found across humans to be those that are also more found across macaques.

Answer: We appreciate the reviewer for raising questions about our work. In order to provide a more systematic comparison for the conclusion that ‘shared peaks are mainly located in lowerorder cortical areas whereas unique peaks are located in higher-order regions’, we have conducted two additional experiments. Following the reviewers’ suggestions, we conducted a statistical analysis of the ratio of shared and unique peaks within different brain networks (as depicted in Figure 2 (b)), and also presented the specific distribution quantities of the two types of peaks in both low- and high-order brain networks (as detailed in the corresponding Table 1). Through these three experiments, we have obtained a more systematic and comprehensive conclusion that ‘shared peaks are more distributed in lower-order networks, while unique peaks are more in higher-order networks’.

In order to identify if unique and shared peaks could be identified based on the structural features of the cortical regions containing them, the authors compared them with t-tests. A correction for multiple comparisons should be applied and t-values reported. Graph-theoretical measures were applied to functional connectivity datasets (resting-state fMRI) and compared between unique and shared peak regions for each species separately. Again the absence of multiple comparison correction and t-values make the results hard to interpret. The same comment applies to the analysis reporting that shared peaks are surrounded by a larger number of brain regions than unique peaks. Finally, the potentially extremely interesting results about differential human gene expression of shared and unique peaks regions are not systematically reported e.g. the 28 genes identified are not listed and the selection procedure of 7 genes is not fully reported.

Answer: We appreciate the reviewer for their suggestions about the statistical analysis in our manuscript. Firstly, we applied False Discovery Rate (FDR) correction to all experiments involving multiple comparisons throughout the entire manuscript, and the corrected t-values are reported (Table 2-5 and A5-A6). Additionally, in response to the reviewers’ guidance regarding the gene analysis section, we provided a list of 28 genes (Table A7) selected by lasso, along with the t-values obtained from Welch’s t-test for the expression of the two type of peaks. The functions corresponding to the seven genes with final t-values below 0.05 are reported in Table 6.

The paper is well written and the methods used for data processing are very compelling i.e. the peak cluster extraction pipeline and cross-species registration. However, the analysis and especially the reporting of statistics, as they stand now, constitutes the main weakness of the paper. Some aspects of the statistical analysis need to be clarified.

Reviewer 2

The authors compared the cortical folding of human brains with folding in macaque monkey brains to reveal shared and unique locations of gyral peaks. The shared gyral peaks were located in cortical regions that are functionally similar and less changed in humans from those in macaques, while the locations of unique peaks in humans are in regions that have changed or expanded functions. These findings are important in that they suggest where human brains have changed more than macaque brains in their subsequent evolution from a common ancestor. The massive analysis of comparative results provides evidence of where humans and macaques are similar or different in cortical markers, as well as noting some of the variations within each of the two primates.

Answer: Gratitude to the reviewer for his/her summary and appreciation of our cross-species work.

Strengths:

The study includes massive detail.

Weaknesses:

The manuscript is too long and there is not enough focus on the main points.

Answer: We appreciate the reviewer for pointing out the shortcomings in our manuscript. Firstly, considering the manuscript is too long, we have chosen to retain only the core experiments and relevant analyses in the main text. Relatively minor conclusions have been moved to the supplementary information, such as original Table 1 is now moved to the Supplementary Information as Table A1 (locations of all shared clusters). Additionally, some non-essential expressions in the original manuscript have been removed.

Our experiments primarily revealed the existence of partially shared cortical landmarks, known as gyral peaks, in both humans and macaques. We found that these shared and unique peaks are mainly distributed across low- and high-order brain networks. To emphasize this main point, we added two experiments on top of the existing ones to provide a more systematic explanation of this conclusion. We conducted a statistical analysis of the ratio of shared and unique peaks within different brain networks (as depicted in Figure 2 (b)), and also presented the specific distribution quantities of the two types of peaks in both low- and high-order brain networks (as detailed in the corresponding Table 1). By combining the results of these two experiments with the original manuscript’s statistical findings on the proportions of the two type of peaks in different brain networks, the conclusion that ‘shared and unique peaks are predominantly located in low-order and high-order brain networks’ becomes more prominent.

A brief listing of previous views on why fissures form and what factors are important would be helpful.

Answer: In response to this suggestion from the reviewer, we have incorporated some previous views on why fissures form and what factors are important into the ‘Introduction’ section.

‘Cortical folds are important features of primate brains. The primary driver of cortical folding is the differential growth between cortical and subcortical layers. During the gyrification process in the cortex, areas with high-density stiff axonal fiber bundles towards gyri. The brain’s folding pattern formed through a series of complex processes. The folding patterns in the brain, formed through a series of complex processes, are found to play a crucial role in various cognitive and behavioral processes, including perception, action, and cognition (Fornito et al. 2004; Cachia et al. 2018; Yang et al. 2019; Whittle et al. 2009).’

Reviewer 1 (Recommendations For The Authors):

(1) Figure 3b shows a non-significant trend for shared peaks that are more consistently found across humans to be those that are also more found across macaques. In the discussion, lines 218-219, the fact that the correlation is not significant should be reported more clearly.

Answers: We thank the reviewer for this question. We revised the Line 218-219 (now Line 257-259) as follows: ‘2. Consistency: The inter-individual consistency of shared peaks within each species was greater than that of unique peaks. The consistency of shared peaks in the human and macaque brains exhibits a positive correlation (non-significant though).’

(2) It is not fully clear how much shared peaks are mostly distributed in the higher-order cortex, especially in the macaque. It is reported in the results lines 132-133 that ‘In the macaque brain, shared peak cluster centers most distributed in the V2, DMN, and CON (Figure.2 (d)), while unique peak cluster centers most distributed in the DMN, Language (Lan), and Dorsal-attention (DAN)’ but not further discussed. Please develop this point in the discussion. Further, the results presented in Figures 2 and A1 are actually quite different and this shall be better described in the results. Given that shared and unique peaks can be found in the same region, this analysis would gain importance by applying a comparison test for the selection of regions where the most shared or unique peaks are found. The sentence lines 306-308 should be accordingly revised.

It is hard to understand what the 0-3% corresponds to in Figures 2 and A1?

Please also correct in both legends and in the text the labeling of panels and add in the legends a brief description of panel (c). In the legend of Figure 2, ‘shared peaks’ in the second sentence shall be replaced by ‘unique peaks’.

Answers: We thank the reviewer for these questions and suggestions. Our responses to them are itemized as follows:

A1: In general, to clarify the distribution of shared and unique peaks in the high-order and loworder networks, we divided 12 brain networks in Cole-Anticevic atlas into the low-order networks (visual 1 (V1), visual 2 (V2), auditory (Aud), somatomotor (SMN), posterior multimodal (PMN), ventral multimodal (VMN), and orbito-affective networks (OAN)) and higher-order networks (include cingulo-opercular (CON), dorsal attention (DAN), language (Lan), frontoparietal (FPN), default mode network (DMN)) based on previous research (Golesorkhi et al. 2022; Ito, Hearne, and Cole 2020). On this lower/higher -order division, we reported the number of shared and unique peaks in both species in Author response table 1. It is found that, whether in humans or macaques, shared peaks are more distributed in lower-order networks, while unique peaks are more in higher-order networks. This observation is particularly pronounced in humans.

Author response table 1.

The number of shared and unique peaks in lower- and higher-order brain networks of the two species. Lower-order networks include visual 1 (V1), visual 2 (V2), auditory (Aud), somatomotor (SMN), posterior multimodal (PMN), ventral multimodal (VMN), and orbito-affective networks (OAN), higher-order networks include cingulo-opercular (CON), dorsal attention (DAN), language (Lan), frontoparietal (FPN), default-mode network (DMN).

In the main text, Figure 2 (referring to Author response figure 1 later in the text.) illustrates the proportions of shared and unique peaks across 12 brain networks in both species. In each pie chart, we have specifically highlighted the top three ranked brain regions. Although the pie chart also generally supports the above results, two brain networks deserve further discussion. They are DMN and CON, two higher-order networks that have higher ranks in terms of shared peak count (the second-ranked and the third-ranked on macaque shared peaks; the fourth-ranked and the fifth-ranked on human shared peaks).

The cingulo-opercular network (CON) is a brain network associated with action, goal, arousal, and pain. However, a study found three newly discovered areas of the primary motor cortex that exhibit strong functional connectivity with the CON region, forming a novel network known as the somato-cognitive action network (SCAN) (Gordon et al. 2023). The SCAN integrates body control (motor and autonomic) and action planning, consistent with the findings that aspects of higher-level executive control might derive from movement coordination (Llinás 2002; Gordon et al. 2023). CON may be shared in the form of the SCAN network across these two species. This could explain in part the results in Author response figure 1 that shared peaks are more on CONs.

Author response image 1.

Pie chart shows the count of shared and unique peaks across different brain networks for both human and macaque. Right panel shows the Cole-Anticevic (CA) networks (Ji et al. 2019) on human surface as a reference.

Default-mode network (DMN) is a ensemble of brain regions that are active in passive tasks, including the anterior and posterior cingulate cortex, medial and lateral parietal cortex, and medial prefrontal cortex (Buckner, Andrews-Hanna, and Schacter 2008). Although DMN is considered a higher-order brain network, numerous studies have provided evidence of its homologous presence in both humans and macaques. Many existing studies have confirmed the similarity between the DMN regions in humans and macaques from various perspectives, including cytoarchitectonic (Parvizi et al. 2006; Buckner, Andrews-Hanna, and Schacter 2008; Caminiti et al. 2010) and anatomical tracing (Vincent et al. 2007). These studies all support the notion that some elements of the DMN may be conserved across primate species (Mantini et al. 2011). In general, the partial sharing of DMN between humans and macaques may be attributed to the higher occurrence of shared peaks within the DMN.

These results have been added to Table 2 along with corresponding text and discussion section.

A2: The difference between the results of Figure 2 and Figure A1 (now Figure A2) is whether the peak count is normalized by cortical area, which hugely varies across networks. For example, among the 12 brain networks, the three networks with the largest surface areas are the DMN, SMN and CON, and the three networks with the smallest area are OAN, PMN and VMN. The area difference between networks can be as large as 18-fold. Therefore, it is not difficult to find that, although the DMN ranks high in both shared and unique peak counts during statistical analysis (Figure 2 (a)), it is relatively small in Figure A2 after area normalization. In contrast, VMN ranks lower in peak count statistics but exhibits a substantial proportion after area normalization (For example, 38% of macaque shared peaks are distributed in the VMN region, but there are actually only four peaks). However, the two pie charts deliver the same message that there are more shared peaks in lower-order networks, while unique peaks are more in higher-order networks (except for macaques, where shared peaks are also distributed significantly in DMN and CON).

Following the suggestion from the reviewer, we adopted a new approach to present the ratio between shared peak count and unique peak count for each network (see Author response figure 2), such that the networks where the most shared or unique peaks are found can be easily highlighted. To mitigate potential imbalances in proportions caused by differences in the absolute numbers of each category (shared or unique) of peak, the proportions of peaks within their respective categories were utilized in the calculations. In Author response figure 2, the pink and green color bins represent ratios of shared and unique peaks, respectively. The dark blue dashed line represents the 50% reference line. In general, from left to right in the figure, the ratio of shared peaks decreases gradually while the ratio of unique peaks increases, suggesting that shared peaks are more (>0.5, above the dashed line) on lower-order networks (orange font), while unique peaks are generally more on higher-order networks (blue font). In specific, in human brains, the networks with a higher abundance of shared peaks are Aud, VMN, V1, SMN, and V2; whereas in macaques, they are CON, VMN, V1, V2, FPN, and SMN. Again, in the human brains, the disparity between shared and unique peaks tends to be more significant (further away from the reference line), for both lower-order and higher-order networks, respectively. In contrast, in the macaque brains, the disparity between shared and unique peaks is less significant (closer to the reference line). The ratio of shared and unique peaks is around 0.5 for 6 out of all 10 networks (including both lower and higher-order ones).

Author response image 2.

The ratio of shared and unique peaks in each brain network in the Cole-Anticevic (CA) atlas. The pink and green color bins represent ratios of shared and unique peaks, respectively. The dark blue dashed line represents the 50% reference line. For each brain region, the sum of the ratios of shared and unique peaks is equal to 1.

Based on these analyses, the sentence lines 306-308 (now Line 368-370) has been revised as follows: ‘In the human brain, the more shared peaks (about 65%) are located in lower-order brain regions, while unique peaks are mainly (about 74%) located in higher-order regions. However, this trend is relatively less pronounced in the macaque brain.’

These results have been added to Figure 2 (b) along with corresponding text and discussion section.

A3: In response to the third suggestion from the reviewer, we have clearly labeled the brain region names corresponding to 0% to 3% in Figure 2 (now Figure 2 (a)) and Figure A1 (now Figure A2).

Author response image 3.

Pie chart shows the count of shared and unique peaks across different brain networks for both human and macaque. Right panel shows the Cole-Anticevic (CA) networks (Ji et al. 2019) on human surface as a reference.

A4: Finally, we would like to express our gratitude to the reviewer for pointing out our mistakes.

We have made improvements to Figure 2 and revised the figure captions accordingly.

(3) The conclusions regarding the spatial relationship between peaks and functional regions shall be revised (Lines 187-188, 228-229, and 329-330). In the macaque, the results are opposite in the two atlases used. Further, in the human, it is not clear how multiple comparison corrections will impact statistics and some atlases show opposite results, although conclusions hold true in the majority of human atlases.

Answers: We thank the reviewer very much for this suggestion. We have added the results of the Cole-Anticevic atlas for macaques in the main text, which also has the observation that shared>unique (Author response table 2, corresponds to Table 5 in main text), namely, there are more diverse brain regions around shared peaks than around unique peaks. Therefore, out of the commonly used three macaque atlases, two (Markov91 and Cole-Anticevic) conform to this observation, while BA05 does not. We utilized false discovery rate (FDR) correction for multiple comparisons, and the corrected p-values are reported in Tables (in the revised main text and are shown below). Results on atlas with multiple resolutions are reported in Author response table 4) (Table A6 in the Supplementary Information). The observation that more diverse brain regions around shared peaks than around unique peaks, holds for human atlases in Author response table 3) (Table 4 in main text), where the atlas resolutions ranges from 7 parcels to 300 parcels, demonstrating the robustness of the conclusion. It is noted that the observation is not consistent on atlases with relatively lower resolutions (e.g., BA05 for macaque, n=30 and Yeo2011 for human, n=7) or, in particular, higher resolutions (e.g., Schaefer-500, and Vosdewael-400, n>300). This inconsistency could be reasonable since the resolution of the parcellation itself will largely determines the chance of a cortical region appear in a peak’s neighborhood, if the parcellation is too coarse or too fine. For example, if n=1 (the entire cortex is the only one region) or n=30k (each vertex is a region), each peak will has the same number of neighboring regions for these two extreme cases (one brain region for each peak for n=1; around 30 vertices for each peak for n=30k).

In conclusion, we observed that there are more diverse brain regions around shared peaks than around unique peaks for multiple brain atlases with a median parcellation resolution. These results have been added to Tables 4, 5, and A6 along with corresponding text and discussion section.

Author response table 2.

The mean values (±SD) of brain regions that appeared within a 3-ring neighborhood for shared and unique peaks in 3 common macaque atlases. For both Markov91 and Cole-Anticevic atlas, the shared peaks has more variety of functional regions around it than the unique peaks. But for the altas BA05, the conclusion was reversed. The bold font represent the larger values between the shared peak and unique peaks. All p<0.001, after false discovery rate (FDR) corrected.

(4) For Tables 2-4, A4, and Figure 3a, please indicate in all the legends if values correspond to Mean plus minus Standard Deviation, report t-value, and n in the legend or in the text.

Answers: We thank the reviewer very much for this suggestion. We added the ‘mean (±SD)’ in the notes of Tables 2-4, A4 (now A6), and Figure 3 (a). All the t and n values of t-test are reported in tables or in the main text.

(5) Please create a statistical section in the Methods to describe more precisely the tests used e.g. for t-tests, if datasets follow a normal distribution with unknown variance. In the case of multiple comparisons like in e.g. Table 2-4, A4, please report what multiple comparisons correction was used to adjust the significance level.

Author response table 3.

The mean values (±SD) of brain regions that appeared within a 3-ring neighborhood for shared and unique peaks in 10 common human atlases. All the shared peaks in the table have a greater number of neighboring brain regions compared to the unique peaks. All p<0.001, false discovery rate (FDR) corrected.

Author response table 4.

The mean values (±SD) of brain regions where shared and unique peaks appeared within a 3-ring neighborhood in 21 common human atlases. The p-values were corrected by FDR.

Answers: Thanks for the reviewer’s suggestion, we added a ‘Statistic Analysis’ section in the ‘Materials and Methods’ part:

‘All variables used in the two-samples t-test follow a normal distribution check and all p-values were corrected for multiple comparisons using the false discovery rate (FDR) method. Moreover, in order to identify differently expressed genes between shared and unique peaks, we employed the Welch’s t-test, given the unequal sample sizes for shared and unique peaks. For all tests, a p-value <0.05 was considered significant (FDR corrected).’

For the experiments of multiple comparisons such as Table 2-4, A4 (now A6), etc., we have added explanations in the main text, multiple comparisons correction has been corrected by false discovery rate (FDR), p-value<0.05 is considered significant.

(6) It would be of great interest to provide the full list of the 28 genes that significantly contributed to the classification of shared and unique peaks. Please provide a description of the Welch’s t-test results. From the 7 genes selected, only two are discussed. Could the authors please describe briefly the function of the other genes? Although we understand that they are not associated with neuronal activity and brain function.

Answers: We thank the reviewer for these suggestions. We have provided a complete list of 28 genes selected by LASSO in the Author response table 5. Additionally, Welch’s t-test was employed to calculate p-values for the expression differences of each gene in shared and unique peak clusters, and the results are also reported in the Author response table 5.

Author response table 5.

The 28 genes selected by LASSO and their corresponding p-values from Welch’s t-test.

Seven genes showed significant differential expression between shared and unique peaks in Welch’s t-test. These genes were PECAM1, TLR1, SNAP29, DHRS4, BHMT2, PLBD1, KCNH5. Brief descriptions of their functions are listed in Author response table 6. All gene function descriptions were derived from the NCBI website (https://www.ncbi.nlm.nih.gov/).

These results have been added to Tables 6 and A7 along with corresponding text.

(6) For comparison, could the authors provide a supplementary figure of shared peak clusters like in Figure 1b but displayed on the surface of the macaque brain template?

Answers: We thank the reviewer very much for this suggestion and we have incorporated a display of shared peak clusters on the macaque brain template surface (Author response figure 4, corresponds to Figure A1 of Supplementary Information.)

(7) Could the author develop or rephrase the sentence lines 69-72 which remains unclear?

Answers: We appreciate the reviewer’s feedback and have revised this sentence to ensure clarity. The sentences from line 69 to 72 have been revised to ‘In the study of macaques, it has been observed that the peak consistently present across individuals is located on more curved gyri (S. Zhang, Chavoshnejad, et al. 2022). Similar conclusions have been drawn in human brain research (S. Zhang, T. Zhang, et al. 2023).’ Now, this sentence corresponds to lines 74-77 in the main text.

(8) Line 99: please indicate which section.

Author response table 6.

Seven genes were selected using LASSO that showed significant differential expression in shared and unique peaks.

Answers: We thank the reviewer very much for this suggestion and we revised this sentence to ‘The definition of peaks and the method for extracting peak clusters within each species are described in the Materials and Methods section’.

(9) In Figure 3b, please report R2 and p-value. A semi-log might be more appropriate given the overdispersion of Human Peak Counts.

Answers: We thank the reviewer very much for this suggestion. Linear regression analysis was conducted on the average counts of all corresponding shared peak clusters of human and macaque. The horizontal and vertical axes of the Author response figure 5 (b) represent the average count of shared peaks in the macaque and human brains, respectively. The Pearson correlation coefficient (PCC) of the interspecies consistency of the left and right brain is 0.20 and 0.26 (p>0.05 for both), respectively. The result of linear regression shows that there is a positive correlation in the inter-individual consistency of shared peaks between macaque and human brains, but it is not statistically significant (with R2 for the left and right brain are 0.07 and 0.01, respectively).

Author response image 4.

Shared peak clusters of macaque, shows on macaque brain template.

The goodness of fit (R2), pearson correlation coefficient (PCC), and their respective p-values were indicated in Author response figure 5 (b). To avoid overdispersion, the peak count of the human brain is displayed in a semi-log format.

The updated Figure and results are presented in Figure 3 of the main text.

(10) Line 177: please indicate where in the Supplementary Information.

Answers: Thank you for the reviewer’s reminder. We have incorporated the results of the human brain structural connectivity matrix into Table A5 in the Supplementary Information and provided corresponding indications in the main text.

(11) Line 226: please correct ‘(except for betweeness [and efficiency] of the’.

Answers: We thank the reviewer very much for this suggestion and we added ‘and efficiency’ in original Line 173 and 226 (now Line 206 and 267) after ‘betweeness’.

(12) The gene expression dataset used is from the Allen Human Brain Atlas (AHBA). Reference to Hawrylycz et al., 2012 Nature. 2012 Sep 20;489(7416):391-399. doi: 10.1038/nature11405 shall be made and abbreviation defined at first use in the text.

Answers: We added the full name ‘Allen Human Brain Atlas’ when AHBA is first mentioned, along with the reference suggested by the reviewer.

Author response image 5.

(a) Mean peak count (±SD) covered by shared and unique peak clusters in two species. ***indicates p<0.001. The t-values for the t-tests in humans and macaques are 4.74 and 2.67, respectively. (b) Linear regression results of the consistency of peak clusters shared between macaque and human brains. The pink and blue colors represent the left and right hemispheres, respectively. The results of the linear regression are depicted in the figure. While there was a positive correlation observed in the consistency of gyral peaks between macaque and human, the obtained p-value for the fitted results exceeded the significance threshold of 0.05.

(13) Line 17: remove ‘are’.

Answers: We thank the reviewer very much for this suggestion and we removed ‘are’ in Line 17 (now Line 18).

(14) Line 201: remove ‘is used’.

Answers: We thank the reviewer very much for this suggestion and we removed ‘is used’ in Line 201 (now Line 237).

References

Buckner, Randy L, Jessica R Andrews-Hanna, and Daniel L Schacter (2008). “The brain’s default network: anatomy, function, and relevance to disease”. In: Annals of the new York Academy of Sciences 1124.1, pp. 1–38.

Cachia, Arnaud et al. (2018). “How interindividual differences in brain anatomy shape reading accuracy”. In: Brain Structure and Function 223, pp. 701–712.

Caminiti, Roberto et al. (2010). “Understanding the parietal lobe syndrome from a neurophysiological and evolutionary perspective”. In: European Journal of Neuroscience 31.12, pp. 2320–2340.

Fornito, Alexander et al. (2004). “Individual differences in anterior cingulate/paracingulate morphology are related to executive functions in healthy males”. In: Cerebral cortex 14.4, pp. 424–431.

Golesorkhi, Mehrshad et al. (2022). “From temporal to spatial topography: hierarchy of neural dynamics in higher-and lower-order networks shapes their complexity”. In: Cerebral Cortex 32.24, pp. 5637–5653.

Gordon, Evan M et al. (2023). “A somato-cognitive action network alternates with effector regions in motor cortex”. In: Nature, pp. 1–9.

Ito, Takuya, Luke J Hearne, and Michael W Cole (2020). “A cortical hierarchy of localized and distributed processes revealed via dissociation of task activations, connectivity changes, and intrinsic timescales”. In: NeuroImage 221, p. 117141.

Ji, Jie Lisa et al. (2019). “Mapping the human brain’s cortical-subcortical functional network organization”. In: Neuroimage 185, pp. 35–57.

Llinás, Rodolfo R (2002). I of the vortex: From neurons to self. MIT press.

Mantini, Dante et al. (2011). “Default mode f brain function in monkeys”. In: Journal of Neuroscience 31.36, pp. 12954–12962.

Parvizi, Josef et al. (2006). “Neural connections of the posteromedial cortex in the macaque”. In:Proceedings of the National Academy of Sciences 103.5, pp. 1563–1568.

Vincent, Justin L et al. (2007). “Intrinsic functional architecture in the anaesthetized monkey brain”.In: Nature 447.7140, pp. 83–86.

Whittle, Sarah et al. (2009). “Variations in cortical folding patterns are related to individual differences in temperament”. In: Psychiatry Research: Neuroimaging 172.1, pp. 68–74.

Yang, Shimin et al. (2019). “Temporal variability of cortical gyral-sulcal resting state functional activity correlates with fluid intelligence”. In: Frontiers in neural circuits 13, p. 36.

Zhang, Songyao, Poorya Chavoshnejad, et al. (2022). “Gyral peaks: Novel gyral landmarks in developing macaque brains”. In: Human Brain Mapping 43.15, pp. 4540–4555.

Zhang, Songyao, Tuo Zhang, et al. (2023). “Gyral peaks and patterns in human brains”. In: Cerebral Cortex.

Author response:

The following is the authors’ response to the original reviews.

Public Reviews:

Reviewer #1 (Public Review):  

In this study, Hunt et al investigated the role of the ubiquitin-conjugating enzyme UBE2D/effete (eff) in maintaining proteostasis during aging. Utilizing Drosophila as a model, the researchers observed diverse roles of E2 ubiquitinconjugating enzymes in handling the aggregation-prone protein huntingtin-polyQ in the retina. While some E2s facilitated aggregate assembly, UBE2D/eff and other E2s were crucial for degradation of hL-polyQ. The study also highlights the significance of UBE2D/eff in skeletal muscle, showing that declining levels of eff during aging correlate with proteostasis disruptions. Knockdown of eff in muscle led to accelerated accumulation of poly-ubiquitinated proteins, shortened lifespan, and mirrored proteomic changes observed in aged muscles. The introduction of human UBE2D2, analogous to eff, partially rescued the deficits in lifespan and proteostasis caused by eff-RNAi expression in muscles. 

The conclusions of this paper are mostly well supported by data, although a more precise mechanistic explanation of phenotypes associated with UBE2D/eff deficiency would have strengthened the study. Additionally, some aspects of image quantification and data analysis need to be clarified and/or extended.  

We thank reviewer #1 for the thoughtful assessment of our work. We have amended the discussion to better explain the phenotypes associated with UBE2D/eff deficiency. We have also improved the methods describing the procedures for image quantification and data analysis.

Reviewer #2 (Public Review):  

Important findings: 

- Knockdown of UBE2D increases HTT aggregation. 

- Knockdown of UBE2D leads to an accumulation of ubiquitinated proteins and reduces the lifespan of Drosophila, which is rescued by an ectopic expression of the human homolog. 

- UBE2D protein levels decline with aging. 

- UBE2D knockdown is associated with an up- and downregulation of several different cellular pathways, including proteostasis components. 

Thank you for reviewing our manuscript.

Caveats: 

- The readout of HTT aggregation (with methods that are not suitable) as a proxy for the role of UBE2D in proteostasis is not convincing. It would probably improve the manuscript to start with the proteomic analysis of UBE2D to demonstrate that its protein levels decrease with aging. The authors could then induce UBE2D in aged animals to assess the role of UBE2D in the proteome with aging.  

While presenting the data in a different order would be possible, we prefer to keep the current order in which from a general screen with a proteostasis readout (HTT aggregates; see the answer below for a discussion on the methods) we proceed to identify a candidate (UBE2D) which is then studied in more detail with additional focused analyses in the retina and skeletal muscle during aging. Concerning the induction of UBE2D in aged animals, our analyses in Figure 4E demonstrate that muscle-specific induction of UBE2D2 throughout life does not increase lifespan alone: this could be explained by UBE2D2 only partially recapitulating the function and substrate diversity of Drosophila eff/UBE2D due to divergence from a single Drosophila UBE2D enzyme (eff) to multiple UBE2D enzymes in humans (UBE2D1/2/3/4).

- UBE2D knockdown increases the number of HTT foci (Figure 1A), but the quantification is less convincing as depicted in Figure 1B, and other E2 enzymes show a stronger effect (e.g. Ubc6 that is only studied in Figures 1 and 2 without an explanation and Ubc84D). The graph is hard to interpret. What is the sample size and which genetic conditions show a significant change? P values and statistical analyses are missing.  

The full data underlying this genetic screen is reported in Supplementary Table 1. The role of UBC6/UBE2A/B is thoroughly examined in Hunt et al 2021 (PMID: 33658508). We agree that Ubc84D has an important effect and that it should be considered for future studies. We have amended the legend of Figure 1 to indicate that each data point in the graph represents a single RNAi line targeting the corresponding gene. The mean of 5 biological replicates is shown for each RNAi, with each biological replicate representing a single eye imaged from a distinct fly. Therefore, the data points that do not show large magnitude changes may indicate RNAi lines that were not effective at knocking down the target protein (or that did not affect HTT aggregates). The E2s worth pursuing were identified because of multiple RNAi lines scoring consistently: this is the case of UBC6 (studied previously in PMID: 33658508) and eff/UBE2D (pursued in this study). This screen was therefore utilized to identify and select candidate genes (i.e. eff/UBE2D) for more in-depth studies on proteostasis.

- The quantification of the HTT fluorescence cannot be used as a proxy for HTT aggregation. The authors should assess HTT aggregation by e.g. SDD-AGE, FRAP, filter retardation, etc. The quantification of the higher MW species of HTT in the SDS-PAGE is not ideal either as this simply reflects material that is stuck in the wells that could not enter the gel. Aggregation and hence high MW size could be one reason, but it can also be HTT trapped in cell debris, etc.  

We agree that the use of multiple methods is a good way to assess the impact of E2 enzymes on HTT protein aggregation. In this regard, we estimated HTT aggregates by fluorescence microscopy and by western blot. Microscopy-based analyses demonstrate both the accumulation of the HTT-GFP pathogenic protein into aggregates (HTT polyQ polypeptides aggregating into one spatial region; Fig. 1 and Fig. 2B) as well as their potential cytotoxicity, resulting in the disruption of the ommatidial ultrastructure and cellular degeneration (Fig. 2A). Similar to native gels and filter retardation, we have utilized SDS-PAGE and western blotting of cellular samples isolated with strong chaotropic and denaturing reagents (8M urea plus detergents and reducing reagents used in the lysis). These experimental conditions maintain the higher-order organization of HTT into high-molecular-weight aggregates that are not broken down into individual polypeptides and that therefore do not readily travel through a gel or filter. Therefore, the biochemical methods we have used are equivalent to those proposed by the reviewer. In addition to combining microscopy-based and biochemical approaches to examine the impact of eff/UBE2D on the HTT aggregates, we have analyzed eff/UBE2D during skeletal muscle aging and found consistent phenotypes as those observed in the HTT model: RNAi for eff/UBE2D leads to the accumulation of detergent-insoluble ubiquitinated proteins that associate with protein aggregates.

- Does UBE2D ubiquitinate HTT? And thus, is HTT accumulation a suitable readout for the functional assessment of the E2 enzyme UBE2D? 

We propose that the accumulation of HTT in response to eff/UBE2D RNAi may be due to a generalized loss of protein quality control rather than to a direct decline in the ubiquitination of HTT by eff/UBE2D. In a previous study that examined the UBE2D interactome (Hunt et al. 2023; PMID: 37963875), we did not find an interaction between UBE2D and HTT, suggesting that HTT may not be directly modulated by eff/UBE2D via ubiquitination.

- The proteomic analyses could help to identify potential substrates for UBE2D.

The proteomic analyses in Figure 5 identify several proteins that are modulated by RNAi for eff and by its human homolog, UBE2D2. Such eff/UBE2D2-modulated proteins may indeed be potential substrates for UBE2D-mediated ubiquitination. For example, this is the case for Pex11 and Pex13, which were found to be upregulated upon UBE2D RNAi also in human cells, where they are ubiquitinated in a UBE2D-dependent manner (Hunt et al. 2023; PMID: 37963875).

- Are there mutants available for UBE2D or conditional mutants? One caveat of RNAi is: first not complete knockdown and second, variable knockdown efficiencies that increase variability.

There are potential hypomorphic alleles of eff/UBE2D that may be available, but they would present the same caveats of incomplete loss of eff/UBE2D function as RNAi. Given the strong phenotype that we find with partial eff knockdown, a caveat of full eff/UBE2D knockout is that this could be lethal.

- The analysis of the E3 enzymes does not add anything to this manuscript. 

The analysis of E3 enzymes relates to our recent publication (Hunt et al. 2023; PMID: 37963875) that reports the physical interactions between E2 and E3 enzymes. Analysis of these E2-E3 pairs in the genetic screen in Fig.1 therefore follows this IP-MS study to provide insight into the functional interaction between these E2-E3 pairs in proteostasis.

- Figure 2B: the fluorescence intensities in images 2 and 4 are rather similar, yet the quantification shows significant differences. 

Please note that some of the GFP fluorescence in image 4 is not punctate, but rather diffuse fluorescence that is not related to HTT-GFP aggregates. Our image quantitation methods utilized thresholding to identify GFP-positive puncta while eliminating background fluorescence not corresponding to HTT-GFP puncta.

- The proteomic analyses could provide insights into the functional spectrum of UBE2D or even the identification of substrates. Yet apart from a DAVID analysis, none of the hits were followed up. In addition, only a few hits were labelled in the volcano plots (Figure 5). On what basis did the authors select those?

Please see the previous answer above regarding the identification of eff/UBE2D protein substrates from our proteomic analysis in Fig. 5. Only some of the top-regulated hits could be labeled in Fig.5 to avoid overcrowding.

- The manuscript remains at this stage rather descriptive. 

Our study has demonstrated a key role for the eff/UBE2D ubiquitin-conjugating enzyme in regulating protein quality control during aging in the Drosophila retina and skeletal muscle. Our study has identified key proteins that are modulated by eff/UBE2D RNAi in Drosophila muscle, that are rescued by expression of human UBE2D2, and that may underlie the accelerated decline in proteostasis that occurs upon eff/UBE2D RNAi. While more could be known about the regulation of these eff/UBE2D-modulated proteins in Drosophila, we have previously demonstrated that some of the proteins that are upregulated by UBE2DRNAi in human cells (e.g. some peroxins) are indeed direct ubiquitination targets of UBE2D via associated E3 ubiquitin ligases (Hunt et al. 2023; PMID: 37963875).

Reviewer #3 (Public Review):  

This is a potentially quite interesting paper that defines E2 and E3 genes in Drosophila that can impact the accumulation of the Q72-GFP protein in the fly eye. The authors then focus on the eff gene, showing which human homolog can rescue fly knockdown. They extend to skeletal muscle, from the hL protein, to show that eff by TMT mass spec decreases with age normally in the fly muscle and that there is a significant overlap of proteins that are disrupted with eff knockdown in young animals in muscle vs aged animals normally in muscle. 

Overall these data suggest eff decrease with age may contribute to the increase in ubiquitinated proteins in muscle with age, and that upregulation of eff activity might be of interest to extending lifespan. Because eff function can be performed by a human homologue, the findings may also apply to human situations of aging. 

These data are overall interesting and are of relevance for those interested in neurodegenerative disease and aging, although a number of points from the figures seem confusing and need more explanation or clarity. 

Thank you for reviewing our manuscript, we have improved the explanations and clarity of the manuscript.

Recommendations for the authors:

We would like to keep the manuscript title as it is currently to report the partial overlap in the proteomic changes induced by aging and effRNAi (Fig. 6).

Reviewer #1 (Recommendations For The Authors): 

(1) A significant concern arises from the unexpected outcome observed in the UBE2D/eff loss-of-function experiments. Despite its role as a ubiquitin-conjugating enzyme (E2), the reduction in UBE2D/eff levels paradoxically increased polyubiquitinated proteins and p62 accumulation, presenting a more intricate and seemingly unrelated phenotype to its anticipated function. 

eff/UBE2D represents one out of 21 different Drosophila E2 ubiquitin-conjugating enzymes and therefore eff RNAi alone is unlikely to reduce the total pool of ubiquitinated proteins. The generalized increase in insoluble polyubiquitinated proteins results from an overall derangement of protein quality control caused by effRNAi. In agreement with this scenario, the protein categories that were found to be modulated by effRNAi (Fig. 5) include proteins associated with protein quality control such as proteasome components and chaperones. Therefore, derangement in the levels of a wide range of regulators of proteostasis may lead to a generalized loss of protein quality control upon effRNAi.

I believe elucidating the mechanisms underlying the impact of UBE2D/eff deficiency on the observed phenotypes would contribute to a more comprehensive understanding of the study's implications. For instance, investigating whether the loss of UBE2D/eff influences muscle proteostasis by impeding proteasome assembly or function, modulating autophagy, etc. 

We have previously utilized luciferase assays to measure the proteolytic activity of the proteasome in human cells treated with siRNAs targeting UBE2D1/2/3/4 but found no effect of UBE2D knockdown compared to control nontargeting siRNAs (Hunt et al. 2023; PMID: 37963875). In Drosophila muscles, we have examined the levels of GFP-CL1 (a GFP fused with a proteasomal degron) and found that effRNAi does not impact GFP-CL1 levels (data shown in author response image 1). Overall, these results suggest that effRNAi reduces protein quality control without affecting proteasome activity.

Author response image 1.

(2) Related to Figures 1B-C: It is not clear to this reviewer the quantification methodology used in the experiment. Does each point represent the Average +/- SD for each replicate? If so, it appears that not all cases align with the n=5 as indicated in the figure legend. Additionally, how many animals per replicate were quantified? 

We have amended the legend of Figure 1 to indicate that each data point in the graph represents a single RNAi line targeting the corresponding gene. The mean of 5 biological replicates is shown for each RNAi line, with each biological replicate representing a single eye imaged from a distinct fly. Therefore, the data points that do not show large magnitude changes may indicate RNAi that were not effective at knocking down the target protein (or with no effect on HTT aggregates).  

(3) Related to the previous point: The analysis of pathogenic Huntingtin aggregation in the Materials and Methods section lacks information regarding the number of individuals, replicates, etc. 

Please see the response above.

(4) Related to Figure 1 B: In the case of eff/UBE2D, it appears that 3 out of 9 replicates demonstrate a significant increase in HL-polyQ aggregates. Considering the strength of this result, it raises questions about whether it justifies using eff for future analyses. 

Please see the response to point (2) above. These results indicate that 3 distinct UAS-RNAi lines targeting eff/UBE2D produced the same effect whereas 6 other effRNAi lines did not, possibly because they are less efficacious in knocking down eff/UBE2D. We have now amended the legend of Fig. 1B to better explain these results.

(5) Related to Figure 1 D-E: Could the authors provide clarification regarding the tissue type and animal age utilized in these experiments? 

Whole flies were utilized at 1 week of age.

(6) Related to Figure 3: Incorporating the normal accumulation of poly-ubiquitinated proteins during aging could provide context to better interpret the effect of eff/UBE2D KD at 3 weeks of age. 

Several papers from us and others have previously demonstrated a progressive increase in the insoluble levels of poly-ubiquitinated proteins during aging in Drosophila skeletal muscle (PMID: 36640359; PMID: 31249065; PMID: 33773104; PMID: 33658508; PMID: 24092876; PMID: 21111239; PMID: 24244197; PMID: 25199830; PMID: 28878259; PMID: 36213625). Our analyses now indicate that such age-related loss of protein quality control is accelerated by eff/UBE2D knockdown.

(7) Related to Figure 3: Would it be possible for the authors to include a list or table detailing the specific E2, deubiquitinating enzymes, and E3s identified in the comparative analysis of the old vs young proteome? This would provide a clear reference for the identified regulatory proteins involved in the age-related proteomic changes. 

We have added a tab to Supplementary Table 2 to report the list of age-regulated deubiquitinating enzymes (DUBs) and E1, E2, and E3 enzymes.

(8) Related to Figures 3 and 4: Given that the comparative analysis of the old versus young proteome identified 10 out of 21 E2 ubiquitin-conjugating enzymes, exploring the impact of eff/UBE2D overexpression becomes pivotal to understanding its role in age-related changes in proteostasis and lifespan. Conducting an experiment involving eff overexpression could provide valuable insights into whether restoring eff levels mitigates aging-related phenotypes. 

Although we have not done this experiment with eff overexpression, Fig. 4E reports that the overexpression of human UBE2D2 in skeletal muscle does not appear to influence lifespan by itself (green line in Fig. 4E), although it can partially rescue the short lifespan of flies with muscle-specific effRNAi (purple line in Fig. 4E).

(9) Providing a more detailed description of the Supplementary Tables would significantly enhance the reader's comprehension of their content. 

A description has been added at the end of the methods.

Reviewer #2 (Recommendations For The Authors): 

In addition, to the points listed above: 

- The title does not reflect the content of the manuscript and should be changed. There is no evidence that UBE2D maintains a "youthful" (needs to be changed as well) proteome. Rather, its expression declines with aging and its depletion leads to an increase of ubiquitinated proteins. This is true for essentially the entire proteostasis network. 

While proteostasis generally declines with aging, it is incompletely understood what specific components of the proteostasis network are dysregulated with aging. Our study now identifies the E2 ubiquitin-conjugating enzyme eff/UBE2D as a key regulator of proteostasis that is transcriptionally downregulated with aging. Comparison of the proteomic changes induced by aging versus those induced by effRNAi in young age indicates a partial overlap (Fig. 6), indicating that eff/UBE2D is, at least in part, necessary to maintain the proteome composition that is found in young age (“youthful”). On this basis, we would like to keep the current title but have amended the manuscript to indicate that such regulation of the proteome composition is only in part dependent on eff/UBE2D.

- Molecular weight markers are missing for the gels/western blot depicted in Fig 1E, 2C, 3E, and 4A. 

Thank you for pointing this out, these have been added.

- Fig. 4A, the Ponceau staining for the detergent insoluble samples shows almost no signal for lane 7 and the data should hence not be analyzed. 

The western blot membrane in Fig. 4A shows a reliable signal in all lanes (including lane 7) when probed with antibodies for ubiquitin, Ref(2)P, and tubulin. Therefore, there is no reason for excluding lane 7 from the analysis. Ponceau S staining is provided as an additional loading control but was not used to normalize the data.

Reviewer #3 (Recommendations For The Authors): 

There are a number of confusing or not sufficiently explained points in the figures that require clarity. 

In Figure 1, panels B and C, one assumes the gray broad line across means no difference from control. For the genes, many have points that are scattered both above and below that control line. What do the dots and range represent for each gene, and why are the data so scattered. How do the authors explain data ranging from no effect, to a negative effect to a positive effect, all for the same gene? Akt1 and Hsp83 are controls but are not quantitated to appreciate how variable the assay is. Can they explain the figure better, and also why the data for any one gene are so variable?

We have amended the legend of Figure 1 to indicate that each data point in the graph represents a single RNAi line targeting the corresponding gene. The mean of 5 biological replicates is shown for each RNAi line, with each biological replicate representing a single eye imaged from a distinct fly. Therefore, the data points that do not show large magnitude changes may indicate RNAi lines that were not effective at knocking down the target protein (or that did not affect HTT aggregates). Therefore, the variability in the analysis of a single gene arises because different RNAi lines targeting that gene may have different efficacy. RNAi lines for Akt1 and Hsp83 are merely used as controls (these have been quantified in Jiao et al. 2023; PMID: 36640359).

In Figure 2A, it is not clear which animals have the hL-Q72-GFP (which eyes are "rough eyes"?). Also, do ubc6-RNAi and eff-RNAi have an impact on the normal eye? That is, can they explain the images and genotypes more clearly. 

UBC6 and eff RNAi produce these rough eye phenotypes in the absence of HTT-polyQ and these are rescued by the expression of their human homologs. The panel images indicated in bold here below are those that have “rough eye” phenotypes: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12 (a green R has been added to these panels in Fig. 2A).

In Figure 2B, panel 3 looks very different from 1 and 4 and yet is not different from them by quantitation. Can they replace it with a more representative panel or is 3 lower (but not significantly so)? 

Please note that some of the GFP fluorescence in image 4 is not punctate, but rather diffuse fluorescence that is not related to HTT-GFP aggregates. Our image quantitation methods utilized thresholding to identify GFP-positive puncta while eliminating background fluorescence not corresponding to HTT-GFP puncta.

In Figures 3E and F, it would be helpful in F to put the detergent soluble bar graphs all on the left so that those data are on the left in both E and F, and then detergent-insoluble in E and F to the right. This would make the figure and quantitation easier to follow. 

Done.

The same point as above for Figures 4 A and B. 

Done.

In Figure 3A, CG7656 is nearly as reduced with age as eff. One wonders if that gene would give a different or similarly overlapping proteome with age as eff. Was CG7656 not focused on because not conserved? 

As indicated in Figure 1B, CG7656 is orthologous to UBE2R1 (also called CDC34) and UBE2R2 in humans. In this screen, however, RNAi targeting CG7656 did not appear to influence HTT aggregates and therefore was not selected for further analyses. However, it may play a role in skeletal muscle proteostasis during aging.

In Figure 6, the R2 value correlating age with eff-RNAi is weak. Although they discuss this in the text, it might also be helpful to include Venn diagrams for gene overlaps and the significance to make the argument more clear that there is a significant correlation in proteins up and down to indicate that eff largely recapitulates the changes of aging. Correlating this with proteins that are restored with UBE2D in muscle in a more clear manner may also be helpful for readers interested in aging. 

We have amended the text to indicate that this relatively low correlation (R2\=~0.2, but corresponding to a consistent regulation of 70% of proteins by aging and effRNAi) could indicate that eff/UBE2D is only in part responsible for maintaining a youthful composition of the muscle proteome during aging. Other changes that occur with aging likely account for non-correlated alterations in protein levels. We have also added Venn diagrams (Fig. 6E) to further display the overlap in protein regulation by aging vs. effRNAi.

In Figure 7, they might indicate that the accumulated insoluble protein is ubiquitinated. That is left out of the figure, although indicated in the legend. 

Done.

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public Review):

Summary:

The authors study the variability of patient response of NSCLC patients on immune checkpoint inhibitors using single-cell RNA sequencing in a cohort of 26 patients and 33 samples (primary and metastatic sites), mainly focusing on 11 patients and 14 samples for association analyses, to understand the variability of patient response based on immune cell fractions and tumor cell expression patterns. The authors find immune cell fraction, clonal expansion differences, and tumor expression differences between responders and non-responders. Integrating immune and tumor sources of signal the authors claim to improve prediction of response markedly, albeit in a small cohort.

Strengths:

The problem of studying the tumor microenvironment, as well as the interplay between tumor and immune features is important and interesting and needed to explain the heterogeneity of patient response and be able to predict it.

Extensive analysis of the scRNAseq data with respect to immune and tumor features on different axes of hypothesis relating to immune response and tumor immune evasion using state-of-the-art methods.

The authors provide an interesting scRNAseq data set linked to outcomes data.

Integration of TCRseq to confirm subtype of T-cell annotation and clonality analysis.

Interesting analysis of cell programs/states of the (predicted) tumor cells and characterization thereof.

Weaknesses:

Generally, a very heterogeneous and small cohort where adjustments for confounding are hard. Additionally, there are many tests for association with outcome, where necessary multiple testing adjustments would negate signal and confirmation bias likely, so biological takeaways have to be questioned.

Thank you for your comment. We made multiple testing adjustments as suggested in “Recommendations for Authors.”

RNAseq is heavily influenced by the tissue of origin (both cell type and expression), so the association with the outcome can be confounded. The authors try to argue that lymph node T-cell and NK content are similar, but a quantitative test on that would be helpful.

Following the reviewer’s suggestion, we performed principal component analysis (PCA) to assess the influence of tissue of origin on immune and stromal cell populations. In the revised Figure S1g, we quantified the similarity using Euclidean distances of centroids between sample groups based on their tissue of origin in the PC1 and PC3 plot.

The authors claim a very high "accuracy" performance, however, given the small cohort and lack of information on the exact evaluation it is not clear if this just amounts to overfitting the data.

We acknowledge the concern about the high “accuracy” potentially indicating overfitting. To address this, we revised the manuscript to clarify the use of 'accuracy,' 'AUC,' and 'performance' with clearer expressions in the following sections: Abstract (Line 57), Results (Line 264), Discussion (Lines 320-321), Methods (Lines 546-547), Legends for Figure 5c and Figure S8b.

Especially for tumor cell program/state analysis the specificity to the setting of ICIs is not clear and could be prognostic.

Thank you for your comments. As outlined in the ‘Table 2 in the revised manuscript’, we conducted a multivariate survival analysis of tumor signature candidates using the TCGA lung adenocarcinoma (LUAD, n = 533) and squamous cell carcinoma (LUSC, n = 502) cohorts to evaluate their prognostic potential. No tumor cell programs or states were found to be associated with overall survival in either LUAD or LUSC. We added descriptions related to Table 2 in the Results (Lines 249-251) and Methods (Lines 530-542) section.

Due to the small cohort with a lot of variability, more external validation is needed to be convincingly reproducible, especially when talking about AUC/accuracy of a predictor.

Expanding the cohort size was difficult due to limited resources. We recognize the challenges posed by the small and heterogeneous cohort. We have acknowledged these limitations and applied statistical corrections to address them.

Reviewer #2 (Public Review):

Summary:

The authors have utilised deep profiling methods to generate deeper insights into the features of the TME that drive responsiveness to PD-1 therapy in NSCLC.

Strengths:

The main strengths of this work lie in the methodology of integrating single-cell sequencing, genetic data, and TCRseq data to generate hypotheses regarding determinants of IO responsiveness.

Some of the findings in this study are not surprising and well precedented eg. association of Treg, STAT3, and NFkB with ICI resistance and CD8+ activation in ICI responders and thus act as an additional dataset to add weight to this prior body of evidence. Whilst the role of Th17 in PD-1 resistance has been previously reported (eg. Cancer Immunol Immunother 2023 Apr;72(4):1047-1058, Cancer Immunol Immunother 2024 Feb 13;73(3):47, Nat Commun. 2021; 12: 2606 ) these studies have used non-clinical models or peripheral blood readouts. Here the authors have supplemented current knowledge by characterization of the TME of the tumor itself.

Weaknesses:

Unfortunately, the study is hampered by the small sample size and heterogeneous population and whilst the authors have attempted to bring in an additional dataset to demonstrate the robustness of their approach, the small sample size has limited their ability to draw statistically supported conclusions. There is also limited validation of signatures/methods in independent cohorts, no functional characterization of the findings, and the discussion section does not include discussion around the relevance/interpretation of key findings that were highlighted in the abstract (eg. role of Th17, TRM, STAT3, and NFKb). Because of these factors, this work (as it stands) does have value to the field but will likely have a relatively low overall impact.

We acknowledge the challenges posed by the small and heterogeneous cohort. To address this, we tempered our claims related to accuracy by applying statistical testing corrections. We also appreciate the feedback on functional characterization and have expanded the discussion in the revised manuscript to include an overview of specific cell populations and genes.

Related to the absence of discussion around prior TRM findings, the association between TRM involvement in response to IO therapy in this manuscript is counter to what has been previously demonstrated (Cell Rep Med. 2020;1(7):100127, Nat Immunol. 2017;18(8):940-950., J Immunol. 2015;194(7):3475-3486.). However, it should be noted that the authors in this manuscript chose to employ alternative markers of TRM characterisation when defining their clusters and this could indicate a potential rationale for differences in these findings. TRM population is generally characterised through the inclusion of the classical TRM markers CD69 (tissue retention marker) and CD103 (TCR experienced integrin that supports epithelial adhesion), which are both absent from the TRM definition in this study. Additional markers often used are CD44, CXCR6, and CD49a, of which only CXCR6 has been included by the authors. Conversely, the majority of markers used by the authors in the cell type clustering are not specific to TRM (eg. CD6, which is included in the TRM cluster but is expressed at its lowest in cluster 3 which the authors have highlighted as the CD8+ TRM population). Therefore, whilst there is an interesting finding of this particular cell cluster being associated with resistance to ICI, its annotation as a TRM cluster should be interpreted with caution.

Single-cell RNA sequencing (scRNA-seq) can sometimes fail to detect the expression of classical cell type markers due to incomplete capture of a cell’s transcriptome. To determine cell identity, we utilized cell type markers established in previous scRNA-seq studies. In response to your comments, we have added the expression levels of classical TRM markers, including CD69, CD103 (ITGAE), CD44, CXCR6, and CD49a (ITGA1), in the revised Figure 2c. Although these markers were not exclusively expressed in TRM clusters, TRM clusters exhibited relatively high levels of these genes while lacking other clusters’ specific marker genes.

Reviewer #1 (Recommendations For The Authors):

General suggestions:

When analyzing the association of cell type proportions with outcomes, some adjustment for multiple testing should be considered (either sampling-based, e.g. permutation test, or adjustment based on assumptions of independence of tests, e.g. Bonferroni).

Thank you for your comments. As suggested, we calculated the adjusted p-value using the False Discovery Rate for the association of cell type proportions with outcomes in Figure 3a. The heatmap in Reviewer's ONLY Figure 1, using the adjusted p-value consistently showed the expected grouping of cell types and outcomes. However, the significance did not meet the conventional statistical cutoff criteria. We acknowledge this limitation, which results from statistical testing based on ratio values.

Author response image 1.

Heat map with unsupervised hierarchical clustering of proportional changes in cell subtypes within total immune cells. Proportional changes were compared across multiple ICI response groups. The color represents the adjusted -log (p-value) calculated using the False Discovery Rate.

A formal test of clonotype differences (normalized to cell type fraction) would be great as the shown plot 2e could be confounded by cell number and type differences between responders and non-responders.

Thank you for your suggestion. We have revised Figure 2e to display the relative clonotype differences versus CD4+ and CD8+ T cell fractions in each sample. The relative clone size of each cell was calculated by dividing the size of each clone by the total number of CD4+ or CD8+ T cells, respectively.

It could be made a bit more clear when the core group of patients was used (only when associating with outcomes?) and when all other patients were used as well (only cell type annotation?).

As the reviewer correctly noted, we performed scRNA-seq analysis on all specimens, but only the core group of patients was used for the comparative analysis between the responder and non-responder groups. This information has been detailed in the manuscript (Lines 103-105).

For immune cells, it would be interesting to look at expression patterns (NMF, scINSIGHT) as well, not just immune cell fractions and expansion.

In contrast to tumor signatures, immune cell programs are more directly tied to their functional characteristics. Therefore, we focused on annotating immune cells based on their functional properties and conducted comparative analyses between responders and non-responders.

Multiple testing is necessary for the univariate association analysis. Some adjustments for confounders in a multivariate model (despite the size) could be informative.

As shown in ‘Reviewer's ONLY Table 1’, we conducted a multivariate regression analysis of immune and tumor signatures for ICI response, adjusting for clinical variables such as tissue origin, cancer subtype, pathological stage, and smoking status. However, the results were not significant, likely due to the heterogeneity and small size of the cohort.

Author response table 1.

P-values from univariate and multivariate regression analysis of immune and tumor signatures for ICI response.

It is not clear from the manuscript how "accuracy" is measured. The terms "accuracy" and "AUC", as well as "performance" are used interchangeably, a section in the methods with the precise definition is needed.

We have revised the manuscript to clarify the terms 'accuracy,' 'AUC,' and 'performance' by using clearer expressions in the following sections: Abstract (Line 57), Results (Line 264), Discussion (Lines 320-321), Methods (Lines 546-547), Legends for Figure 5c and Figure S8b.

Furthermore, it has to be clear if this is in-sample performance or if there was some train/test split or cross-validation used. Given the small cohort size and wealth of features finding some combination of predictors that could overfit on responders/non-responders would not be surprising.

As the reviewer has noted, we acknowledge the statistical limitations due to the small cohort size. We have revised the sentence on Lines 545-547 “Classification models of responders and non-responders for PC signatures and combinatorial indexes between tumor and/or immune cells were generated based on in-sample performance…”.

Suggestions to improve readability:

Line 84: The sentence should be reformulated to improve understanding.

We have revised sentences in lines 81-93.

Line 86: missing a "the".

We have revised the sentences in lines 81-93.

Reviewer #2 (Recommendations For The Authors):

"Tumor-infiltrating PD-1 positive T cells have higher capacity of tumor recognition than PD-1 negative T cells" Please look to rephrase this sentence as this is not entirely accurate: PD-1 is upregulated in tumor-experienced T cells as a consequence of antigen recognition ie those cells that recognise tumor will increase PD-1, whereas the sentence as it's currently written indicates that PD1+ cells have an intrinsically increased capacity to kill tumors, which is incorrect.

We have revised the sentence “Tumor-infiltrating PD-1 positive T cells have higher capacity of tumor recognition than PD-1 negative T cells” in lines 86-88 as “More specifically, PD-1 expression is upregulated upon antigen recognition (PMID29296515), indicating that certain T cells in the tumor microenvironment are actively engaged as tumor-specific T cells.” in the revised manuscript.

Cancer subtype abbreviations (eg. SQ, ADC, NUT) are used in figures in the main article and so should be defined in the main text (they are currently only explained in the legend for the supplementary table).

As per the reviewer’s suggestion, the manuscript has been revised to include definitions of cancer type abbreviations in lines 108-110.

Figure S1d-f does not appear to corroborate the statement that "Although there were differences in tissue-specific resident populations, we found that the immune cell profiles, especially T/NK cells of mLN were similar to those of primary tumor tissues indicating the activation of immune responses were 118 consistently observed at metastatic sites (Figure S1d-f)." The diagrams are complex (please explain all abbreviations) and it is not clear how the authors have come to this conclusion. Additionally, cell quantity does not indicate that the 'activation of immune responses' is consistently observed at metastatic sites as these cells could be dysfunctional/bystander.

In the revision, we have quantified the diagrams (Figure S1f) to more clearly highlight the differences in tissue-specific resident populations. We performed principal component analysis (PCA) to evaluate the impact of tissue origin on immune and stromal cell populations. In the revised Figure S1g, we illustrated the quantitative similarity between sample groups using Euclidean distances in the PC plot based on their tissue of origin. Additionally, the legends for Figures S1d and S1e have been updated to include definitions for all abbreviations.

We agree with the reviewer's comment that cell quantity alone may not fully reflect activation of antigen-specific immune responses, even though we annotated the functional T cell subtypes. To better focus on the comparisons of cellular profiles between metastatic sites (mLN) and primary tumors (tLung and tL/B), we removed the sentence “…indicating the activation of immune responses were consistently observed at metastatic sites (Fig. S1d-f).” from the revised manuscript.

In Figure 2c, classical markers for TRM (CD103, CD69) should be included in the description for the definition of the TRM clusters, or their exclusion appropriately explained. The findings regarding the negative correlation between follicular B cells and ICI response are surprising. Figure S3, the cluster identified as Follicular B cells contains MS4A1 (CD20) and HLA-DRA. Classical markers are CD20 (pan-B cell), CD21 (CR2), CD23, and IgD/IgM (double positive), and as such it is not clear if the authors have appropriately annotated this cluster as representing follicular B cells. These classical markers should be included in the interpretation of the cell clustering or their exclusion appropriately explained.

We appreciate your comments. In response, we have added the expression levels of classical TRM markers such as CD69, CD103 (ITGAE), CD44, CXCR6, and CD49a (ITGA1), in the revised Figure 2c. Additionally, we revised the dot plot showing the mean expression of marker genes in each cell cluster for B/Plasma cells (revised Figure S3b) by incorporating classical markers for Follicular B cells, such as CD21 (CR2), CD23 (FCER2), IgD (IGHD), IgM (IGHM).

Figure 2f is rather confusing for the reader. I would recommend changing to an alternative plot that shows logP and response in a different way. If keeping to this plot type please clarify why plotting response vs PD, and whether the lower left quadrant indicates patients with progressive disease and the top right indicates responders as the interpretation is not clear currently.

Thank you for your feedback. To address the concerns raised, we have updated the figure legend for Figure 2f to clarify the interpretation of the quadrants: “The lower left quadrant shows cell types overrepresented in the poor responder groups, while the upper right quadrant indicates cell types overrepresented in the better responder groups”. This clarification aims to help readers understand that the lower left quadrant reflects cell types associated with worse treatment outcomes, while the upper right quadrant reflects cell types associated with improved therapeutic responses.

The terms "PC7.neg, INT.down, and UNION.down" are included in the results with no explanation to the reader of what they are or how to interpret them. The methods description "We constructed DEGs with 470 intersections (INT) and union (UNION) of up- or down-regulated genes for comparisons" does not sufficiently describe how they were generated/calculated and, therefore, this is difficult for the reader to interpret in the final results section. Please add an additional explanation for the reader in the final section of the results/Figure 5 and in the methods.

Following the reviewer’s suggestion, we added additional explanation in the Results section (lines 258-261): “PC7.neg denotes genes negatively correlated with PC7, a principal component extracted from PCA that distinguishes tumor cells in poor response groups. INT.down and UNION.down represent the intersection and union of down-regulated genes in the responder group, respectively.”. We also explained the details in the Methods section (lines 489-495): “We reconstructed DEGs as four groups: INT.up, INT.down, UNION,up, and UNION.down, based on with the intersection (INT) and union (UNION) of up- or down-regulated genes for pairwise comparisons between responder versus non-responder, PR versus PD, and PR versus SD. INT.up and INT.down represent the intersection of up- and down-regulated genes in the responder group, respectively. UNION.up and UNION.down represent the union of up- and down-regulated genes in the responder group, respectively.”

The TRM and Th17+ T cell populations are highlighted in the abstract as being related to ICI resistance, but these populations of cells are not even mentioned in the discussion. Likewise, STAT3 and NFkb pathways are also highlighted in the abstract but absent in the discussion section. Please discuss the relevance of these findings, particularly given the prior studies demonstrating the opposite impact of TRM populations in NSCLC.

We have expanded the discussion in the revised manuscript (Lines 295-313) to address the roles of TRM and Th17+ T cell, as well as the STAT3 and NF-κB pathways, in association with ICI resistance in NSCLC.

“The identification of an abundance of CD4+ TRM cells as a negative predictor of ICI response is an unexpected finding, considering that higher frequencies of TRM cells in lung tumor tissues are generally associated with better clinical outcomes in NSCLC (PMID28628092). This is largely due to their role in sustaining high densities of tumor-infiltrating lymphocytes and promoting anti-tumor responses. Additionally, previous studies have demonstrated that TRM cell subsets coexpressing PD-1 and TIM-3 are relatively enriched in patients who respond to PD-1 inhibitors (PMID31227543). However, recent findings suggest that pre-existing TRM-like cells in lung cancer may promote immune evasion mechanisms, contributing to resistance to immune checkpoint blockade therapies (PMID37086716). These observations suggest that the roles of TRM subsets in tumor immunity are highly context-dependent.

Similarly, CD4+ TH17 cells, which were overrepresented in the non-responder groups, exhibit context-dependent roles in tumor immunity and may be associated with both unfavorable and favorable outcomes (PMID34733609; PMID30941641). In exploring tumor cell signatures linked to ICI response, non-responder attributes were regulated by STAT3 and NFKB1. The STAT3 and NF-κB pathways are crucial for Th17 cell differentiation and T cell activation (PMID24605076; PMID32697822). Notably, STAT3 activation in lung cancer orchestrates immunosuppressive characteristics by inhibiting T-cell mediated cytotoxicity (PMID31848193). The combined influence of the Th17/STAT3 axis and TRM cell activity in predicting ICI response underscores the complexity of these pathways and suggests that their roles in tumor immunity and therapy response warrants further investigation.”

Author Response

The following is the authors’ response to the original reviews.

Response to reviewer’s comments

Reviewer #1 (Public Review):

In this study, the structural characteristics of plant AlaDC and SerDC were analyzed to understand the mechanism of functional differentiation, deepen the understanding of substrate specificity and catalytic activity evolution, and explore effective ways to improve the initial efficiency of theanine synthesis.

On the basis of previous solid work, the authors successfully obtained the X-ray crystal structures of the precursors of theanine synthesis-CsAlaDC and AtSerDC, which are key proteins related to ethylamine synthesis, and found a unique zinc finger structure on these two crystal structures that are not found in other Group II PLP-dependent amino acid decarboxylases. Through a series of experiments, it is pointed out that this characteristic zinc finger motif may be the key to the folding of CsAlaDC and AtSerDC proteins, and this discovery is novel and prospective in the study of theine synthesis.

In addition, the authors identified Phe106 of CsAlaDC and Tyr111 of AtSerDC as key sites of substrate specificity by comparing substrate binding regions and identified amino acids that inhibit catalytic activity through mutation screening based on protein structure. It was found that the catalytic activity of CsAlaDCL110F/P114A was 2.3 times higher than that of CsAlaDC. At the same time, CsAlaDC and AtSerDC substrate recognition key motifs were used to carry out evolutionary analysis of the protein sequences that are highly homologous to CsAlaDC in embryos, and 13 potential alanine decarboxylases were found, which laid a solid foundation for subsequent studies related to theanine synthesis.

In general, this study has a solid foundation, the whole research idea is clear, the experimental design is reasonable, and the experimental results provide strong evidence for the author's point of view. Through a large number of experiments, the key links in the theanine synthesis pathway are deeply studied, and an effective way to improve the initial efficiency of theanine synthesis is found, and the molecular mechanism of this way is expounded. The whole study has good novelty and prospectivity, and sheds light on a new direction for the efficient industrial synthesis of theanine

Response: Thank you very much for taking time to review this manuscript. We appreciate all your insightful comments and constructive suggestions.

Reviewer #1 (Recommendations For The Authors):

(1) If some test methods are not original, references or method basis should be indicated.

Response: Thank you very much for your careful reading of the manuscript and valuable suggestions. We have added references for the enzymatic activity experiments performed to measure the synthesis of theanine in the revised manuscript.

(2) The conclusion is a little lengthy, and the summary of the whole study is not well condensed.

Response: Thank you very much for your valuable suggestions. We have refined the conclusion in the revised manuscript, and it is as follows:

In conclusion, our structural and functional analyses have significantly advanced understanding of the substrate-specific activities of alanine and serine decarboxylases, typified by CsAlaDC and AtSerDC. Critical amino acid residues responsible for substrate selection were identified—Tyr111 in AtSerDC and Phe106 in CsAlaDC—highlighting pivotal roles in enzyme specificity. The engineered CsAlaDC mutant (L110F/P114A) not only displayed enhanced catalytic efficiency but also substantially improved L-theanine yield in a synthetic biosynthesis setup with PsGS or GMAS. Our research expanded the repertoire of potential alanine decarboxylases through the discovery of 13 homologous enzyme candidates across embryophytic species and uncovered a special motif present in serine protease-like proteins within Fabale, suggesting a potential divergence in substrate specificity and catalytic functions. These insights lay the groundwork for the development of industrial biocatalytic processes, promising to elevate the production of L-theanine and supporting innovation within the tea industry.

Reviewer #2 (Public Review)

Summary:

The manuscript focuses on the comparison of two PLP-dependent enzyme classes that perform amino acyl decarboxylations. The goal of the work is to understand the substrate specificity and factors that influence the catalytic rate in an enzyme linked to theanine production in tea plants.

Strengths:

The work includes x-ray crystal structures of modest resolution of the enzymes of interest. These structures provide the basis for the design of mutagenesis experiments to test hypotheses about substrate specificity and the factors that control catalytic rate. These ideas are tested via mutagenesis and activity assays, in some cases both in vitro and in plants.

Weaknesses:

The manuscript could be more clear in explaining the contents of the x-ray structures and how the complexes studied relate to the reactant and product complexes. The structure and mechanism section would also be strengthened by including a diagram of the reaction mechanism and including context about reactivity. As it stands, much of the structural results section consists of lists of amino acids interacting with certain ligands without any explanation of why these interactions are important or the role they play in catalysis. The experiments testing the function of a novel Zn(II)-binding domain also have serious flaws. I don't think anything can be said at this point about the function of the Zn(II) due to a lack of key controls and problems with experimental design.

Response: Thank you very much for your thoughtful comments and feedback on our manuscript. We are pleased to hear that the work's strengths, such as the X-ray crystal structures and the mutagenesis experiments tied to the catalytic rate and substrate specificity, align with the goals of our research.

We recognize the areas identified for improvement and appreciate the suggestions provided. We have emphasized how we use the structural information obtained to infer the roles of key amino acid residues in the reaction. Additionally, we have added a diagram of the reaction mechanism in the Supplementary figure to provide clearer context on reactivity and improve the overall understanding of the catalytic process. Regarding the structural results section, we have included a discussion that contextualizes the list of amino acids and their interactions with the ligands by explaining their significance and roles in catalysis. We acknowledge the weaknesses you've pointed out in the experiments concerning the novel Zn(II)-binding domain, but we would like to clarify that the focus of our study was not primarily on the zinc structure. While we agree that there may be limitations in the experimental design and controls for the zinc binding domain, we believe that these flaws do not significantly impact the overall findings of the study. The experiment served as a preliminary exploration of the potential functionality of the domain, and further studies are required to fully understand its role and mechanism.

Reviewer #2 (Recommendations For The Authors):

(1) In addition to the points raised in the public review, it would be ideal to provide some context for the enzymatic characterization. Why are the differences in kinetic parameters for AlaDC and SerDC significant?

Response: Thank you for your comments and suggestions. The Km values for CsAlaDC and SerDCs are comparable, suggesting similar substrate affinities. However, CsAlaDC exhibits a significantly lower Vmax compared to AtSerDC and CsSerDC. This discrepancy implies that CsAlaDC and SerDCs may differ in the rates at which they convert substrate to product when saturated with substrate. SerDCs may have a faster turnover rate, meaning they convert substrate to product and release the enzyme more quickly, resulting in a higher Vmax. Differences in the stability or correct folding of the enzymes under assay conditions can also affect their Vmax. If SerDCs are more stable, they might maintain their catalytic activity better at higher substrate concentrations, contributing to a higher Vmax. We have added these to the part of “Enzymatic properties of CsAlaDC, AtSerDC, and CsSerDC” in our revised manuscript.

(2) Why is Phe106/Tyr111 pair critical for substrate specificity? Does the amino acid contact the side chain? It might be helpful to a reader to formulate a hypothesis for this interaction.

Response: Thank you for the question and comments. We conducted a comparison between the active sites of CsAlaDC and AtSerDC and observed a distinct difference in only two amino acids: F106 in CsAlaDC and Y111 in AtSerDC. The remaining amino acids were found to be identical. Expanding on previous research concerning Group II PLP-dependent amino acid decarboxylases, it was postulated and subsequently confirmed that these specific amino acids play a crucial role in substrate recognition. However, since we lack the structure of the enzyme-substrate complex, we are unable to elucidate the precise interactions occurring between the substrate and the amino acids at this particular site based solely on structural information.

(3) Line 55 - Define EA again.

Response: Thank you very much for your careful reading of the manuscript and valuable suggestions. We have redefined “EA” as the abbreviation for ethylamine in the revised manuscript.

(4) Line 58 - The meaning of "determined by the quality formation of tea" is not clear.

Response: Thank you very much for your careful reading of the manuscript and valuable suggestions. We have modified it in the revised manuscript.

(5) Line 65 - Missing words between "despite they".

Response: Thank you very much for your careful reading of the manuscript. We have corrected it in the revised manuscript.

(6) Line 67 - Need a reference for the statement about lower activity?

Response: Thank you for the question and comments. We have provided the following reference to support this statement in the revised manuscript.

Reference: Bai, P. et al. (2021) Biochemical characterization of specific Alanine Decarboxylase (ADC) and its ancestral enzyme Serine Decarboxylase (SDC) in tea plants (Camellia sinensis). BMC Biotechnol. 21,17.

(7) Line 100-101 - The meaning of "its closer relationship was Dicots plants." is not clear.

Response: We have revised the sentence in the revised manuscript, as follows: “Phylogenetic analysis indicated that CsAlaDC is homologous with SerDCs in Dicots plants.”

(8) Line 139 - Missing a word between "as well as" and "of".

Response: Thank you very much for your careful reading of the manuscript and valuable suggestions. We have corrected it in the revised manuscript.

(9) Line 142 - The usage of comprised here is not correct. It would be more correct to say "The overall architecture of CsAlaDC and AtSerDC is homodimeric with the two subunits...".

Response: Thank you very much for your careful reading of the manuscript and valuable suggestions. We have corrected it in the revised manuscript.

(10) Line 148-149 - I didn't understand the statement about the "N-terminal structures" Are these structures obtained from protein samples that have a truncated N-terminus?

Response: Group II PLP-dependent amino acid decarboxylases are comprised of three distinct structural domains: the N-terminal domain, the large domain, and the C-terminal domain. Each of these domains possesses unique structural features. Similarly, CsAlaDC and AtSerDC can also be classified into three structural domains based on their specific characteristics. To achieve more stable proteins for further experiments, we conducted truncation on both of these proteins. The truncated section pertains to a subsection of the N-terminal domain and is truncated from the protein's N-terminus.

(11) Line 153 - Say "is composed of" instead of "composes of".

Response: Thank you very much for your careful reading of the manuscript and valuable suggestions. We have corrected it in the revised manuscript.

(12) Line 156 - I didn't understand the statement about the cofactor binding process. What is the cofactor observed? And how can we say anything about the binding process from a single static structure of the enzyme? It might be better to say that the cofactor binding site is located at the subunit junction - but the identity of the cofactor still needs to be defined first.

Response: Thank you for your comments and suggestions. The cofactor mentioned here is PLP. We aim to elucidate the binding state of PLP at the active site, excluding the binding process. The description has been revised in the revised manuscript.

(13) Lines 157-158 - I didn't understand the conclusion about the roles of each monomer. In the images in Figure 3 - both monomers appear to bind PLP but the substrate is not present - so it's not clear how conclusions can be drawn about differential substrate binding in the two subunits.

Response: Thank you very much for your careful reading and valuable suggestions. The main idea we want to convey is that this protein possesses two active sites. At each active site, the two monomers carry out distinct functions. Of course, our previous conclusion is inaccurate due to the non-existence of the substrate. So, we have made the necessary amendments in the revised manuscript.

(14) Line 161 - I would say loop instead of ring.

Response: Thank you very much for your careful reading of the manuscript and valuable suggestions. We have corrected it in the revised manuscript.

(15) Line 165 - Please provide some references for this statement. It would also be ideal to state the proximity of the Zn-binding motif to the active site or otherwise provide some information about the role of the motif based on its location.

Response: Thank you for your comments and suggestions. We have provided the following references to support this statement in the revised manuscript.

Author response image 1.

(A) Structure of histidine decarboxylase. (B) Structure of glutamate decarboxylase.

Reference:

30 Komori, H. et al. (2012) Structural study reveals that Ser-354 determines substrate specificity on human Histidine Decarboxylase. J Biol Chem. 287, 29175-83.

31 Huang, J. et al. (2018) Lactobacillus brevis CGMCC 1306 glutamate decarboxylase: Crystal structure and functional analysis. Biochem Biophys Res Co. 503, 1703-1709

In CsAlaDC, the zinc is positioned at a distance of 29.6 Å from the active center, whereas in AtSerDC, the zinc is situated 29 Å away from the active center. Hence, we hypothesize that this structure does not impact the enzyme's catalytic activity but might be correlated with its stability.

(16) Lines 166-178 - This paragraph appears to be a list of all of the interactions between the protein, PLP, and the EA product. It would be ideal to provide some text to explain why these interactions are important and what we can learn from them.

Response: Thank you very much for your careful reading of the manuscript and valuable suggestions. We have been conducting additional analysis on the functional roles of amino acid residues involved in the interaction between the active site and PLP. This analysis focuses on aiding PLP binding, determining its orientation, and understanding enzyme catalytic mechanisms. These details are mentioned in the revised manuscript.

(17) Line 192 - Bond not bound.

Response: Thank you very much for your careful reading of the manuscript and valuable suggestions. We have made corrections in the revised manuscript.

(18) Lines 201-207 - It would be ideal to verify that the inclusion of 5 mM DTT affects Zn binding. It's not clear to me that this reagent would necessarily disrupt Zn binding. Under certain circumstances, it could instead promote Zn association. For example, if the Cys ligands are oxidized initially but then become reduced? I don't think the current experiment really provides any insight into the role of the Zn.

Response: Thank you for your valuable insights regarding the role of DTT and its potential effects on Zn binding in our experiments. The main function of DTT is to protect or restore the reduced state of proteins and other biological molecules, particularly by disrupting the crosslinking formed by thiol (-SH) groups and disulfide bonds to maintain the function and structure of proteins. Therefore, the reason for DTT's inhibition of enzyme activity is unknown, and we cannot provide a reasonable explanation for this phenomenon. As a result, we have removed the section discussing the inhibition of enzyme activity by DTT in our revised manuscript.

Reviewer #3 (Public Review):

In the manuscript titled "Structure and Evolution of Alanine/Serine Decarboxylases and the Engineering of Theanine Production," Wang et al. solved and compared the crystal structures of Alanine Decarboxylase (AlaDC) from Camellia sinensis and Serine Decarboxylase (SerDC) from Arabidopsis thaliana. Based on this structural information, the authors conducted both in vitro and in vivo functional studies to compare enzyme activities using site-directed mutagenesis and subsequent evolutionary analyses. This research has the potential to enhance our understanding of amino acid decarboxylase evolution and the biosynthetic pathway of the plant-specialized metabolite theanine, as well as to further its potential applications in the tea industry. Response: Thank you very much for taking the time to review this manuscript. We appreciate all your insightful comments.

Reviewer #3 (Recommendations For The Authors):

Page 6, Figure 2, Page 23 (Methods)

"The supernatants were purified with a Ni-Agarose resin column followed by size-exclusion chromatography."

What kind of SEC column did the authors use? Can the authors provide the SEC elution profile comparison results and size standard curve?

Response: We use a Superdex 200 (Hiload 16/600) column for size exclusion chromatography. The comparison results of SEC elution profiles for AtSerDC and CsAlaDC, along with the standard curve of SEC column, are presented below.

Author response image 2.

(A) Comparison of elution profiles of CsAlaDC and AtSerDC. (B) Elution profile of Blue Dextron 2000. (C) Elution profile of mixed protein (Aldolase, 158000 Da,71.765ml; Conalbumin, 75000 Da,79.391ml; Ovalbumin, 44000 Da,83.767ml; Carbonic anhydrase, 29000 Da,90.019ml; Ribonuclease A, 13700 Da,98.145ml). (D) Size standard curves of Superdex 200 (Hiload 16/600) column.

Page 6 & Page 24 (Methods)

"The 100 μL reaction mixture, containing 20 mM substrate (Ala or Ser), 100 mM potassium phosphate, 0.1 mM PLP, and 0.025 mM purified enzyme, was prepared and incubated at standard conditions (45 ℃ and pH 8.0 for CsAlaDC, 40 ℃ and pH 8.0 for AtSerDC for 30 min)."

(1) The enzymatic activities of CsAldDC and AtSerDC were measured at two different temperatures (45 and 40 ℃, but their activities were directly compared. Is there a reason for experimenting at different temperatures?

Response: We determined that the optimal reaction temperature for AtSerDC is 40°C and for CsAlaDC is 45°C through our verification process. Consequently, all subsequent experiments were performed at these specific temperatures.

Author response image 3.

(A) Relative activity of CsAlaDC at different temperatures. (B) Relative activity of AtSerDC at different temperatures.

(2) Enzyme activities were measured at temperatures above 40℃, which is not a physiologically relevant temperature and may affect the stability or activity of the proteins. At the very least, the authors should provide temperature-dependent protein stability data (e.g., CD spectra analysis) or, if possible, temperature-dependent enzyme activities, to show that their experimental conditions are suitable for studying the activities of these enzymes.

Response: Thank you very much for your careful reading. We have already validated that the experimental temperature we used did not significantly affect the stability of the protein before experimenting. The results are shown in the figure below:

Author response image 4.

Place the two proteins individually into water baths set at temperatures of 25°C, 37°C, 45°C, 60°C, and 80°C for 15 minutes. Subsequently, carry out enzymatic reactions utilizing a standard reaction system, with untreated enzymes serving as the experimental control within the said system. The experimental results suggest that the temperature at which we experimented does not have a significant impact on the stability of the enzyme.

(3) The authors used 20 mM of substrate. What are the physiological concentrations of alanine and serine typically found in plants?

Response: The content of alanine in tea plant roots ranges from 0.28 to 4.18 mg/g DW (Yu et al., 2021; Cheng et al., 2017). Correspondingly, the physiological concentration of alanine is 3.14 mM to 46.92 mM, in tea plant roots. The content of serine in plants ranges from 0.014 to 17.6 mg/g DW (Kumar et al., 2017). Correspondingly, the physiological concentration of serine is 0.13 mM to 167.48 mM in plants. In this study, the substrate concentration of 20 mM was close to the actual concentrations of alanine and serine in plants.

Yu, Y. et al. (2021) Glutamine synthetases play a vital role in high accumulation of theanine in tender shoots of albino tea germplasm "Huabai 1". J. Agric. Food Chem. 69 (46),13904-13915.

Cheng, S. et al. (2017) Studies on the biochemical formation pathway of the amino acid L-theanine in tea (Camellia sinensis) and other plants.” J. Agric. Food Chem. 65 (33), 7210-7216.

Kumar, V. et al. (2017) Differential distribution of amino acids in plants. Amino Acids. 49(5), 821-869.

Pages 6-7 & Table 1

(1) Use the correct notation for Km and Vmax. Also, the authors show kinetic parameters and use multiple units (e.g., mmol/L or mM for Km).

Response: Thank you very much for your careful reading of the manuscript and valuable suggestions. We have corrected this in the revised manuscript.

(2) When comparing the catalytic efficiency of enzymes, kcat/Km (or Vmax/Km) is generally used. The authors present a comparison of catalytic activity from results to conclusion. A clarification of what results are being compared is needed.

Response: Thank you for your comments and suggestions. The catalytic activity is assessed by comparing reaction rates.

Page 7 & Figure 3

In Figure 3A, the authors describe the overall structure, but a simple explanation or labeling within the figure should be added.

Response: Thank you very much for your suggestions, we have made modifications to Figure 3A as follows:

Author response image 5.

Crystal structures of CsAlaDC and AtSerDC. (A) Dimer structure of CsAlaDC. The color display of the N-terminal domain, large domain, and C-terminal domains of chain A is shown in light pink, khaki and sky blue, respectively. Chain B is shown in spring green. The PLP molecule is shown as a sphere model. The zinc finger structure at the C-terminus of CsAlaDC is indicated by the red box. The gray spheres represent zinc ions, while the red dotted line depicts the coordination bonds formed by zinc ions with cysteine and histidine.

Figures 3F & 4A

In these figures, the two structures are overlaid and compared, but the colors are very similar to see the differences. The authors should use a different color scheme.

Response: Thank you very much for your suggestions, we have made modifications to the Figure 3F & 4A as follows:

Author response image 6.

(Figure 3F) - The monomers of CsAlaDC and AtSerDC are superimposed. CsAlaDC is depicted in spring green, while AtSerDC is shown in plum. The conserved amino acid catalytic ring is indicated by the red box. (Figure 4A) - Superposition of substrate binding pocket amino acid residues in CsAlaDC and AtSerDC. The amino acid residues of CsAlaDC are shown in spring green, the amino acid residues of AtSerDC are shown in plum, with the substrate specificity-related amino acid residue highlighted in a red ellipse.

Pages 7 & 8

Figures 3 and 4 do not include illustrations of what the authors describe in the text. The reader will not be able to understand the descriptions until they download and view the structures themselves. The authors should create additional figures to make it easier for readers to understand the structures.

Response: Thank you very much for your suggestions, we have included supplementary figure 1 in the revised manuscript, which presents more elaborate structural depictions of the two proteins.

Pages 9 & 10

"This result suggested this Tyr is required for the catalytic activity of CsAlaDC and AtSerDC."

The author's results are interesting, but it is recommended to perform the experiments in a specific order. First, experiments should determine whether mutagenesis affects the protein's stability (e.g., CD, as discussed earlier), and second, whether mutagenesis affects ligand binding (e.g., ITC, SPR, etc.), before describing how site-directed mutagenesis alters enzyme activity. In particular, the authors' hypothesis would be much more convincing if they could show that the ligand binding affinity is similar between WT and mutants.

Response: Thank you for your insightful feedback on our manuscript, which we greatly appreciate. Your suggestion to methodically sequence the experiments provides a clear pathway to bolster the strength and conclusiveness of our results.

We agree that it is crucial to first assess the stability of the mutant proteins, as changes therein could inadvertently affect catalytic activity. To this end, we have employed circular dichroism (CD) to study the potential structural alterations in the proteins induced by mutations. The experimental results are shown in the following figure:

Author response image 7.

(A) Circular Dichroism Spectra of CsAlaDC (WT). (B) Circular Dichroism Spectra of CsAlaDC (Y336F). (C) Circular Dichroism Spectra of CD of AtSerDC (WT). (D) Circular Dichroism Spectra of AtSerDC (Y341F).

The experimental results indicate that the secondary structure of the mutant proteins remains unchanged, which means the mutations do not alter the protein's stability.

The ligand PLP forms a Schiff base structure with the ε-amino group of a lysine residue in the protein, with maximum absorbance around 420-430 nm. Since we have already added PLP during the protein purification process, as long as the absorbance of mutant proteins and wild-type proteins is the same at 420-430 nm at equivalent concentrations, it indicates that the mutant proteins do not affect the binding of the ligand PLP. Therefore, we scanned the UV-visible absorption spectra of both the wild-type and mutant proteins, and the results are as presented in the following figure:

Author response image 8.

(A) UV-Visible Absorption Spectra of CsAlaDC (WT) compared to CsAlaDC (Y336F). (B) UV-Visible Absorption Spectra of AtSerDC (WT) compared to AtSerDC (Y341F).

The mutant protein and the wild-type protein exhibit similar absorbance at 420-430 nm, indicating that the mutation does not affect the binding of PLP to the protein.

The above experiments have confirmed that the mutations do not significantly affect the stability of the protein or the affinity for the ligand, so we can more confidently attribute changes in enzyme activity to the specific role of the tyrosine residue in question. We believe this comprehensive approach will substantiate our hypothesis and illustrate the necessity of this Tyr residue for the catalytic activity of CsAlaDC and AtSerDC enzymes.

Figure 3

In the 3D structure figure provided by the authors, the proposed reaction mechanism of the enzyme and the involved amino acids are not included. Can the authors add a supplementary figure with a schematic drawing that includes more information, such as distances?

Response: Thank you for your valuable feedback on our manuscript. We completely agree that a schematic drawing with additional details, including distances, would enhance the clarity and understanding of the enzymatic mechanism. In response to your suggestion, we have added a supplementary figure 2 in the revised manuscript that accurately illustrates the proposed reaction pathway, highlighting the key amino acids involved.

Page 10

"The results showed that 5 mM L-DTT reduced the relative activity of CsAlaDC and AtSerDC to 22.0% and 35.2%, respectively"

The authors primarily use relative activity to compare WT and mutants. Can the authors specify the exact experiments, units, and experimental conditions? Is it Vmax or catalytic efficiency? If so, under what specific experimental conditions?

Response: Thank you for your attention and review of our research paper, we appreciate your suggestions and feedback. The experimental protocol employed to evaluate the influence of DTT on protein catalytic efficiency is outlined as follows:

The 100 μL reaction mixture, containing 20 mM substrate (Ala or Ser), 100 mM potassium phosphate, 0.1 mM PLP, 5 mM L-DTT, and 0.025 mM purified enzyme, was prepared and incubated at standard conditions (45 °C and pH 8.0 for CsAlaDC for 5 min, 40 °C and pH 8.0 for AtSerDC for 2 min). DTT is absent as a control in the reaction system. Then the reaction was stopped with 20 μL of 10% trichloroacetic acid. The product was derivatized with 6-aminoquinolyl-N-hydroxy-succinimidyl carbamate (AQC) and subjected to analysis by UPLC. All enzymatic assays were performed in triplicate.

However, due to the unknown mechanism of DTT inhibition on protein activity, we have removed this part of the content in the revised manuscript.

Pages 10-12

The identification of 'Phe106 in CsAlaDC' and 'Tyr111 in AtSerDC,' along with the subsequent mutagenesis and enzymatic activity assays, is intriguing. However, the current manuscript lacks an explanation and discussion of the underlying reasons for these results. As previously mentioned, it would be helpful to gain insights and analysis from WT-ligand and mutant-ligand binding studies (e.g., ITC, SPR, etc.). Furthermore, the authors' analysis would be more convincing with accompanying structural analysis, such as steric hindrance analysis.

Response: Thank you for your insightful comments and constructive feedback on our manuscript. We appreciate the interest you have expressed in the identification of 'Phe106 in CsAlaDC' and 'Tyr111 in AtSerDC' and their functional implications based on mutagenesis and enzymatic assays.

In order to investigate the binding status of the mutant protein and the ligand PLP,we scanned the UV-visible absorption spectra of both the wild-type and mutant proteins, and the results are as presented in the following figure:

Author response image 9.

(A) UV-Visible Absorption Spectra of CsAlaDC (WT) compared to CsAlaDC (F106Y). (B) UV-Visible Absorption Spectra of AtSerDC (WT) compared to AtSerDC (Y111F).

The mutant protein and the wild-type protein exhibit similar absorbance at 420-430 nm, indicating that the mutation does not affect the binding of PLP to the protein. Therefore, we can conclude that the change in activity of the mutant protein is caused by the substitution of the amino acid at that site, i.e., the amino acid at that site affects substrate specificity. By combining the structure of the two proteins, we can see that the Lys at position 111 of AtSerDC is a hydrophilic amino acid, which increases the hydrophilicity of the active site, and thus the substrate is the hydrophilic amino acid Ser. In contrast, the amino acid at the corresponding site in CsAlaDC is Phe, which, lacking a hydroxyl group compared to Lys, increases the hydrophobicity of the active site, making the substrate lean towards the hydrophobic amino acid Ala. We have added a discussion of the potential reasons for this result to the revised manuscript's discussion section.

Page 5 & Figure 1B

"As expected, CsSerDC was most closed to AtSerDC, which implies that they shared similar functions. However, CsAlaDC is relatively distant from CsSerDC."

In Figure 1B, CsSerDC and AtSerDC are in different clades, and this figure does not show that the two enzymes are closest. To provide another quantitative comparison, please provide a matrix table showing amino acid sequence similarities as a supplemental table.

Response: Many thanks for your constructive suggestion. We added a matrix table showing amino acid sequence similarities in the supplemental materials. The results showed that the similarity of amino acid sequences between CsSerDC and AtSerDC is 86.21%, which is higher than that between CsAlaDC and CsSerDC (84.92%). This data exactly supports the description of Figure 1B. We added the description of the amino acid sequence similarities analysis in the revised manuscript. The description of "As expected, CsSerDC was most closed to AtSerDC, which implies that they shared similar functions. " is not accurate enough, so we revised it to "As expected, CsSerDC was closer to AtSerDC, which implies that they shared similar functions.", in the revised manuscript.

Page 5 & Figure 1C

Figure 1C, which shows a multiple sequence alignment with the amino acid sequences of the 6 SerDCs and CsAlaDC, clearly shows the differences between the sequences of AlaDC and other SerDCs. However, the authors' hypothesis would be more convincing if they showed that this difference is also conserved in AlaDCs from other plants. Can the authors show a new multiple-sequence alignment by adding more amino acid sequences of other AlaDCs?

Response: Thank you for your comments and suggestions. We aim to discover additional alanine decarboxylase. However, at present, the only experimentally confirmed alanine decarboxylase is CsAlaDC. No experimentally verified alanine decarboxylases have been found in other plant species.

Figure 5A

Figure 5A is missing the error bar.

Response: Figure 5A serves as a preliminary screening for these mutants, without conducting repeated experiments. Subsequently, only the L110F and P114A mutants, which exhibited significantly improved activity, underwent further experimental verification to confirm their enhanced functionality.

Author response:

The following is the authors’ response to the original reviews.

Public Reviews:

Reviewer #1 (Public Review):

This work from Cui, Pan, Fan, et al explores memory impairment in chronic pain mouse models, a topic of great interest in the neurobiology field. In particular, the work starts from a very interesting observation, that WT mice can be divided into susceptible and unsusceptible to memory impairment upon modelling chronic pain with CCI. This observation represents the basis of the work where the authors identify the sphingosine receptor S1PR1 as down-regulated in the dentate gyrus of susceptible animals and demonstrate through an elegant range of experiments involving AAV-mediated knockdown or overexpression of S1PR1 that this receptor is involved in the memory impairment observed with chronic pain. Importantly for translational purposes, they also show that activation of S1PR1 through a pharmacological paradigm is able to rescue the memory impairment phenotype.

The authors also link these defects to reduced dendritic branching and a reduced number of mature excitatory synapses in the DG to the memory phenotype.

They then proceed to explore possible mechanisms downstream of S1PR1 that could explain this reduction in dendritic spines. They identify integrin α2 as an interactor of S1PR1 and show a reduction in several proteins involved in actin dynamic, which is crucial for dendritic spine formation and plasticity.

They thus hypothesize that the interaction between S1PR1 and Integrin α2 is fundamental for the activation of Rac1 and Cdc42 and consequently for the polymerisation of actin; a reduction in this pathway upon chronic pain would thus lead to impaired actin polymerisation, synapse formation, and thus impaired memory.

The work is of great interest and the experiments are of very good quality with results of great importance. I have however some concerns. The main concern I have relates to the last part of the work, namely Figures 8 and 9, which I feel are not at the same level as the results presented in the previous 7 Figures, which are instead outstanding.

In particular:

- In Figure 8, given the reduction in all the proteins tested, the authors need to check some additional proteins as controls. One good candidate could be RhoA, considering the authors say it is activated by S1PR2 and not by S1PR1;

Thanks for your suggestion. We tested the expression level of RhoA in mice 7 days and 21 days post CCI as negative controls (Supplemental Figure 9).

- In addition to the previous point, could the authors also show that the number of neurons is not grossly different between susceptible and unsusceptible mice? This could be done by simply staining for NeuN or performing a western blot for a neuronal-specific protein (e.g. Map2 or beta3-tubulin);

As suggested, we performed immunofluorescence using NeuN antibody to detect the number of neurons in susceptible and unsusceptible mice. The number is not significantly different between the two populations (Supplementary Figure 7).

- In Figure 8, the authors should also evaluate the levels of activated RAC1 and activated Cdc42, which are much more important than just basal levels of the proteins to infer an effect on actin dynamics. This is possible through kits that use specific adaptors to pulldown GTP-Rac1 and GTP-Cdc42;

Thanks for your constructive suggestion. An elevated level and hyperactivation of Rac1 protein are both associated with actin dynamics and dendritic development [1]. We agree that showing the levels of activated RAC1 is better to infer its effect on actin dynamics. Here in Figure 8, the purpose of this experiment is to prove the levels of actin organization related proteins are altered according to the expression level of S1PR1, thus drawing a conclusion that the actin organization was disrupted, but not to specifically emphasize that S1PR1 activated these proteins. We apologize for the confusion made but we think the current data is enough to support the conclusion.

Thanks again for your advice. Your understanding is greatly appreciated.

- In Figure 9C, the experiment is performed in an immortalised cell line. I feel this needs to be performed at least in primary hippocampal neurons;

Thanks for your suggestion. As suggested, we performed the experiment in primary hippocampal neurons. Knockdown of S1pr1 in primary hippocampal neurons induced reduction in the number of branches and filamentous actin. Please refer to the updated Figure 9C.

- In Figure 9D, the authors use a Yeast two-hybrid system to demonstrate the interaction between S1PR1 and Integrin α2. However, as the yeast two-hybrid system is based on the proximity of the GAL4 activating domain and the GAL4 binding domain, which are used to activate the transcription of reporter genes, the system is not often used when probing the interaction between transmembrane proteins. Could the authors use other transmembrane proteins as negative controls?;

Thanks for your question. We apologize for the unclear description in the method part. Traditional yeast two-hybrid system can only detect protein interactions that occur in the nucleus, but cannot detect ones between membrane proteins. Here, we utilized the split-ubiquitin membrane-based Yeast two-hybrid system. Briefly, in the ubiquitin system, ubiquitin, a protein composed of 76 amino acid residues that can mediate the ubiquitination degradation of target proteins by proteasomes, is split into two domains, namely Cub at the C-terminus and NbuG at the N-terminus, which are fused and expressed with the bait protein “Bait” and the prey protein “Prey”, respectively. At the same time, Cub is also fused with transcription factors. If Bait and Prey proteins could bind, Cub and NbuG would be brought together and a complete ubiquitin would be formed, which would be recognized by the proteasome and the fused transcription factor would be cut off and enter the cell nucleus to activate the expression of the reporter gene. We then determine whether the Bait and Prey proteins interact with each other through the growth of the yeast.

Thanks again for pointing this out. We reworded the method in M&M (Line 678-696).

- In Figure 9E, the immunoblot is very unconvincing. The bands in the inputs are very weak for both ITGA2 and S1PR1, the authors do not show the enrichment of S1PR1 upon its immunoprecipitation and the band for ITGA2 in the IP fraction has a weird appearance. Were these experiments performed on DG lysates only? If so, I suggest the authors repeat the experiment using the whole brain (or at least the whole hippocampus) so as to have more starting material. Alternatively, if this doesn't work, or in addition, they could also perform the immunoprecipitation in heterologous cells overexpressing the two proteins;

Thanks for the question and suggestion. We used DG lysates from both the dentate gyrus of a single mouse as the starting material. We updated the result which showed clearer bands (Figure 9E).

- About the point above, even if the results were convincing, the authors can't say that they demonstrate an interaction in vivo. In co-IP experiments, the interaction is much more likely to occur in the lysate during the incubation period rather than being conserved from the in vivo state. These co-IPs demonstrate the ability of proteins to interact, not necessarily that they do it in vivo. If the authors wanted to demonstrate this, they could perform a Proximity ligation assay in primary hippocampal neurons, using antibodies against S1PR1 and ITGA2.

Thanks for your concern. Co-immunoprecipitation (Co-IP) is the gold standard to identify protein-protein interactions [2], and it is one of the most efficient techniques to study these protein-protein interactions in vivo [3]. We repeated the experiment and followed the experimental procedure exactly to avoid the protein interaction due to over-incubation. Over-incubation, particularly at room temperature, may result in non-specific binding and therefore high background, thus we performed Co-IPs at 4°C to preserve protein interactions. We agree that Proximity ligation assay is better suited for studies of endogenously expressed proteins in primary cells [4]. Since we optimized the experiment procedure to avoid non-specific binding and particularly, Co-IP utilized proteins from DG lysates which could validate the specificity of the protein interaction in native tissue, we prefer to keep the Co-IP result in Figure 9E.

Thanks again for your suggestion. We appreciate your understanding on this matter.

- In Figure 9H, could the authors increase the N to see if shItga2 causes further KD in the CCI?

As suggested, we repeated the experiment and increased the N to 6. As shown in the following picture, shItga2 did not cause further KD in the CCI.

Author response image 1.

- To conclusively demonstrate that S1PR1 and ITGA2 participate in the same pathway, they could show that knocking down the two proteins at the same time does not have additive effects on behavioral tests compared to the knockdown of each one of them in isolation.

Thanks for your suggestion. As suggested, we knocked down the two proteins at the same and did not observe additive effects on behavioral tests compared to the knockdown of each one of them in isolation. Please refer to Figure 9L-O.

Other major concerns:

- Supplementary Figure 5: the image showing colocalisation between S1PR1 and CamKII is not very convincing. Is the S1PR1 antibody validated on Knockout or knockdown in immunostaining?;

S1PR1 is a membrane receptor and the S1P1 antibody (PA1-1040, Invitrogen) shows membranous staining with diffuse dot-like signals (Please refer to the image “A” provided by ThermoFisher Scientific). Here, we utilized the antibody to detect the expression of S1PR1 in DG granule cells. We can see the diffuse dot-like signals aggregated in each single granule cell. CaMKII shows intense staining around the border of the granule cell soma (Image “B”) [5]. According to the images shown in Supplementary Figure 5B, we concluded that S1PR1 is expressed in CaMKII+ cells.

Besides, as suggested, we validated the S1PR1 antibody on knockdown in immunostaining (Image “C” and “D”). The expression of S1PR1 is significantly decreased compared with the control.

Author response image 2.

- It would be interesting to check S1PR2 levels as a control in CCI-chronic animals;

As suggested, we quantified the S1PR2 levels in Sham and CCI animals, and there is no significant difference between groups (Supplementary Figure 9).

- Figure 1: I am a bit concerned about the Ns in these experiments. In the chronic pain experiments, the N for Sham is around 8 whereas is around 20 for CCI animals. Although I understand higher numbers are necessary to see the susceptible and unsusceptible populations, I feel that then the same number of Sham animals should be used;

Thanks for your concern. In the preliminary experiment, we noticed that the ratio of susceptible and unsusceptible populations is around 1:1. After the behavioral tests, we need to further take samples to investigate molecular and cellular changes of each group. Thus, we set sham around 8 and CCI around 20 to ensure that after characterization into susceptible and unsusceptible groups, each group has relatively equal numbers for further investigations.

- Figures 1E and 1G have much higher Ns than the other panels. Why is that? If they have performed this high number of animals why not show them in all panels?;

Thanks for your concern. For Figure 1B, C, D and F, we showed the data for each batch of experiment, while for Figure 1E and 1G, we used data collected from all batches of experiment. To show the data from a single batch, we would like to demonstrate the ratio of susceptible to unsusceptible is relatively stable, but not only based on a big sample size.

- In the experiments where viral injection is performed, the authors should show a zoomed-out image of the brain to show the precision of the injection and how spread the expression of the different viruses was;

As suggested, we showed the zoomed-out image in Supplementary Figure 6. The viruses are mainly expressed in the hippocampal DG.

- The authors should check if there is brain inflammation in CCI chronic animals. This would be interesting to explain if this could be the trigger for the effects seen in neurons. In particular, the authors should check astrocytes and microglia. This is of interest also because the pathways altered in Figure 8A are related to viral infection.

- If the previous point shows increased brain inflammation, it would be interesting for the authors to check whether a prolonged anti-inflammatory treatment in CCI animals administered before the insurgence of memory impairment could stop it from happening;

- In addition, the authors should speculate on what could be the signal that can induce these molecular changes starting from the site of injury;

- Also, as the animals are all WT, the authors should speculate on what could render some animals prone to have memory impairments and others resistant.<br />

Thanks for the above four suggestions. We have observed inflammation including T cell infiltration and microglia activation in the hippocampal DG in CCI chronic animals and also used S1PR1 modulator which has anti-lymphocyte mediated inflammatory effect to prevent the insurgence of memory impairment from happening. We also examined the alteration in the numbers of peripheral T-lymphocyte subsets and the serum levels of cytokines. Furthermore, we found a neuron-microglia dialogue in the DG which may promote the resilience to memory impairment in CCI animals. Since these are unpublished results, we apologize that we would not give much detailed information to the public at the current stage. We will publish these data as soon as possible. Thanks for your understanding.

Reviewer #2 (Public Review):

Summary:

The study investigates the molecular mechanisms underlying chronic pain-related memory impairment by focusing on S1P/S1PR1 signaling in the dentate gyrus (DG) of the hippocampus. Through behavioural tests (Y-maze and Morris water maze) and RNA-seq analysis, the researchers segregated chronic pain mice into memory impairment-susceptible and -unsusceptible subpopulations. They discovered that S1P/S1PR1 signaling is crucial for determining susceptibility to memory impairment, with decreased S1PR1 expression linked to structural plasticity changes and memory deficits.

Knockdown of S1PR1 in the DG induced a susceptible phenotype, while overexpression or pharmacological activation of S1PR1 promoted resistance to memory impairment and restored normal synaptic structure. The study identifies actin cytoskeleton-related pathways, including ITGA2 and its downstream Rac1/Cdc42 signaling, as key mediators of S1PR1's effects, offering new insights and potential therapeutic targets for chronic pain-related cognitive dysfunction.

This manuscript consists of a comprehensive investigation and significant findings. The study provides novel insights into the molecular mechanisms of chronic pain-related memory impairment, highlighting the critical role of S1P/S1PR1 signaling in the hippocampal dentate gyrus. The clear identification of S1P/S1PR1 as a potential therapeutic target offers promising avenues for future research and treatment strategies. The manuscript is well-structured, methodologically sound, and presents valuable contributions to the field.

Strengths:

(1) The manuscript is well-structured and written in clear, concise language. The flow of information is logical and easy to follow.

(2) The segregation of mice into memory impairment-susceptible and -unsusceptible subpopulations is innovative and well-justified. The statistical analyses are robust and appropriate for the data.

(3) The detailed examination of S1PR1 expression and its impact on synaptic plasticity and actin cytoskeleton reorganization is impressive. The findings are significant and contribute to the understanding of chronic pain-related memory impairment.

Weaknesses:

(1) Results: While the results are comprehensive, some sections are data-heavy and could be more reader-friendly with summarized key points before diving into detailed data.

Thanks for the suggestion. For the first sentence in each part/paragraph, we used statement that summarises what will be investigating in the following experiments to make it more reader-friendly. They are labeled as blue in the main text.

(2) Discussion: There is a need for a more balanced discussion regarding the limitations of the study. For example, addressing potential biases in the animal model or limitations in the generalizability of the findings to humans would strengthen the discussion. Also, providing specific suggestions for follow-up studies would be beneficial.

As suggested, we discussed more on the limitations of this study and outlined some directions for future research (Line 481-498).

(3) Conclusion: The conclusion, while concise, could better highlight the study's broader impact on the field and potential clinical implications.

Thanks. We reworded the conclusion to better highlight the impacts of this study (Line 501-505).

Reviewer #3 (Public Review):

Summary of the Authors' Objectives:

The authors aimed to delineate the role of S1P/S1PR1 signaling in the dentate gyrus in the context of memory impairment associated with chronic pain. They sought to understand the molecular mechanisms contributing to the variability in memory impairment susceptibility and to identify potential therapeutic targets.

Major Strengths and Weaknesses of the Study:

The study is methodologically robust, employing a combination of RNA-seq analysis, viral-mediated gene manipulation, and pharmacological interventions to investigate the S1P/S1PR1 pathway. The use of both knockdown and overexpression approaches to modulate S1PR1 levels provides compelling evidence for its role in memory impairment. The research also benefits from a comprehensive assessment of behavioral changes associated with chronic pain.

However, the study has some weaknesses. The categorization of mice into 'susceptible' and 'unsusceptible' groups based on memory performance requires further validation. Additionally, the reliance on a single animal model may limit the generalizability of the findings. The study could also benefit from a more detailed exploration of the impact of different types of pain on memory impairment.

Assessment of the Authors' Achievements:

The authors successfully identified S1P/S1PR1 signaling as a key factor in chronic pain-related memory impairment and demonstrated its potential as a therapeutic target. The findings are supported by rigorous experimental evidence, including biochemical, histological, and behavioral data. However, the study's impact could be enhanced by further exploration of the molecular pathways downstream of S1PR1 and by assessing the long-term effects of S1PR1 manipulation.

Impact on the Field and Utility to the Community:

This study is likely to have a significant impact on pain research by providing a novel perspective on the mechanisms underlying memory impairment in chronic pain conditions. The identification of the S1P/S1PR1 pathway as a potential therapeutic target could guide the development of new treatments.

Additional Context for Readers:

The study's approach to categorizing susceptibility to memory impairment could inspire new methods for stratifying patient populations in clinical settings.

Recommendations:

(1) A more detailed explanation of the k-means clustering algorithm and its application in categorizing mice should be provided.

As suggested, we explained the k-means clustering algorithm in details (Line 697-711).

(2) The discussion on the potential influence of different pain types or sensitivities on memory impairment should be expanded.

Thanks for your suggestion. We discussed this point in the limitations of this study (Line 484-491).

(3) The protocol for behavioral testing should be clarified and the potential for learning or stress effects should be addressed.

Thanks for your suggestion. We clarified the order of the battery of behavioral tests in this study (Line 537-542). We start with the least stressful test (Y-maze) and leave the most stressful of all for last (Morris Water maze) [6]. Besides, we also conducted behavioral assays to prove that a one-day rest is enough to decrease carryover effects from prior test (Y-maze). We examined the stress related behaviors one day after Y-maze (23d post CCI) using open field test (OFT) and elevated plus maze (EPM). As shown in Author response image 3, the tests did not reflect the mice were under stressful circumstances. Thus, the order in which the tests were performed are appropriate in this study.

Author response image 3.

(4) Conduct additional behavioral assays for other molecular targets implicated in the study.

We agree that other molecular targets on susceptibility to memory impairment would be interesting to know. Our study was designed to focus specifically on ITGA2 this time and we'd like to keep the focus intact, but we have included your point as a consideration for future study (Lines 496-498). Thank you for the suggestion.

(5) The effective drug thresholds and potential non-specific effects of pharmacological interventions should be discussed in more detail.

As suggested, we emphasized this point of drug SEW2871 in Line 242-245.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Minor concerns:

- In Figure 6E the lines of the different groups are not visible. Showing the errors as error bars for each point would probably be better;

We apologize for the mistake of using mean±SD here instead of mean±SEM. After changing to mean±SEM, the lines of Figure 6E, Figure 7E and 7L become much clearer. It looks a little bit messy to show the error bars since there are numerous points, so we prefer to keep the line style.

- Do the authors have any speculation on why the % time in the quadrant is not further affected in the KD Itga2 in CCI animals (Figure 9K)?;

In CCI animals, the level of S1PR1 expression is decreased. ITGA2 may participate in the same pathway with S1PR1. Thus, knocking down ITGA2 in CCI animals will not further affect the animal behaviors. This has been proved by knocking down the two proteins at the same time and no additive effects were observed on behavioral tests compared to the knockdown of each one of them in isolation (Figure 9L-O).

- In the methods, it's unclear if in the multiple infusion, the animals were anaesthetised or kept awake;

We have clarified this point in the method. mice were deeply anesthetized by 1% pentobarbital sodium (40 mg/kg, i.p.). (Line 649-650)

- As the DG is quite small, could the authors clarify if, when performing western blots, they used the two DGs from one animal for each sample or if they pulled together the DGs of several animals?;

We used the two DGs from one animal for each sample. The amount of protein extracted from each sample is enough for 20-30 times of Western Blot assays. We have now added this to the method for clarity (Line 612).

- Is it possible to check the correlation between performance in the YM and MWM with S1PR1 levels?;

We would also be interested in this point. The data that we have cannot reveal this for it is difficult to manipulate the S1PR1 levels by using KD and overexpression viruses.

- EM images have a poor resolution in the figures, could the authors show higher-resolution images?;

We have inserted 300 DPI images for high resolution output.

- In line 268 there is a mention of an "ShLamb1"?

We apologize for the mistake and it was revised.

Reviewer #3 (Recommendations For The Authors):

This study explored the role of S1P/S1PR1 signaling within the dentate gyrus (DG) in chronic pain-related memory impairment using a murine model. The authors identified decreased expression of S1PR1 in the DG of mice susceptible to memory deficits. They demonstrated that S1PR1 knockdown increased susceptibility to memory deficits, whereas its overexpression or pharmacological activation mitigated these effects. Further biochemical and immunofluorescence analyses indicated that disruptions in S1P/S1PR1 signaling were related to disruptions in actin cytoskeleton dynamics, influenced by molecular pathways involving ITGA2, Rac1/Cdc42 signaling, and the Arp2/3 complex. These findings offer intriguing insights and suggest a potential therapeutic target for treating memory impairment in chronic pain.

Major Concerns:

The following five major concerns are the same with the five recommendations from Reviewer 3 on Page 9-10. Please refer to the answers above.

(1) The division of subjects into 'susceptible' and 'unsusceptible' categories requires further clarification regarding the methodologies and rationale employed, particularly concerning the use of the k-means clustering algorithm in data analysis. This explanation will strengthen the scientific grounding of the categorization process.

(2) The categorization of 'susceptible' and 'unsusceptible' groups might also benefit from a more detailed analysis or discussion concerning the influence of different pain sensitivities or types of pain assessments. Although the study mentions that memory impairment stands independent of pain thresholds, a more nuanced exploration could provide deeper insights.

(3) The article could benefit from more clarity on the protocol of behavioral testing, especially regarding the potential effects of repeated testing on performance outcomes due to learning or stress.

(4) While the connection between S1P/S1PR1 signaling and the molecular pathways highlighted (ITGA2, Rac1/Cdc42, Arp2/3) is intriguing, only ITGA2 underwent further behavioral validation in vivo. Conducting additional behavioral assays for one or more of the molecular targets could substantially strengthen these findings.

(5) Discussions regarding effective drug thresholds and the potential for non-specific effects are essential to fully evaluate the implications of pharmacological interventions utilized in the study.

Minor Concerns:

(1) Clarification of evidence of the specific infusion sites in pharmacological experiments would enhance the transparency and replicability of these methods.

For the infusion of S1PR1 agonist, guide cannula (internal diameter 0.34 mm, RWD) was unilaterally implanted into DG of hippocampus (-1.3 A/P, -1.95 M/L, and -2.02 D/V) as evidenced by Figure 5B.

(2) It would be beneficial if the manuscript provided details regarding the efficiency and reach of viral transfection within the neuronal population. This information would help in assessing the impact of genetic manipulations.

S1PR1 immunostaining showed that the efficiency is quite high and the reach of viral transfection is sufficient.

Author response image 4.

(3) The manuscript should make explicit the normalization techniques used in quantitative assessments such as Western blotting, including the housekeeping genes or proteins used for this purpose.

Here, we used housekeeping protein normalization for normalizing Western blot data. GAPDH was used as the internal control. First, the stained blot is imaged, a rectangle is drawn around the target protein in each lane, and the signal intensity inside the rectangle is measured by using ImageJ. The signal intensity obtained can then be normalized by being divided by the signal intensity of the loading internal control (GAPDH) detected on the same blot. The average of the ratios from the control group is calculated, and all individual ratios are divided by this average to obtain a new set of values, which represent the normalized values (Line 619-625).

(4) Details about the control groups in behavioral assessments were subjected to comparable handling and experimental conditions as the chronic pain groups are crucial, barring nerve injury, for maintaining the integrity of the comparative analysis.

We agree that a control group and an experimental group is identical in all respects except for one difference-nerve injury. We have added this point in the method (Line 520-522).

Minor Recommendations:

The following four minor recommendations are the same with the four minor concerns from Reviewer 3 on Page 12-13. Please refer to the answers above.

(1) Clarify the specifics of infusion site verification in pharmacological experiments.

(2) Provide details on the efficiency and neuronal reach of viral transfections.

(3) Explicitly describe the normalization techniques used in quantitative assessments.

(4) Ensure that control groups in behavioral assessments undergo comparable handling to maintain analysis integrity.

References

(1) Gualdoni, S., et al., Normal levels of Rac1 are important for dendritic but not axonal development in hippocampal neurons. Biology of the Cell, 2007. 99(8): p. 455-464.

(2) Alam, M.S., Proximity Ligation Assay (PLA). Curr Protoc Immunol, 2018. 123(1): p. e58.

(3) Song, P., S. Zhang, and J. Li, Co-immunoprecipitation Assays to Detect In Vivo Association of Phytochromes with Their Interacting Partners. Methods Mol Biol, 2021. 2297: p. 75-82.

(4) Krieger, C.C., et al., Proximity ligation assay to study TSH receptor homodimerization and crosstalk with IGF-1 receptors in human thyroid cells. Frontiers in Endocrinology, 2022. 13.

(5) Arruda-Carvalho, M., et al., Conditional Deletion of α-CaMKII Impairs Integration of Adult-Generated Granule Cells into Dentate Gyrus Circuits and Hippocampus-Dependent Learning. The Journal of Neuroscience, 2014. 34(36): p. 11919-11928.

(6) Wolf, A., et al., A Comprehensive Behavioral Test Battery to Assess Learning and Memory in 129S6/Tg2576 Mice. PLoS One, 2016. 11(1): p. e0147733.

Author response:

The following is the authors’ response to the previous reviews

Responses to Editors:

We appreciate the editors’ concern regarding the difficulty of disentangling the contributions of tightly-coupled brain regions to the speech-gesture integration process—particularly due to the close temporal and spatial proximity of the stimulation windows and the potential for prolonged disruption. While we agree with that stimulation techniques, such as transcranial magnetic stimulation (TMS), can evoke or modulate neuronal activity both locally within the target region and in remote connected areas of the network. This complex interaction makes drawing clear conclusions about the causal relationship between stimulation and cognitive function more challenging. However, we believe that cause-and-effect relationships in cognitive neuroscience studies using non-invasive brain stimulation (NIBS) can still be robustly established if key assumptions are explicitly tested and confounding factors are rigorously controlled (Bergmann & Hartwigsen et al., 2021, J Cogn Neurosci).

In our experiment, we addressed these concerns by including a sham TMS condition, an irrelevant control task, and multiple control time points. The results showed that TMS selectively disrupted the IFG-pMTG interaction during specific time windows of the task related to gesture-speech semantic congruency, but not in the sham TMS condition or the control task (gender congruency effect) (Zhao et al., 2021, JN). This selective disruption provides strong evidence for a causal link between IFG-pMTG connectivity and gesture-speech integration in the targeted time window.

Regarding the potential for transient artifacts from TMS, we acknowledge that previous research has demonstrated that single-pulse TMS induces brief artifacts (0–10 ms) due to direct depolarization of cortical neurons, which momentarily disrupts electrical activity in the stimulated area (Romero et al., 2019, NC). However, in the case of paired-pulse TMS (ppTMS), the interaction between the first and second pulses is more complex. The first pulse increases membrane conductance in the target neurons via shunting inhibition mediated by GABAergic interneurons. This effectively lowers neuronal membrane resistance, “leaking” excitatory current and diminishing the depolarization induced by the second pulse, leading to a reduction in excitability during the paired-pulse interval. This mechanism suppresses the excitatory response to the second pulse, which is reflected in a reduced motor evoked potential (MEP) (Paulus & Rothwell, 2016, J Physiol).

Furthermore, ppTMS has been widely used in previous studies to infer causal temporal relationships and explore the neural contributions of both structurally and functionally connected brain regions, across timescales as brief as 3–60 ms. We have reviewed several studies that employed paired-pulse TMS to investigate neural dynamics in regions such as the tongue and lip areas of the primary motor cortex (M1), as well as high-level semantic regions like the pMTG, PFC, and ATL (Table 1). These studies consistently demonstrate the methodological rigor and precision of double-pulse TMS in elucidating the temporal dynamics between different brain regions within short temporal windows.

Given these precedents and the evidence provided, we respectfully assert the validity of the methods employed in our study. We therefore kindly request the editors to reconsider the assessment that “the methods are insufficient for studying tightly-coupled brain regions over short timescales.” We hope that the editors’ concerns about the complexities of TMS-induced effects have been adequately addressed, and that our study’s design and results provide a clear and convincing causal argument for the role of IFG-pMTG in gesture-speech integration.

Author response table 1.

Double-pulse TMS studies on brain regions over 3-60 ms time interval

Reference

Teige, C., Mollo, G., Millman, R., Savill, N., Smallwood, J., Cornelissen, P. L., & Jefferies, E. (2018). Dynamic semantic cognition: Characterising coherent and controlled conceptual retrieval through time using magnetoencephalography and chronometric transcranial magnetic stimulation. Cortex, 103, 329-349.

Amemiya, T., Beck, B., Walsh, V., Gomi, H., & Haggard, P. (2017). Visual area V5/hMT+ contributes to perception of tactile motion direction: a TMS study. Scientific reports, 7(1), 40937.

Muessgens, D., Thirugnanasambandam, N., Shitara, H., Popa, T., & Hallett, M. (2016). Dissociable roles of preSMA in motor sequence chunking and hand switching—a TMS study. Journal of Neurophysiology, 116(6), 2637-2646.

Vernet, M., Brem, A. K., Farzan, F., & Pascual-Leone, A. (2015). Synchronous and opposite roles of the parietal and prefrontal cortices in bistable perception: a double-coil TMS–EEG study. Cortex, 64, 78-88.

Pitcher, D. (2014). Facial expression recognition takes longer in the posterior superior temporal sulcus than in the occipital face area. Journal of Neuroscience, 34(27), 9173-9177.

Bardi, L., Kanai, R., Mapelli, D., & Walsh, V. (2012). TMS of the FEF interferes with spatial conflict. Journal of cognitive neuroscience, 24(6), 1305-1313.

D’Ausilio, A., Bufalari, I., Salmas, P., & Fadiga, L. (2012). The role of the motor system in discriminating normal and degraded speech sounds. Cortex, 48(7), 882-887.

Pitcher, D., Duchaine, B., Walsh, V., & Kanwisher, N. (2010). TMS evidence for feedforward and feedback mechanisms of face and body perception. Journal of Vision, 10(7), 671-671.

Gagnon, G., Blanchet, S., Grondin, S., & Schneider, C. (2010). Paired-pulse transcranial magnetic stimulation over the dorsolateral prefrontal cortex interferes with episodic encoding and retrieval for both verbal and non-verbal materials. Brain Research, 1344, 148-158.

Kalla, R., Muggleton, N. G., Juan, C. H., Cowey, A., & Walsh, V. (2008). The timing of the involvement of the frontal eye fields and posterior parietal cortex in visual search. Neuroreport, 19(10), 1067-1071.

Pitcher, D., Garrido, L., Walsh, V., & Duchaine, B. C. (2008). Transcranial magnetic stimulation disrupts the perception and embodiment of facial expressions. Journal of Neuroscience, 28(36), 8929-8933.

Til Ole Bergmann, Gesa Hartwigsen; Inferring Causality from Noninvasive Brain Stimulation in Cognitive Neuroscience. J Cogn Neurosci 2021; 33 (2): 195–225. https://doi.org/10.1162/jocn_a_01591

Romero, M.C., Davare, M., Armendariz, M. et al. Neural effects of transcranial magnetic stimulation at the single-cell level. Nat Commun 10, 2642 (2019). https://doi.org/10.1038/s41467-019-10638-7

Paulus W, Rothwell JC. Membrane resistance and shunting inhibition: where biophysics meets state-dependent human neurophysiology. J Physiol. 2016 May 15;594(10):2719-28. doi: 10.1113/JP271452. PMID: 26940751; PMCID: PMC4865581.

Staat, C., Gattinger, N., & Gleich, B. (2022). PLUSPULS: A transcranial magnetic stimulator with extended pulse protocols. HardwareX, 13. https://doi.org/10.1016/j.ohx.2022.e00380

Zhao, W., Li, Y., and Du, Y. (2021). TMS reveals dynamic interaction between inferior frontal gyrus and posterior middle temporal gyrus in gesture-speech semantic integration. The Journal of Neuroscience, 10356-10364. https://doi.org/10.1523/jneurosci.1355-21.2021.

Reviewer #1 (Public review):

Summary:

The authors quantified information in gesture and speech, and investigated the neural processing of speech and gestures in pMTG and LIFG, depending on their informational content, in 8 different time-windows, and using three different methods (EEG, HD-tDCS and TMS). They found that there is a time-sensitive and staged progression of neural engagement that is correlated with the informational content of the signal (speech/gesture).

Strengths:

A strength of the paper is that the authors attempted to combine three different methods to investigate speech-gesture processing.

We sincerely thank the reviewer for recognizing our efforts in conducting three experiments to explore the neural activity linked to the amount of information processed during multisensory gesture-speech integration. In Experiment 1, we observed that the extent of inhibition in the pMTG and LIFG was closely linked to the overlapping gesture-speech responses, as quantified by mutual information. Building on the established roles of the pMTG and LIFG in our previous study (Zhao et al., 2021, JN), we then expanded our investigation to determine whether the dynamic neural engagement between the pMTG and LIFG during gesture-speech processing was also associated with the quality of the information. This hypothesis was further validated through high-temporal resolution EEG, where we examined ERP components related to varying information contents. Notably, we observed a close time alignment between the ERP components and the time windows of the TMS effects, which were associated with the same informational matrices in gesture-speech processing.

Weaknesses:

(1) One major issue is that there is a tight anatomical coupling between pMTG and LIFG. Stimulating one area could therefore also result in stimulation of the other area (see Silvanto and Pascual-Leone, 2008). I therefore think it is very difficult to tease apart the contribution of these areas to the speech-gesture integration process, especially considering that the authors stimulate these regions in time windows that are very close to each other in both time and space (and the disruption might last longer over time).

Response 1: We greatly appreciate the reviewer’s careful consideration. We trust that the explanation provided above has clarified this issue (see Response to Editors for detail).

(2) Related to this point, it is unclear to me why the HD-TDCS/TMS is delivered in set time windows for each region. How did the authors determine this, and how do the results for TMS compare to their previous work from 2018 and 2023 (which describes a similar dataset+design)? How can they ensure they are only targeting their intended region since they are so anatomically close to each other?

Response 2: The current study builds on a series of investigations that systematically examined the temporal and spatial dynamics of gesture-speech integration. In our earlier work (Zhao et al., 2018, J. Neurosci), we demonstrated that interrupting neural activity in the IFG or pMTG using TMS selectively disrupted the semantic congruency effect (reaction time costs due to semantic incongruence), without affecting the gender congruency effect (reaction time costs due to gender incongruence). These findings identified the IFG and pMTG as critical hubs for gesture-speech integration. This informed the brain regions selected for subsequent studies.

In Zhao et al. (2021, J. Neurosci), we employed a double-pulse TMS protocol, delivering stimulation within one of eight 40-ms time windows, to further examine the temporal involvement of the IFG and pMTG. The results revealed time-window-selective disruptions of the semantic congruency effect, confirming the dynamic and temporally staged roles of these regions during gesture-speech integration.

In Zhao et al. (2023, Frontiers in Psychology), we investigated the semantic predictive role of gestures relative to speech by comparing two experimental conditions: (1) gestures preceding speech by a fixed interval of 200 ms, and (2) gestures preceding speech at its semantic identification point. We observed time-window-selective disruptions of the semantic congruency effect in the IFG and pMTG only in the second condition, leading to the conclusion that gestures exert a semantic priming effect on co-occurring speech. These findings underscored the semantic advantage of gesture in facilitating speech integration, further refining our understanding of the temporal and functional interplay between these modalities.

The design of the current study—including the choice of brain regions and time windows—was directly informed by these prior findings. Experiment 1 (HD-tDCS) targeted the entire gesture-speech integration process in the IFG and pMTG to assess whether neural activity in these regions, previously identified as integration hubs, is modulated by changes in informativeness from both modalities (i.e., entropy) and their interactions (mutual information, MI). The results revealed a gradual inhibition of neural activity in both areas as MI increased, evidenced by a negative correlation between MI and the tDCS inhibition effect in both regions. Building on this, Experiments 2 and 3 employed double-pulse TMS and ERPs to further assess whether the engaged neural activity was both time-sensitive and staged. These experiments also evaluated the contributions of various sources of information, revealing correlations between information-theoretic metrics and time-locked brain activity, providing insights into the ‘gradual’ nature of gesture-speech integration.

We acknowledge that the rationale for the design of the current study was not fully articulated in the original manuscript. In the revised version, we provided a more comprehensive and coherent explanation of the logic behind the three experiments, as well as the alignment with our previous findings in Lines 75-102:

‘To investigate the neural mechanisms underlying gesture-speech integration, we conducted three experiments to assess how neural activity correlates with distributed multisensory integration, quantified using information-theoretic measures of MI. Additionally, we examined the contributions of unisensory signals in this process, quantified through unisensory entropy. Experiment 1 employed high-definition transcranial direct current stimulation (HD-tDCS) to administer Anodal, Cathodal and Sham stimulation to either the IFG or the pMTG. HD-tDCS induces membrane depolarization with anodal stimulation and membrane hyperpolarization with cathodal stimulation[26], thereby increasing or decreasing cortical excitability in the targeted brain area, respectively. This experiment aimed to determine whether the overall facilitation (Anodal-tDCS minus Sham-tDCS) and/or inhibitory (Cathodal-tDCS minus Sham-tDCS) of these integration hubs is modulated by the degree of gesture-speech integration, as measure by MI.

Given the differential involvement of the IFG and pMTG in gesture-speech integration, shaped by top-down gesture predictions and bottom-up speech processing [23], Experiment 2 was designed to further assess whether the activity of these regions was associated with relevant informational matrices. Specifically, we applied inhibitory chronometric double-pulse transcranial magnetic stimulation (TMS) to specific temporal windows associated with integration processes in these regions[23], assessing whether the inhibitory effects of TMS were correlated with unisensory entropy or the multisensory convergence index (MI).

Experiment 3 complemented these investigations by focusing on the temporal dynamics of neural responses during semantic processing, leveraging high-temporal event-related potentials (ERPs). This experiment investigated how distinct information contributors modulated specific ERP components associated with semantic processing. These components included the early sensory effects as P1 and N1–P2[27,28], the N400 semantic conflict effect[14,28,29], and the late positive component (LPC) reconstruction effect[30,31]. By integrating these ERP findings with results from Experiments 1 and 2, Experiment 3 aimed to provide a more comprehensive understanding of how gesture-speech integration is modulated by neural dynamics.’

Although the IFG and pMTG are anatomically close, the consistent differentiation of their respective roles, as evidenced by our experiment across various time windows (TWs) and supported by previous research (see Response to editors for details), reinforces the validity of the stimulation effect observed in our study.

References

Zhao, W.Y., Riggs, K., Schindler, I., and Holle, H. (2018). Transcranial magnetic stimulation over left inferior frontal and posterior temporal cortex disrupts gesture-speech integration. Journal of Neuroscience 38, 1891-1900. 10.1523/Jneurosci.1748-17.2017.

Zhao, W., Li, Y., and Du, Y. (2021). TMS reveals dynamic interaction between inferior frontal gyrus and posterior middle temporal gyrus in gesture-speech semantic integration. The Journal of Neuroscience, 10356-10364. https://doi.org/10.1523/jneurosci.1355-21.2021.

Zhao, W. (2023). TMS reveals a two-stage priming circuit of gesture-speech integration. Front Psychol 14, 1156087. 10.3389/fpsyg.2023.1156087.

Bikson, M., Inoue, M., Akiyama, H., Deans, J.K., Fox, J.E., Miyakawa, H., and Jefferys, J.G.R. (2004). Effects of uniform extracellular DC electric fields on excitability in rat hippocampal slices. J Physiol-London 557, 175-190. 10.1113/jphysiol.2003.055772.

Federmeier, K.D., Mai, H., and Kutas, M. (2005). Both sides get the point: hemispheric sensitivities to sentential constraint. Memory & Cognition 33, 871-886. 10.3758/bf03193082.

Kelly, S.D., Kravitz, C., and Hopkins, M. (2004). Neural correlates of bimodal speech and gesture comprehension. Brain and Language 89, 253-260. 10.1016/s0093-934x(03)00335-3.

Wu, Y.C., and Coulson, S. (2005). Meaningful gestures: Electrophysiological indices of iconic gesture comprehension. Psychophysiology 42, 654-667. 10.1111/j.1469-8986.2005.00356.x.

Fritz, I., Kita, S., Littlemore, J., and Krott, A. (2021). Multimodal language processing: How preceding discourse constrains gesture interpretation and affects gesture integration when gestures do not synchronise with semantic affiliates. J Mem Lang 117, 104191. 10.1016/j.jml.2020.104191.

Gunter, T.C., and Weinbrenner, J.E.D. (2017). When to take a gesture seriously: On how we use and prioritize communicative cues. J Cognitive Neurosci 29, 1355-1367. 10.1162/jocn_a_01125.

Ozyurek, A., Willems, R.M., Kita, S., and Hagoort, P. (2007). On-line integration of semantic information from speech and gesture: Insights from event-related brain potentials. J Cognitive Neurosci 19, 605-616. 10.1162/jocn.2007.19.4.605.

(3) As the EEG signal is often not normally distributed, I was wondering whether the authors checked the assumptions for their Pearson correlations. The authors could perhaps better choose to model the different variables to see whether MI/entropy could predict the neural responses. How did they correct the many correlational analyses that they have performed?

Response 3: We greatly appreciate the reviewer’s thoughtful comments.

(1) Regarding the questioning of normal distribution of EEG signals and the use of Pearson correlation, in Figure 5 of the manuscript, we have already included normal distribution curves to illustrate the relationships between average ERP amplitudes across each ROI or elicited cluster and the three information models.

Additionally, we performed the Shapiro-Wilk test, a widely accepted method for assessing bivariate normality, on both the MI/entropy and averaged ERP data. The p-values for all three combinations were greater than 0.05, indicating that the sample data from all bivariate combinations were normally distributed (Author response table 2).

Author response table 2.

Shapiro-Wilk results of bivariable normality test

To further consolidate the relationship between entropy/MI and various ERP components, we also conducted a Spearman rank correlation analysis (Author response table 3-5). While the correlation between speech entropy and ERP amplitude in the P1 component yielded a p-value of 0.061, all other results were consistent with those obtained from the Pearson correlation analysis across the three experiments. Therefore, our conclusion that progressive neural responses reflected the degree of information remains robust. Although the Spearman rank and Pearson correlation analyses yielded similar results, we opted to report the Pearson correlation coefficients throughout the manuscript to maintain consistency.

Author response table 3.

Comparison of Pearson and Spearman results in Experiment 1

Author response table 4.

Comparison of Pearson and Spearman results in Experiment 2

Author response table 5.

Comparison of Pearson and Spearman results in Experiment 3

(2) Regarding the reviewer’s comment ‘choose to model the different variables to see whether MI/entropy could predict the neural responses’, we employed Representational Similarity Analysis (RSA) (Popal et.al, 2019) with MI and entropy as continuous variables. This analysis aimed to build a model to predict neural responses based on these feature metrics.

To capture dynamic temporal features indicative of different stages of multisensory integration, we segmented the EEG data into overlapping time windows (40 ms in duration with a 10 ms step size). The 40 ms window was chosen based on the TMS protocol used in Experiment 2, which also employed a 40 ms time window. The 10 ms step size (equivalent to 5 time points) was used to detect subtle shifts in neural responses that might not be captured by larger time windows, allowing for a more granular analysis of the temporal dynamics of neural activity.

Following segmentation, the EEG data were reshaped into a four-dimensional matrix (42 channels × 20 time points × 97 time windows × 20 features). To construct a neural similarity matrix, we averaged the EEG data across time points within each channel and each time window. The resulting matrix was then processed using the pdist function to compute pairwise distances between adjacent data points. This allowed us to calculate correlations between the neural matrix and three feature similarity matrices, which were constructed in a similar manner. These three matrices corresponded to (1) gesture entropy, (2) speech entropy, and (3) mutual information (MI). This approach enabled us to quantify how well the neural responses corresponded to the semantic dimensions of gesture and speech stimuli at each time window.

To determine the significance of the correlations between neural activity and feature matrices, we conducted 1000 permutation tests. In this procedure, we randomized the data or feature matrices and recalculated the correlations repeatedly, generating a null distribution against which the observed correlation values were compared. Statistical significance was determined if the observed correlation exceeded the null distribution threshold (p < 0.05). This permutation approach helps mitigate the risk of spurious correlations, ensuring that the relationships between the neural data and feature matrices are both robust and meaningful.

Finally, significant correlations were subjected to clustering analysis, which grouped similar neural response patterns across time windows and channels. This clustering allowed us to identify temporal and spatial patterns in the neural data that consistently aligned with the semantic features of gesture and speech stimuli, thus revealing the dynamic integration of these multisensory modalities across time. Results are as follows:

(1) Two significant clusters were identified for gesture entropy (Author response image 1 left). The first cluster was observed between 60-110 ms (channels F1 and F3), with correlation coefficients (r) ranging from 0.207 to 0.236 (p < 0.001). The second cluster was found between 210-280 ms (channel O1), with r-values ranging from 0.244 to 0.313 (p < 0.001).

(2) For speech entropy (Author response image 1 middle), significant clusters were detected in both early and late time windows. In the early time windows, the largest significant cluster was found between 10-170 ms (channels F2, F4, F6, FC2, FC4, FC6, C4, C6, CP4, and CP6), with r-values ranging from 0.151 to 0.340 (p = 0.013), corresponding to the P1 component (0-100 ms). In the late time windows, the largest significant cluster was observed between 560-920 ms (across the whole brain, all channels), with r-values ranging from 0.152 to 0.619 (p = 0.013).

(3) For mutual information (MI) (Author response image 1 right), a significant cluster was found between 270-380 ms (channels FC1, FC2, FC3, FC5, C1, C2, C3, C5, CP1, CP2, CP3, CP5, FCz, Cz, and CPz), with r-values ranging from 0.198 to 0.372 (p = 0.001).

Author response image 1.

Results of RSA analysis.

These additional findings suggest that even using a different modeling approach, neural responses, as indexed by feature metrics of entropy and mutual information, are temporally aligned with distinct ERP components and ERP clusters, as reported in the current manuscript. This alignment serves to further consolidate the results, reinforcing the conclusion we draw. Considering the length of the manuscript, we did not include these results in the current manuscript.

(3) In terms of the correction of multiple comparisons, in Experiment 1, two separate participant groups were recruited for HD-tDCS applied over either the IFG or pMTG. FDR correction was performed separately for each group, resulting in six comparisons for each brain region (three information matrices × two tDCS effects: anodal-sham or cathodal-sham). In Experiment 2, six comparisons (three information matrices × two sites: IFG or pMTG) were submitted for FDR correction. In Experiment 3, FDR correction was applied to the seven regions of interest (ROIs) within each component, resulting in five comparisons.

Reference:

Wilk, M.B. (2015). The Shapiro Wilk And Related Tests For Normality.

Popal, H., Wang, Y., & Olson, I. R. (2019). A guide to representational similarity analysis for social neuroscience. Social cognitive and affective neuroscience, 14(11), 1243-1253.

(4) The authors use ROIs for their different analyses, but it is unclear why and on the basis of what these regions are defined. Why not consider all channels without making them part of an ROI, by using a method like the one described in my previous comment?

Response 4: For the EEG data, we conducted both a traditional ROI analysis and a cluster-based permutation approach. The ROIs were defined based on a well-established work (Habets et al., 2011), allowing for hypothesis-driven testing of specific regions. In addition, we employed a cluster-based permutation methods, which is data-driven and helps enhance robustness while addressing multiple comparisons. This method serves as a complement to the hypothesis-driven ROI analysis, offering an exploratory, unbiased perspective. Notably, the results from both approaches were consistent, reinforcing the reliability of our findings.

To make the methods more accessible to a broader audience, we clarified the relationship between these approaches in the revised manuscript in Lines 267-270: ‘To consolidate the data, we conducted both a traditional region-of-interest (ROI) analysis, with ROIs defined based on a well-established work40, and a cluster-based permutation approach, which utilizes data-driven permutations to enhance robustness and address multiple comparisons’

Additionally, we conducted an RSA analysis without defining specific ROIs, considering all channels in the analysis. This approach yielded consistent results, further validating the robustness of our findings across different analysis methods. See Response 3 for detail.

Reference:

Habets, B., Kita, S., Shao, Z.S., Ozyurek, A., and Hagoort, P. (2011). The Role of Synchrony and Ambiguity in Speech-Gesture Integration during Comprehension. J Cognitive Neurosci 23, 1845-1854. 10.1162/jocn.2010.21462

(5) The authors describe that they have divided their EEG data into a "lower half" and a "higher half" (lines 234-236), based on entropy scores. It is unclear why this is necessary, and I would suggest just using the entropy scores as a continuous measure.

Response 5: To identify ERP components or spatiotemporal clusters that demonstrated significant semantic differences, we split each model into higher and lower halves based on entropy scores. This division allowed us to capture distinct levels of information processing and explore how different levels of entropy or mutual information (MI) related to neural activity. Specifically, the goal was to highlight the gradual activation process of these components and clusters as they correlate with changes in information content. Remarkably, consistent results were observed between the ERP components and clusters, providing robust evidence that semantic information conveyed through gestures and speech significantly influenced the amplitude of these components or clusters. Moreover, the semantic information was shown to be highly sensitive, varying in tandem with these amplitude changes.

Reviewer #2 (Public review):

Comment:

Summary:

The study is an innovative and fundamental study that clarified important aspects of brain processes for integration of information from speech and iconic gesture (i.e., gesture that depicts action, movement, and shape), based on tDCS, TMS, and EEG experiments. They evaluated their speech and gesture stimuli in information-theoretic ways and calculated how informative speech is (i.e., entropy), how informative gesture is, and how much shared information speech and gesture encode. The tDCS and TMS studies found that the left IFG and pMTG, the two areas that were activated in fMRI studies on speech-gesture integration in the previous literature, are causally implicated in speech-gesture integration. The size of tDC and TMS effects are correlated with the entropy of the stimuli or mutual information, which indicates that the effects stem from the modulation of information decoding/integration processes. The EEG study showed that various ERP (event-related potential, e.g., N1-P2, N400, LPC) effects that have been observed in speech-gesture integration experiments in the previous literature, are modulated by the entropy of speech/gesture and mutual information. This makes it clear that these effects are related to information decoding processes. The authors propose a model of how the speech-gesture integration process unfolds in time, and how IFG and pMTG interact with each other in that process.

Strengths:

The key strength of this study is that the authors used information theoretic measures of their stimuli (i.e., entropy and mutual information between speech and gesture) in all of their analyses. This made it clear that the neuro-modulation (tDCS, TMS) affected information decoding/integration and ERP effects reflect information decoding/integration. This study used tDCS and TMS methods to demonstrate that left IFG and pMTG are causally involved in speech-gesture integration. The size of tDCS and TMS effects are correlated with information-theoretic measures of the stimuli, which indicate that the effects indeed stem from disruption/facilitation of the information decoding/integration process (rather than generic excitation/inhibition). The authors' results also showed a correlation between information-theoretic measures of stimuli with various ERP effects. This indicates that these ERP effects reflect the information decoding/integration process.

We sincerely thank the reviewer for recognizing our efforts and the innovation of employing information-theoretic measures to elucidate the brain processes underlying the multisensory integration of gesture and speech.

Weaknesses:

The "mutual information" cannot fully capture the interplay of the meaning of speech and gesture. The mutual information is calculated based on what information can be decoded from speech alone and what information can be decoded from gesture alone. However, when speech and gesture are combined, a novel meaning can emerge, which cannot be decoded from a single modality alone. When example, a person produces a gesture of writing something with a pen, while saying "He paid". The speech-gesture combination can be interpreted as "paying by signing a cheque". It is highly unlikely that this meaning is decoded when people hear speech only or see gestures only. The current study cannot address how such speech-gesture integration occurs in the brain, and what ERP effects may reflect such a process. Future studies can classify different types of speech-gesture integration and investigate neural processes that underlie each type. Another important topic for future studies is to investigate how the neural processes of speech-gesture integration change when the relative timing between the speech stimulus and the gesture stimulus changes.

We greatly appreciate Reviewer2 ’s thoughtful concern regarding whether "mutual information" adequately captures the interplay between the meanings of speech and gesture. We would like to clarify that the materials used in the present study involved gestures that were performed without actual objects, paired with verbs that precisely describe the corresponding actions. For example, a hammering gesture was paired with the verb “hammer”, and a cutting gesture was paired with the verb “cut”. In this design, all gestures conveyed redundant information relative to the co-occurring speech, creating significant overlap between the information derived from speech alone and that from gesture alone.

We understand the reviewer’s concern about cases where gestures and speech might provide complementary, rather than redundant, information. To address this, we have developed an alternative metric for quantifying information gains contributed by supplementary multisensory cues, which will be explored in a subsequent study. However, for the present study, we believe that the observed overlap in information serves as a key indicator of multisensory convergence, a central focus of our investigation.

Regarding the reviewer’s concern about how neural processes of speech-gesture integration may change with varying relative timing between speech and gesture stimuli, we would like to highlight findings from our previous study (Zhao, 2023, Frontiers in Psychology). In that study, we explored the semantic predictive role of gestures relative to speech under two timing conditions: (1) gestures preceding speech by a fixed interval of 200 ms, and (2) gestures preceding speech at its semantic identification point. Interestingly, only in the second condition did we observe time-window-selective disruptions of the semantic congruency effect in the IFG and pMTG. This led us to conclude that gestures play a semantic priming role for co-occurring speech. Building on this, we designed the present study with gestures deliberately preceding speech at its semantic identification point to reflect this semantic priming relationship. Additionally, ongoing research in our lab is exploring gesture and speech interactions in natural conversational settings to investigate whether the neural processes identified here remain consistent across varying contexts.

To address potential concerns and ensure clarity regarding the limitations of the MI measurement, we have included a discussion of tthis in the revised manuscript in Lines 543-547: ‘Furthermore, MI quantifies overlap in gesture-speech integration, primarily when gestures convey redundant meaning. Consequently, the conclusions drawn in this study are constrained to contexts in which gestures serve to reinforce the meaning of the speech. Future research should aim to explore the neural responses in cases where gestures convey supplementary, rather than redundant, semantic information.’ This is followed by a clarification of the timing relationship between gesture and speech: ‘Note that the sequential cortical involvement and ERP components discussed above are derived from a deliberate alignment of speech onset with gesture DP, creating an artificial priming effect with gesture semantically preceding speech. Caution is advised when generalizing these findings to the spontaneous gesture-speech relationships, although gestures naturally precede speech[34].’ (Lines 539-543).

Reviewer #3 (Public review):

In this useful study, Zhao et al. try to extend the evidence for their previously described two-step model of speech-gesture integration in the posterior Middle Temporal Gyrus (pMTG) and Inferior Frontal Gyrus (IFG). They repeat some of their previous experimental paradigms, but this time quantifying Information-Theoretical (IT) metrics of the stimuli in a stroop-like paradigm purported to engage speech-gesture integration. They then correlate these metrics with the disruption of what they claim to be an integration effect observable in reaction times during the tasks following brain stimulation, as well as documenting the ERP components in response to the variability in these metrics.

The integration of multiple methods, like tDCS, TMS, and ERPs to provide converging evidence renders the results solid. However, their interpretation of the results should be taken with care, as some critical confounds, like difficulty, were not accounted for, and the conceptual link between the IT metrics and what the authors claim they index is tenuous and in need of more evidence. In some cases, the difficulty making this link seems to arise from conceptual equivocation (e.g., their claims regarding 'graded' evidence), whilst in some others it might arise from the usage of unclear wording in the writing of the manuscript (e.g. the sentence 'quantitatively functional mental states defined by a specific parser unified by statistical regularities'). Having said that, the authors' aim is valuable, and addressing these issues would render the work a very useful approach to improve our understanding of integration during semantic processing, being of interest to scientists working in cognitive neuroscience and neuroimaging.

The main hurdle to achieving the aims set by the authors is the presence of the confound of difficulty in their IT metrics. Their measure of entropy, for example, being derived from the distribution of responses of the participants to the stimuli, will tend to be high for words or gestures with multiple competing candidate representations (this is what would presumptively give rise to the diversity of responses in high-entropy items). There is ample evidence implicating IFG and pMTG as key regions of the semantic control network, which is critical during difficult semantic processing when, for example, semantic processing must resolve competition between multiple candidate representations, or when there are increased selection pressures (Jackson et al., 2021). Thus, the authors' interpretation of Mutual Information (MI) as an index of integration is inextricably contaminated with difficulty arising from multiple candidate representations. This casts doubt on the claims of the role of pMTG and IFG as regions carrying out gesture-speech integration as the observed pattern of results could also be interpreted in terms of brain stimulation interrupting the semantic control network's ability to select the best candidate for a given context or respond to more demanding semantic processing.

Response 1: We sincerely thank the reviewer for pointing out the confound of difficulty. The primary aim of this study is to investigate whether the degree of activity in the established integration hubs, IFG and pMTG, is influenced by the information provided by gesture-speech modalities and/or their interactions. While we provided evidence for the differential involvement of the IFG and pMTG by delineating their dynamic engagement across distinct time windows of gesture-speech integration and associating these patterns with unisensory information and their interaction, we acknowledge that the mechanisms underlying these dynamics remain open to interpretation. Specifically, whether the observed effects stem from difficulties in semantic control processes, as suggested by the reviewer, or from resolving information uncertainty, as quantified by entropy, falls outside the scope of the current study. Importantly, we view these two interpretations as complementary rather than mutually exclusive, as both may be contributing factors. Nonetheless, we agree that addressing this question is a compelling avenue for future research.

In the revised manuscript, we have included an additional analysis to assess whether the confounding effects of lexical or semantic control difficulty—specifically, the number of available responses—affect the neural outcomes. To address this, we performed partial correlation analyses, controlling for the number of responses.

We would like to clarify an important distinction between the measure of entropy derived from the distribution of responses and the concept of response diversity. Entropy, in our analysis, is computed based on the probability distribution of each response, as captured by the information entropy formula. In contrast, response diversity refers to the simple count of different responses provided. Mutual Information (MI), by its nature, is also an entropy measure, quantifying the overlap in responses. For reference, although we observed a high correlation between the three information matrices and the number of responses (gesture entropy & gesture response number: r = 0.976, p < 0.001; speech entropy & speech response number: r = 0.961, p < 0.001; MI & total response number: r = 0.818, p < 0.001), it is crucial to emphasize that these metrics capture different aspects of the semantic information represented. In the revised manuscript, we have provided a table detailing both entropy and response numbers for each stimulus, to allow for greater transparency and clarity.

Furthermore, we have added a comprehensive description of the partial correlation analysis conducted across all three experiments in the methodology section: for Experiment 1, please refer to Lines 213–222: ‘To account for potential confounds related to multiple candidate representations, we conducted partial correlation analyses between the tDCS effects and gesture entropy, speech entropy, and MI, controlling for the number of responses provided for each gesture and speech, as well as the total number of combined responses. Given that HD-tDCS induces overall disruption at the targeted brain regions, we hypothesized that the neural activity within the left IFG and pMTG would be progressively affected by varying levels of multisensory convergence, as indexed by MI. Moreover, we hypothesized that the modulation of neural activity by MI would differ between the left IFG and pMTG, as reflected in the differential modulation of response numbers in the partial correlations, highlighting their distinct roles in semantic processing[37].’

Experiment 2: ‘To control for potential confounds, partial correlations were also performed between the TMS effects and gesture entropy, speech entropy, and MI, controlling for the number of responses for each gesture and speech, as well as the total number of combined responses. By doing this, we can determine how the time-sensitive contribution of the left IFG and pMTG to gesture–speech integration was affected by gesture and speech information distribution.’ (Lines 242–246).

Experiment 3: ‘Additionally, partial correlations were conducted, accounting for the number of responses for each respective metric’ (Lines 292–293).

As anticipated by the reviewer, we observed a consistent modulation of response numbers across both regions as well as across the four ERP components and associated clusters. The detailed results are presented below:

Experiment 1: ‘However, partial correlation analysis, controlling for the total response number, revealed that the initially significant correlation between the Cathodal-tDCS effect and MI was no longer significant (r = -0.303, p = 0.222, 95% CI = [-0.770, 0.164]). This suggests that the observed relationship between Cathodal-tDCS and MI may be confounded by semantic control difficulty, as reflected by the total number of responses. Specifically, the reduced activity in the IFG under Cathodal-tDCS may be driven by variations in the difficulty of semantic control rather than a direct modulation of MI.’ (Lines 310-316) and ‘’Importantly, the reduced activity in the pMTG under Cathodal-tDCS was not influenced by the total response number, as indicated by the non-significant correlation (r = -0.253, p = 0.295, 95% CI = [-0.735, 0.229]). This finding was further corroborated by the unchanged significance in the partial correlation between Cathodal-tDCS and MI, when controlling for the total response number (r = -0.472, p = 0.048, 95% CI = [-0.903, -0.041]). (Lines 324-328).

Experiment 2:’ Notably, inhibition of pMTG activity in TW2 was not influenced by the number of speech responses (r = -0.539, p = 0.087, 95% CI = [-1.145, 0.067]). However, the number of speech responses did affect the modulation of speech entropy on the pMTG inhibition effect in TW2. This was evidenced by the non-significant partial correlation between pMTG inhibition and speech entropy when controlling for speech response number (r = -0.218, p = 0.545, 95% CI = [-0.563, 0.127]).

In contrast, the interrupted IFG activity in TW6 appeared to be consistently influenced by the confound of semantic control difficulty. This was reflected in the significant correlation with both gesture response number (r = -0.480, p = 0.032, 95% CI = [-904, -0.056]), speech response number (r = -0.729, p = 0.011, 95% CI = [-1.221, -0.237]), and total response number (r = -0.591, p = 0.008, 95% CI = [-0.993, -0.189]). Additionally, partial correlation analyses revealed non-significant relationship between interrupted IFG activity in TW6 and gesture entropy (r = -0.369, p = 0.120, 95% CI = [-0.810, -0.072]), speech entropy (r = -0.455, p = 0.187, 95% CI = [-1.072, 0.162]), and MI (r = -0.410, p = 0.091, 95% CI = [-0.856, -0.036]) when controlling for response numbers.’ (Lines 349-363)

Experiment 3: ‘To clarify potential confounds of semantic control difficulty, partial correlation analyses were conducted to examine the relationship between the elicited ERP components and the relevant information matrices, controlling for response numbers. Results consistently indicated modulation by response numbers in the relationship of ERP components with the information matrix, as evidenced by the non-significant partial correlations between the P1 amplitude (P1 component over ML: r = -0.574, p = 0.082, 95% CI = [-1.141, -0.007]) and the P1 cluster (r = -0.503, p = 0.138, 95% CI = [-1.102, 0.096]) with speech entropy; the N1-P2 amplitude (N1-P2 component over LA: r = -0.080, p = 0.746, 95% CI = [-0.554, 0.394]) and N1-P2 cluster (r \= -0.179, p = 0.464, 95% CI = [-0.647, 0.289]) with gesture entropy; the N400 amplitude (N400 component over LA: r = 0.264, p = 0.247, 95% CI = [-0.195,0.723]) and N400 cluster (r = 0.394, p = 0.095, 95% CI = [-0.043, 0.831]) with gesture entropy; the N400 amplitude (N400 component over LA: r = -0.134, p = 0.595, 95% CI = [-0.620, 0.352]) and N400 cluster (r = -0.034, p = 0.894, 95% CI = [-0.524,0.456]) with MI; and the LPC amplitude (LPC component over LA: r \= -0.428, p = 0.217, 95% CI = [-1.054, 0.198]) and LPC cluster (r \= -0.202, p = 0.575, 95% CI = [-0.881, 0.477]) with speech entropy.’ (Lines 424-438)

Based on the above results, we conclude that there is a dynamic interplay between the difficulty of semantic representation and the control pressures that shape the resulting neural responses. Furthermore, while the role of the IFG in control processes remains consistent, the present study reveals a more segmented role for the pMTG. Specifically, although the pMTG is well-established in the processing of distributed speech information, the integration of multisensory convergence, as indexed by MI, did not elicit the same control-related modulation in pMTG activity. A comprehensive discussion of the control process in shaping neural responses, as well as the specific roles of the IFG and pMTG in this process, is provided in the Discussion section in Lines (493-511): ‘Given that control processes are intrinsically integrated with semantic processing50, a distributed semantic representation enables dynamic modulation of access to and manipulation of meaningful information, thereby facilitating flexible control over the diverse possibilities inherent in a concept. Accordingly, an increased number of candidate responses amplifies the control demands necessary to resolve competing semantic representations. This effect was observed in the present study, where the association of the information matrix with the tDCS effect in IFG, the inhibition of pMTG activity in TW2, disruption of IFG activity in TW6, and modulation of four distinct ERP components collectively demonstrated that response quantity modulated neural activity. These results underscore the intricate interplay between the difficulty of semantic representation and the control pressures that shape the resulting neural responses. 

The IFG and pMTG, central components of the semantic control network, have been extensively implicated in previous research 50-52. While the role of the IFG in managing both unisensory information and multisensory convergence remains consistent, as evidenced by the confounding difficulty results across Experiments 1 and 2, the current study highlights a more context-dependent function for the pMTG. Specifically, although the pMTG is well-established in the processing of distributed speech information, the multisensory convergence, indexed by MI, did not evoke the same control-related modulation in pMTG activity. These findings suggest that, while the pMTG is critical to semantic processing, its engagement in control processes is likely modulated by the specific nature of the sensory inputs involved’

Reference:

Tesink, C.M.J.Y., Petersson, K.M., van Berkum, J.J.A., van den Brink, D., Buitelaar, J.K., and Hagoort, P. (2009). Unification of speaker and meaning in language comprehension: An fMRI study. J Cognitive Neurosci 21, 2085-2099. 10.1162/jocn.2008.21161

Jackson, R.L. (2021). The neural correlates of semantic control revisited. Neuroimage 224, 117444. 10.1016/j.neuroimage.2020.117444.

Jefferies, E. (2013). The neural basis of semantic cognition: converging evidence from neuropsychology, neuroimaging and TMS. Cortex 49, 611-625. 10.1016/j.cortex.2012.10.008.

Noonan, K.A., Jefferies, E., Visser, M., and Lambon Ralph, M.A. (2013). Going beyond inferior prefrontal involvement in semantic control: evidence for the additional contribution of dorsal angular gyrus and posterior middle temporal cortex. J Cogn Neurosci 25, 1824-1850. 10.1162/jocn_a_00442.

In terms of conceptual equivocation, the use of the term 'graded' by the authors seems to be different from the usage commonly employed in the semantic cognition literature (e.g., the 'graded hub hypothesis', Rice et al., 2015). The idea of a graded hub in the controlled semantic cognition framework (i.e., the anterior temporal lobe) refers to a progressive degree of abstraction or heteromodal information as you progress through the anatomy of the region (i.e., along the dorsal-to-ventral axis). The authors, on the other hand, seem to refer to 'graded manner' in the context of a correlation of entropy or MI and the change in the difference between Reaction Times (RTs) of semantically congruent vs incongruent gesture-speech. The issue is that the discourse through parts of the introduction and discussion seems to conflate both interpretations, and the ideas in the main text do not correspond to the references they cite. This is not overall very convincing. What is it exactly the authors are arguing about the correlation between RTs and MI indexes? As stated above, their measure of entropy captures the spread of responses, which could also be a measure of item difficulty (more diverse responses imply fewer correct responses, a classic index of difficulty). Capturing the diversity of responses means that items with high entropy scores are also likely to have multiple candidate representations, leading to increased selection pressures. Regions like pMTG and IFG have been widely implicated in difficult semantic processing and increased selection pressures (Jackson et al., 2021). How is this MI correlation evidence of integration that proceeds in a 'graded manner'? The conceptual links between these concepts must be made clearer for the interpretation to be convincing.

Response 2: Regarding the concern of conceptual equivocation, we would like to emphasize that this study represents the first attempt to focus on the relationship between information quantity and neural engagement, a question addressed in three experiments. Experiment 1 (HD-tDCS) targeted the entire gesture-speech integration process in the IFG and pMTG to assess whether neural activity in these regions, previously identified as integration hubs, is modulated by changes in informativeness from both modalities (i.e., entropy) and their interactions (MI). The results revealed a gradual inhibition of neural activity in both areas as MI increased, evidenced by a negative correlation between MI and the tDCS inhibition effect in both regions. Building on this, Experiments 2 and 3 employed double-pulse TMS and ERPs to further assess whether the engaged neural activity was both time-sensitive and staged. These experiments also evaluated the contributions of various sources of information, revealing correlations between information-theoretic metrics and time-locked brain activity, providing insights into the ‘gradual’ nature of gesture-speech integration.

Therefore, the incremental engagement of the integration hub of IFG and pMTG along with the informativeness of gesture and speech during multisensory integration is different from the "graded hub," which refers to anatomical distribution. We sincerely apologize for this oversight. In the revised manuscript, we have changed the relevant conceptual equivocation in Lines 44-60: ‘Consensus acknowledges the presence of 'convergence zones' within the temporal and inferior parietal areas [1], or the 'semantic hub' located in the anterior temporal lobe[2], pivotal for integrating, converging, or distilling multimodal inputs. Contemporary theories frame the semantic processing as a dynamic sequence of neural states[3], shaped by systems that are finely tuned to the statistical regularities inherent in sensory inputs[4]. These regularities enable the brain to evaluate, weight, and integrate multisensory information, optimizing the reliability of individual sensory signals[5]. However, sensory inputs available to the brain are often incomplete and uncertain, necessitating adaptive neural adjustments to resolve these ambiguities [6]. In this context, neuronal activity is thought to be linked to the probability density of sensory information, with higher levels of uncertainty resulting in the engagement of a broader population of neurons, thereby reflecting the brain’s adaptive capacity to handle diverse possible interpretations[7,8]. Although the role of 'convergence zones' and 'semantic hubs' in integrating multimodal inputs is well established, the precise functional patterns of neural activity in response to the distribution of unified multisensory information—along with the influence of unisensory signals—remain poorly understood.

To this end, we developed an analytic approach to directly probe the cortical engagement during multisensory gesture-speech semantic integration.’  

Furthermore, in the Discussion section, we have replaced the term 'graded' with 'incremental' (Line 456,). Additionally, we have included a discussion on the progressive nature of neural engagement, as evidenced by the correlation between RTs and MI indices in Lines 483-492: ‘The varying contributions of unisensory gesture-speech information and the convergence of multisensory inputs, as reflected in the correlation between distinct ERP components and TMS time windows (TMS TWs), are consistent with recent models suggesting that multisensory processing involves parallel detection of modality-specific information and hierarchical integration across multiple neural levels[4,48]. These processes are further characterized by coordination across multiple temporal scales[49]. Building on this, the present study offers additional evidence that the multi-level nature of gesture-speech processing is statistically structured, as measured by information matrix of unisensory entropy and multisensory convergence index of MI, the input of either source would activate a distributed representation, resulting in progressively functioning neural responses.’

Reference:

Damasio, H., Grabowski, T.J., Tranel, D., Hichwa, R.D., and Damasio, A.R. (1996). A neural basis for lexical retrieval. Nature 380, 499-505. DOI 10.1038/380499a0.

Patterson, K., Nestor, P.J., and Rogers, T.T. (2007). Where do you know what you know? The representation of semantic knowledge in the human brain. Nature Reviews Neuroscience 8, 976-987. 10.1038/nrn2277.

Brennan, J.R., Stabler, E.P., Van Wagenen, S.E., Luh, W.M., and Hale, J.T. (2016). Abstract linguistic structure correlates with temporal activity during naturalistic comprehension. Brain and Language 157, 81-94. 10.1016/j.bandl.2016.04.008.

Benetti, S., Ferrari, A., and Pavani, F. (2023). Multimodal processing in face-to-face interactions: A bridging link between psycholinguistics and sensory neuroscience. Front Hum Neurosci 17, 1108354. 10.3389/fnhum.2023.1108354.

Noppeney, U. (2021). Perceptual Inference, Learning, and Attention in a Multisensory World. Annual Review of Neuroscience, Vol 44, 2021 44, 449-473. 10.1146/annurev-neuro-100120-085519.

Ma, W.J., and Jazayeri, M. (2014). Neural coding of uncertainty and probability. Annu Rev Neurosci 37, 205-220. 10.1146/annurev-neuro-071013-014017.

Fischer, B.J., and Pena, J.L. (2011). Owl's behavior and neural representation predicted by Bayesian inference. Nat Neurosci 14, 1061-1066. 10.1038/nn.2872.

Ganguli, D., and Simoncelli, E.P. (2014). Efficient sensory encoding and Bayesian inference with heterogeneous neural populations. Neural Comput 26, 2103-2134. 10.1162/NECO_a_00638.

Meijer, G.T., Mertens, P.E.C., Pennartz, C.M.A., Olcese, U., and Lansink, C.S. (2019). The circuit architecture of cortical multisensory processing: Distinct functions jointly operating within a common anatomical network. Prog Neurobiol 174, 1-15. 10.1016/j.pneurobio.2019.01.004.

Senkowski, D., and Engel, A.K. (2024). Multi-timescale neural dynamics for multisensory integration. Nat Rev Neurosci 25, 625-642. 10.1038/s41583-024-00845-7.

Reviewer #2 (Recommendations for the authors):

I have a number of small suggestions to make the paper more easy to understand.

We sincerely thank the reviewer for their careful reading and thoughtful consideration. All suggestions have been thoroughly addressed and incorporated into the revised manuscript.

(1) Lines 86-87, please clarify whether "chronometric double-pulse TMS" should lead to either excitation or inhibition of neural activities

Double-pulse TMS elicits inhibition of neural activities (see responses to editors), which has been clarified in the revised manuscript in Lines 90-93: ‘we applied inhibitory chronometric double-pulse transcranial magnetic stimulation (TMS) to specific temporal windows associated with integration processes in these regions[23], assessing whether the inhibitory effects of TMS were correlated with unisensory entropy or the multisensory convergence index (MI)’

(2) Line 106 "validated by replicating the semantic congruencey effect". Please specify what the task was in the validation study.

The description of the validation task has been added in Lines 116-119: ‘To validate the stimuli, 30 participants were recruited to replicate the multisensory index of semantic congruency effect, hypothesizing that reaction times for semantically incongruent gesture-speech pairs would be significantly longer than those for congruent pairs.’

(3) Line 112. "30 subjects". Are they Chinese speakers?

Yes, all participants in the present study, including those in the pre-tests, are native Chinese speakers.

(4) Line 122, "responses for each item" Please specify whether you mean here "the comprehensive answer" as you defined in 118-119.

Yes, and this information has been added in Lines 136-137: ‘comprehensive responses for each item were converted into Shannon's entropy (H)’

(5) Line 163 "one of three stimulus types (Anodal, Cathodal or Sham)". Please specify whether the order of the three conditions was counterbalanced across participants. Or, whether the order was fixed for all participants.

The order of the three conditions was counterbalanced across participants, a clearer description has been added in the revised manuscript in Lines 184-189: ‘Participants were divided into two groups, with each group undergoing HD-tDCS stimulation at different target sites (IFG or pMTG). Each participant completed three experimental sessions, spaced one week apart, during which 480 gesture-speech pairs were presented across various conditions. In each session, participants received one of three types of HD-tDCS stimulation: Anodal, Cathodal, or Sham. The order of stimulation site and type was counterbalanced using a Latin square design to control for potential order effects.’

(6) Line 191-192, "difference in reaction time between semantic incongruence and semantic congruent pairs)" Here, please specify which reaction time was subtracted from which one. This information is very crucial; without it, you cannot interpret your graphs.

(17) Figure 3. Figure caption for (A). "The semantic congruence effect was calculated as the reaction time difference between...". You need to specify which condition was subtracted from what condition; otherwise, you cannot interpret this figure. "difference" is too ambiguous.

Corrections have been made in the revised manuscript in Lines 208-211: ‘Neural responses were quantified based on the effects of HD-tDCS (active tDCS minus sham tDCS) on the semantic congruency effect, defined as the difference in reaction times between semantic incongruent and congruent conditions (Rt(incongruent) - Rt(congruent))’ and Line 796-798: ‘The semantic congruency effect was calculated as the reaction time (RT) difference between semantically incongruent and semantically congruent pairs (Rt(incongruent) - Rt(congruent))’.

(7) Line 363 "progressive inhibition of IFG and pMTG by HD-tDCS as the degree of gesture-speech interaction, indexed by MI, advanced." This sentence is very hard to follow. I don't understand what part of the data in Figure 3 speaks to "inhibition of IFG". And what is "HD-tDCS"? I think it is easier to read if you talk about correlation (not "progressive" and "advanced").

High-Definition transcranial direct current stimulation (HD-tDCS) was applied to modulate the activity of pMTG and IFG, with cathodal stimulation inducing inhibitory effects and anodal stimulation facilitating neural activity. In Figure 3, we examined the relationship between the tDCS effects on pMTG and IFG and the three information matrices (entropy and MI). Our results revealed significant correlations between MI and the cathodal-tDCS effects in both regions. We acknowledge that the original phrasing may have been unclear, and in the revised manuscript, we have provided a more explicit explanation to enhance clarity in Lines 443-445: ‘Our results, for the first time, revealed that the inhibition effect of cathodal-tDCS on the pMTG and IFG correlated with the degree of gesture-speech multisensory convergence, as indexed by MI’.

(8) Lines 367-368 I don't understand why gesture is top down and speech is bottom up. Is that because gesture precedes speech (gesture is interpretable at the point of speech onset)?

Yes, since we employed a semantic priming paradigm by aligning speech onset with the gesture comprehension point, we interpret the gesture-speech integration process as an interaction between the top-down prediction from gestures and the bottom-up processing of speech. In the revised manuscript, we have provided a clearer and more coherent description that aligns with the results. Lines 445-449: ‘Moreover, the gradual neural engagement was found to be time-sensitive and staged, as evidenced by the selectively interrupted time windows (Experiment 2) and the distinct correlated ERP components (Experiment 3), which were modulated by different information contributors, including unisensory entropy or multisensory MI’

(9) Line 380 - 381. Can you spell out "TW" and "IP"?

(16) Line 448, NIBS, Please spell out "NIBS".

"TW" have been spelled out in Lines 459: ‘time windows (TW)’,"IP" in Line 460: ‘identification point (IP)’. The term "NIBS" was replaced with "HD-tDCS and TMS" to provide clearer specification of the techniques employed: ‘Consistent with this, the present study provides robust evidence, through the application of HD-tDCS and TMS, that the integration hubs for gesture and speech—the pMTG and IFG—operate in an incremental manner.’ (Lines 454-457). 

(10) Line 419, The higher certainty of gesture => The higher the certainty of gesture is

(13) Line 428, "a larger MI" => "a larger MI is"

(12) Line 427-428, "the larger overlapped neural populations" => "the larger, the overlapped neural populations"

Changes have been made in Line 522 ‘The higher the certainty of gesture is’ , Line 531: ‘a larger MI is’ and Line 530 ‘the larger, overlapped neural populations’

(11) Line 423 "Greater TMS effect over the IFG" Can you describe the TMS effect?

TMS effect has been described as ‘Greater TMS inhibitory effect’ (Line 526)

(14) Line 423 "reweighting effect" What is this? Please describe (and say which experiment it is about).

Clearer description has been provided in Lines 535-538: ‘As speech entropy increases, indicating greater uncertainty in the information provided by speech, more cognitive effort is directed towards selecting the targeted semantic representation. This leads to enhanced involvement of the IFG and a corresponding reduction in LPC amplitude’.

(15) Line 437 "the graded functionality of every disturbed period is not guaranteed" (I don't understand this sentence).

Clearer description has been provided in Lines 552-557: ‘Additionally, not all influenced TWs exhibited significant associations with entropy and MI. While HD-tDCS and TMS may impact functionally and anatomically connected brain regions[55,56], whether the absence of influence in certain TWs can be attributed to compensation by other connected brain areas, such as angular gyrus[57] or anterior temporal lobe[58], warrants further investigation. Therefore, caution is needed when interpreting the causal relationship between inhibition effects of brain stimulation and information-theoretic metrics (entropy and MI).’

References:

Humphreys, G. F., Lambon Ralph, M. A., & Simons, J. S. (2021). A Unifying Account of Angular Gyrus Contributions to Episodic and Semantic Cognition. Trends in neurosciences, 44(6), 452–463. https://doi.org/10.1016/j.tins.2021.01.006

Bonner, M. F., & Price, A. R. (2013). Where is the anterior temporal lobe and what does it do?. The Journal of neuroscience : the official journal of the Society for Neuroscience, 33(10), 4213–4215. https://doi.org/10.1523/JNEUROSCI.0041-13.2013

(18) Figure 4. "TW1", "TW2", etc. are not informative. Either replace them with the actual manuscript or add manuscript information (either in the graph itself or in the figure title).

Information was added into the figure title ‘Figure 4. TMS impacts on semantic congruency effect across various time windows (TW).’ (Line 804), included a detailed description of each time window in Lines 805-807: ‘(A) Five time windows (TWs) showing selective disruption of gesture-speech integration were chosen: TW1 (-120 to -80 ms relative to speech identification point), TW2 (-80 to -40 ms), TW3 (-40 to 0 ms), TW6 (80 to 120 ms), and TW7 (120 to 160 ms).’

(19) Table 2C.

The last column is titled "p(xi, yi)". I don't understand why the authors use this label for this column.

In the formula, at the very end, there is "p(xi|yi). I wonder why it is p(xi|yi), as opposed to p(yi|xi).

Mutual Information (MI) was calculated by subtracting the entropy of the combined gesture-speech dataset (Entropy(gesture + speech)) from the sum of the individual entropies of gesture and speech (Entropy(gesture) + Entropy(speech)). Thus, the p(xi,yi) aimed to describe the entropy of the combined dataset. We acknowledge the potential ambiguity in the original description, and in the revised manuscript, we have changed the formula of p(xi,yi) into ‘p(xi+yi)’ (Line 848) in Table 2C, and the relevant equation of MI ‘’. Also we provided a clear MI calculation process in Lines 143-146: ‘MI was used to measure the overlap between gesture and speech information, calculated by subtracting the entropy of the combined gesture-speech dataset (Entropy(gesture + speech)) from the sum of their individual entropies (Entropy(gesture) + Entropy(speech)) (see Appendix Table 2C)’.

Reviewer #3 (Recommendations for the authors):

(1) The authors should try and produce data showing that the confound of difficulty due to the number of lexical or semantic representations is not underlying high-entropy items if they wish to improve the credibility of their claim that the disruption of the congruency effect is due to speech-gesture integration. Additionally, they should provide more evidence either in the form of experiments or references to better justify why mutual information is an index for integration in the first place.

Response 1: An additional analysis has been conducted to assess whether the number of lexical or semantic representations affect the neural outcomes, please see details in the Responses to Reviewer 3 (public review) response 1.

Mutual information (MI), a concept rooted in information theory, quantifies the reduction in uncertainty about one signal when the other is known, thereby capturing the statistical dependence between them. MI is calculated as the difference between the individual entropies of each signal and their joint entropy, which reflects the total uncertainty when both signals are considered together. This metric aligns with the core principle of multisensory integration: different modalities reduce uncertainty about each other by providing complementary, predictive information. Higher MI values signify that the integration of sensory signals results in a more coherent and unified representation, while lower MI values indicate less integration or greater divergence between the modalities. As such, MI serves as a robust and natural index for assessing the degree of multisensory integration.

To date, the use of MI as an index of integration has been limited, with one notable study by Tremblay et al. (2016), cited in the manuscript, using pointwise MI to quantify the extent to which two syllables mutually constrain each other. While MI has been extensively applied in natural language processing to measure the co-occurrence strength between words (e.g., Lin et al., 2012), its application as an index of multisensory convergence—particularly in the context of gesture-speech integration as employed in this study—is novel. In the revised manuscript, we have clarified the relationship between MI and multisensory convergence: ‘MI assesses share information between modalities[25],indicating multisensory convergence and acting as an index of gesture-speech integration’ (Lines 73-74).

Also, in our study, we calculated MI as per its original definition, by subtracting the entropy of summed dataset of gesture-speech from the combined entropies of gesture and speech. The detailed calculation method is provided in Lines 136-152: ‘To quantify information content, comprehensive responses for each item were converted into Shannon's entropy (H) as a measure of information richness (Figure 1A bottom). With no significant gender differences observed in both gesture (t(20) = 0.21, p = 0.84) and speech (t(20) = 0.52, p = 0.61), responses were aggregated across genders, resulting in 60 answers per item (Appendix Table 2). Here, p(xi) and p(yi) represent the distribution of 60 answers for a given gesture (Appendix Table 2B) and speech (Appendix Table 2A), respectively. High entropy indicates diverse answers, reflecting broad representation, while low entropy suggests focused lexical recognition for a specific item (Figure 2B). MI was used to measure the overlap between gesture and speech information, calculated by subtracting the entropy of the combined gesture-speech dataset (Entropy(gesture + speech)) from the sum of their individual entropies (Entropy(gesture) + Entropy(speech)) (see Appendix Table 2C). For specific gesture-speech combinations, equivalence between the combined entropy and the sum of individual entropies (gesture or speech) indicates absence of overlap in response sets. Conversely, significant overlap, denoted by a considerable number of shared responses between gesture and speech datasets, leads to a noticeable discrepancy between combined entropy and the sum of gesture and speech entropies. Elevated MI values thus signify substantial overlap, indicative of a robust mutual interaction between gesture and speech.’

Additional examples outlined in Appendix Table 2 in Lines 841-848:

This novel application of MI as a multisensory convergence index offers new insights into how different sensory modalities interact and integrate to shape semantic processing.

Reference:

Tremblay, P., Deschamps, I., Baroni, M., and Hasson, U. (2016). Neural sensitivity to syllable frequency and mutual information in speech perception and production. Neuroimage 136, 106-121. 10.1016/j.neuroimage.2016.05.018

Lin, W., Wu, Y., & Yu, L. (2012). Online Computation of Mutual Information and Word Context Entropy. International Journal of Future Computer and Communication, 167-169.

(2) Finally, if the authors wish to address the graded hub hypothesis as posited by the controlled semantic cognition framework (e.g., Rice et al., 2015), they would have to stimulate a series of ROIs progressing gradually through the anatomy of their candidate regions showing the effects grow along this spline, more than simply correlate MI with RT differences.

Response 2: We appreciate the reviewer’s thoughtful consideration. The incremental engagement of the integration hub of IFG and pMTG along with the informativeness of gesture and speech during multisensory integration is different from the concept of "graded hub," which refers to anatomical distribution. See Responses to reviewer 3 (public review) response 2 for details.

(3) The authors report significant effects with p values as close to the threshold as p=0.49 for the pMTG correlation in Experiment 1, for example. How confident are the authors these results are reliable and not merely their 'statistical luck'? Especially in view of sample sizes that hover around 22-24 participants, which have been called into question in the field of non-invasive brain stimulation (e.g., Mitra et al, 2021)?

Response 3: In Experiment 1, a total of 52 participants were assigned to two groups, each undergoing HD-tDCS stimulation over either the inferior frontal gyrus (IFG) or posterior middle temporal gyrus (pMTG), yielding 26 participants per group for correlation analysis. Power analysis, conducted using G*Power, indicated that a sample size of 26 participants per group would provide sufficient power (0.8) to detect a large effect size (0.5) at an alpha level of 0.05, justifying the chosen sample size. To control for potential statistical artifacts, we compared the results to those from the unaffected control condition.

In the Experiment 1, participants were tasked with a gender categorization task, where they responded as accurately and quickly as possible to the gender of the voice they saw, while gender congruency (e.g., a male gesture paired with a male voice or a female gesture with a male voice) was manipulated. This manipulation served as direct control, enabling the investigation of automatic and implicit semantic interactions between gesture and speech. This relevant information was provided in the manuscript in Lines 167-172:‘An irrelevant factor of gender congruency (e.g., a man making a gesture combined with a female voice) was created[22,23,35]. This involved aligning the gender of the voice with the corresponding gender of the gesture in either a congruent (e.g., male voice paired with a male gesture) or incongruent (e.g., male voice paired with a female gesture) manner. This approach served as a direct control mechanism, facilitating the investigation of the automatic and implicit semantic interplay between gesture and speech[35]’. Correlation analyses were conducted to examine the TMS disruption effects on gender congruency, comparing reaction times for gender-incongruent versus congruent trials. No significant correlations were found between TMS disruption effects on either the IFG (Cathodal-tDCS effect with MI: r = 0.102, p = 0.677; Anodal-tDCS effect with MI: r = 0.178, p = 0.466) or pMTG (Cathodal-tDCS effect with MI: r \= -0.201, p = 0.410; Anodal-tDCS effect with MI: r = -0.232, p = 0.338).

Moreover, correlations between the TMS disruption effect on semantic congruency and both gesture entropy, speech entropy, and mutual information (MI) were examined. P-values of 0.290, 0.725, and 0.049 were observed, respectively.  

The absence of a TMS effect on gender congruency, coupled with the lack of significance when correlated with the other information matrices, highlights the robustness of the significant finding at p = 0.049.

(4) The distributions of entropy for gestures and speech are very unequal. Whilst entropy for gestures has high variability, (.12-4.3), that of speech is very low (ceiling effect?) with low variance. Can the authors comment on whether they think this might have affected their analyses or results in any way? For example, do they think this could be a problem when calculating MI, which integrates both measures? L130-131.'

Response 4: We sincerely thank the reviewer for raising this insightful question. The core premise of the current study is that brain activity is modulated by the degree of information provided. Accordingly, the 20 entropy values for gesture and speech represent a subset of the overall entropy distribution, with the degree of entropy correlating with a distributed pattern of neural activity, regardless of the scale of variation. This hypothesis aligns with previous studies suggesting that neuronal activity is linked to the probability density of sensory information, with higher levels of uncertainty resulting in the engagement of a broader population of neurons, thereby reflecting the brain’s adaptive capacity to handle diverse possible interpretations (Fischer & Pena, 2011; Ganguli & Simoncelli, 2014).

Importantly, we conducted another EEG experiment with 30 subjects. Given the inherent differences between gesture and speech, it is important to note that speech, being more structurally distinct, tends to exhibit lower variability than gesture. To prevent an imbalance in the distribution of gesture and speech, we manipulated the information content of each modality. Specifically, we created three conditions for both gesture and speech (i.e., 0.75, 1, and 1.25 times the identification threshold), thereby ensuring comparable variance between the two modalities: gesture (mean entropy = 2.91 ± 1.01) and speech (mean entropy = 1.82 ± 0.71) (Author response table 6).

Full-factorial RSA analysis revealed an early P1 effect (0-100 ms) for gesture and a late LPC effect (734-780 ms) for speech (Author response image 2b). Crucially, the identified clusters showed significant correlations with both gesture (Author response image 2c1) and speech entropy (Author response image 2c3), respectively. These findings replicate the results of the present study, demonstrating that, irrespective of the variance in gesture and speech entropy, both modalities elicited ERP amplitude responses in a progressive manner that aligned with their respective information distributions.

Regarding the influence on MI values, since MI was calculated based on the overlapping responses between gesture and speech, a reduction in uncertainty during speech comprehension would naturally result in a smaller contribution to the MI value. However, as hypothesized above, the MI values were also assumed to represent a subset of the overall distribution, where the contributions of both gesture and speech are expected to follow a normal distribution. This hypothesis was further supported by our replication experiment. When the contributions of gesture and speech were balanced, a correlation between MI values and N400 amplitude was observed (Author response image 2c2), consistent with the results reported in the present manuscript. These findings not only support the idea that the correlation between MI and ERP components is unaffected by the subset of MI values but also confirm the replicability of our results.

Author response table 6.

Quantitative entropy for each gesture stimulus (BD: before discrimination point; DP: discrimination point; AD: after discrimination point) and speech stimulus (BI: before identification point; IP: identification point; AI: after identification point).

Author response image 2.

Results of group-level analysis and full-factorial RSA. a: The full-factorial representational similarity analysis (RSA) framework is illustrated schematically. Within the general linear model (GLM), the light green matrix denotes the representational dissimilarity matrix (RDM) for gesture semantic states, while light blue matrix represents speech semantic states, and the light red matrix illustrates the semantic congruency effect. The symbol ‘e’ indicates the random error term. All matrices, including the neural dissimilarity matrix, are structured as 18 * 18 matrices, corresponding to 18 conditions (comprising 3 gesture semantic states, 3 speech semantic states, and 2 congruency conditions). b: Coding strength for gesture states, speech states and congruency effect. Shaded clusters represent regions where each factor exhibited significant effects. Clusters with lower opacity correspond to areas where the grand-mean ERP amplitudes across conditions showed the highest correlation with unimodal entropy or MI. c1-c6: Topographical correlation maps illustrate the four significant RSA clusters (top), accompanied by the highest correlations between ERP amplitudes within the significant RSA clusters and the information matrices (bottom). Black dots represent electrodes exhibiting significant correlations, while black stars highlight the electrode with the highest correlation coefficient.

(5) L383: Why are the authors calling TW2 pre-lexical and TW6 post-lexical? I believe they must provide evidence or references justifying calling these periods pre- and post-lexical. This seems critical given the argument they're trying to make in this paragraph.

Response 5: The time windows (TWs) selected for the current study were based on our previous work (Zhao et al., 2021, J. Neurosci). In that study, we employed a double-pulse TMS protocol, delivering stimulation across eight 40-ms time windows: three windows preceding the speech identification point (TWs 1-3) and five windows following it (TWs 4-8). The pre-lexical time windows (TWs 1-3) occur before speech identification, while the post-lexical time windows (TWs 4-8) occur after this point. in the revised manuscript, we have made that clear in Lines 462-466:

“In TW2 of gesture-speech integration, which precedes the speech identification point23 and represents a pre-lexical stage, the suppression effect observed in the pMTG was correlated with speech entropy. Conversely, during TW6, which follows the speech identification point23 and represents a post-lexical stage, the IFG interruption effect was influenced by both gesture entropy, speech entropy, and their MI”

Reference:

Zhao, W., Li, Y., and Du, Y. (2021). TMS reveals dynamic interaction between inferior frontal gyrus and posterior middle temporal gyrus in gesture-speech semantic integration. The Journal of Neuroscience, 10356-10364. 10.1523/jneurosci.1355-21.2021.

(6) Below, I recommend the authors improve their description of the criteria employed to select ROIs. This is important for several reasons. For example, the lack of a control ROI presumably not implicated in integration makes the interpretation of the specificity of the results difficult. Additionally, other regions have been proposed more consistently by recent evidence as multimodal integrators, like for example, the angular gyrus (Humphreys, 2021), or the anterior temporal lobe. The inclusion of IFG as a key region for integration and the oversight of angular gyrus seems to me unjustified in the light of recent evidence.

Response 6: We appreciate the reviewer’s thoughtful consideration. The selection of IFG and pMTG as ROIs was based on a meta-analysis of multiple fMRI studies on gesture-speech integration, in which these two locations were consistently identified as activated. See Table 2 for details of the studies and coordinates of brain locations reported.

Author response table 7.

Meta-analysis of previous studies on gesture-speech integration.

Based on the meta-analysis of previous studies, we selected the IFG and pMTG as ROIs for gesture-speech integration. The rationale for selecting these brain regions is outlined in the introduction in Lines 65-68: ‘Empirical studies have investigated the semantic integration between gesture and speech by manipulating their semantic relationship[15-18] and revealed a mutual interaction between them[19-21] as reflected by the N400 latency and amplitude[14] as well as common neural underpinnings in the left inferior frontal gyrus (IFG) and posterior middle temporal gyrus (pMTG)[15,22,23]’.

And further described in Lines 79-80: ‘_Experiment 1 employed high-definition transcranial direct current stimulation (HD-tDCS) to administer Anodal, Cathodal and Sham stimulation to either the IFG or the pMTG ’._ And Lines 87-90: ‘Given the differential involvement of the IFG and pMTG in gesture-speech integration, shaped by top-down gesture predictions and bottom-up speech processing [23], Experiment 2 was designed to assess whether the activity of these regions was associated with relevant informational matrices’.

In the Methods section, we clarified the selection of coordinates in Lines 193-199: ‘Building on a meta-analysis of prior fMRI studies examining gesture-speech integration[22], we targeted Montreal Neurological Institute (MNI) coordinates for the left IFG at (-62, 16, 22) and the pMTG at (-50, -56, 10). In the stimulation protocol for HD-tDCS, the IFG was targeted using electrode F7 as the optimal cortical projection site[36], with four return electrodes placed at AF7, FC5, F9, and FT9. For the pMTG, TP7 was selected as the cortical projection site36, with return electrodes positioned at C5, P5, T9, and P9.’

The selection of IFG or pMTG as integration hubs for gesture and speech has also been validated in our previous studies. Specifically, Zhao et al. (2018, J. Neurosci) applied TMS to both areas. Results demonstrated that disrupting neural activity in the IFG or pMTG via TMS selectively impaired the semantic congruency effect (reaction time costs due to semantic incongruence), while leaving the gender congruency effect unaffected. These findings identified the IFG and pMTG as crucial hubs for gesture-speech integration, guiding the selection of brain regions for our subsequent studies.

In addition, Zhao et al. (2021, J. Neurosci) employed a double-pulse TMS protocol across eight 40-ms time windows to explore the temporal dynamics of the IFG and pMTG. The results revealed time-window-selective disruptions of the semantic congruency effect, further supporting the dynamic and temporally staged involvement of these regions in gesture-speech integration.

While we have solid rationale for selecting the IFG and pMTG as key regions, we acknowledge the reviewer's point that the involvement of additional functionally and anatomically brain areas, cannot be excluded. We have included in the discussion as limitations in Lines 552-557: ‘Additionally, not all influenced TWs exhibited significant associations with entropy and MI. While HD-tDCS and TMS may impact functionally and anatomically connected brain regions[55,56], whether the absence of influence in certain TWs can be attributed to compensation by other connected brain areas, such as angular gyrus[57] or anterior temporal lobe[58], warrants further investigation. Therefore, caution is needed when interpreting the causal relationship between inhibition effects of brain stimulation and information-theoretic metrics (entropy and MI).’

References:

Willems, R.M., Ozyurek, A., and Hagoort, P. (2009). Differential roles for left inferior frontal and superior temporal cortex in multimodal integration of action and language. Neuroimage 47, 1992-2004. 10.1016/j.neuroimage.2009.05.066.

Drijvers, L., Jensen, O., and Spaak, E. (2021). Rapid invisible frequency tagging reveals nonlinear integration of auditory and visual information. Human Brain Mapping 42, 1138-1152. 10.1002/hbm.25282.

Drijvers, L., and Ozyurek, A. (2018). Native language status of the listener modulates the neural integration of speech and iconic gestures in clear and adverse listening conditions. Brain and Language 177, 7-17. 10.1016/j.bandl.2018.01.003.

Drijvers, L., van der Plas, M., Ozyurek, A., and Jensen, O. (2019). Native and non-native listeners show similar yet distinct oscillatory dynamics when using gestures to access speech in noise. Neuroimage 194, 55-67. 10.1016/j.neuroimage.2019.03.032.

Holle, H., and Gunter, T.C. (2007). The role of iconic gestures in speech disambiguation: ERP evidence. J Cognitive Neurosci 19, 1175-1192. 10.1162/jocn.2007.19.7.1175.

Kita, S., and Ozyurek, A. (2003). What does cross-linguistic variation in semantic coordination of speech and gesture reveal?: Evidence for an interface representation of spatial thinking and speaking. J Mem Lang 48, 16-32. 10.1016/S0749-596x(02)00505-3.

Bernardis, P., and Gentilucci, M. (2006). Speech and gesture share the same communication system. Neuropsychologia 44, 178-190. 10.1016/j.neuropsychologia.2005.05.007.

Zhao, W.Y., Riggs, K., Schindler, I., and Holle, H. (2018). Transcranial magnetic stimulation over left inferior frontal and posterior temporal cortex disrupts gesture-speech integration. Journal of Neuroscience 38, 1891-1900. 10.1523/Jneurosci.1748-17.2017.

Zhao, W., Li, Y., and Du, Y. (2021). TMS reveals dynamic interaction between inferior frontal gyrus and posterior middle temporal gyrus in gesture-speech semantic integration. The Journal of Neuroscience, 10356-10364. 10.1523/jneurosci.1355-21.2021.

Hartwigsen, G., Bzdok, D., Klein, M., Wawrzyniak, M., Stockert, A., Wrede, K., Classen, J., and Saur, D. (2017). Rapid short-term reorganization in the language network. Elife 6. 10.7554/eLife.25964.

Jackson, R.L., Hoffman, P., Pobric, G., and Ralph, M.A.L. (2016). The semantic network at work and rest: Differential connectivity of anterior temporal lobe subregions. Journal of Neuroscience 36, 1490-1501. 10.1523/JNEUROSCI.2999-15.2016.

Humphreys, G. F., Lambon Ralph, M. A., & Simons, J. S. (2021). A Unifying Account of Angular Gyrus Contributions to Episodic and Semantic Cognition. Trends in neurosciences, 44(6), 452–463. https://doi.org/10.1016/j.tins.2021.01.006

Bonner, M. F., & Price, A. R. (2013). Where is the anterior temporal lobe and what does it do?. The Journal of neuroscience : the official journal of the Society for Neuroscience, 33(10), 4213–4215. https://doi.org/10.1523/JNEUROSCI.0041-13.2013

(7) Some writing is obscure or unclear, in part due to superfluous words like 'intricate neural processes' on L74. Or the sentence in L47 - 48 about 'quantitatively functional mental states defined by a specific parser unified by statistical regularities' which, even read in context, fails to provide clarity about what a quantitatively functional mental state is, or how it is defined by specific parsers (or what these are), and what is the link to statistical regularities. In some cases, this lack of clarity leads to difficulties assessing the appropriateness of the methods, or the exact nature of the claims. For example, do they mean degree of comprehension instead of comprehensive value? I provide some more examples below:

Response 7: We appreciate the reviewer’s thoughtful consideration. The revised manuscript now includes a clear description and a detailed explanation of the association with the statistical logic, addressing the concerns raised in Lines 47-55: ‘Contemporary theories frame the semantic processing as a dynamic sequence of neural states[3], shaped by systems that are finely tuned to the statistical regularities inherent in sensory inputs[4]. These regularities enable the brain to evaluate, weight, and integrate multisensory information, optimizing the reliability of individual sensory signals [5]. However, sensory inputs available to the brain are often incomplete and uncertain, necessitating adaptive neural adjustments to resolve these ambiguities[6]. In this context, neuronal activity is thought to be linked to the probability density of sensory information, with higher levels of uncertainty resulting in the engagement of a broader population of neurons, thereby reflecting the brain’s adaptive capacity to handle diverse possible interpretations[7,8].’

References:

Brennan, J.R., Stabler, E.P., Van Wagenen, S.E., Luh, W.M., and Hale, J.T. (2016). Abstract linguistic structure correlates with temporal activity during naturalistic comprehension. Brain and Language 157, 81-94. 10.1016/j.bandl.2016.04.008.

Benetti, S., Ferrari, A., and Pavani, F. (2023). Multimodal processing in face-to-face interactions: A bridging link between psycholinguistics and sensory neuroscience. Front Hum Neurosci 17, 1108354. 10.3389/fnhum.2023.1108354.

Noppeney, U. (2021). Perceptual Inference, Learning, and Attention in a Multisensory World. Annual Review of Neuroscience, Vol 44, 2021 44, 449-473. 10.1146/annurev-neuro-100120-085519.

Ma, W.J., and Jazayeri, M. (2014). Neural coding of uncertainty and probability. Annu Rev Neurosci 37, 205-220. 10.1146/annurev-neuro-071013-014017.

Fischer, B.J., and Pena, J.L. (2011). Owl's behavior and neural representation predicted by Bayesian inference. Nat Neurosci 14, 1061-1066. 10.1038/nn.2872.

Ganguli, D., and Simoncelli, E.P. (2014). Efficient sensory encoding and Bayesian inference with heterogeneous neural populations. Neural Comput 26, 2103-2134. 10.1162/NECO_a_00638.

Comment 7.1: a) I am not too sure what they mean by 'response consistently provided by participants for four to six consecutive instances' [L117-118]. They should be clearer with the description of these 'pre-test' study methods.

Response 7.1: Thank you for this insightful question. An example of a participant's response to the gesture 'an' is provided below (Table 3). Initially, within 240 ms, the participant provided the answer "an," which could potentially be a guess. To ensure that the participant truly comprehends the gesture, we repeatedly present it until the participant’s response stabilizes, meaning the same answer is given consistently over several trials. While one might consider fixing the number of repetitions (e.g., six trials), this could lead to participants predicting the rule and providing the same answer out of habit. To mitigate this potential bias, we allow the number of repetitions to vary flexibly between four and six trials. 

We understand that the initial phrase might be ambiguous, in the revised manuscript, we have changed the phrase into: ‘For each gesture or speech, the action verb consistently provided by participants across four to six consecutive repetitions—with the number of repetitions varied to mitigate learning effects—was considered the comprehensive response for the gesture or speech.’ (Lines 130-133)

Author response table 8.

Example of participant's response to the gesture 'an'

Comment 7.2: b) I do not understand the paragraph in L143 - 146. This is important to rephrase for clarification. What are 'stepped' neural changes? What is the purpose of 'aggregating' neural responses with identical entropy / MI values?

Response 7.2: It is important to note that the 20 stimuli exhibit 20 increments of gesture entropy values, 11 increments of speech entropy values, and 19 increments of mutual information values (Appendix Table 3). This discrepancy arises from the calculation of entropy and mutual information, where the distributions were derived from the comprehensive set of responses contributed by all 30 participants. As a result, these values were impacted not only by the distinct nameabilities of the stimuli but also by the entirety of responses provided. Consequently, in the context of speech entropy, 9 items demonstrate the nameability of 1, signifying unanimous comprehension among all 30 participants, resulting in an entropy of 0. Moreover, stimuli 'ning' and 'jiao' share an identical distribution, leading to an entropy of 0.63. Regarding MI, a value of 0.66 is computed for the combinations of stimuli 'sao' (gesture entropy: 4.01, speech entropy: 1.12, Author response image 32) and 'tui' (gesture entropy: 1.62, speech entropy: 0, Author response image 4). This indicates that these two sets of stimuli manifest an equivalent degree of integration.

Author response image 3.

Example of gesture answers (gesture sao), speech answers (speech sao), and mutual information (MI) for the ‘sao’ item

Author response image 4.

Example of gesture answers (gesture tui), speech answers (speech tui), and mutual information (MI) for the ‘tui’ item

To precisely assess whether lower entropy/MI corresponds to a smaller or larger neural response, neural responses (ERP amplitude or TMS inhibition effect) with identical entropy or MI values were averaged before undergoing correlational analysis. We understand that the phrasing might be ambiguous. Clear description has been changed in the revised manuscript in Lines 157-160: ‘To determine whether entropy or MI values corresponds to distinct neural changes, the current study first aggregated neural responses (including inhibition effects of tDCS and TMS or ERP amplitudes) that shared identical entropy or MI values, prior to conducting correlational analyses.’

Comment 7.3: c) The paragraph in L160-171 is confusing. Is it an attempt to give an overview of all three experiments? If so, consider moving to the end or summarising what each experiment is at the beginning of the paragraph giving it a name (i.e., TMS). Without that, it is unclear what each experiment is counterbalancing or what 'stimulation site' refers to, for example, leading to a significant lack of clarity.

Response 7.3: We are sorry for the ambiguity, in the revised manuscript, we have moved the relevant phrasing to the beginning of each experiment.

‘Experiment 1: HD-tDCS protocol and data analysis

Participants were divided into two groups, with each group undergoing HD-tDCS stimulation at different target sites (IFG or pMTG). Each participant completed three experimental sessions, spaced one week apart, during which 480 gesture-speech pairs were presented across various conditions. In each session, participants received one of three types of HD-tDCS stimulation: Anodal, Cathodal, or Sham. The order of stimulation site and type was counterbalanced using a Latin square design to control for potential order effects’ (Lines 183-189)

‘Experiment 2: TMS protocol and data analysis

Experiment 2 involved 800 gesture-speech pairs, presented across 15 blocks over three days, with one week between sessions. Stimulation was administered at three different sites (IFG, pMTG, or Vertex). Within the time windows (TWs) spanning the gesture-speech integration period, five TWs that exhibited selective disruption of integration were selected: TW1 (-120 to -80 ms relative to the speech identification point), TW2 (-80 to -40 ms), TW3 (-40 to 0 ms), TW6 (80 to 120 ms), and TW7 (120 to 160 ms)23 (Figure 1C). The order of stimulation site and TW was counterbalanced using a Latin square design.’ (Lines 223-230)

‘Experiment 3: Electroencephalogram (EEG) recording and data analysis

Experiment 3, comprising a total of 1760 gesture-speech pairs, was completed in a single-day session.’ (Lines 249-250)

Comment 7.4: d) L402-406: This sentence is not clear. What do the authors mean by 'the state of [the neural landscape] constructs gradually as measured by entropy and MI'? How does this construct a neural landscape? The authors must rephrase this paragraph using clearer language since in its current state it is very difficult to assess whether it is supported by the evidence they present.

Response 7.4: We are sorry for the ambiguity, in the revised manuscript we have provided clear description in Lines 483-492: ‘The varying contributions of unisensory gesture-speech information and the convergence of multisensory inputs, as reflected in the correlation between distinct ERP components and TMS time windows (TMS TWs), are consistent with recent models suggesting that multisensory processing involves parallel detection of modality-specific information and hierarchical integration across multiple neural levels[4,48]. These processes are further characterized by coordination across multiple temporal scales[49]. Building on this, the present study offers additional evidence that the multi-level nature of gesture-speech processing is statistically structured, as measured by information matrix of unisensory entropy and multisensory convergence index of MI, the input of either source would activate a distributed representation, resulting in progressively functioning neural responses’

References:

Benetti, S., Ferrari, A., and Pavani, F. (2023). Multimodal processing in face-to-face interactions: A bridging link between psycholinguistics and sensory neuroscience. Front Hum Neurosci 17, 1108354. 10.3389/fnhum.2023.1108354.

Meijer, G.T., Mertens, P.E.C., Pennartz, C.M.A., Olcese, U., and Lansink, C.S. (2019). The circuit architecture of cortical multisensory processing: Distinct functions jointly operating within a common anatomical network. Prog Neurobiol 174, 1-15. 10.1016/j.pneurobio.2019.01.004.

Senkowski, D., and Engel, A.K. (2024). Multi-timescale neural dynamics for multisensory integration. Nat Rev Neurosci 25, 625-642. 10.1038/s41583-024-00845-7.

(8) Some writing suffers from conceptual equivocation. For example, the link between 'multimodal representation' and gesture as a type of multimodal extralinguistic information is not straightforward. What 'multimodal representations' usually refer to in semantic cognition is not the co-occurrence of gesture and speech, but the different sources or modalities that inform the structure of a semantic representation or concept (not the fact we use another modality vision to perceive gestures that enrich the linguistic auditory communication of said concepts). See also my comment in the public review regarding the conceptual conflation of the graded hub hypothesis.

Response 8: We aimed to clarify that the integration of gesture and speech, along with the unified representation it entails, is not merely a process whereby perceived gestures enhance speech comprehension. Rather, there exists a bidirectional influence between these two modalities, affecting both their external forms (Bernaidis et al., 2006) and their semantic content (Kita et al., 2003; Kelly et al., 2010). Given that multisensory processing is recognized as an interplay of both top-down and bottom-up mechanisms, we hypothesize that this bidirectional semantic influence between gesture and speech operates similarly. Consequently, we recorded neural responses—specifically the inhibitory effects observed through TMS/tDCS or ERP components—beginning at the onset of speech, which marks the moment when both modalities are accessible.

We prioritize gesture for two primary reasons. Firstly, from a naturalistic perspective, speech and gesture are temporally aligned; gestures typically precede their corresponding speech segments by less than one second (Morrelsamuls et al., 1992). This temporal alignment has prompted extensive research aimed at identifying the time windows during which integration occurs (Obermeier et al., 2011, 2015). Results indicate that local integration of gesture and speech occurs within a time frame extending from -200 ms to +120 ms relative to gesture-speech alignment, where -200 ms indicates that gestures occur 200 ms before speech onset, and +120 ms signifies gestures occurring after the identification point of speech.

Secondly, in our previous study (Zhao, 2023), we investigated this phenomenon by manipulating gesture-speech alignment across two conditions: (1) gestures preceding speech by a fixed interval of 200 ms, and (2) gestures preceding speech at its semantic identification point. Notably, only in the second condition did we observe time-window-selective disruptions of the semantic congruency effect in the IFG and pMTG. This led us to conclude that gestures serve a semantic priming function for co-occurring speech.

We recognize that our previous use of the term "co-occurring speech" may have led to ambiguity. Therefore, in the revised manuscript, we have replaced those sentences with a detailed description of the properties of each modality in Lines 60-62: ‘Even though gestures convey information in a global-synthetic way, while speech conveys information in a linear segmented way, there exists a bidirectional semantic influence between the two modalities[9,10]’

Conceptual conflation of the graded hub hypothesis has been clarified in the Response to Reviewer 3 (public review) response 2.

References:

Bernardis, P., & Gentilucci, M. (2006). Speech and gesture share the same communication system. Neuropsychologia, 44(2), 178-190

Kelly, S. D., Ozyurek, A., & Maris, E. (2010b). Two sides of the same coin: speech and gesture mutually interact to enhance comprehension. Psychological Science, 21(2), 260-267. doi:10.1177/0956797609357327

Kita, S., & Ozyurek, A. (2003). What does cross-linguistic variation in semantic coordination of speech and gesture reveal?: Evidence for an interface representation of spatial thinking and speaking. Journal of Memory and Language, 48(1), 16-32. doi:10.1016/s0749-596x(02)00505-3

Obermeier, C., & Gunter, T. C. (2015). Multisensory Integration: The Case of a Time Window of Gesture-Speech Integration. Journal of Cognitive Neuroscience, 27(2), 292-307. doi:10.1162/jocn_a_00688

Obermeier, C., Holle, H., & Gunter, T. C. (2011). What Iconic Gesture Fragments Reveal about Gesture-Speech Integration: When Synchrony Is Lost, Memory Can Help. Journal of Cognitive Neuroscience, 23(7), 1648-1663. doi:10.1162/jocn.2010.21498

Morrelsamuels, P., & Krauss, R. M. (1992). WORD FAMILIARITY PREDICTS TEMPORAL ASYNCHRONY OF HAND GESTURES AND SPEECH. Journal of Experimental Psychology-Learning Memory and Cognition, 18(3), 615-622. doi:10.1037/0278-7393.18.3.615

Hostetter, A., and Mainela-Arnold, E. (2015). Gestures occur with spatial and Motoric knowledge: It's more than just coincidence. Perspectives on Language Learning and Education 22, 42-49. doi:10.1044/lle22.2.42.

McNeill, D. (2005). Gesture and though (University of Chicago Press). 10.7208/chicago/9780226514642.001.0001.

Zhao, W. (2023). TMS reveals a two-stage priming circuit of gesture-speech integration. Front Psychol 14, 1156087. 10.3389/fpsyg.2023.1156087.

(9) The last paragraph of the introduction lacks a conductive thread. The authors describe three experiments without guiding the reader through a connecting thread underlying the experiments. Feels more like three disconnected studies than a targeted multi-experiment approach to solve a problem. What is each experiment contributing to? What is the 'grand question' or thread unifying these?

Response 9: The present study introduced three experiments to explore the neural activity linked to the amount of information processed during multisensory gesture-speech integration. In Experiment 1, we observed that the extent of inhibition in the pMTG and LIFG was closely linked to the overlapping gesture-speech responses, as quantified by mutual information. Building on the established roles of the pMTG and LIFG in our previous study (Zhao et al., 2021, JN), we then expanded our investigation to determine whether the dynamic neural engagement between the pMTG and LIFG during gesture-speech processing was also associated with the quality of the information. This hypothesis was further validated through high-temporal resolution EEG, where we examined ERP components related to varying information qualities. Notably, we observed a close time alignment between the ERP components and the time windows of the TMS effects, which were associated with the same informational matrices in gesture-speech processing.

Linkage of the three experiments has been clarified in the introduction in Lines 75-102: ‘

To investigate the neural mechanisms underlying gesture-speech integration, we conducted three experiments to assess how neural activity correlates with distributed multisensory integration, quantified using information-theoretic measures of MI. Additionally, we examined the contributions of unisensory signals in this process, quantified through unisensory entropy. Experiment 1 employed high-definition transcranial direct current stimulation (HD-tDCS) to administer Anodal, Cathodal and Sham stimulation to either the IFG or the pMTG. HD-tDCS induces membrane depolarization with anodal stimulation and membrane hyperpolarization with cathodal stimulation[26], thereby increasing or decreasing cortical excitability in the targeted brain area, respectively. This experiment aimed to determine whether the overall facilitation (Anodal-tDCS minus Sham-tDCS) and/or inhibitory (Cathodal-tDCS minus Sham-tDCS) of these integration hubs is modulated by the degree of gesture-speech integration, as measure by MI.

Given the differential involvement of the IFG and pMTG in gesture-speech integration, shaped by top-down gesture predictions and bottom-up speech processing [23], Experiment 2 was designed to further assess whether the activity of these regions was associated with relevant informational matrices. Specifically, we applied inhibitory chronometric double-pulse transcranial magnetic stimulation (TMS) to specific temporal windows associated with integration processes in these regions[23], assessing whether the inhibitory effects of TMS were correlated with unisensory entropy or the multisensory convergence index (MI).

Experiment 3 complemented these investigations by focusing on the temporal dynamics of neural responses during semantic processing, leveraging high-temporal event-related potentials (ERPs). This experiment investigated how distinct information contributors modulated specific ERP components associated with semantic processing. These components included the early sensory effects as P1 and N1–P2[27,28], the N400 semantic conflict effect[14,28,29], and the late positive component (LPC) reconstruction effect[30,31]. By integrating these ERP findings with results from Experiments 1 and 2, Experiment 3 aimed to provide a more comprehensive understanding of how gesture-speech integration is modulated by neural dynamics’

References:

Bikson, M., Inoue, M., Akiyama, H., Deans, J.K., Fox, J.E., Miyakawa, H., and Jefferys, J.G.R. (2004). Effects of uniform extracellular DC electric fields on excitability in rat hippocampal slices. J Physiol-London 557, 175-190. 10.1113/jphysiol.2003.055772.

Federmeier, K.D., Mai, H., and Kutas, M. (2005). Both sides get the point: hemispheric sensitivities to sentential constraint. Memory & Cognition 33, 871-886. 10.3758/bf03193082.

Kelly, S.D., Kravitz, C., and Hopkins, M. (2004). Neural correlates of bimodal speech and gesture comprehension. Brain and Language 89, 253-260. 10.1016/s0093-934x(03)00335-3.

Wu, Y.C., and Coulson, S. (2005). Meaningful gestures: Electrophysiological indices of iconic gesture comprehension. Psychophysiology 42, 654-667. 10.1111/j.1469-8986.2005.00356.x.

Fritz, I., Kita, S., Littlemore, J., and Krott, A. (2021). Multimodal language processing: How preceding discourse constrains gesture interpretation and affects gesture integration when gestures do not synchronise with semantic affiliates. J Mem Lang 117, 104191. 10.1016/j.jml.2020.104191.

Gunter, T.C., and Weinbrenner, J.E.D. (2017). When to take a gesture seriously: On how we use and prioritize communicative cues. J Cognitive Neurosci 29, 1355-1367. 10.1162/jocn_a_01125.

Ozyurek, A., Willems, R.M., Kita, S., and Hagoort, P. (2007). On-line integration of semantic information from speech and gesture: Insights from event-related brain potentials. J Cognitive Neurosci 19, 605-616. 10.1162/jocn.2007.19.4.605.

Zhao, W., Li, Y., and Du, Y. (2021). TMS reveals dynamic interaction between inferior frontal gyrus and posterior middle temporal gyrus in gesture-speech semantic integration. The Journal of Neuroscience, 10356-10364. 10.1523/jneurosci.1355-21.2021.

(10) The authors should provide a clearer figure to appreciate their paradigm, illustrating clearly the stimulus presentation (gesture and speech).

Response 10: To reduce ambiguity, unnecessary arrows were deleted from Figure 1.

Comment 11.1: (11) Required methodological clarifications to better assess the strength of the evidence presented:

a) Were the exclusion criteria only handedness and vision? Did the authors exclude based on neurological and psychiatric disorders? Psychoactive drugs? If not, do they think the lack of these exclusion criteria might have influenced their results?

Response 11.1: Upon registration, each participant is required to complete a questionnaire alongside the consent form and handedness questionnaire. This procedure is designed to exclude individuals with potential neurological or psychiatric disorders, as well as other factors that may affect their mental state or reaction times. Consequently, all participants reported in the manuscript do not have any of the aforementioned neurological or psychiatric disorders. The questionnaire is attached below:

Author response image 4.

Comment 11.2: b) Are the subjects from the pre-tests (L112-113) and the replication study (L107) a separate sample or did they take part in Experiments 1-3?

Response 11.2: The participants in each pre-test and experiment were independent, resulting in a total of 188 subjects. Since the stimuli utilized in this study were previously validated and reported (Zhao et al., 2021), the 90 subjects who participated in the three pre-tests are not included in the final count for the current study, leaving a total of 98 participants reported in the manuscript in Lines 103-104: ‘Ninety-eight young Chinese participants signed written informed consent forms and took part in the present study’.

Comment 11.3: c) L176. The authors should explain how they selected ROIs. This is very important for the reasons outlined above.

Response 11.3: Please see Response to Comment 6 for details.

Comment 11.4: d) The rationale for Experiment 1 and its analysis approach should be explicitly described. Why perform Pearson correlations? What is the conceptual explanation of the semantic congruency effect and why should it be expected to correlate with the three information-theoretic metrics? What effects could the authors expect to find and what would they mean? There is a brief description in L187-195 but it is unclear.

Response 11.4: We thank the reviewer for their rigorous consideration. The semantic congruency effect is widely used as an index of multisensory integration. Therefore, the effects of HD-tDCS on the IFG and pMTG, as measured by changes in the semantic congruency effect, serve as an indicator of altered neural responses to multisensory integration. In correlating these changes with behavioral indices of information degree, we aimed to assess whether the integration hubs (IFG and pMTG) function progressively during multisensory gesture-speech integration. The rationale for using Pearson correlations is based on the hypothesis that the 20 sets of stimuli used in this study represent a sample from a normally distributed population. Thus, even with changes in the sample (e.g., using another 20 values), the gradual relationship between neural responses and the degree of information would remain unchanged. This hypothesis is supported by the findings from another experiment (see details in Response to Comment 4).

In the revised manuscript, we have provided a clear description of the rationale for Experiment 1 in Lines 206-219: ‘To examine the relationship between the degree of information and neural responses, we conducted Pearson correlation analyses using a sample of 20 sets. Neural responses were quantified based on the effects of HD-tDCS (active tDCS minus sham tDCS) on the semantic congruency effect, defined as the difference in reaction times between semantic incongruent and congruent conditions (Rt(incongruent) - Rt(congruent)). This effect served as an index of multisensory integration[35] within the left IFG and pMTG. The variation in information was assessed using three information-theoretic metrics. To account for potential confounds related to multiple candidate representations, we conducted partial correlation analyses between the tDCS effects and gesture entropy, speech entropy, and MI, controlling for the number of responses provided for each gesture and speech, as well as the total number of combined responses. Given that HD-tDCS induces overall disruption at the targeted brain regions, we hypothesized that the neural activity within the left IFG and pMTG would be progressively affected by varying levels of multisensory convergence, as indexed by MI.’

Additionally, in the introduction, we have rephrased the relevant rationale in Lines 75-86: _‘_To investigate the neural mechanisms underlying gesture-speech integration, we conducted three experiments to assess how neural activity correlates with distributed multisensory integration, quantified using information-theoretic measures of MI. Additionally, we examined the contributions of unisensory signals in this process, quantified through unisensory entropy. Experiment 1 employed high-definition transcranial direct current stimulation (HD-tDCS) to administer Anodal, Cathodal and Sham stimulation to either the IFG or the pMTG. HD-tDCS induces membrane depolarization with anodal stimulation and membrane hyperpolarization with cathodal stimulation[26], thereby increasing or decreasing cortical excitability in the targeted brain area, respectively. This experiment aimed to determine whether the overall facilitation (Anodal-tDCS minus Sham-tDCS) and/or inhibitory (Cathodal-tDCS minus Sham-tDCS) of these integration hubs is modulated by the degree of gesture-speech integration, as measure by MI

Reference:

Kelly, S.D., Creigh, P., and Bartolotti, J. (2010). Integrating speech and iconic gestures in a Stroop-like task: Evidence for automatic processing. Journal of Cognitive Neuroscience 22, 683-694. 10.1162/jocn.2009.21254.

Comment 11.5: e) The authors do not mention in the methods if FDR correction was applied to the Pearson correlations in Experiment 1. There is a mention in the Results Figure, but it is unclear if it was applied consistently. Can the authors confirm, and explicitly state the way they carried out FDR correction for this family of tests in Experiment 1? This is especially important in the light of some of their results having a p-value of p=.049.

Response 11.5: FDR correction was applied to Experiment 1, and all reported p-values were corrected using this method. In the revised manuscript, we have included a reference to FDR correction in Lines 221-222: ‘False discovery rate (FDR) correction was applied for multiple comparisons.’

In Experiment 1, since two separate participant groups (each N = 26) were recruited for the HD-tDCS over either the IFG or pMTG, FDR correction was performed separately for each group. Therefore, for each brain region, six comparisons (three information matrices × two tDCS effects: anodal-sham or cathodal-sham) were submitted for FDR correction.

In Experiment 2, six comparisons (three information matrices × two sites: IFG or pMTG) were submitted for FDR correction. In Experiment 3, FDR correction was applied to the seven regions of interest (ROIs) within each component, resulting in five comparisons

The confidence of a p-value of 0.049 was clarified in Response to Comment 3.

Comment 11.6: f) L200. What does the abbreviation 'TW' stands for in this paragraph? When was it introduced in the main text? The description is in the Figure, but it should be moved to the main text.]

Comment 11.7: g) How were the TWs chosen? Is it the criterion in L201-203? If so, it should be moved to the start of the paragraph. What does the word 'selected' refer to in that description? Selected for what? The explanation seems to be in the Figure, but it should be in the main text. It is still not a complete explanation. What were the criteria for assigning TWs to the IFG or pMTG?

Response 11.6& 11.7: Since the two comments are related, we will provide a synthesized response. 'TW' refers to time window, the selection of which was based on our previous study (Zhao et al., 2021, J. Neurosci). In Zhao et al. (2021), we employed the same experimental protocol—using inhibitory double-pulse transcranial magnetic stimulation (TMS) over the IFG and pMTG in one of eight 40-ms time windows relative to the speech identification point (IP; the minimal length of lexical speech), with three time windows before the speech IP and five after. Based on this previous work, we believe that these time windows encompass the potential gesture-speech integration process. Results demonstrated a time-window-selective disruption of the semantic congruency effect (i.e., reaction time costs driven by semantic conflict), with no significant modulation of the gender congruency effect (i.e., reaction time costs due to gender conflict), when stimulating the left pMTG in TW1, TW2, and TW7, and when stimulating the left IFG in TW3 and TW6. Based on these findings, the present study selected the five time windows that showed a selective disruption effect during gesture-speech integration.

Note that in the present study, we applied stimulation to both the IFG and pMTG across all five time windows, and further correlated the TMS disruption effects with the three information matrices.

We recognize that the rationale for the choice of time windows was not sufficiently explained in the original manuscript. In the revised manuscript, we have added the relevant description in Lines 223-228: ‘Stimulation was administered at three different sites (IFG, pMTG, or Vertex). Within the time windows (TWs) spanning the gesture-speech integration period, five TWs that exhibited selective disruption of integration were selected: TW1 (-120 to -80 ms relative to the speech identification point), TW2 (-80 to -40 ms), TW3 (-40 to 0 ms), TW6 (80 to 120 ms), and TW7 (120 to 160 ms)[23] (Figure 1C). The order of stimulation site and TW was counterbalanced using a Latin square design.’

Comment 11.8: h) Again, the rationale for the Pearson correlations of semantic congruency with information-theoretic metrics should be explicitly outlined. What is this conceptually?

Response 11.8: Given that the rationale behind Experiment 1 and Experiment 2 is similar—both investigating the correlation between interrupted neural effects and the degree of information—we believe that the introduction of the Pearson correlation between semantic congruency and information-theoretic metrics, as presented in Experiment 1 (see Response to Comment 11.4 for details), is sufficient for both experiments.

Comment 11.9: i)What does 'gesture stoke' mean in the Figure referring to Experiment 3? Figure 1D is not clear. What are the arrows referring to?

Response 11.9: According to McNeill (1992), gesture phases differ based on whether the gesture depicts imagery. Iconic and metaphoric gestures are imagistic and typically consist of three phases: a preparation phase, a stroke phase, and a retraction phrase. Figure 4 provides an example of these three phases using the gesture ‘break’. In the preparation phase, the hand and arm move away from their resting position to a location in gesture space where the stroke begins. As illustrated in the first row of Figure 4, during the preparation phase of the ‘break’ gesture, the hands, initially in a fist and positioned downward, rise to a center-front position. In the stroke phase, the meaning of the gesture is conveyed. This phase occurs in the central gesture space and is synchronized with the linguistic segments it co-expresses. For example, in the stroke phase of the ‘break’ gesture (second row of Figure 4), the two fists move 90 degrees outward before returning to a face-down position. The retraction phase involves the return of the hand from the stroke position to the rest position. In the case of the ‘break’ gesture, this involves moving the fists from the center front back into the resting position (see third row of Figure 4).

Therefore, in studies examining gesture-speech integration, gestures are typically analyzed starting from the stroke phase (Habets et al., 2011; Kelly et al., 2010), a convention also adopted in our previous studies (Zhao et al., 2018, 2021, 2023). We acknowledge that this should be explained explicitly, and in the revised manuscript, we have added the following clarification in Lines 162-166: ‘Given that gestures induce a semantic priming effect on concurrent speech[33], this study utilized a semantic priming paradigm in which speech onset was aligned with the DP of each gesture[23,33], the point at which the gesture transitions into a lexical form[34]. The gesture itself began at the stroke phase, a critical moment when the gesture conveys its primary semantic content[34].’

Additionally, Figure 1 has been revised in the manuscript to eliminate ambiguous arrows. (see Response 10 for detail).

Author response image 5.

An illustration of the gesture phases of the 'break' gesture.

References：

Habets, B., Kita, S., Shao, Z. S., Ozyurek, A., & Hagoort, P. (2011). The Role of Synchrony and Ambiguity in Speech-Gesture Integration during Comprehension. Journal of Cognitive Neuroscience, 23(8), 1845-1854. doi:10.1162/jocn.2010.21462

Kelly, S. D., Creigh, P., & Bartolotti, J. (2010). Integrating Speech and Iconic Gestures in a Stroop-like Task: Evidence for Automatic Processing. Journal of Cognitive Neuroscience, 22(4), 683-694. doi:DOI 10.1162/jocn.2009.21254

Comment 11.10: j) L236-237: "Consequently, four ERP components were predetermined" is very confusing. Were these components predetermined? Or were they determined as a consequence of the comparison between the higher and lower halves for the IT metrics described above in the same paragraph? The description of the methods is not clear.

Response 11.10: The components selected were based on a comparison between the higher and lower halves of the information metrics. By stating that these components were predetermined, we aimed to emphasize that the components used in our study are consistent with those identified in previous research on semantic processing. We acknowledge that the phrasing may have been unclear, and in the revised manuscript, we have provided a more explicit description in Lines 267-276: ‘To consolidate the data, we conducted both a traditional region-of-interest (ROI) analysis, with ROIs defined based on a well-established work[40], and a cluster-based permutation approach, which utilizes data-driven permutations to enhance robustness and address multiple comparisons.

For the traditional ROI analysis, grand-average ERPs at electrode Cz were compared between the higher (≥50%) and lower (<50%) halves for gesture entropy (Figure 5A1), speech entropy (Figure 5B1), and MI (Figure 5C1). Consequently, four ERP components were determined: the P1 effect observed within the time window of 0-100 ms[27,28], the N1-P2 effect observed between 150-250ms[27,28], the N400 within the interval of 250-450ms[14,28,29], and the LPC spanning from 550-1000ms[30,31].’

Reference: Habets, B., Kita, S., Shao, Z.S., Ozyurek, A., and Hagoort, P. (2011). The Role of Synchrony and Ambiguity in Speech-Gesture Integration during Comprehension. J Cognitive Neurosci 23, 1845-1854. 10.1162/jocn.2010.21462.

(12) In the Results section for Experiment 2 (L292-295), it is not clear what the authors mean when they mention that a more negative TMS effect represents a stronger interruption of the integration effect. If I understand correctly, the correlation reported for pMTG was for speech entropy, which does not represent integration (that would be MI).

Response 12: Since the TMS effect was defined as active TMS minus Vertex TMS, the inhibitory TMS effect is inherently negative. A greater inhibitory TMS effect corresponds to a larger negative value, such that a more negative TMS effect indicates a stronger disruption of the integration process. We acknowledge that the previous phrasing was somewhat ambiguous. In the revised manuscript, we have rephrased the sentence as follows: ‘a larger negative TMS effect signifies a greater disruption of the integration process’ (Lines 342-343)

Multisensory integration transcends simple data amalgamation, encompassing complex interactions at various hierarchical neural levels and the parallel detection and discrimination of raw data from each modality (Benetti et al., 2023; Meijer et al., 2019). Therefore, we regard the process of gesture-speech integration as involving both unisensory processing and multisensory convergence. The correlation of gesture and speech entropy reflects contributions from unisensory processing, while the mutual information (MI) index indicates the contribution of multisensory convergence during gesture-speech integration. The distinction between these various source contributions will be the focus of Experiment 2 and Experiment 3, as described in the revised manuscript Lines 87-102: ‘Given the differential involvement of the IFG and pMTG in gesture-speech integration, shaped by top-down gesture predictions and bottom-up speech processing [23], Experiment 2 was designed to further assess whether the activity of these regions was associated with relevant informational matrices. Specifically, we applied inhibitory chronometric double-pulse transcranial magnetic stimulation (TMS) to specific temporal windows associated with integration processes in these regions[23], assessing whether the inhibitory effects of TMS were correlated with unisensory entropy or the multisensory convergence index (MI).

Experiment 3 complemented these investigations by focusing on the temporal dynamics of neural responses during semantic processing, leveraging high-temporal event-related potentials (ERPs). This experiment investigated how distinct information contributors modulated specific ERP components associated with semantic processing. These components included the early sensory effects as P1 and N1–P2[27,28], the N400 semantic conflict effect[14,28,29], and the late positive component (LPC) reconstruction effect[30,31]. By integrating these ERP findings with results from Experiments 1 and 2, Experiment 3 aimed to provide a more comprehensive understanding of how gesture-speech integration is modulated by neural dynamics’.  

References:

Benetti, S., Ferrari, A., and Pavani, F. (2023). Multimodal processing in face-to-face interactions: A bridging link between psycholinguistics and sensory neuroscience. Front Hum Neurosci 17, 1108354. 10.3389/fnhum.2023.1108354.

Meijer, G.T., Mertens, P.E.C., Pennartz, C.M.A., Olcese, U., and Lansink, C.S. (2019). The circuit architecture of cortical multisensory processing: Distinct functions jointly operating within a common anatomical network. Prog Neurobiol 174, 1-15. 10.1016/j.pneurobio.2019.01.004.

(13) I find the description of the results for Experiment 3 very hard to follow. Perhaps if the authors have decided to organise the main text by describing the components from earliest to latest, the Figure organisation should follow suit (i.e., organise the Figure from the earliest to the latest component, instead of gesture entropy/speech entropy / mutual information). This might make the description of the results easier to follow.

Response 13: As suggested, we have reorganized the results of experiment 3 based on components from earliest to latest, together with an updated Figure 5.

The results are detailed in Lines 367-423: ‘Topographical maps illustrating amplitude differences between the lower and higher halves of speech entropy demonstrate a central-posterior P1 amplitude (0-100 ms, Figure 5B). Aligning with prior findings[27], the paired t-tests demonstrated a significantly larger P1 amplitude within the ML ROI (t(22) = 2.510, p = 0.020, 95% confidence interval (CI) = [1.66, 3.36]) when contrasting stimuli with higher 50% speech entropy against those with lower 50% speech entropy (Figure 5D1 left). Subsequent correlation analyses unveiled a significant increase in the P1 amplitude with the rise in speech entropy within the ML ROI (r = 0.609, p = 0.047, 95% CI = [0.039, 1.179], Figure 5D1 right). Furthermore, a cluster of neighboring time-electrode samples exhibited a significant contrast between the lower 50% and higher 50% of speech entropy, revealing a P1 effect spanning 16 to 78 ms at specific electrodes (FC2, FCz, C1, C2, Cz, and CPz, Figure 5D2 middle) (t(22) = 2.754, p = 0.004, 95% confidence interval (CI) = [1.65, 3.86], Figure 5D2 left), with a significant correlation with speech entropy (r = 0.636, p = 0.035, 95% CI = [0.081, 1.191], Figure 5D2 right).

Additionally, topographical maps comparing the lower 50% and higher 50% gesture entropy revealed a frontal N1-P2 amplitude (150-250 ms, Figure 5A). In accordance with previous findings on bilateral frontal N1-P2 amplitude[27], paired t-tests displayed a significantly larger amplitude for stimuli with lower 50% gesture entropy than with higher 50% entropy in both ROIs of LA (t(22) = 2.820, p = 0.011, 95% CI = [2.21, 3.43]) and RA (t(22) = 2.223, p = 0.038, 95% CI = [1.56, 2.89]) (Figure 5E1 left).  Moreover, a negative correlation was found between N1-P2 amplitude and gesture entropy in both ROIs of LA (r = -0.465, p = 0.039, 95% CI = [-0.87, -0.06]) and RA (r = -0.465, p = 0.039, 95% CI = [-0.88, -0.05]) (Figure 5E1 right). Additionally, through a cluster-permutation test, the N1-P2 effect was identified between 184 to 202 ms at electrodes FC4, FC6, C2, C4, C6, and CP4 (Figure 5E2 middle) (t(22) = 2.638, p = 0.015, 95% CI = [1.79, 3.48], (Figure 5E2 left)), exhibiting a significant correlation with gesture entropy (r = -0.485, p = 0.030, 95% CI = [-0.91, -0.06], Figure 5E2 right).

Furthermore, in line with prior research[42], a left-frontal N400 amplitude (250-450 ms) was discerned from topographical maps of gesture entropy (Figure 5A). Specifically, stimuli with lower 50% values of gesture entropy elicited a larger N400 amplitude in the LA ROI compared to those with higher 50% values  (t(22) = 2.455, p = 0.023, 95% CI = [1.95, 2.96], Figure 5F1 left). Concurrently, a negative correlation was noted between the N400 amplitude and gesture entropy (r = -0.480, p = 0.032, 95% CI = [-0.94, -0.03], Figure 5F1 right) within the LA ROI. The identified clusters showing the N400 effect for gesture entropy (282 – 318 ms at electrodes FC1, FCz, C1, and Cz, Figure 5F2 middle) (t(22) = 2.828, p = 0.010, 95% CI = [2.02, 3.64], Figure 5F2 left) also exhibited significant correlation between the N400 amplitude and gesture entropy (r = -0.445, p = 0.049, 95% CI = [-0.88, -0.01], Figure 5F2 right).

Similarly, a left-frontal N400 amplitude (250-450 ms) [42] was discerned from topographical maps for MI (Figure 5C). A larger N400 amplitude in the LA ROI was observed for stimuli with lower 50% values of MI compared to those with higher 50% values (t(22) = 3.00, p = 0.007, 95% CI = [2.54, 3.46], Figure 5G1 left). This was accompanied by a significant negative correlation between N400 amplitude and MI (r = -0.504, p = 0.028, 95% CI = [-0.97, -0.04], Figure 5G1 right) within the LA ROI. The N400 effect for MI, observed in the 294–306 ms window at electrodes F1, F3, Fz, FC1, FC3, FCz, and C1 (Figure 5G2 middle) (t(22) = 2.461, p = 0.023, 95% CI = [1.62, 3.30], Figure 5G2 left), also showed a significant negative correlation with MI (r = -0.569, p = 0.011, 95% CI = [-0.98, -0.16], Figure 5G2 right).

Finally, consistent with previous findings[30], an anterior LPC effect (550-1000 ms) was observed in topographical maps comparing stimuli with lower and higher 50% speech entropy (Figure 5B). The reduced LPC amplitude was evident in the paired t-tests conducted in ROIs of LA (t(22) = 2.614, p = 0.016, 95% CI = [1.88, 3.35]); LC (t(22) = 2.592, p = 0.017, 95% CI = [1.83, 3.35]); RA (t(22) = 2.520, p = 0.020, 95% CI = [1.84, 3.24]); and ML (t(22) = 2.267, p = 0.034, 95% CI = [1.44, 3.10]) (Figure 5H1 left). Simultaneously, a marked negative correlation with speech entropy was evidenced in ROIs of LA (r = -0.836, p =   0.001, 95% CI = [-1.26, -0.42]); LC (r = -0.762, p = 0.006, 95% CI = [-1.23, -0.30]); RA (r = -0.774, p = 0.005, 95% CI = [-1.23, -0.32]) and ML (r = -0.730, p = 0.011, 95% CI = [-1.22, -0.24]) (Figure 5H1 right). Additionally, a cluster with the LPC effect (644 - 688 ms at electrodes Cz, CPz, P1, and Pz, Figure 5H2 middle) (t(22) = 2.754, p = 0.012, 95% CI = [1.50, 4.01], Figure 5H2 left) displayed a significant correlation with speech entropy (r = -0.699, p = 0.017, 95% CI = [-1.24, -0.16], Figure 5H2 right).’

(14) In the Discussion (L394 - 395) the authors mention for the first time their task being a semantic priming paradigm. This idea of the task as a semantic priming paradigm allowing top-down prediction of gesture over speech should be presented earlier in the paper, perhaps during the final paragraph of the introduction (as part of the rationale) or during the explanation of the task. The authors mention top-down influences earlier and this is impossible to understand before this information about the paradigm is presented. It would also make the reading of the paper significantly clearer. Critically, an appropriate description of the paradigm is missing in the Methods (what are the subjects asked to do? It states that it replicates an effect in Ref 28, but this manuscript does not contain a clear description of the task). To further complicate things, the 'Experimental Procedure' section of the methods states this is a semantic priming paradigm of gestures onto speech (L148) and proceeds to provide two seemingly irrelevant references (for example, the Pitcher reference is to a study that employed faces and houses as stimuli). How is this a semantic priming paradigm? The study where I found the first mention of this paradigm seems to clearly classify it as a Stroop-like task (Kelly et al, 2010).

We appreciate the reviewer’s thorough consideration. The experimental paradigm employed in the current study differs from the Stroop-like task utilized by Kelly et al. (2010). In their study, the video presentation started with the stroke phase of the gesture, while speech occurred 200 ms after the gesture onset.

As detailed in our previous study (Zhao et al., 2023, Frontiers in Psychology), we confirmed the semantic predictive role of gestures in relation to speech by contrasting two experimental conditions: (1) gestures preceding speech by a fixed 200 ms interval, and (2) gestures preceding speech at the semantic identification point of the gesture. Our findings revealed time-window-selective disruptions in the semantic congruency effect in the IFG and pMTG, but only in the second condition, suggesting that gestures exert a semantic priming effect on concurrent speech.

This work highlighted the semantic priming role of gestures in the integration of speech found in Zhao et al. (2021, Journal of Neuroscience). In the study, a comparable approach was adopted by segmenting speech into eight 40-ms time windows based on the speech discrimination point, while manipulating the speech onset to align with the gesture identification point. The results revealed time-window-selective disruptions in the semantic congruency effect, providing support for the dynamic and temporally staged roles of the IFG and pMTG in gesture-speech integration.

Given that the present study follows the same experimental procedure as our prior work (Zhao et al., 2021, Journal of Neuroscience; Zhao et al., 2023, Frontiers in Psychology), we refer to this design as a "semantic priming" of gesture upon speech. We agree with the reviewer that a detailed description should be clarified earlier in the manuscript. To address this, we have added a more explicit description of the semantic priming paradigm in the methods section of the revised manuscript in Lines 162-166: ‘Given that gestures induce a semantic priming effect on concurrent speech[33], this study utilized a semantic priming paradigm in which speech onset was aligned with the DP of each gesture[23,33], the point at which the gesture transitions into a lexical form[34]. The gesture itself began at the stroke phase, a critical moment when the gesture conveys its primary semantic content [34].’

The task participants completed was outlined immediately following the explanation of the experimental paradigm: ‘Gesture–speech pairs were presented randomly using Presentation software (www.neurobs.com). Participants were asked to look at the screen but respond with both hands as quickly and accurately as possible merely to the gender of the voice they heard’ (Lines:177-180).

Wrongly cited references have been corrected.

(15) L413-417: How do the authors explain that they observe this earlier ERP component and TMS effect over speech and a later one over gesture in pMTG when in their task they first presented gesture and then speech? Why mention STG/S when they didn't assess this?

(19) L436-440: This paragraph yields the timing of the findings represented in Figure 6 even more confusing. If gesture precedes speech in the paradigm, why are the first TMS and ERP results observed in speech?

Response 15 &19: Since these two aspects are closely related, we offer a comprehensive explanation. Although gestures were presented before speech, the integration process occurs once both modalities are available. Consequently, ERP and TMS measurements were taken after speech onset to capture the integration of the two modalities. Neural responses were used as the dependent variable to reflect the degree of integration—specifically, gesture-speech semantic congruency in the TMS study and high-low semantic variance in the ERP study. Therefore, the observed early effect can be interpreted as an interaction between the top-down influence of gesture and the bottom-up processing of speech.

To isolate the pure effect of gesture, neural activity would need to be recorded from gesture onset. However, if one aims to associate the strength of neural activity with the degree of gesture information, recording from the visual processing areas would be more appropriate.

To avoid unnecessary ambiguity, the phrase "involved STG/S" has been removed from the manuscript.

(16) L427-428: I find it hard to believe that MI, a behavioural metric, indexes the size of overlapped neural populations activated by gesture and speech. The authors should be careful with this claim or provide evidence in favour.

Response 16: Mutual information (MI) is a behavioral metric that indexes the distribution of overlapping responses between gesture and speech (for further details, please see the Response to Comment 1). In the present study, MI was correlated with neural responses evoked by gesture and speech, with the goal of demonstrating that neural activity progressively reflects the degree of information conveyed, as indexed by MI.

(17) Why would you have easier integration (reduced N400) with larger gesture entropy in IFG (Figure 6(3))? Wouldn't you expect more difficult processing if entropy is larger?

(18) L431-432: The claim that IFG stores semantic information is controversial. The authors provide two references from the early 2000s that do not offer support for this claim (the IFG's purported involvement according to these is in semantic unification, not storage).

Response 17 &18: As outlined in the Responses to Comment 1 of the public review, we have provided a re-explanation of the IFG as a semantic control region. Additionally, we have clarified the role of the IFG in relation to the various stages of gesture-speech integration in Lines 533-538: ‘Last, the activated speech representation would disambiguate and reanalyze the semantic information and further unify into a coherent comprehension in the pMTG[12,37]. As speech entropy increases, indicating greater uncertainty in the information provided by speech, more cognitive effort is directed towards selecting the targeted semantic representation. This leads to enhanced involvement of the IFG and a corresponding reduction in LPC amplitude’

(20) Overall, the grammar makes some parts of the discussion hard to follow (e.g. the limitation in L446-447: 'While HD tDCS and TMS may impact functionally and anatomically connected brain regions, the graded functionality of every disturbed period is not guaranteed')

Response 20: Clear description has been provided in the revised manuscript in Lines 552-557: ‘Additionally, not all influenced TWs exhibited significant associations with entropy and MI. While HD-tDCS and TMS may impact functionally and anatomically connected brain regions[55,56],  whether the absence of influence in certain TWs can be attributed to compensation by other connected brain areas, such as angular gyrus[57] or anterior temporal lobe[58], warrants further investigation. Therefore, caution is needed when interpreting the causal relationship between inhibition effects of brain stimulation and information-theoretic metrics (entropy and MI).’

References:

Hartwigsen, G., Bzdok, D., Klein, M., Wawrzyniak, M., Stockert, A., Wrede, K., Classen, J., and Saur, D. (2017). Rapid short-term reorganization in the language network. Elife 6. 10.7554/eLife.25964.

Jackson, R.L., Hoffman, P., Pobric, G., and Ralph, M.A.L. (2016). The semantic network at work and rest: Differential connectivity of anterior temporal lobe subregions. Journal of Neuroscience 36, 1490-1501. 10.1523/JNEUROSCI.2999-15.2016

Humphreys, G. F., Lambon Ralph, M. A., & Simons, J. S. (2021). A Unifying Account of Angular Gyrus Contributions to Episodic and Semantic Cognition. Trends in neurosciences, 44(6), 452–463. https://doi.org/10.1016/j.tins.2021.01.006

Bonner, M. F., & Price, A. R. (2013). Where is the anterior temporal lobe and what does it do?. The Journal of neuroscience : the official journal of the Society for Neuroscience, 33(10), 4213–4215. https://doi.org/10.1523/JNEUROSCI.0041-13.2013

(21) Inconsistencies between terminology employed in Figures and main text (e.g., pre-test study in text, gating study in Figure?)

Response 21: Consistence has been made by changing the ‘gating study’ into ‘pre-tests’ in Figure 1 (Lines 758).

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public review):

Summary:

Lejeune et al. demonstrated sex-dependent differences in the susceptibility to MRSA infection. The authors demonstrated the role of the microbiota and sex hormones as potential determinants of susceptibility. Moreover, the authors showed that Th17 cells and neutrophils contribute to sex hormone-dependent protection in female mice.

Strengths:

The role of microbiota was examined in various models (gnotobiotic, co-housing, microbiota transplantation). The identification of responsible immune cells was achieved using several genetic knockouts and cell-specific depletion models. The involvement of sex hormones was clarified using ovariectomy and the FCG model.

Weaknesses:

The mechanisms by which specific microbiota confer female-specific protection remain unclear.

We thank the reviewer for highlighting the strengths of the manuscript including the models and techniques we employ. We agree that the relationship between the microbiota and sex-dependent protection is less developed compared with other aspects of the study. As detailed below, we are attempting to identify specific microbes that confer femalespecific protection and links with sex hormones. We have promising but preliminary results. Thus, in our revised manuscript, we added new data on the host response as suggested by the detailed comments from the Reviewers. We also elaborate on the potential role of the microbiota in the discussion section.

Reviewer #1 (Recommendations for the authors):

(1) The authors nicely showed that the transfer of the protective phenotype by FMT requires the female sex in recipients (Figure 2E). However, it remains unclear whether the female sex is required to develop protective microbiota in donor mice, as only the female NYU donor-male Jax recipient combination was tested. What happens if the microbiota from male NYU mice is transplanted into female Jax mice? If sex hormones act only on the downstream of the microbiota, such mice would show the protective phenotype. However, if sex hormones are required to establish a protective microbiota, the transplantation of microbiota from male NYU mice will not confer protection in recipient female Jax mice.

The Reviewer’s comment is well taken. We have not conducted the suggested experiment of FMT from male NYU mice to JAX female mice yet because we are pursuing an in vitro approach that we hope will eventually provide a more definitive answer. We observed that stool from female NYU mice and not JAX mice inhibits MRSA when cultured under anaerobic conditions, and this inhibitory activity is eliminated by filtration (Author response image 1A). We also observed that stool from male NYU mice inhibits MRSA growth to a similar extent as stool from female NYU mice (Author response image 1B). This result suggests that the protective role of sex hormones is downstream of the microbiota. We are in the process of identifying the specific microbiota member to support this conclusion.

Author response image 1.

Stool from NYU mice inhibits MRSA growth in vitro. (A) MRSA CFU/mL in media (TSB) following culture with unfiltered or filtered stool homogenate from female NYU or JAX mice. Stool homogenate or TSB alone was added in a 1:1 ratio to 1x106 CFU/mL MRSA and cultured anaerobically for up to 24 hours. (B) MRSA CFU/mL in TSB following culture with unfiltered stool homogenate from NYU male or female mice. Stool homogenate or TSB alone was added in a 1:1 ratio to 1x106 CFU/mL MRSA. 3 experimental replicates performed; stool taken from 6 individual mice per condition. Mean MRSA burden ± SEM. Area under the curve analysis + One way ANOVA with Sidak’s multiple comparisons test. ns: not significant.

(2) The results clearly showed the involvement of the specific microbiota in NYU mice in the sex-dependent bias in susceptibility to MRSA. However, the mechanisms by which specific microbiota promotes female sex-mediated protection need to be better described. Is this simply attributed to the different Th17 cell numbers in NYU and Jax mice (i.e., increased commensalspecific Th17 cells in NYU like Taconic mice)? Or is it possible that NYU microbiota impacts the regulation of sex hormones or their downstream signaling? What about the level of sex hormones in NYU and Jax mice? Are these levels equivalent or different? Do NYU and Jax microbiotas regulate the expression of sex hormone receptors in immune cells differently?

These are great questions. We do not observe baseline differences in Th17 cells like JAX versus Taconic mice (Figure 5B), suggesting that the mechanism is different. However, it is quite possible that an antigen-specific T cells, or Th17 cell specifically, is present at low levels and expands rapidly upon MRSA colonization. We have added this possibility to the discussion in the revised manuscript. To address the Reviewer’s question about the effect of the microbiota on sex hormones, we first sought to determine which sex hormone is necessary. Using estrogen receptor knockouts (Esr1^-/-), we were able to implicate estrogen and have added this important finding to the manuscript (Fig 6C). Then, we measured levels of estradiol in stool samples but did not observe a difference between NYU and JAX female mice (Author response image 2). We provide the results below but did not add it to the revised manuscript because we found it difficult to draw a conclusion without more extensive profiling as well as quantification of the receptor on specific immune cell subsets and cell-type specific knockouts. Also, see our response to Reviewer #3 regarding receptor expression. Although we have yet to explain the role of the microbiota, we hope the Reviewer agrees that we have promising yet preliminary results and that the new experiments we added to the manuscript have further strengthened the mechanism on the host-side. 

Author response image 2.

Estradiol levels in stool samples prior to MRSA inoculation. (A) Estradiol levels in stool samples collected prior to MRSA inoculation in male and female mice bred at NYU or purchased from Jackson Labs. Frozen stool samples were normalized by weight and processed using the DetectX® Estradiol ELISA Kit (Arbor Assays).

(3) The authors claimed that Th17-mediated recruitment of neutrophils likely promotes the clearance of MRSA in female NYU mice. However, the experimental evidence supporting this claim could be stronger. The authors should show the neutrophil recruitment in the gut mucosa in female and male NYU mice. Also, the levels of neutrophils between NYU and Jax female mice should be examined. To further strengthen the link between Th17 and neutrophils, it would be ideal to analyze neutrophil recruitment in mice lacking Th17 cells (i.e., Rag2-/-, anti-CD4 treated, Rorgt-/- mice).

We agree and now include a more detailed analyses of neutrophils. We found that the number of neutrophils in the intestine were not higher in NYU female mice compared with NYU male mice, with or without MRSA. Instead, we show that neutrophils in NYU female mice display higher levels of surface CD11b, a sign of activation, compared to males following inoculation with MRSA . We have added these findings to the revised manuscript (Fig5 H and I). IL-17 can activate neutrophils and increase their antimicrobial activity. Consistent with this possibility, we now show that female mice lacking the IL-17 receptor lose the enhanced colonization resistance. Based on these findings, we have modified this aspect of the conclusion, and thank the reviewer for the helpful suggestion.

Reviewer #2 (Public review):

The current study by Lejeune et al. investigates factors that allow for persistent MRSA infection in the GI tract. They developed an intriguing model of intestinal MRSA infection that does not use the traditional antibiotic approach, thereby allowing for a more natural infection that includes the normal intestinal microbiota. This model is more akin to what might be expected to be observed in a healthy human host. They find that biological sex plays a clear role in bacterial persistence during infection but only in mice bred at an NYU Facility and not those acquired from Jackson Labs. This clearly indicates a role for the intestinal microbiome in affecting female bacterial persistence but not male persistence which was unaffected by the origin of the mice and thus the microbiome. Through a series of clever microbiome-specific transfer experiments, they determine that the NYU-specific microbiome plays a role in this sexual dimorphism but is not solely responsible. Additional experiments indicate that Th17 cells, estrogen, and neutrophils also participate in the resistance to persistent infection. Notably, they assess the role of sex chromosomes (X/Y) using the established four core genotype model and find that these chromosomes appear to play little role in bacterial persistence.

Overall, the paper nicely adds to the growing body of literature investigating how biological sex impacts the immune system and the burden of infectious disease. The conclusions are mostly supported by the data although there are some aspects of the data that could be better addressed and clarified.

We thank the Reviewer for appreciating our contribution and these supportive comments. We have added several experiments to fill-in gaps and text revisions to increase clarity and acknowledge limitations. 

(1) There is something of a disconnect between the initial microbiome data and the later data that analyzes sex hormones and chromosomes. While there are clearly differences in microbial species across the two sites (NYU and JAX) how these bacterial species might directly interact with immune cells to induce female-specific responses is left unexplored. At the very least it would help to try and link these two distinct pieces of data to try and inform the reader how the microbiome is regulating the sex-specific response. Indeed, the reader is left with no clear exploration of the microbiota's role in the persistence of the infection and thus is left wanting.

We agree. This comment is similar to Reviewer #1’s feedback. As mentioned above, we are attempting to clarify the association between sex differences and the microbiota and have included preliminary results for the Reviewers. However, addressing this disconnect will require substantially more investigation. Instead, we have added insightful new data that elaborate on aspects of the host response.  We hope the Reviewer agrees that revised manuscript is stronger and that further delineation of the microbiota can be addressed by future studies.

(2) While the authors make a reasonable case that Th17 T cells are important for controlling infection (using RORgt knockout mice that cannot produce Th17 cells), it is not clear how these cells even arise during infection since the authors make most of the observations 2 days postinfection which is longer before a normal adaptive immune response would be expected to arise. The authors acknowledge this, but their explanation is incomplete. The increase in Th17 cells they observe is predicated on mitogenic stimulation, so they are not specific (at least in this study) for MRSA. It would be helpful to see a specific restimulation of these cells with MRSA antigens to determine if there are pre-existing, cross-reactive Th17 cells specific for MRSA and microbiota species which could then link these two as mentioned above.

We acknowledge that this is a limitation of our study. Although an experiment demonstrating pre-existing, cross-reactive T cells would help support our conclusion, aspects of MRSA biology may make the results of this experiment difficult to interpret. We have consulted with an expert on MRSA virulence factors, co-lead author Dr. Victor Torres, about the feasibility of this experiment. MRSA possess superantigens, such as Staphylococcal enterotoxin B, which bind directly to specific Vβ regions of T-cell receptors (TCR) and major histocompatibility complex (MHC) class II on antigen-presenting cells, resulting in hyperactivation of T lymphocytes and monocytes/macrophages. Additionally, other MRSA virulence factors, such as α-hemolysin and LukED, induce cell death of lymphocytes. MRSA’s enterotoxins are heat stable, so heat-inactivation of the bacterium may not help in this matter.  For these reasons, it is unlikely that we can perform a simple restimulation of lymphocytes with MRSA antigens. 

A study by Shao et al. provides an example of a host commensal species inducing Th17 cells with cross-reactivity against MRSA. Upon intestinal colonization, the intestinal fungus Candida albicans influences T cell polarization towards a Th17 phenotype in the spleen and peripheral lymph nodes which provided protection to the host against systemic candidemia. Interestingly, this induction of protective Th17 cells, increased IL-17 and responsiveness in circulating Ly6G+ neutrophils also protected mice from intravenous infection with MRSA, indicating that T cell activation and polarization by intestinal C. albicans leads to non-specific protective responses against extracellular pathogens.

Shao TY, Ang WXG, Jiang TT, Huang FS, Andersen H, Kinder JM, Pham G, Burg AR, Ruff B, Gonzalez T, Khurana Hershey GK, Haslam DB, Way SS. Commensal Candida albicans Positively Calibrates Systemic Th17 Immunological Responses. Cell Host & Microbe. 2019 Mar 13;25(3):404-417.e6. doi: 10.1016/j.chom.2019.02.004. PMID: 30870622; PMCID: PMC6419754.

We have added a brief version of the above discussion in the revised manuscript. Also, as mentioned earlier, we have added new data strengthening the axis between Th17 and neutrophils, including showing that IL-17 receptor is necessary and that neutrophils display signs of heightened activation in female mice during MRSA colonization.   

(3) The ovariectomy experiment demonstrates a role for ovarian hormones; however, it lacks a control of adding back ovarian hormones (or at least estrogen) so it is not entirely obvious what is causing the persistence in this experiment. This is especially important considering the experiments demonstrating no role for sex chromosomes thus demonstrating that hormonal effects are highly important. Here it leaves the reader without a conclusive outcome as to the exact hormonal mechanism.

This is a great suggestion. Rather than adding back ovarian hormones, we performed the more direct experiment and tested whether the estrogen receptor (ERα, encoded by Esr1) is necessary for the enhanced colonization resistance. Indeed, we observed that Esr1^-/- female mice have increased MRSA burden compared to Esr1^+/- littermates. We have added this new result (Figure 6C) and thank the Reviewer for their guidance. 

4) The discussion is underdeveloped and is mostly a rehash of the results. It would greatly enhance the manuscript if the authors would more carefully place the results in the context of the current state of the field including a more enhanced discussion of the role of estrogen, microbiome, and T cells and how the field might predict these all interact and how they might be interacting in the current study as well.

Author response: We thank the Reviewer for their feedback in improving the scholarship on the manuscript. We have expanded on the literature and the mechanistic model in both the discussion section and other parts to provide better context for our findings. 

Reviewer #3 (Public review):

Summary:

Using a mouse model of Staphylococcus aureus gut colonization, Lejeune et al. demonstrate that the microbiome, immune system, and sex are important contributing factors for whether this important human pathogen persists in the gut. The work begins by describing differential gut clearance of S. aureus in female B6 mice bred at NYU compared to those from Jackson Laboratories (JAX). NYU female mice cleared S. aureus from the gut but NYU male mice and mice of both sexes from JAX exhibited persistent gut colonization. Further experimentation demonstrated that differences between staphylococcal gut clearance in NYU and JAX female mice were attributed to the microbiome. However, NYU male and female mice harbor similar microbiomes, supporting the conclusion that the microbiome cannot account for the observed sex-dependent clearance of S. aureus gut colonization. To identify factors responsible for female clearance of S. aureus, the authors performed RNAseq on intestinal epithelial cells and cells enriched within the lamina propria. This analysis revealed sexdependent transcriptional responses in both tissues. Genes associated with immune cell function and migration were distinctly expressed between the sexes. To determine which immune cell types contribute to S. aureus clearance Lejeune et al employed genetic and antibody-mediated immune cell depletion. This experiment demonstrated that CD4+ IL17+ cells and neutrophils promote the elimination of S. aureus from the gut. Subsequent experiments, including the use of the 'four core genotype model' were conducted to discern between the roles of sex chromosomes and sex hormones. This work demonstrated that sex-chromosome-linked genes are not responsible for clearance, increasing the likelihood that hormones play a dominant role in controlling S. aureus gut colonization.

Strengths:

A strength of the work is the rigorous experimental design. Appropriate controls were executed and, in most cases, multiple approaches were conducted to strengthen the authors' conclusions. The conclusions are supported by the data.

The following suggestions are offered to improve an already strong piece of scholarship.

Weaknesses:

The correlation between female sex hormones and the elimination of S. aureus from the gut could be further validated by quantifying sex hormones produced in the four core genotype mice in response to colonization. Additionally, and this may not be feasible, but according to the proposed model administering female sex hormones to male mice should decrease colonization. Finally, knowing whether the quantity of IL-17a CD4+ cells change in the OVX mice has the potential to discern whether abundance/migration of the cells or their activation is promoted by female sex hormones.

In the Discussion, the authors highlight previous work establishing a link between immune cells and sex hormone receptors, but whether the estrogen (and progesterone) receptor is differentially expressed in response to S. aureus colonization could be assessed in the RNAseq dataset. Differential expression of known X and Y chromosome-linked genes were discussed but specific sex hormones or sex hormone receptors, like the estrogen receptor, were not. This potential result could be highlighted.

We appreciate the comment on the scholarship and thank the Reviewer for the insightful suggestions to improve this manuscript. We apologize for not including references that address some of the Reviewer’s questions. Other research groups have compared the levels of hormones between XX and XY males and females in the four core genotypes model and have found similar levels of circulating testosterone in adult XX and XY males. No difference was found in circulating estradiol levels in XX vs XY- females when tested at 4-6 or 79 months of age. 

Karen M. Palaszynski, Deborah L. Smith, Shana Kamrava, Paul S. Burgoyne, Arthur P. Arnold, Rhonda R. Voskuhl, A Yin-Yang Effect between Sex Chromosome Complement and Sex Hormones on the Immune Response. Endocrinology, Volume 146, Issue 8, 1 August 2005, Pages 3280–3285, https://doi.org/10.1210/en.2005-0284

Sasidhar MV, Itoh N, Gold SM, Lawson GW, Voskuhl RR. The XX sex chromosome complement in mice is associated with increased spontaneous lupus compared with XY. Ann Rheum Dis. 2012 Aug;71(8):1418-22. doi: 10.1136/annrheumdis-2011-201246. Epub 2012 May 12. PMID: 22580585; PMCID: PMC4452281.

Administering female sex hormones to males is a good idea. We did not observe an effect of injecting males with estrogen on MRSA colonization (data not shown), perhaps due to the dose or timing, or because it is not sufficient (i.e., additional hormones and factors may be required). Therefore, we analyzed the necessity of estrogen signaling and found that Esr1^-/- female mice impairs colonization resistance to MRSA. We have added this new experiment to the revised manuscript (Fig6 C).

Examination of the levels of estrogen, progesterone, and androgen receptors in our cecalcolonic lamina propria RNA-seq dataset is an excellent idea. We observed a significant increase in the G-protein coupled estrogen receptor 1 (Gper1) and a non-significant increase in Estrogen receptor alpha (Esr1) following MRSA inoculation in the immune cell compartment. This analysis has been added to the revised manuscript (Supplemental Fig6).

Reviewer #3 (Recommendations for the authors)

Minor editing issues:

The topic sentence of the last paragraph in the Results section states - 'male sex defining gene sex determining region Y (Sry) has been moved from the Y chromosome to an autosome'. 'Sex defining gene' and sex-determining region seems redundant in this context. A sex-defining gene would presumably be located within a sex-determining region.

Bold the letter 'F' in the Figure 5 legend.

It's not clear from the Figure 6E legend when the IL-17A+ CD4+ cells were quantified, 2 dpi?

In the third sentence of the second paragraph of the Discussion, the two references are merged together.

We thank the Reviewer for pointing out these editing issues. They have been addressed in the revised manuscript.

Author Response

The following is the authors’ response to the original reviews.

Reviewer #1:

This paper describes the role of WRNIP1 AAA+ ATPase, particularly its UBZ domain for ubiquitinbinding, but not ATPase, to prevent the formation of the R-loop when DNA replication is mildly perturbated. By combining cytological analysis for DNA damage, R-loop, and chromosome aberration with the proximity ligation assay for colocalization of various proteins involved in DNA replication and transcription, the authors provide solid evidence to support the claim. The authors also revealed a distinct role of WRNIP1 in the prevention of R-loop-induced DNA damage from FANCD2, which is inconsistent with the known relationship between WRNIP1 and FANCD2 in the repair of crosslinks.

One concern is the relationship between WRNIP1 and FANCD2 (Figure 6) in the suppression of Rloop-induced DNA damage. This is different from the relationship in inter-crosslink (ICL) repair (Socha et al. 2020), which shows the epistatic relationship between WRNIP1 as well as its UBZ domain and FANCD2 in the ICL repair. The authors need to re-evaluate the role of FNACD2 in Rloop suppression under mild replication stress (MRS) by analyzing R-loop formation in the FANCD2 knockdown (KD) cells as well as colocalization of FANCD2 with PCNA and RNA polymerase II by the PLA method and restarting the forks by the DNA coming.

In this line, it is important to show PLA signal between FANCD2 and R-loop depends on WRNIP1 since WRINP1 recruits FANCD2 in ICL repair (Socha et al. 2020).

In the study referenced by the reviewer, the authors implicated WRNIP1 in repairing interstrand crosslinks (ICLs) induced by agents, such as TMP/UVA, MMC, and Cisplatin (Socha et al., 2020). For the repair of ICLs, the FANCD2/FANCI complex, the central component of the FA pathway, must be recruited to DNA. The study suggests a potential role for WRNIP1 in loading the FANCD2/FANCI complex onto DNA immediately after ICL formation. However, even in the absence of WRNIP1, a residual recruitment of the FANCD2/FANCI complex to DNA was observed, possibly due to alternative mechanisms, as proposed by the authors. Interestingly, the study did not establish a similar relationship between WRNIP1 and FANCD2 after treatments that does not induce ICLs, demonstrating that WRNIP1 and FANCD2 may also play independent roles. Hence, our data demonstrating a distinct role of WRNIP1 from the FA pathway in response to R-loop-associated replication stress are not inconsistent with prior findings. Additionally, considering the UBZ domain ability to interact with ubiquitin in both its free form and when conjugated to other proteins, thereby regulating protein functions, it is not surprising that the UBZ domain of WRNIP1 may also play a role in the response to R-loop accumulation.

Therefore, to address the reviewer's request for a more in-depth exploration of the role of FANCD2 in the regulation of R-loops, we chose to examine the impact of FANCD2 loss on the accumulation of R-loops in WRNIP1-deficient and WRNIP1 UBZ mutant cells, as well as on the dynamics of stalled forks following aphidicolin-induced MRS. Additionally, we investigated the colocalization between FANCD2 and R-loops in shWRNIP1WT, shWRNIP1 and shWRNIP1D37A cells. Details are provided below.

In agreement with our observations, the analysis of R-loop formation upon MRS, in WRNIP1deficient cells depleted of FANCD2, revealed a significantly higher accumulation of R-loops in cells with a concomitant loss of both WRNIP1 and FANCD2 compared to those with a single deficiency (see Fig. 6D of the revised manuscript). Similar results were observed in the WRNIP1 UBZ mutant cells in which FANCD2 was abrogated (see Fig. 6D of the revised manuscript). It is important to note that, to eliminate contaminant-free RNA, particularly dsRNA, which could interfere with the binding of RNA-DNA hybrids by the S9.6 antibody (Hartono et al., 2018), and to determine the proximity between FANCD2 and R-loops more accurately, cells were treated with RNase III, following established protocols (Crossley et al., 2020).

Furthermore, we examined the interaction of FANCD2 with R-loops using a proximity ligation assay (PLA). Our findings revealed significant colocalization between FANCD2 and R-loops in the absence of WRNIP1 and in WRNIP1 UBZ mutant cells following low-dose aphidicolin treatment and RNase III exposure, showing a significant increase compared to the control counterpart (shWRNIP1WT cells; see Fig. 6B of the revised manuscript). Consequently, we conclude that neither WRNIP1 nor its UBZ domain is necessary for FANCD2 recruitment under conditions of MRS.

We also performed a DNA fiber assay to evaluate restarting replication forks in shWRNIP1WT, shWRNIP1 and shWRNIP1D37A cells in which FANCD2 was abrogated. Our results show that FANCD2 depletion slightly decreased the ability of the cells to restart forks from MRS (see Fig. 6E of the revised manuscript).

Given a low number (2-4) of PLA foci for WRNIP1-RNA polymerase II or WRNIP1 and R-loop (Figure 4B and 4D), how does this colocalization reflect the functional significance?

The data from the PLA of Figures 4B and 4D are reported as the mean of three independent experiments. It is important to note that we have introduced a new Figure 4D. To selectively assess R-loop structures, cells were treated with RNase III, a double-stranded RNA-specific endoribonuclease, following established protocols (Crossley et al., 2020). Our PLA analysis confirms the localization of WRNIP1 at/near R-loops in shWRNIP1 and shWRNIP1D37A cells, and this phenomenon is more evident in WRNIP1 UBZ mutant cells (see Fig. 4D of the revised manuscript). Specifically, the new protocol allows us to visualize a higher number of PLA foci, and we observed that Aph increased the spots per nucleus in shWRNIP1D37A cells compared to the previous experiment.

Regarding the Fig. 4B, it is not uncommon for a low number of PLA spots per nucleus to correspond to a phenotypic effect. For instance, a similar low average in the colocalization of PCNA or RNA pol II with FANCD2 has been observed in a prior paper as well, suggesting that transcription-replication collisions occur upon Aph-induced MRS (Okamoto et al., 2019). Also, not all R-loops could be “targeted” by WRNIP1.

It would be helpful to readers if the authors were to provide a summary figure of this paper.

As suggested by the reviewer, we have developed a model to summarize the findings obtained in our study (see Fig. 6F of the revised manuscript).

Minor points:

(1) Most of the cytological images in the paper show only colocalized ones, which makes it hard to see a signal. Please show a single-color image.

For a better visualization of nuclei signals in the figures, single-color images have been provided for Figs. 2A; 3B; 4A, B, C, D and E; 6B and D; Suppl. Fig. 2A and B of the revised manuscript.

(2) In Figure 2A, only one or two S9.6 focus(foci) can be seen. Why 1 or 2? This focus marks a specific chromosomal locus such as the centromere or telomere.

We agree with the reviewer that the observed foci in nuclei may indicate a specific chromosomal locus, such as telomeres or centromeres.

(3) Figure 3A, graph: Why this graph does not use a dot plot like Figure 1B and Figure 3C?

The graph in Figure 3A has been represented as a dot plot, as requested.

(4) Figure 1C: P values between unperturbed conditions should be provided.

In Figure 1C, P values comparing unperturbed conditions were already included. The results showed no significance between shWRNIP1 and shWRNIP1D37A cells when compared to MRC5SV cells and, similarly, to shWRNIP1T294A cells, as indicated in the corresponding legend.

(5) Figure 2B: Please provide the quantification or show the reproducibility of the data.

The quantification of R-loops using the S9.6 monoclonal antibody is not accurate, as the specificity for RNA-DNA hybrids is questionable (Hartono et al., 2018). Therefore, to demonstrate the reproducibility of the findings in Fig. 2B, we conducted a repeat of the dot-blot experiment. We treated the samples with RNase H to degrade RNA-DNA hybrids and hybridized the membrane with an anti-dsDNA to quantify R-loop levels more accurately. Our analysis confirms that the S9.6 signal strongly accumulates in shWRNIP1 cells compared to shWRNIP1WT cells (see Fig. 2B of the revised manuscript). Additionally, a graph illustrating the fold-change values of the S9.6/dsDNA signal relative to wild-type untreated cells is provided.

(6) Figure 4A: the expression of RNaseH under aphidicolin addition increased colocalization of PCNA and RNA pol II. It is important to mention the result and provide an explanation of why it is increasing in the main text.

Although the result may appear unexpected, and we lack experiments that explain the nature of this phenotype, a previous study reported that overexpression of RNase H1 in mammalian cells may lead to a dose-dependent reduction of certain proteins of the repair pathway, resulting in a significant accumulation of DNA damage (Shen et al., 2017). Consequently, the observed increase in TRCs upon RNase H1 overexpression in wild-type cells may be attributed to the disruption of proteins that, by impairing the repair process, can potentially cause more fork stalling and, consequently, more conflicts. We have introduced a comment in the text.

Reviewer #2:

This paper aims at establishing the role of WRN-interacting protein 1 (WRNIP1) and its UBZ domain (an N-terminal ubiquitin-binding zinc finger domain) on genome instability caused by mild inhibition of DNA synthesis by aphidicolin. The authors used human MRC5 fibroblasts investigated with standard methods in the field. The results clearly showed that WRNIP1 silencing and UBZ-mutation (D37A) increased DNA damage, chromosome aberrations, and transcription-replication conflicts caused by aphidicolin. The conclusions of the paper are overall well supported by results, however, aspects of some data analyses would need to be clarified and/or extended.

(1) The methods (immunofluorescence microscopy and dot-blots) to determine R-loop levels can lack sensitivity and specificity. In particular, since the S9.6 antibody can bind to other structures besides heteroduplex, dot-blot analyses only grossly assess R-loop levels in cellular samples of purified nucleic acids, which are constituted by many different types of DNA/RNA structures.

To eliminate contaminant-free RNA, particularly dsRNA, which could interfere with the capture of RNA-DNA hybrids by the S9.6 antibody (Hartono et al., 2018), and to determine R-loop levels more accurately, we treated cells with RNase III, following established protocols (Crossley et al., 2020). Under our experimental conditions, RNase III treatment significantly reduced the amount of dsRNA, nearly eliminating it, as evaluated using a specific antibody against dsRNA (see Suppl Fig 2 of the revised manuscript). To better appreciate the effect of the loss of WRNIP1 or its UBZ domain on Rloop accumulation and the amount of DNA damage, we have reproduced key data (see Figs 2B; 3B; 4D and E; 6B of the revised manuscript). Our analysis from immunofluorescence experiments, performed using a dsRNA ribonuclease (RNase III), confirms higher R-loop accumulation in WRNIP1-deficient or WRNIP1 UBZ mutant cells compared to control cells (Fig 3B). Additionally, proximity ligation assay (PLA) data are consistent with those previously presented and, in some cases, are more readily interpretable (see Figs 4D and E; 6B of the revised manuscript). Finally, we performed a new dot-blot experiment (see Fig. 2B of the revised manuscript). We treated with RNase H to degrade RNA-DNA hybrids and hybridized the membrane with an anti-dsDNA antibody to quantify R-loop levels more accurately. Our analysis confirms a significant accumulation of the S9.6 signal in shWRNIP1 cells compared to shWRNIP1WT cells. Additionally, a graph illustrating the foldchange values of the S9.6/dsDNA signal relative to wild-type untreated cells is provided.

(2) Experimental plan has analyzed the impact of WRNIP1 lack or mutations at steady-state conditions. Thus, the possible role of WRNIP1 at an early step of the mechanism would require some sort of kinetics analysis of the molecular process, therefore not at steady-state conditions. The findings of a co-localization of R-loops and WRNIP1 have been obtained with the S9.6 antibody, which recognizes DNA-RNA heteroduplexes. Since WRNIP1 is known to be recruited at stalled forks and DNA cleavage sites, it is not surprising that WRNIP1 is very close to heteroduplexes, abundant structures at replication forks and cleavage sites. Similar interpretations may also be valid for Rad51/S9.6 co-localization findings.

Investigating the potential role of WRNIP1 at an early step in the mechanism is undoubtedly very interesting and requires separate investigation. Our decision to explore the relevance of the loss of WRNIP1 or WRNIP1 mutations under steady-state conditions is based on a preliminary alkaline comet assay (provided below). The comet assay, performed at various exposure times of aphidicolin at a concentration of 0.4 micromolar, clearly indicates that the most significant effect on DNA damage accumulation in WRNIP1-deficient cells occurs after 24 hours of treatment. Therefore, we have chosen to study the transcription-associated genomic instability in our cells by treating them with a low-dose of aphidicolin for 24 hours to maximize the effect.

Author response image 1.

We agree that the presence of WRNIP1 or RAD51 in proximity to R-loops is consistent with their roles and may not be surprising. However, these experiments formally demonstrate their proximity to R-loops under our conditions. Notably, the new graphs, obtained from experiments repeated by treating with RNase III to reduce the amount of dsRNA and improve the specificity of the S9.6 antibody, show increased interaction of the mutated form of WRNIP1 in the UBZ domain with Rloops when compared to the wild-type form. Additionally, it is more evident that the presence of RAD51 at/near R-loops is reduced in WRNIP1 UBZ mutant cells both in untreated conditions and after MRS (see Figs 4D and E of the revised manuscript).

(3) Determination of DNA damage, chromosome aberration, and co-localization data are reported as means of measurements with appropriate statistics. However, the fold-change values relative to corresponding untreated samples are not reported. In some instances, it seems that WRNIP1 silencing or mutations actually reduce or do not affect aphidicolin effects. That leaves open the interpretation of specific results.

To better evaluate the significance of the data presented in the study, we have introduced the foldchange values calculated with respect to the untreated samples, as requested by the reviewer. This allowed us to conclude that the loss of WRNIP1 or the expression of the UBZ mutant form of WRNIP1 does not reduce in any case the effects of aphidicolin-induced mild replication stress.

I would suggest some additional experiments or analyses to get more convincing results:

(1) DNA damage should be verified also with other methods, such as DNA damage markers pH2AX and 53BP1.

The quantification of DNA damage was also corroborated by determining the percentage of gammaH2AX-positive cells, as reported in Supplementary Figure 1B. This result is consistent with the findings from the comet assay, confirming transcription-dependent DNA accumulation in shWRNIP1 and shWRNIP1D37A cells. Regarding the 53BP1 marker, we believe that the existing data sufficiently demonstrate DNA damage accumulation in the absence of WRNIP1 or when its UBZ domain is mutated, providing comprehensive support to the study without necessitating additional results.

(2) Repair foci may also be detected with Rad51 foci. That will also provide evidence for increased DNA damage levels under the tested conditions.

Our prior study identified WRNIP1 as a crucial factor for RAD51 function (Leuzzi et al., 2016). Loss of WRNIP1 indeed results in a defective relocalization of RAD51 to chromatin. Consequently, the analysis of RAD51 foci may be not a useful readout to evaluate DNA damage levels under our conditions.

(3) WRNIP1 effects should be presented as FC (fold-changes) of DNA damage, PLA results, chromosomal errors, etc, to provide evidence of the level of effects on the tested phenotypes.

We have introduced the fold-change values calculated with respect to the untreated samples, as requested by the reviewer, for a more comprehensive analysis in the graph of Figs. 1B, C and D; 2A and B; 3A, B and C; 4A, B, C, D and E; 6B, C and D.

(4) R-loop detection ideally should be performed by one of the several types of immunoprecipitation techniques. Alternatively, dot-blot assays should be performed with a 1:2 dilution series of each sample. Then, heteroduplexes should be detected with S9.6 along with a general aspecific dye for DNA quantity in each spot. Next, densitometric analyses of S9.6 signal should be normalized over DNA quantity.

We acknowledge that the quantification of R-loops using the S9.6 monoclonal antibody is not accurate, as the specificity for RNA-DNA hybrids is questionable (Hartono et al., 2018). Therefore, to overcome this issue, we repeated the experiment shown in Fig. 2B. We treated the samples with RNase H to degrade RNA-DNA hybrids and hybridized the membrane with an anti-dsDNA antibody to quantify R-loop levels more accurately. Our analysis confirms that the S9.6 signal strongly accumulates in shWRNIP1 cells compared to shWRNIP1WT cells (see Fig. 2B of the revised manuscript). Additionally, a graph illustrating the fold-change values of the S9.6/dsDNA signal relative to wild-type untreated cells is provided.

(5) A major focus on WRNIP1 D37A and T294A mutations may also make the paper overall more convincing. For instance: do the mutations affect protein recruitment at damaged chromatin? Do they increase repair foci? Do they affect the recruitment of WRN or BLM helicases or specific nucleases at chromatin under the tested conditions of MRS?

To address this point raised by the reviewer, we performed a chromatin experiment to assess the ability of WRNIP1 and its mutated forms to translocate to chromatin upon MRS. Our analysis shows that the mutated forms of WRNIP1 do not exhibit any defects in recruitment to chromatin, although the levels of the WRNIP1 ATPase mutant appear lower than the others (see Western blotting provided below for the reviewer’s use only, Fig. A). Additionally, we tested the presence of WRN helicase, which does not show any difference between cells lines (see Western blot provided below, Author Response image 2B).

Author response image 2.

(6) I suggest revising the text for spelling errors.

The manuscript has been carefully revised to identify and correct any spelling errors that may have occurred.

Reviewer #3:

In the manuscript by Valenzisi et al., the authors report on the role of WRNIP1 to prevent R-loop and TRC-associated DNA damage. The authors claim WRNIP1 localizes to TRCs in response to replication stress and prevents R-loop accumulation, TRC formation, replication fork stalling, and subsequent DNA damage. While the findings are of potential significance to the field, the strength of evidence in support of the conclusions is lacking.

Weaknesses:

(1) The authors fail to utilize the proper controls throughout the manuscript in regard to the shWRNIP1, WT, and mutant cell lines. It is unclear why the authors failed to use the shWRNIP1WT line in the comet assay, DNA fiber assay, and the FANCD2 assays. This is a key control for i) the use of only a single shRNA (most studies will use at least 2 different shRNAs) and ii) the use of the mutant WRNIP1 lines. In several figures, the authors only show the effect of the UBZ mutant, but don't include the ATPase mutant or WT for comparison. Including these is essential.

We agree with the reviewer's criticism that the use of shWRNIP1WT cells as a control is more appropriate. Therefore, all the new experiments presented in the revised version of the manuscript have been performed using the shWRNIP1WT cells. Notably, new results are in line with those obtained using the MRC5SV cells, rendering us confident that our findings are reliable overall. By contrast, we do not feel that including the WRNIP1 ATPase mutant cells is always essential, since our data clearly demonstrate that the loss of ATPase activity of WRNIP1 does not affect transcriptionassociated genome instability.

(2) The authors use the S9.6 antibody to conclude the loss of WRNIP1 causes more R-loops; however, it has been shown that this antibody detects dsRNA in addition to RNA-DNA hybrids. Accordingly, it cannot be ruled out that the increased S9.6 signal is due to increased dsRNA.

To eliminate contaminant-free RNA, particularly dsRNA, which could interfere with the capture of RNA-DNA hybrids by the S9.6 antibody (Hartono et al., 2018), and to determine R-loop levels more accurately, we treated cells with RNase III, following established protocols (Crossley et al., 2020). Under our experimental conditions, RNase III treatment significantly reduced the amount of dsRNA, nearly eliminating it, as evaluated using a specific antibody against dsRNA (see Suppl Fig 2 of the revised manuscript). To better appreciate the effect of the loss of WRNIP1 or its UBZ domain on Rloop accumulation and the amount of DNA damage, we have reproduced key data (see Figs 3B; 4D and E; 6B, D and E of the revised manuscript). Our analysis from immunofluorescence experiments, performed using a dsRNA ribonuclease, confirms higher R-loop accumulation in WRNIP1-deficient or UBZ WRNIP1 mutant cells compared to control cells (Fig. 3B). Additionally, proximity ligation assay (PLA) data are consistent with those previously presented and, in some cases, are more readily interpretable (see Figs 4D and E; 6B of the revised manuscript).

(3) Multiple pieces of data do not support the conclusions. For example, Figure 1D shows shWRNIP1 to reduce damage in Aph+DRB cells compared to MRC5SV cells with Aph+DRB. This result suggests that WRNIP1 actually increases DNA damage in stressed cells with transcription blocked. Another result is seen in Figure 4a, where the number of PLA spots (presumably TRCs) increases in the shWRNIP1WT cells with Aph+RNH1 compared to Aph alone. If R-loops are required for TRC accumulation, then the RNH1 should decrease the PLA foci. This result instead suggests that WRNIP leads to increased TRCs in stressed cells with R-loops cleared by RNH1.

Regarding Figure 1D, in MRC5SV cells, DRB does not significantly increase DNA damage upon Aph treatment. Therefore, it is not correct to conclude that WRNIP1 exacerbates DNA damage in stressed cells with transcription blocked.

Regarding Figure 4A, while the outcome may appear unexpected, and we do not provide data that explain the nature of this phenotype, a previous study demonstrated that overexpression of RNase H1 in mammalian cells may lead to a dose-dependent reduction of certain proteins of the repair pathway, leading to a significant accumulation of DNA damage (Shen et al., 2017). Accordingly, the observed increase in TRCs upon RNase H1 overexpression in wild-type cells may be attributed to the disruption of proteins that, by impairing the repair process, can potentially cause more fork stalling and, consequently, more conflicts. We have introduced a comment in the text.

(4) The data are mostly phenomenological and fail to yield mechanistic insight. For example, the authors state that "it remains unclear whether WRNIP1 is directly involved in the mechanisms of Rloop removal/resolution". Unfortunately, the data presented in this manuscript do not provide new insights into this unresolved question.

We agree with the reviewer that elucidating the mechanism by which WRNIP1 contributes to R-loop suppression would be of interest. Nevertheless, the findings presented here provide compelling evidence of a novel role for WRNIP1 in preventing R-loop accumulation. Investigating how WRNIP1 accomplishes this function will require significant effort, which we are committed to undertaking.

(5) The authors only show merged images making it impossible to visualize differences in PLA foci.

For a better visualization of nuclei signals in the PLA panels of Figs 4A, B, C, D and E; 6B, singlecolor images have been provided.

In addition to including the controls I mentioned in the public review, I recommend investigating the mechanism of how WRNIP1 prevents R-loop accumulation. If it is indeed related to its UBZ domain, then does that mean ubiquitination is an important step in R-loop removal? I believe elucidating this would be a novel and significant contribution. If it's not related to ubiquitination, then how does the UBZ domain regulate R-loops?

We agree with the reviewer that investigating the precise role of the UBZ domain of WRNIP1 in Rloop prevention would be of interest, and several experiments are required to adequately address this issue. However, as discussed, we hypothesize that the UBZ domain might contribute to directing WRNIP1 to DNA at TRC sites through RAD18.

I recommend using purified RNH1-dead-GFP to detect R-loops as opposed to the S9.6 antibody. The Cimprich lab has published this recently as a tool for detecting R-loops in fixed cells.

As explained in point 2), to eliminate contaminant-free RNA, particularly dsRNA, which could interfere with the capture of RNA-DNA hybrids by the S9.6 antibody (Hartono et al., 2018), and to determine R-loop levels more accurately, we used treatment with RNase III, following established protocols (Crossley et al., 2020). New experiments are reported in the revised version of the manuscript for R-loops in all cell lines (see Fig. 3B of the revised manuscript).

Additionally, colocalization by PLA of WRNIP1/R-loops, RAD51/R-loops, FANCD2/R-loops, and R-loop accumulation by anti-S9.6 antibody in cells depleted of FANCD2 are presented (see Figs. 4D and E; 6B and D of the revised manuscript).

Furthermore, we repeated the dot-blot experiment (see Fig. 2B of the revised manuscript). We treated the samples with RNase H to degrade RNA-DNA hybrids and hybridized the membrane with an antidsDNA antibody to quantify R-loop levels more accurately. Our analysis confirms that the S9.6 signal strongly accumulates in shWRNIP1 cells compared to shWRNIP1WT cells. Additionally, a graph illustrating the fold-change values of the S9.6/dsDNA signal relative to wild-type untreated cells is provided.

Importantly, overall, our findings suggest that treatment with RNase III does not substantially change the results obtained without it, but in some cases, such as in Fig. 4D, makes them are more readily interpretable. Specifically, the new protocol allows us to visualize a higher number of PLA foci, and Aph increased the spots per nucleus in shWRNIP1D37A cells compared to the previous experiment (see Fig. 4D of the revised manuscript).

Author Response

The following is the authors’ response to the original reviews.

We would like to thank the reviewers for their thoughtful evaluation of our manuscript. We considered all the comments and prepared the revised version. The following are our responses to the reviewers’ comments. All references, including those in the original manuscript are included at the end of this point-by-point response.

Reviewer #1 (Public Review):

Weaknesses:

1) The authors should better review what we know of fungal Drosophila microbiota species as well as the ecology of rotting fruit. Are the microbiota species described in this article specific to their location/setting? It would have been interesting to know if similar species can be retrieved in other locations using other decaying fruits. The term 'core' in the title suggests that these species are generally found associated with Drosophila but this is not demonstrated. The paper is written in a way that implies the microbiota members they have found are universal. What is the evidence for this? Have the fungal species described in this paper been found in other studies? Even if this is not the case, the paper is interesting, but there should be a discussion of how generalizable the findings are.

The reviewer inquires as to whether the microbial species described in this article are ubiquitously associated with Drosophila or not. Indeed, most of the microbes described in this manuscript are generally recognized as species associated with Drosophila spp. For example, yeasts such as Hanseniaspora uvarum, Pichia kluyveri, and Starmerella bacillaris have been detected in or isolated from Drosophila spp. collected in European countries as well as the United States and Oceania (Chandler et al., 2012; Solomon et al., 2019). As for bacteria, species belonging to the genera Pantoea, Lactobacillus, Leuconostoc, and Acetobacter have also previously been detected in wild Drosophila spp. (Chandler et al., 2011). These statements have been incorporated into our revised manuscript (lines 391-397). Nevertheless, the term “core” in the manuscript and title may lead to misunderstanding, as the generality does not ensure the ubiquitous presence of these microbial species in every individual fly. Considering this point, we replaced the “core” with “key,” a term that is more appropriate to our context.

2) Can the authors clearly demonstrate that the microbiota species that develop in the banana trap are derived from flies? Are these species found in flies in the wild? Did the authors check that the flies belong to the D. melanogaster species and not to the sister group D. simulans?

Can the authors clearly demonstrate that the microbiota species that develop in the banana trap are derived from flies? Are these species found in flies in the wild?

The reviewer asked whether the microbial species detected from the fermented banana samples were derived from flies. To address this question, additional experiments under more controlled conditions would be needed, such as artificially introducing wild flies onto fresh bananas in the laboratory. Nevertheless, the microbes potentially originate from wild flies, as supported by the literature cited in our response to the Weakness 1).

Alternative sources of microbes also merit consideration. For example, microbes may have been introduced to unfermented bananas by penetration through peel injuries (lines 1300-1301). In addition, they could be introduced by insects other than flies, given that rove beetles (Staphylinidae) and sap beetles (Nitidulidae) were observed in some of the traps. The explanation of these possibilities have been incorporated into DISCUSSION (lines 414427) of our revised manuscript.

Did the authors check that the flies belong to the D. melanogaster species and not to the sister group D. simulans?

Our sampling strategy was designed to target not only D. melanogaster but also other domestic Drosophila species, such as D. simulans, that inhabit human residential areas. For the traps where adult flies were caught, we identified the species of the drosophilids as shown in Table S1, thereby showing the presence of either or both D. melanogaster and D. simulans. We added these descriptions in MATERIALS AND METHODS (lines 511-512 and 560-562), and DISCUSSION (lines 378-379).

3) Did the microarrays highlight a change in immune genes (ex. antibacterial peptide genes)? Whatever the answer, this would be worth mentioning. The authors described their microarray data in terms of fed/starved in relation to the Finke article. They should clarify if they observed significant differences between species (differences between species within bacteria or fungi, and more generally differences between bacteria versus fungi).

Did the microarrays highlight a change in immune genes (ex. antibacterial peptide genes)? Whatever the answer, this would be worth mentioning.

Regarding the antimicrobial peptide genes, statistical comparisons of our RNA-seq data across different conditions were impracticable because most of the genes showed low expression levels. The RNA-seq data of the yeast-fed larvae is shown in Author response Table 1. While a subset of genes exhibited significantly elevated expression in the nonsupportive conditions relative to the supportive ones, this can be due to intra-sample variability rather than the difference in the nutritional conditions. Similar expression profiles were observed in the bacteria-fed larvae as well (data not shown). Therefore, it is difficult to discuss a change in immune genes in the paper. Additionally, the previous study that conducted larval microarray analysis (Zinke et al., 2002) did not explicitly focus on immune genes.

Author response table 1.

Antimicrobial peptide genes are not up-regulated by any of the microbes. Antimicrobial peptides gene expression profiles of whole bodies of first-instar larvae fed on yeasts. TPM values of all samples and comparison results of gene expression levels in the larvae fed on supportive and non-supportive yeasts are shown. Antibacterial peptide genes mentioned in Hanson and Lemaitre, 2020 are listed. NA or na, not available.

They should clarify if they observed significant differences between species (differences between species within bacteria or fungi, and more generally differences between bacteria versus fungi).

We did not observe significant differences in the gene expression profiles of the larvae fed on different microbial species within bacteria or fungi, or between those fed on bacteria and those fed on fungi. For example, the gene expression profiles of larvae fed on the various supportive microbes showed striking similarities to each other, as evidenced by the heat map showing the expression of all genes detected in larvae fed either yeast or bacteria (Author response image 1). Similarities were also observed among larvae fed on various nonsupportive microbes.

Only a handful of genes showed different expression patterns between larvae fed on yeast and those fed on bacteria. Thus, it is challenging to discuss the potential differential impacts of yeast and bacteria on larval growth, if any.

Author response image 1.

Gene expression profiles of larvae fed on the various supporting microbes show striking similarities to each other. Heat map showing the gene expression of the first-instar larvae that fed on yeasts or bacteria. Freshly hatched germ-free larvae were placed on banana agar inoculated with each microbe and collected after 15 h feeding to examine gene expression of the whole body. Note that data presented in Figures 3A and 4C in the original manuscript, which are obtained independently, are combined to generate this heat map. The labels under the heat map indicate the microbial species fed to the larvae, with three samples analyzed for each condition. The lactic acid bacteria (“LAB”) include Lactiplantibacillus plantarum and Leuconostoc mesenteroides, while the lactic acid bacterium (“AAB”) represents Acetobacter orientalis. “LAB + AAB” signifies mixtures of the AAB and either one of the LAB species. The asterisks in the label highlight “LAB + AAB” or “LAB” samples clustered separately from the other samples in those conditions; “” indicates a sample in a “LAB + AAB” condition (Lactiplantibacillus plantarum + Acetobacter orientalis), and “*” indicates a sample in a “LAB” condition (Leuconostoc mesenteroides). Brown abbreviations of scientific names are for the yeast-fed conditions. H. uva, Hanseniaspora uvarum; K. hum, Kazachstania humilis; M. asi, Martiniozyma asiatica; Sa. cra, Saccharomycopsis crataegensis; P. klu, Pichia kluyveri; St. bac, Starmerella bacillaris; BY4741, Saccharomyces cerevisiae BY4741 strain.

4) The whole paper - and this is one of its merits - points to a role of the Drosophila larval microbiota in processing the fly food. Are these bacterial and fungal species found in the gut of larvae/adults? Are these species capable of establishing a niche in the cardia of adults as shown recently in the Ludington lab (Dodge et al.,)? Previous studies have suggested that microbiota members stimulate the Imd pathway leading to an increase in digestive proteases (Erkosar/Leulier). Are the microbiota species studied here affecting gut signaling pathways beyond providing branched amino acids?

The whole paper - and this is one of its merits - points to a role of the Drosophila larval microbiota in processing the fly food. Are these bacterial and fungal species found in the gut of larvae/adults? Are these species capable of establishing a niche in the cardia of adults as shown recently in the Ludington lab (Dodge et al.,)?

Although we did not investigate the microbiota in the gut of either larvae or adults, we did compare the microbiota within surface-sterilized larvae or adults with the microbiota in food samples. We found that adult flies and early-stage foods, as well as larvae and late-stage foods, harbored similar microbial species (Figure 1F). Additionally, previous studies examining the gut microbiota in wild adult flies have detected microbes belonging to the same species or taxa as those isolated from our foods (Chandler et al., 2011; Chandler et al., 2012). We have elaborated on this in our response to Weakness 1).

While we did not investigate whether these species are capable of establishing a niche in the cardia of adults, we have cited the study by Dodge et al., 2023 in our revised manuscript and discussed the possibility that predominant microbes in adult flies may show a propensity for colonization (lines 410-413).

Previous studies have suggested that microbiota members stimulate the Imd pathway leading to an increase in digestive proteases (Erkosar/Leulier). Are the microbiota species studied here affecting gut signaling pathways beyond providing branched amino acids?

The reviewer inquires whether the supportive microbes in our study stimulate gut signaling pathways and induce the expression of digestive protease genes, as demonstrated in a previous study (Erkosar et al., 2015). Based on our RNA-seq data, this is unlikely. The aforementioned study demonstrated that seven protease genes are upregulated through Imd pathway stimulation by a bacterium that promotes the larval growth. In our RNA-seq analysis, these seven genes did not exhibit a consistent upregulation in the presence of the supportive microbes (H. uva or K. hum in Author response table 2A; Le. mes + A. ori in Author response table 2B). Rather, they exhibited a tendency to be upregulated by the presence of non-supportive microbes (St. bac or Pi. klu in Author response table 2A; La. pla in Author Response Table 2B).

Author response table 2.

Most of the peptidase genes reported by Erkosar et al., 2015 are more highly expressed under the non-supportive conditions than the supportive conditions. Comparison of the expression levels of seven peptidase genes derived from the RNA-seq analysis of yeast-fed (A) or bacteria-fed (B) first-instar larvae. A previous report demonstrated that the expression of these genes is upregulated upon association with a strain of Lactiplantibacillus plantarum, and that the PGRP-LE/Imd/Relish signaling pathway, at least partially, mediates the induction (Erkosar et al., 2015). H. uva, Hanseniaspora uvarum; K. hum, Kazachstania humilis; P. klu, Pichia kluyveri; S. bac, Starmerella bacillaris; La. pla, Lactiplantibacillus plantarum; Le. mes, Leuconostoc mesenteroides; A. ori, Acetobacter orientalis; ns, not significant.

Reviewer #2 (Public Review):

Weaknesses:

The experimental setting that, the authors think, reflects host-microbe interactions in nature is one of the key points. However, it is not explicitly mentioned whether isolated microbes are indeed colonized in wild larvae of Drosophila melanogaster who eat bananas. Another matter is that this work is rather descriptive and a few mechanical insights are presented. The evidence that the nutritional role of BCAAs is incomplete, and molecular level explanation is missing in "interspecies interactions" between lactic acid bacteria (or yeast) and acetic acid bacteria that assure their inhabitation. Apart from these matters, the future directions or significance of this work could be discussed more in the manuscript.

The experimental setting that, the authors think, reflects host-microbe interactions in nature is one of the key points. However, it is not explicitly mentioned whether isolated microbes are indeed colonized in wild larvae of Drosophila melanogaster who eat bananas.

The reviewer asks whether the isolated microbes were colonized in the larval gut. Previous studies on microbial colonization associated with Drosophila have predominantly focused on adults (Pais et al. PLOS Biology, 2018), rather than larval stages. Developing larvae continually consume substrates which are already subjected to microbial fermentation and abundant in live microbes until the end of the feeding larval stage. Therefore, we consider it difficult to discuss microbial colonization in the larval gut. We have mentioned this point in DISCUSSION of the revised manuscript (lines 408-410).

Another matter is that this work is rather descriptive and a few mechanical insights are presented. The evidence that the nutritional role of BCAAs is incomplete, and molecular level explanation is missing in "interspecies interactions" between lactic acid bacteria (or yeast) and acetic acid bacteria that assure their inhabitation.

While we recognize the importance of comprehensive mechanistic analysis, elucidation of more detailed molecular mechanisms lies beyond the scope of this study and will be a subject of future research.

Regarding the nutritional role of BCAAs, the incorporation of BCAAs enabled larvae fed with the non-supportive yeast to grow to the second-instar stage. This observation implies that consumption of BCAAs upregulates diverse genes involved in cellular growth processes in larvae. We mentioned a previously reported interaction between lactic acid bacteria (LAB) and acetic acid bacteria (AAB) in the manuscript (lines 433-436). LAB may facilitate lactate provision to AAB, consequently enhancing the biosynthesis of essential nutrients such as amino acids. To test this hypothesis, future experiments will include the supplementation of lactic acid to AAB culture plates, and the co-inoculation of AAB with LAB mutant strains defective in lactate production to assess both larval growth and continuous larval association with AAB. With respect to AAB-yeast interactions, metabolites released from yeast cells might benefit AAB growth, and this possibility will be investigated through the supplementation of AAB culture plates with candidate metabolites identified in the cell suspension supernatants of the late-stage yeasts.

Apart from these matters, the future directions or significance of this work could be discussed more in the manuscript.

We appreciate the reviewer's recommendations. The explanation of the universality of our findings has been included in the revised DISCUSSION (lines 391-397). We have also added descriptions on the implication of compositional shifts occurring in adult microbiota (lines 404413), possible inoculation routes of different microbes (lines 414-427), and hypotheses on the mechanism of larval growth promotion by yeasts (lines 469-476), all of which could be the focus of our future study.

Reviewer #3 (Public Review):

Weaknesses:

Despite describing important findings, I believe that a more thorough explanation of the experimental setup and the steps expected to occur in the exposed diet over time, starting with natural "inoculation" could help the reader, in particular the non-specialist, grasp the rationale and main findings of the manuscript. When exactly was the decision to collect earlystage samples made? Was it when embryos were detected in some of the samples? What are the implications of bacterial presence in the no-fly traps? These samples also harbored complex microbial communities, as revealed by sequencing. Were these samples colonized by microbes deposited with air currents? Were they the result of flies that touched the material but did not lay eggs? Could the traps have been visited by other insects? Another interesting observation that could be better discussed is the fact that adult flies showed a microbiome that more closely resembles that of the early-stage diet, whereas larvae have a more late-stage-like microbiome. It is easy to understand why the microbiome of the larvae would resemble that of the late-stage foods, but what about the adult microbiome? Authors should discuss or at least acknowledge the fact that there must be a microbiome shift once adults leave their food source. Lastly, the authors should provide more details about the metabolomics experiments. For instance, how were peaks assigned to leucine/isoleucine (as well as other compounds)? Were both retention times and MS2 spectra always used? Were standard curves produced? Were internal, deuterated controls used?

When exactly was the decision to collect early-stage samples made? Was it when embryos were detected in some of the samples?

We collected traps and early-stage samples 2.5 days after setting up the traps. This duration was determined from pilot experiments. A shorter collection time resulted in a lower likelihood of obtaining traps visited by adult flies, whereas a longer collection time caused overcrowding of larvae as well as deaths of adults from drowning in the liquid seeping out of the fruits. These procedural details have been included in the MATERIALS AND METHODS section of the revised manuscript (lines 523-526).

What are the implications of bacterial presence in the no-fly traps? These samples also harbored complex microbial communities, as revealed by sequencing. Were these samples colonized by microbes deposited with air currents? Were they the result of flies that touched the material but did not lay eggs? Could the traps have been visited by other insects?

We assume that the origins of the microbes detected in the no-fly trap foods vary depending on the species. For instance, Colletotrichum musae, the fungus that causes banana anthracnose, may have been present in fresh bananas before trap placement. The filamentous fungi could have originated from airborne spores, but they could also have been introduced by insects that feed on these fungi. We have included these possibilities in the DISCUSSION section of the revised manuscript (lines 417-421).

Another interesting observation that could be better discussed is the fact that adult flies showed a microbiome that more closely resembles that of the early-stage diet, whereas larvae have a more late-stage-like microbiome. It is easy to understand why the microbiome of the larvae would resemble that of the late-stage foods, but what about the adult microbiome? Authors should discuss or at least acknowledge the fact that there must be a microbiome shift once adults leave their food source.

We are grateful for the reviewer's insightful suggestion regarding shifts in the adult microbiome. We have included in the DISCUSSION section of the revised manuscript the possibility that the microbial composition may change substantially during pupal stages or after adult eclosion (lines 404-413).

Lastly, the authors should provide more details about the metabolomics experiments. For instance, how were peaks assigned to leucine/isoleucine (as well as other compounds)? Were both retention times and MS2 spectra always used?

In this metabolomic analysis, LC-MS/MS with triple quadrupole MS monitors the formation of fragment ions from precursor ions specific to each target compound. The use of PFPP columns, which provide excellent separation of amino acids and nucleobases, allows chromatographic peaks of many structural isomers to be separated into independent peaks. In addition, all measured compounds are compared with data from a standard library to confirm retention time agreement. Structural isomers were separated either by retention time on the column or by compound-specific MRM signals (in fact, leucine and isoleucine have both unique MRM channels and column separations). Detailed MRM conditions are identical to the previously published study (Oka et al., 2017). These have been included in the revised ‘LC-MS/MS measurement’ section in MATERIALS AND METHODS (lines 810-824).

Were standard curves produced?

Since relative quantification of metabolite amounts was performed in this study, no standard curve was generated to determine absolute concentrations. However, a standard compound of known concentration (single point) was measured to confirm retention time and relative area values.

Were internal, deuterated controls used?

Internal standards for deuterium-labeled compounds were not used in this study. This is because it is not realistic to obtain deuterium-labeled compounds for all compounds since a large number of compounds are measured. However, an internal standard (L-methionine sulfone) is added to the extraction solvent to calculate the recovery rate. This has been included in the revised ‘LC-MS/MS measurement’ section in MATERIALS AND METHODS (lines 824-825).

Reviewer #1 (Recommendations For The Authors):

Additional comments 1. The authors should do a better job of presenting their data. It took me quite a while to understand the protocol of Figure 1. Panel 1A, B, C could be improved. For instance, 1A suggests that flies are transferred to the lab while this is in fact the banana trap. Indicate 'Banana trap colonized by flies' rather 'wild-type flies in the trap'. 1C: should indicate that the food suspension comes from the banana trap. 1B,D,D: do not use pale color as legend. Avoid the use of indices in Figure 2 (Y1 rather than Y1). Grey colors are difficult to distinguish in Figure 2. Etc. It is a pain for reviewers that figure legends are on the verso of each figure and not just below.

We thank the reviewer for the detailed suggestions to improve the clarity and comprehensibility of our figures. We have improved the figures according to the suggestions. As for the figure legends, we have placed them below each respective figure whenever possible.

1. Clarify in the text if 'sample' means food substratum or flies/larvae (ex. line 116 and elsewhere).

We have revised the word “sample” throughout our manuscript and eliminated the confusion.

1. Line 170 - clarify what you mean by fermented food.

We have replaced the “fermented larval foods” with “fermented bananas” in our revised manuscript (line 165).

1. Line 199 - what is the meaning of 'stocks'.

We have replaced the “stocks” with “strains” (line 195).

1. Line 320 - explain more clearly what the yeast-conditioned banana-agar plate and cell suspension supernatant are, and what the goals of using these media are. This will help in understanding the subsequent text.

We have added a supplemental figure illustrating the sample preparation for the metabolomic analysis (Figure S6), with the following legend describing the procedure (lines 1335-1346): “Sample preparation process for the metabolomic analysis. We suspected that the supportive live yeast cells may release critical nutrients for larval growth, whereas the non-supportive yeasts may not. To test this possibility, we made three distinct sample preparations of individual yeast strains (yeast cells, yeast-conditioned banana-agar plates, and cell suspension supernatants). Yeast cells were for the analysis of intracellular metabolites, whereas yeast-conditioned banana-agar plates and cell suspension supernatants were for that of extracellular metabolites. The samples were prepared as the following procedures. Yeasts were grown on banana-agar plates for 2 days at 25°C, and then scraped from the plates to obtain “yeast cells.” Next, the remaining yeasts on the resultant plates were thoroughly removed, and a portion from each plate was cut out (“yeast-conditioned banana agar”). In addition, we suspended yeast cells from the agar plates into sterile PBS, followed by centrifugation and filtration to eliminate the yeast cells, to prepare “cell suspension supernatants.”

1. Figure 5 is difficult to understand. Provide more explanation. Consider moving the 'all metabolites panel' to Supp. Better explain what this holidic medium is.

The holidic medium is a medium that has been commonly used in the Drosophila research community, which contains ~40 known nutrients, and supports the larval development to pupariation (Piper et al., 2014; Piper et al., 2017). We have introduced this explanation to the RESULTS section of the manuscript (lines 322-327). However, the scope of our research reaches beyond the analysis of the holidic medium components, because feeding the holidic medium alone causes a significant delay in larval growth, suggesting a lack of nutritional components (Piper et al., 2014). Thus, we believe the "All Metabolites" panels should be placed alongside the corresponding “The holidic medium components” panels.

1. I could not access Figure 6 when downloading the PDF. The page is white and an error message appears - it is problematic to review a paper lacking a figure.

We regret any inconvenience caused, perhaps due to a system error. Please refer to the Author response image 2, which is identical to Figure 6 of our original manuscript.

Author response image 2.

Supportive yeasts facilitate larval growth by providing nutrients, including branched-chain amino acids, by releasing them from their cells (Figure 6 from the original manuscript). (A and B) Growth of larvae feeding on yeasts on banana agar supplemented with leucine and isoleucine. (A) The mean percentage of the live/dead individuals in each developmental stage. n=4. (B) The percentage of larvae that developed into second instar or later stages. The “Not found” population in Figure 6A was omitted from the calculation. Each data point represents data from a single tube. Unique letters indicate significant differences between groups (Tukey-Kramer test, p < 0.05). (C) The biosynthetic pathways for leucine and isoleucine with S. cerevisiae gene names are shown. The colored dots indicate enzymes that are conserved in the six isolated species, while the white dots indicate those that are not conserved. Abbreviations of genera are given in the key in the upper right corner. LEU2 is deleted in BY4741. (D-G) Representative image of Phloxine B-stained yeasts. The right-side images are expanded images of the boxed areas. The scale bar represents 50 µm. (H) Summary of this study. H. uvarum is predominant in the early-stage food and provides Leu, Ile, and other nutrients that are required for larval growth. In the late-stage food, AAB directly provides nutrients, while LAB and yeasts indirectly contribute to larval growth by enabling the stable larva-AAB association. The host larva responds to the nutritional environment by dramatically altering gene expression profiles, which leads to growth and pupariation. H. uva, Hanseniaspora uvarum; K. hum, Kazachstania humilis; Pi. klu, Pichia kluyveri; St. bac, Starmerella bacillaris; GF, germ-free.

1. Line 323 - Consider rewriting this sentence (too long, explain what the holidic medium is and why this is interesting). "In the yeast-conditioned banana-agar plates, which were anticipated to contain yeast-derived nutrients, many well-known nutrients included in a chemically defined synthetic (holidic) medium for Drosophila melanogaster (Piper et al., 2014, 2017) were not increased compared to the sterile banana-agar plates; instead, they exhibited drastic decreases irrespective of the yeast species."

We thank the reviewer's suggestion to improve the readability of our manuscript. We have rewritten the sentence in the revised manuscript (lines 320-328) as follows: “The yeastconditioned banana-agar plates were expected to contain yeast-derived nutrients. On the contrary, the result revealed a depletion of various metabolites originally present in the sterile banana agar (Figure 5A). This result prompted us to focus on the metabolites in the chemically defined (holidic) medium for Drosophila melanogaster Piper et al., 2014; Piper et al., 2017. This medium contains ~40 known nutrients, and supports the larval development to pupariation, albeit at the half rate compared to that on a yeast-containing standard laboratory food Piper et al., 2014; Piper et al., 2017. Therefore, the holidic medium could be considered to contain the minimal essential nutrients required for larval growth. Our analysis indicated a substantial reduction of these known nutrients in the yeast-conditioned plates compared to their original quantities (Figure 5B).”

Reviewer #2 (Recommendations For The Authors):

Suggestions for improved or additional experiments, data or analyses.

1. It should be clearly shown (or stated) that isolated microbes, such as H. uvarum and Pa. agglomerans, are indigenous microbes in wild Drosophila melanogaster in their outdoor sampling.

We thank the reviewer for the suggestions. Addressing the presence of isolated microbes within wild D. melanogaster adults is important, but cannot be feasible with our data for the following reasons. Our microbiota analysis of adults was conducted using pooled individuals of multiple Drosophila species, rather than using D. melanogaster exclusively. Moreover, the microbial isolation and the analysis of adult microbiota were carried out in two independent samplings (Figures 1A and 1E in the original manuscript, respectively). As a result, the microbial species detected in the adults were slightly different from those isolated from the food samples collected in the previous sampling. Nevertheless, it is worth noting that H. uvarum dominated in 2 out of the 3 adult samples, constituting >80% of the fungal composition. Pantoea agglomerans was not detected in the adults, although Enterobacterales accounted for >59% in 2 out of the 3 samples. Therefore, these isolated microbial species, or at least their phylogenetically related species, are presumed to be indigenous to wild D. melanogaster.

If the reviewer’s suggestion was to state the dominance of H. uvarum and Pantoea agglomerans in early-stage foods, we have added a supplemental figure showing the species-level microbial compositions corresponding to Figure 1B of the original manuscript (Figure S1), and further revised the manuscript (lines 180-186).

1. The reviewer supposes that the indigenous microbes of flies may differ from what they usually eat. In this study, the authors use banana-based food, but is it justified in terms of the natural environment of the places where those microbes were isolated? In other words, did sampled wild flies eat bananas outside the laboratory at Kyoto University?

Drosophila spp. inhabit human residential areas and feed on various fermented fruits and vegetables. In the areas surrounding Kyoto University, they can be found in garbage in residential dwellings as well as supermarkets. In this regard, fruits are natural food sources of wild Drosophila in the area.

Among various fruits, bananas were selected based on the following two reasons. Firstly, bananas were commonly used in previous Drosophila studies as a trap bait or a component of Drosophila food (Anagnostou et al., 2010; Stamps et al., 2012; Consuegra et al., 2020). Secondly, and rather practically, bananas can be obtained in Japan all year at a relatively low cost. Previous studies have used various fruits such as grapes (Quan and Eisen, 2018), figs (Pais et al., 2018), and raspberries (Cho and Rohlfs, 2023). However, these fruits are only available during limited seasons and are more expensive per volume than bananas. Thus, they were not practical for our study, which required large amounts of fruit-based culture media. We have included a brief explanation regarding this point in MATERIALS AND METHODS (lines 514-518).

1. In Fig. 6B, the Leu and Ile experiment, is the added amount of those amino acids appropriate in the context that they mention "...... supportive yeasts had concentrations of both leucine and isoleucine that were at least four-fold higher than those of non-supportive yeasts"?

We acknowledge that the supplementation should be carried out ideally in a quantity equivalent to the difference between the released amounts of supportive and non-supportive species. However, achieving this has been highly challenging. Previous studies determined the amount of amino acid supplementation by quantifying their concentration in the bacteriaconditioned media (Consuegra et al., 2020; Henriques et al., 2020). However, we found that quantifying the exact concentrations of the amino acids is not feasible with our yeasts. As shown in Figure 5B in the original manuscript, the amino acid contents were markedly reduced in the yeast-conditioned banana agar compared to the agar without yeasts, presumably because of the uptake by the yeasts. Thus, the amino acids released from yeast cells on the banana-agar plate are not expected to accumulate in the medium. As this reviewer pointed out, in the cell suspension supernatants of the supportive yeasts, concentrations of both leucine and isoleucine were at least four-fold higher compared to those of non-supportive yeasts (Figures 5G-H in the original submission), However, this measurement does not give the absolute amount of either amino acid available for larvae. Given these constraints, we opted for the amino acid concentrations in the holidic medium, which support larval growth under axenic conditions (Piper et al., 2014). We also showed that the supplementation of the amino acids at that concentration to the bananaagar plate was not detrimental to larval growth (Figures 6A-B in the original manuscript). These rationales have been included in the revised ‘Developmental progression with BCAA supplementation’ section in MATERIALS AND METHODS of our manuscript (lines 840-847).

1. In addition to the above, it can be included other amino acids or nutrients as control experiments.

As mentioned in our manuscript (lines 365-368), we did supplement other amino acids, lysine and asparagine, which failed to rescue the larval growth.

1. In the experiment of Fig. 2E, how about examining larval development using heat-killed LAB or yeast with live AAB? The reviewer speculates that one possibility is that AAB needs nutrients from LAB.

We did not feed larvae with heat-killed LAB and live AAB for the following reasons. LAB grows very poorly on banana agar compared to yeasts, and preparation of LAB required many banana-agar plates even when we fed live bacteria to larvae. Adding dead LAB to banana-agar tubes would require far more plates, but this preparation is impractical. Furthermore, heat-killing may not allow the investigation of the contribution of heat-unstable or volatile compounds.

As for the reviewer's suggestion regarding the addition of heat-killed yeast with AAB, heat-killed yeast itself promotes larval growth, as shown in Figures 4G and 4H in the original manuscript, so the contribution of yeast cannot be examined using this method.

Recommendations for improving the writing and presentation.

1. It would be good to mention that during sample collection, other insects (other than Drosophila species) were not found in the food if this is true.

Insects other than Drosophila spp. were found in several traps in the sampling shown in Figures 1C-F. These insects, rove beetles (Staphylinidae) and sap beetles (Nitidulidae), seemed to share a niche with Drosophila in nature. Therefore, we believe that the contamination of these insects did not interfere with our goal of obtaining larval food samples. We added these descriptions and explanations to MATERIALS AND METHODS (lines 527531).

1. There are many different kinds of bananas. It should be mentioned the detailed information.

We had included the information on the banana in MATERIALS AND METHODS section (line 622).

1. Concerning the place of sample collection, detailed longitude, and latitude information can be provided (this is easily obtained from Google Maps). When the collection was performed should also be mentioned. This may suggest the environment of the "wild flies" they collected.

We added a table listing the dates of our collections, along with the longitude and latitude of each sampling place (Table S1A).

1. The reviewer could not find how the authors conducted heat killing of yeast.

We added the following procedure to the ‘Quantification of larval development’ section in MATERIALS AND METHODS (lines 680-688). “When feeding heat-killed yeasts to larvae, yeasts were added to the banana-agar tubes and subsequently heated as following procedures. The yeasts were revived from frozen stocks on banana-agar plates, incubated at 25°C, and then streaked on fresh agar plates. After 2-day incubation, yeast cells were scraped from the plates and suspended in PBS at the concentration of 400 mg of yeast cells in 500 µL of PBS. 125 µL of the suspensions were added to banana-agar tubes prepared as described, and after centrifugation at 3,000 x g for 5 min, the supernatants were removed. The amount of cells in each tube is ~50x compared to that when feeding live yeasts, which compensates for the reduced amount due to their inability to proliferate. The tubes were subsequently heated at 80°C for 30 min before adding germ-free larvae.”

1. The reviewer prefers that all necessary information on how to see figures be provided in figure legends. For example, an explanation of some abbreviations is missing.

We carefully re-examined the figure legends and added necessary information.

1. Many of the figures are not kind to readers, i.e., one needs to refer to the legends and main text very frequently. Adding subheadings (titles) to each figure may help.

We added subheadings to our figures to improve the comprehensibility.

Reviewer #3 (Recommendations For The Authors):

I have some minor questions/suggestions about the manuscript that, if addressed, may increase the clarity and quality of the work.

1. Please, when referring to microbial species in the abbreviated form, use only the first letter of the genus. For example, P. agglomerans should be used, not Pa. agglomerans.

We are concerned about the potential confusion caused by using only the first letter of genera, since several genera mentioned in our work share the first letters, such as P (Pichia and Pantoea), S (Starmerella, Saccharomyces, and Saccharomycopsis), or L (Lactiplantibacillus and Leuconostoc). Therefore, we used only the unabbreviated form of the above seven genera in our revised manuscript. We have also made every effort to avoid abbreviations in our figures and tables, but found it necessary to retain two-letter abbreviations when spaces are particularly limiting.

1. In lines 294-298, how exactly was the experiment where yeasts were killed by anti-fungal agents performed? If these agents killed the yeast, how was the microbial growth on plates required to have biomass for fly inoculation obtained? Please, clarify this section.

The yeasts were grown on normal banana-agar plates before the addition onto the anti-fungal agents-containing banana agar. We added the following procedure to MATERIALS AND METHODS (lines 689-695). “When feeding yeasts on banana agar supplemented with antifungal agents, the yeasts were individually grown on normal banana agar twice before being suspended in PBS at the concentration of 400 mg of yeast cells in 500 µL of PBS. 125 µL of the suspensions was introduced onto the anti-fungal agents (10 mL/L 10% p-hydroxybenzoic acid in 70% ethanol and 6 mL/L propionic acid, following the concentration described in Kanaoka et al., 2023)-containing banana agar in 1.5 mL tubes. After centrifugation, the supernatants were removed. The amount of cells in each tube is ~50x compared to that when feeding live yeasts.”

1. In lines 557-558, please clarify how rDNA copy numbers can be calculated in this way.

Considering the results of the ITS and 16S sequencing analysis, it was highly likely that rDNAs from bananas and Drosophila were amplified along with microbial rDNA in this qPCR. To estimate the microbial rDNA copy number, we assumed that the proportion of microbial rDNA within the total amplification products remains consistent between the qPCR and the corresponding sequencing analysis, because the template DNA samples and amplified regions were shared between the analyses. Based on this, the copy number of microbial rDNA was estimated by multiplying the qPCR results with the microbial rDNA ratio observed in the ITS or 16S sequencing analysis of each sample. This methodology has been detailed in the MATERIALS AND METHODS section (lines 609-615).

1. In lines 609-611, how did you check for cells left from the previous day? Microscopy? Or do you mean that if there was liquid still in the sample you would not add more bacterial cultures? Please, clarify.

We observed with the naked eye from outside the tubes to determine if additional AAB should be introduced. Since we placed AAB on the banana agar in a lump, we examined whether the lumps were gone or not. We have added these procedures in MATERIALS AND METHODS (lines 671-673).

1. In Figure 2A, it is hard to differentiate between the gray tones. Please, improve this.

We have distinguished the plots for different conditions by changing the shape of the markers on the graphs.

1. In the legend of Figure 4, line 1101, I believe the panel letters are incorrect.

We have corrected the manuscript (lines 1241-1242) from “heat-killed yeasts on banana agar (H and I) or live yeasts on a nutritionally rich medium (J and K)” to “heat-killed yeasts on banana agar (G and H) or live yeasts on a nutritionally rich medium (I and J).”

1. In Figure S1, authors showed that bananas that were not inoculated still had detectable rDNA signal. Is this really because bacteria can penetrate the peel? Or could this be the “reagent microbiome”? Alternatively, could these microbes have been introduced during sample prep, such as cutting the bananas?

The detection of rDNA in bananas that were not inoculated with microbes was unlikely to be due to microbial contamination during experimental manipulation. The reviewer pointed out the possibility that the “reagent microbiome”, presumably the microbes in PBS, are detected from the uninoculated bananas. This seems to be unlikely, considering the PBS was sterilized by autoclaving before use. To ensure that no viable microbe was left in the autoclaved PBS, we applied a portion of the PBS onto a banana-agar plate and confirmed no colony was formed after incubation for a few days. DNA derived from dead microbes might be present in the PBS, but the PBS-added bananas were incubated for 4 days, so it is also unlikely that a detectable amount of DNA remained until sample collection. Furthermore, we believe that no contamination occurred during sample preparation. Banana peels were treated with 70% ethanol before removing them extremely carefully to avoid touching the fruit inside. All tools were sterilized before use. Taking all of these into account, we speculate that the microbes were already present in the bananas before peeling. We added the details of the sample preparation processes in MATERIALS AND METHODS (lines 518-521 and 540).

Other major revisions

1. We deposited our yeast genome annotation data in the DDBJ Annotated/Assembled Sequences database, and the accession numbers have been added to the ‘Data availability’ section in MATERIALS AND METHODS (lines 868-873).

2. The bacterial composition data in Figure 1B was corrected, because in the original version, the data for Place 3 and Place 4 was plotted in reverse. The original and revised plots are shown side by side in Author response image 3. We hope that the reviewers agree that this replacement of the plots does not affect our conclusion (p5, lines 117-120).

Author response image 3.

Comparison of the original and revised version of bacterial composition graph in Figure 1B. Comparison of the original (left) and revised (right) version of the graph at the bottom of Figure 1B, which shows the result of bacterial composition analysis. The color key, which is unmodified, is placed below the revised version.

1. The plot data and labels in the RNA-seq result heatmaps (Figures 3A and 4C) have been corrected. In these figures, row Z-scores of log2(TPM + 1) were to be plotted, as indicated by the key in each figure. However, in the original version, row Z-scores of TPM was erroneously plotted. Thus, Figures 3A and 4C of the original version have been replaced with the correct plots, and the original and revised plots are shown side by side in Author response images 4A and 4B. We hope that the reviewers agree that this replacement of the plots does not affect our conclusion (p7, lines 222-226 and p9, lines 277-281).

Author response image 4.

Comparison of the original and revised version of Figures 3A and 4C. (A and B) Comparison of the original (left) and revised (right) version of Figures 3A (A) or 4C (B).

1. The keys in the original Figures 3D and 4F indicate that log2(fold change) was used to plot all data. However, when plotting the data from the previous study (Zinke et al., 2002), their “fold change value” was used. We have corrected the keys, plots, and legend of Figure 3D to reflect the different nature of the data from our RNA-seq analysis and those from microarray analysis by Zinke et al. The original and revised plots are shown side by side in Author response image 5. We hope that the reviewers agree that this replacement of the plots does not affect our conclusion (p7, lines 228230 and p9, 277-284).

Author response image 5.

Comparison of the original and revised version of Figures 3D and 4F. (A and B) Comparison of the original (left) and revised (right) version of Figures 3D (A) or 4F (B).

1. The labels in Figure S5C and S5D (Figure S4C and S4D in the original version) have been corrected (they are "Pichia kluyveri > Supportive" and "Starmerella bacillaris > Supportive" rather than "Non-support. > H. uva" and "Non-support. > K. hum"). Additionally, we have reintroduced the circle indicating the number of “dme04070: Phosphatidylinositol signaling system” DEGs in Figure S5D, which was missing in Figure S4D of the original version. The original and revised figures are shown in Author response image 6.

Author response image 6.

Comparison of the original and revised version of Figures S5C and S5D. (A and B) Comparison of the original (left) and revised (right) versions of Figures S5C (A) or S5D (B). The original figures corresponding to the aforementioned figures were Figures S4C and S4D, respectively.

1. The "Fermentation stage" column in Table 1, which indicated whether each microbe was considered an early-stage microbe or a late-stage microbe, has been removed to avoid confusion. This is because some of the microbes (Hanseniaspora uvarum, Pichia kluyveri, and Pantoea agglomerans) were employed in both of the feeding experiments using the microbes detected from the early-stage foods (Figures 2A, 2B, S2A, and S2B) and those from the late-stage foods (Figures 2C, 2D, S2C, and S2D).

2. The leftmost column in Table S7 has been edited to indicate species names rather than “Sample IDs,” because the IDs were not used in anywhere else in the paper.

Reference

Chandler, J. A., Lang, J., Bhatnagar, S., Eisen, J. A. and Kopp, A. (2011). Bacterial communities of diverse Drosophila species: Ecological context of a host-microbe model system. PLoS Genetics 7, e1002272.

Chandler, J. A., Eisen, J. A. and Kopp, A. (2012). Yeast communities of diverse Drosophila species: Comparison of two symbiont groups in the same hosts. Applied and Environmental Microbiology 78, 7327–7336.

Cho, H. and Rohlfs, M. (2023). Transmission of beneficial yeasts accompanies offspring production in Drosophila—An initial evolutionary stage of insect maternal care through manipulation of microbial load? Ecology and Evolution 13, e10184.

Consuegra, J., Grenier, T., Akherraz, H., Rahioui, I., Gervais, H., da Silva, P. and Leulier, F. (2020). Metabolic Cooperation among Commensal Bacteria Supports Drosophila Juvenile Growth under Nutritional Stress. iScience 23, 101232.

Dodge, R., Jones, E. W., Zhu, H., Obadia, B., Martinez, D. J., Wang, C., Aranda-Díaz, A., Aumiller, K., Liu, Z., Voltolini, M., et al. (2023). A symbiotic physical niche in Drosophila melanogaster regulates stable association of a multi-species gut microbiota. Nat Commun 14, 1557.

Erkosar, B., Storelli, G., Mitchell, M., Bozonnet, L., Bozonnet, N. and Leulier, F. (2015). Pathogen Virulence Impedes Mutualist-Mediated Enhancement of Host Juvenile Growth via Inhibition of Protein Digestion. Cell Host & Microbe 18, 445–455.

Hanson, M. A. and Lemaitre, B. (2020). New insights on Drosophila antimicrobial peptide function in host defense and beyond. Current Opinion in Immunology 62, 22–30.

Henriques, S. F., Dhakan, D. B., Serra, L., Francisco, A. P., Carvalho-Santos, Z., Baltazar, C., Elias, A. P., Anjos, M., Zhang, T., Maddocks, O. D. K., et al. (2020). Metabolic cross-feeding in imbalanced diets allows gut microbes to improve reproduction and alter host behaviour. Nat Commun 11, 4236.

Oka, M., Hashimoto, K., Yamaguchi, Y., Saitoh, S., Sugiura, Y., Motoi, Y., Honda, K., Kikko, Y., Ohata, S., Suematsu, M., et al. (2017). Arl8b is required for lysosomal degradation of maternal proteins in the visceral yolk sac endoderm of mouse embryos. Journal of Cell Science jcs.200519.

Pais, I. S., Valente, R. S., Sporniak, M. and Teixeira, L. (2018). Drosophila melanogaster establishes a species-specific mutualistic interaction with stable gut-colonizing bacteria. PLOS Biology 16, e2005710.

Piper, M. D. W., Blanc, E., Leitão-Gonçalves, R., Yang, M., He, X., Linford, N. J., Hoddinott, M. P., Hopfen, C., Soultoukis, G. A., Niemeyer, C., et al. (2014). A holidic medium for Drosophila melanogaster. Nature Methods 11, 100–105.

Piper, M. D. W., Soultoukis, G. A., Blanc, E., Mesaros, A., Herbert, S. L., Juricic, P., He, X., Atanassov, I., Salmonowicz, H., Yang, M., et al. (2017). Matching Dietary Amino Acid Balance to the In Silico-Translated Exome Optimizes Growth and Reproduction without Cost to Lifespan. Cell Metab 25, 610–621.

Quan, A. S. and Eisen, M. B. (2018). The ecology of the drosophila-yeast mutualism in wineries. PLOS ONE 13, e0196440.

Solomon, G. M., Dodangoda, H., McCarthy-Walker, T. T., Ntim-Gyakari, R. R. and Newell, P. D. (2019). The microbiota of Drosophila suzukii influences the larval development of Drosophila melanogaster. PeerJ 7, e8097.

Zinke, I., Schütz, C. S., Katzenberger, J. D., Bauer, M. and Pankratz, M. J. (2002). Nutrient control of gene expression in Drosophila: microarray analysis of starvation and sugar-dependent response. The EMBO Journal 21, 6162–6173.

Author response:

The following is the authors’ response to the original reviews

Reviewer #1:

Comment 0: Summary: This work presents an Interpretable protein-DNA Energy Associative (IDEA) model for predicting binding sites and affinities of DNA-binding proteins. Experimental results demonstrate that such an energy model can predict DNA recognition sites and their binding strengths across various protein families and can capture the absolute protein-DNA binding free energies.

We appreciate the reviewer’s careful assessment of the paper, and we thank the reviewer for the insightful suggestions and comments.

Comment 1: Strengths: (1) The IDEA model integrates both structural and sequence information, although such an integration is not completely original. (2) The IDEA predictions seem to have agreement with experimental data such as ChIP-seq measurements.

We appreciate the reviewer’s positive comments on the strength of the paper.

Comment 2: Weaknesses: (1) The authors claim that the binding free energy calculated by IDEA, trained using one MAX-DNA complex, correlates well with experimentally measured MAX-DNA binding free energy (Figure 2) based on the reported Pearson Correlation of 0.67. However, the scatter plot in Figure 2A exhibits distinct clustering of the points and thus the linear fit to the data (red line) may not be ideal. As such. the use of the Pearson correlation coefficient that measures linear correlation between two sets of data may not be appropriate and may provide misleading results for non-linear relationships.

We thank the reviewer for the insightful comments and agree that a linear fit between our predictions and the experimental data may not be the best measure of performance. The primary utility of the IDEA model is to predict high-affinity DNA-binding sequences for a given DNA-binding protein by assessing the relative binding affinities across different DNA sequences. In this regard, the ranked order of predicted sequence binding affinities serves as a better metric for evaluating the success of this model. To evaluate this, we calculated both Spearman’s rank correlation coefficient, which does not rely on linear correlation, and the Pearson correlation coefficient between our predictions and the experimental results. As shown in Figure 2, our computation shows a Spearman’s rank correlation coefficient of 0.65 for the MAX-based predictions using one MAX-DNA complex (PDB ID: 1HLO), supporting the model’s capability to effectively distinguish strong from weak binders.

Although our model generally captures the relative binding affinities across different DNA sequences, its predictive accuracy diminishes for low-affinity sequences (Figure 2).

This could be due to two limitations of the current modeling framework: (1) The model is residue-based and estimates binding free energy as the additive sum of contributions from individual contacting amino-acid-nucleotide pairs. This assumption does not account for cooperative effects caused by simultaneous changes at multiple nucleotide positions. One potential direction to further improve the model would be to use a finergrained representation by incorporating more atom types within contacting residues, and to use a many-body potential to better capture cooperative effects from multiple mutations. (2) The model assumes that the target DNA adopts the same binding interface as in the reference crystal structure. However, sequence-dependent DNA shape has been shown to be important in determining protein-DNA binding affinity [1]. To address this limitation, a future direction is to use deep-learning-based methods to incorporate predicted DNA shape or protein-DNA complex structures based on their sequences [2, 3] into our model prediction.

To fully evaluate the predictive power of IDEA, we have included Spearman’s rank correlation coefficient for every correlation plot in this manuscript and have updated the relevant texts. Across all our analyses, the Spearman’s rank correlation coefficients reveal similar predictive performance as the Pearson correlation coefficients. Additionally, we have included in our discussion the current limitations of our model and potential directions for future improvement.

We have edited our Discussion Section to include a discussion on the limitations of the current model. Specifically, the added texts are:

“Although IDEA has proved successful in many examples, it can be improved in several aspects. The model currently assumes the training and testing sequences share the same protein-DNA structure. While double-stranded DNA is generally rigid, recent studies have shown that sequence-dependent DNA shape contributes to their binding specificity [1, 2, 4]. To improve predictive accuracy, one could incorporate predicted DNA shapes or structures into the IDEA training protocol. In addition, the model is residue-based and evaluates the binding free energy as the additive sum of contributions from individual amino-acid-nucleotide contacts. This assumption does not account for cooperative effects that may arise from multiple nucleotide changes. A potential refinement could utilize a finer-grained model that includes more atom types within contacting residues and employs a many-body potential to account for such cooperative effects.”

Comment 3: (2) In the same vein, the linear Pearson Correlation analysis performed in Figure 5A and the conclusion drawn may be misleading.

We thank the reviewer for the insightful comments. As noted in our response to the previous comment, we have added Spearman’s rank correlation coefficient in addition to the Pearson correlation coefficient to all correlation plots, including Figure 5A.

Comment 4: (3) The authors included the sequences of the protein and DNA residues that form close contacts in the structure in the training dataset, whereas a series of synthetic decoy sequences were generated by randomizing the contacting residues in both the protein and DNA sequences. In particular, synthetic decoy binders were generated by randomizing either the DNA (1000 sequences) or protein sequences (10,000 sequences) from the strong binders. However, the justification for such randomization and how it might impact the model’s generalizability and transferability remain unclear.

We thank the reviewer for the insightful comments. The number of randomizing sequences was chosen to strike a balance between sufficient sequence coverage and computational feasibility. Because proteins have more types of amino acids than four nucleotides in DNA, we utilized more protein decoy sequences than DNA decoys. To examine the robustness of our choice against different number of decoy sequences, we repeated the transferability analysis within the bHLH superfamily (Figure 3A) and the generalizability analysis across 12 protein families (Figure 2E) using two additional decoy sequence combinations: (1) 1000 DNA sequences and 1000 protein sequences; (2) 100 DNA sequences and 1000 protein sequences. As shown in Figure S15, we achieved similar results to those reported using the original decoy set, demonstrating the robustness of our model prediction against the variations in the number of decoys. We have included this figure as Figure S15.

Comment 5: (4) The authors performed Receiver Operating Characteristic (ROC) analysis and reported the Area Under the Curve (AUC) scores in order to quantitate the successful identification of the strong binders by IDEA. It would be beneficial to analyze the precision-recall (PR) curve and report the PRAUC metric which could be more robust.

We agree with the reviewer that more robust statistical metrics should be used to evaluate our model’s performance. We have included the PRAUC score as an additional evaluation metric of the model’s performance. Due to a significant imbalance in the number of strong and weak binders from the experimental data [5], where the experimentally identified strong binders are far fewer than the weak binders, we reweighted the sample to achieve a balanced evaluation [6], using 0.5 as the baseline for randomized prediction. As shown in Figure S5, IDEA achieves successful predictions in 18 out of 22 cases, demonstrating its predictive accuracy.

The updated PRAUC result has been included as Figure S5 in the manuscript. We have also included the detailed precision-recall curves for each case in Figure S4.

In addition, we have provided PRAUC scores for comparing the performance of IDEA with other models, and have summarized these results in Table S2.

Reviewer #2:

Comment 0: Summary: Zhang et al. present a methodology to model protein-DNA interactions via learning an optimizable energy model, taking into account a representative bound structure for the system and binding data. The methodology is sound and interesting. They apply this model for predicting binding affinity data and binding sites in vivo. However, the manuscript lacks discussion of/comparison with state-of-the-art and evidence of broad applicability. The interpretability aspect is weak, yet over-emphasized.

We appreciate the reviewer’s excellent summary of the paper, and we thank the reviewer for the insightful suggestions and comments.

Comment 1: Strengths: The manuscript is well organized with good visualizations and is easy to follow. The methodology is discussed in detail. The IDEA energy model seems like an interesting way to study a protein-DNA system in the context of a given structure and binding data. The authors show that an IDEA model trained on one system can be transferred to other structurally similar systems. The authors show good performance in discriminating between binding-vs-decoy sequences for various systems, and binding affinity prediction. The authors also show evidence of the ability to predict genome-wide binding sites.

We appreciate the reviewer’s strong assessment of the strengths of this paper. We have further refined our Methods Section to ensure all modeling details are clearly presented.

Comment 2: Weaknesses: An energy-based model that needs to be optimized for specific systems is inherently an uncomfortable idea. Is this kind of energy model superior to something like Rosetta-based energy models, which are generally applicable? Or is it superior to family-specific knowledge-based models? It is not clear.

We thank the reviewer for the insightful comments. The protein-DNA energy model facilitates the calculation of protein-DNA binding free energy based on protein-DNA structures and sequences. Because this model is optimized using the structure-sequence relationship of given protein-DNA complexes, it features specificity based on the conserved structural interface characteristic of each protein family. Because of that, its predictive accuracy depends on the degree of protein-DNA interface similarity between the training and target protein-DNA pairs, and is distinct from a general protein-DNA energy model, such as a Rosetta-based energy model. The model has some connections to the familyspecific energy model. As shown in Author response image 1, systems belonging to the same protein superfamily (MAX and PHO4) exhibit similar patterns in their learned energy models, in contrast to those from a different superfamily (PDX1).

Author response image 1:

Comparison of learned energy models for different protein-DNA complexes: MAX (A), PHO4 (B), and PDX1 (C). MAX and PHO4 are members of the Helixloop-helix (HLH) CATH protein superfamily (4.10.280.100), while PDX1 belongs to another Homeodomain-like CATH protein superfamily (1.10.10.60).

To compare our approach with both general and family-specific knowledge-based energy models, we conducted two studies. First, we incorporated a knowledge-based generic protein-DNA energy model (DBD-Hunter) learned from the protein-DNA database, reported by Skoinick and coworkers [7], into our prediction protocol. This model assigns interaction energies to different functional groups within each DNA nucleotide (e.g., phosphate (PP), sugar (SU), pyrimidine (PY), and imidazole (IM) groups). For our comparison, we averaged the energy contributions of these groups within each nucleotide and replaced the IDEA-learned energy model with this generic one to test its ability to differentiate strong binders from weak binders in the HT-SELEX dataset [5]. As shown in Figure S6, the IDEA model generally achieves better performance than the generic energy model.

Additionally, we compared IDEA with rCLAMPS, a family-specific energy model developed to predict protein-DNA binding specificity in the C2H2 and homeodomain families.

As shown in Table S1 and Table S2, IDEA also shows better performance than rCLAMPS in most cases across the C2H2 and homeodomain families, demonstrating that it has better predictive accuracy than both state-of-the-art family-specific and generic knowledgebased models.

We have included relevant texts in Appendix Section Comparison of IDEA predictive performance Using HT-SELEX data to clarify this point. The added texts are:

In addition, we compared the performance of IDEA with both general and family-specific knowledge-based energy models. First, we incorporated a knowledgebased generic protein-DNA energy model (DBD-Hunter) learned from the protein-DNA database, reported by Skoinick and coworkers [7], into our prediction protocol. This model assigns interaction energies to different functional groups within each DNA nucleotide, including phosphate (PP), sugar (SU), pyrimidine (PY), and imidazole (IM) groups. For our comparison, we averaged the energy contributions of these groups within each nucleotide and replaced the IDEA-learned energy model with the DBD-Hunter model to assess its ability to differentiate strong binders from weak binders in the HTSELEX dataset [5]. Additionally, we compared IDEA with rCLAMPS, a familyspecific energy model developed to predict protein-DNA binding specificity in the C2H2 and homeodomain families. rCLAMPS learns a position-dependent amino-acid-nucleotide interaction energy model. To incorporate this model into the binding free energy calculation, we averaged the energy contributions across all occurrences of each amino-acid-nucleotide pair, which resulted in a 20-by-4 residue-type-specific energy matrix. This matrix is structurally analogous to the IDEA-trained energy model and can be directly integrated into the binding free energy calculations. As shown in Figure S6, Table S1, and Table S2, the IDEA model generally outperforms DBD-Hunter and rCLAMPS, demonstrating that it can achieve better predictive accuracy than both generic and family-specific knowledge-based models.

Comment 3: Prediction of binding affinity is a well-studied domain and many competitors exist, some of which are well-used. However, no quantitative comparison to such methods is presented. To understand the scope of the presented method, IDEA, the authors should discuss/compare with such methods (e.g. PMID 35606422).

We thank the reviewer for the insightful comments. As detailed in our response to Comment 5, we previously misused the term “binding specificity”, and would like to clarify that our model is designed to predict protein-DNA binding affinity. To compare the performance of IDEA with state-of-the-art protein-DNA predictive models, we examined the predictive accuracies of two additional popular computational models: ProBound [8] and DeepBind [9]. ProBound has been shown to have a better performance than several earlier predictive protein-DNA models, including JASPAR 2018 [11], HOCOMOCO [12], Jolma et al. [13], and DeepSELEX [14]. To benchmark these models’ performance, we examine each method’s capability to identify strong binders with the HT-SELEX datasets covering 22 proteins from 12 protein families [5]. As suggested by Reviewer 1, we also calculated the PRAUC score, reweighted to account for data imbalance [6], as a complementary metric for evaluating the model performance.

As shown in Figure S6, Table S1, and Table S2, IDEA ranked second among the three predictive methods. It is important to note that both ProBound and DeepBind were trained on a curated version of the HT-SELEX data [13], which overlaps with the testing data [5]. Compared with them, IDEA was trained only on the given structural and sequence information from a single protein-DNA complex, thus independent of the testing data. In order to assess how IDEA performs when incorporating knowledge from HT-SELEX data, we augmented the training by randomly including half of the HT-SELEX data (see the Methods Section Enhanced Modeling Prediction with SELEX Data). The augmented IDEA model achieved the best performance among all the models. Overall, IDEA can be used to predict protein-DNA affinities in the absence of known binding sequence data, thereby filling a critical gap when such experimental datasets are unavailable.

Additionally, we have conducted a 10-fold cross-validation using the same HT-SELEX data [5] and found that IDEA outperformed a recent regression model that considers the shape of DNA with different sequences [5].

We have revised our text to include the comparison between IDEA and other predictive models. Specifically, we revised the text in Section: IDEA Generalizes across Various Protein Families.

The revised text reads:

“To examine IDEA’s predictive accuracy across different DNA-binding protein families, we applied it to calculate protein-DNA binding affinities using a comprehensive HT-SELEX dataset [5]. We focused on evaluating the capability of IDEA to distinguish strong binders from weak binders for each protein with an experimentally determined structure. We calculated the probability density distribution of the top and bottom binders identified in the SELEX experiment. A well-separated distribution indicates the successful identification of strong binders by IDEA (Figure 2D and S4). Receiver Operating Characteristic (ROC) analysis was performed to calculate the Area Under the Curve (AUC) and the precision-recall curve (PRAUC) scores for these predictions. Further details are provided in the Methods Section Evaluation of IDEA Prediction Using HT-SELEX Data. Our analysis shows that IDEA successfully differentiates strong from weak binders for 80% of the 22 proteins across 12 protein families, achieving AUC and balanced PRAUC scores greater than 0.5 (Figure 2D and S5). To benchmark IDEA’s performance against other leading methods, we compared its predictions with several popular models, including the sequence-based predictive models ProBound [8] and DeepBind [9], the familybased energy model rCLAMPS [10], and the knowledge-based energy model DBD-Hunter [7]. IDEA demonstrates performance comparable to these stateof-the-art approaches, and incorporating sequence features further improves its prediction accuracy (Figure S6, Table S1, and Table S2). We also performed 10-fold cross-validation on the binding affinities of protein–DNA pairs in this dataset and found that IDEA outperforms a recent regression model that considers the shape of DNA with different sequences [5] (Figure S7). Details are provided in Section: Comparison of IDEA predictive performance Using HT-SELEX data.”

We also added one section Comparison of IDEA predictive performance Using HT-SELEX data in the Appendix to fully explain the comparison between IDEA and other popular models. The added texts are:

“To benchmark the performance of IDEA against state-of-the-art protein-DNA predictive models, we evaluated its ability to recognize strong binders with the HT-SELEX datasets across 22 proteins from 12 families [5]. Specifically, we compare IDEA with two widely used sequence-based models: ProBound [8] and DeepBind [9]. ProBound has demonstrated superior performance over many other predictive protein-DNA models, including JASPAR 2018 [11], HOCOMOCO [12], Jolma et al. [13], and DeepSELEX [14]. To use ProBound, we retrieved the trained binding model for each protein from motifcentral.org and used the GitHub implementation of ProBoundTools to infer the binding scores between protein and target DNA sequences. Except for POU3F1, binding models are available for all proteins. Therefore, we excluded POU3F1 and evaluated the protein-DNA binding affinities for the remaining 21 proteins. To use DeepBind, sequence-specific binding affinities were predicted directly with its web server. The Area Under the Curve (AUC) and the Precision-Recall AUC (PRAUC) scores were used as metrics for comparison. An AUC score of 1.0 indicates a perfect separation between the strong- and weak-binder distributions, while an AUC score of 0.5 indicates no separation. Because there is a significant imbalance in the number of strong and weak binders from the experimental data [5], where the strong binders are far fewer than the weak binders, we reweighted the samples to achieve a balanced evaluation, using 0.5 as the baseline for randomized prediction [6]. As summarized in Figure S6, Table S1, and Table S2, IDEA ranked second among the three predictive models. In order to assess the performance of IDEA when augmented with additional protein-DNA binding data, we augmented IDEA using randomly selected half of the HT-SELEX data (see the Methods Section Enhanced Modeling Prediction with SELEX Data). The augmented IDEA model achieved the best performance among all the models.”

“We also performed 10-fold cross-validation using the same HT-SELEX datasets, following the protocol described in the Methods Section Enhanced Modeling Prediction with SELEX Data. For each protein, we divided the entire dataset into 10 equal, randomly assigned folds. In each iteration, we used randomly selected 9 of the 10 folds as the training dataset and the remaining fold as the testing dataset. This process was repeated 10 times so that each fold served as the test set once. We then reported the average R2 scores across these iterations to evaluate IDEA’s predictive performance. Our results are compared with the 1mer and 1mer+shape methods from [5], the latest regression model that considers the shape of DNA with different sequences (Figure S7). This comparative analysis shows IDEA achieved higher predictive accuracy than the state-of-the-art sequence-based protein-DNA binding predictors for proteinDNA complexes that have available experimentally resolved structures.”

“Overall, these results demonstrate that IDEA can be used to predict the proteinDNA pairs in the absence of known binding sequence data, thus filling an important gap in protein-DNA predictions when experimental binding sequence data are unavailable.”

Comment 4: The term “interpretable” has been used lavishly in the manuscript while providing little evidence on the matter. The only evidence shown is the family-specific residue-nucleotide interaction/energy matrix and speculations on how these values are biologically sensible. Recent works already present more biophysical, fine-grained, and sometimes family-independent interpretability (e.g. PMID 39103447, 36656856, 38352411, etc.). The authors should put into context the scope of the interpretability of IDEA among such works.

We thank the reviewer for the insightful comment and agree that “interpretability” should be discussed in a relevant context. In our work, interpretability refers to the familyspecific amino-acid-nucleotide interaction energies identified from the model training, which reveal interaction preferences within protein-DNA binding interfaces. As detailed in our response to Comment 6, we performed principal component analysis (PCA) on the learned energy models and observed clustering of learned energy models corresponding to protein families. Therefore, the IDEA-learned energy models can be used as a signature to capture the energetic preferences of amino-acid-nucleotide interactions within a given protein family. This preference can be used to infer preferred sequence binding motifs, similar to those identified by other computational tools [10, 4, 15, 16].

We have revised the text to clarify the “interpretability” as the family-specific aminoacid-nucleotide interactions that govern sequence-dependent protein-DNA binding, and to discuss IDEA’s interoperability within the context of recent works, including those suggested by the reviewers.

We have revised the text in Introduction. The new text reads:

“Here, we introduce the Interpretable protein-DNA Energy Associative (IDEA) model, a predictive model that learns protein-DNA physicochemical interactions by fusing available biophysical structures and their associated sequences into an optimized energy model (Figure 1). We show that the model can be used to accurately predict the sequence-specific DNA binding affinities of DNA-binding proteins and is transferrable across the same protein superfamily. Moreover, the model can be enhanced by incorporating experimental binding data and can be generalized to enable base-pair resolution predictions of genomic DNA-binding sites. Notably, IDEA learns a family-specific interaction matrix that quantifies energetic interactions between each amino acid and nucleotide, allowing for a direct interpretation of the “molecular grammar” governing sequence-specific protein-DNA binding affinities. This interpretable energy model is further integrated into a simulation framework, facilitating mechanistic studies of various biomolecular functions involving protein-DNA dynamics.”

We have revised the text in Results. The new text reads:

“IDEA is a coarse-grained biophysical model at the residue resolution for investigating protein-DNA binding interactions (Figure 1). It integrates both structures and corresponding sequences of known protein-DNA complexes to learn an interpretable energy model based on the interacting amino acids and nucleotides at the protein-DNA binding interface. The model is trained using available protein-DNA complexes curated from existing databases [17, 18].

Unlike existing deep-learning-based protein-DNA binding prediction models, IDEA aims to learn a physicochemical-based energy model that quantitatively characterizes sequence-specific interactions between amino acids and nucleotides, thereby interpreting the “molecular grammar” driving the binding energetics of protein-DNA interactions. The optimized energy model can be used to predict the binding affinity of any given protein-DNA pair based on its structures and sequences. Additionally, it enables the prediction of genomic DNA binding sites by a given protein, such as a transcription factor. Finally, the learned energy model can be incorporated into a simulation framework to study the dynamics of DNA-binding processes, revealing mechanistic insights into various DNA-templated processes. Further details of the optimization protocol are provided in Methods Section Energy Model Optimization.”

The revised text in Section: Discussion now reads:

“Another highlight of IDEA is its ability to present an interpretable, familyspecific amino acid-nucleotide interaction energy model for given proteinDNA complexes. The optimized IDEA energy model can not only predict sequence-specific binding affinities of protein-DNA pairs but also provide a residue-specific interaction matrix that dictates the preferences of amino acidnucleotide interactions within specific protein families (Figure S11). This interpretable energy matrix would facilitate the discovery of sequence binding motifs for target DNA-binding proteins, complementing both sequencebased [24, 16, 25] and structure-based approaches [10, 26, 4, 15]. Additionally, we integrated this physicochemical-based energy model into a simulation framework, thereby improving the characterization of protein-DNA binding dynamics. IDEA-based simulation enables the investigation into dynamic interactions between various proteins and DNA, facilitating molecular-level understanding of the physical mechanisms underlying many DNA-binding processes, such as transcription, epigenetic regulations, and their modulation by sequence variations, such as single-nucleotide polymorphisms (SNPs) [22, 23].”

Comment 5: The manuscript disregards subtle yet important differences in commonly used terminology in the field. For example, the authors use the term ”specificity” and ”affinity” almost interchangeably (for example, the caption for Figure 3A uses ”specificity” although the Methods text describes the prediction as about ”affinity”). If the authors are looking to predict specificity, IDEA needs to be put in the context of the corresponding state-of-the-art (PMID 36123148, 39103447, 38867914, 36124796, etc).

We really appreciate the reviewer for pointing out the conflation of “specificity” and “affinity” in our manuscript. To clarify, the primary function of IDEA is to predict the binding affinities of protein-DNA pairs in a sequence-specific manner. We have revised the text to clarify the distinction between affinity and specificity and acknowledge prior works, including those provided by the reviewers, that focus on predicting protein-DNA binding specificity.

We have revised the Section title IDEA Accurately Predicts Protein-DNA Binding Specificity to IDEA Accurately Predicts Sequence-Specific Protein-DNA Binding Affinity; and ResidueLevel Protein-DNA Energy Model for Predicting Protein-DNA Recognition Specificities to Predictive Protein-DNA Energy Model at Residue Resolution.

We have revised the text in Introduction. The revised text reads:

“Computational methods complement experimental efforts by providing the initial filter for assessing sequence-specific protein-DNA binding affinity. Numerous methods have emerged to enable predictions of binding sites and affinities of DNA-binding proteins [27, 9, 1, 5, 28, 29, 30, 31, 8]. These methods often utilized machine-learning-based training to extract sequence preference information from DNA or protein by utilizing experimental high-throughput (HT) assays [27, 9, 1, 5, 28, 8], which rely on the availability and quality of experimental binding assays. Additionally, many approaches employ deep neural networks [29, 30, 31], which could obscure the interpretation of interaction patterns governing protein-DNA binding specificities. Understanding these patterns, however, is crucial for elucidating the molecular mechanisms underlying various DNA-recognition processes, such as those seen in TFs [32].”

We have revised the text in Section: IDEA Demonstrates Transferability across Proteins in the Same CATH Superfamily.

The revised text reads:

“Since IDEA relies on the sequence-structure relationship of given protein-DNA complexes to reach predictive accuracy, we inquired whether the trained energy model from one protein-DNA complex could be generalized to predict the sequence-specific binding affinities of other complexes. To test this, we assessed the transferability of IDEA predictions across all 11 structurally available protein-DNA complexes within the MAX TF-associated CATH superfamily (CATH ID: 4.10.280.10, Helix-loop-helix DNA-binding domain). We trained IDEA based on each of these 11 complexes and then used the trained model to predict the MAX-based MITOMI binding affinity. Our results show that IDEA generally makes correct predictions of the binding affinity when trained on proteins that are homologous to MAX, with Pearson and Spearman Correlation coefficients larger than 0.5 (Figure 3A and Figure S10).”

We have revised the caption of Figure 3: The revised text reads:

“IDEA prediction shows transferability within the same CATH superfamily. (A) The predicted MAX binding affinity, trained on other protein-DNA complexes within the same protein CATH superfamily, correlates well with experimental measurement. The proteins are ordered by their probability of being homologous to the MAX protein, determined using HHpred [33]. Training with a homologous protein (determined as a hit by HHpred) usually leads to better predictive performance (Pearson Correlation coefficient > 0.5) compared to non-homologous proteins. (B) Structural alignment between 1HLO (white) and 1A0A (blue), two protein-DNA complexes within the same CATH Helix-loop-helix superfamily. The alignment was performed based on the Ebox region of the DNA [34]. (C) The optimized energy model for 1A0A, a protein-DNA complex structure of the transcription factor PHO4 and DNA, with 33.41% probability of being homologous to the MAX protein. The optimized energy model is presented in reduced units, as explained in the Methods Section: Training Protocol.”

We have revised the text in Section Discussion: The revised text now reads:

“The protein-DNA interaction landscape has evolved to facilitate precise targeting of proteins towards their functional binding sites, which underlie essential processes in controlling gene expression. These interaction specifics are determined by physicochemical interactions between amino acids and nucleotides. By integrating sequences and structural data from available proteinDNA complexes into an interaction matrix, we introduce IDEA, a data-driven method that optimizes a system-specific energy model. This model enables high-throughput in silico predictions of protein-DNA binding specificities and can be scaled up to predict genomic binding sites of DNA-binding proteins, such as TFs. IDEA achieves accurate de novo predictions using only proteinDNA complex structures and their associated sequences, but its accuracy can be further enhanced by incorporating available experimental data from other binding assay measurements, such as the SELEX data [35, 36, 37], achieving accuracy comparable or better than state-of-the-art methods (Figures S2 and S7, Table S1 and S2). Despite significant progress in genome-wide sequencing techniques [38, 39, 40, 41], determining sequence-specific binding affinities of DNA-binding biomolecules remains time-consuming and expensive. Therefore, IDEA presents a cost-effective alternative for generating the initial predictions before pursuing further experimental refinement.”

We have revised the text in Discussion to clarify that the acquired binding affinities of target DNA sequences can be used to help existing models to infer specific DNA binding motifs.

The revised text now reads:

Another highlight of IDEA is its ability to present an interpretable, familyspecific amino acid-nucleotide interaction energy model for given proteinDNA complexes. The optimized IDEA energy model can not only predict sequence-specific binding affinities of protein-DNA pairs but also provide a residue-specific interaction matrix that dictates the preferences of amino acidnucleotide interactions within specific protein families (Figure S11). This interpretable energy matrix would facilitate the discovery of sequence binding motifs for target DNA-binding proteins, complementing both sequencebased [24, 16, 25] and structure-based approaches [10, 26, 4, 15]. Additionally, we integrated this physicochemical-based energy model into a simulation framework, thereby improving the characterization of protein-DNA binding dynamics. IDEA-based simulation enables the investigation into dynamic interactions between various proteins and DNA, facilitating molecular-level understanding of the physical mechanisms underlying many DNA-binding processes, such as transcription, epigenetic regulations, and their modulation by sequence variations, such as single-nucleotide polymorphisms (SNPs) [22, 23].

Comment 6: It is not clear how much the learned energy model is dependent on the structural model used for a specific system/family. It would be interesting to see the differences in learned model based on different representative PDB structures used. Similarly, the supplementary figures show a lack of discriminative power for proteins like PDX1 (homeodomain family), POU, etc. Can the authors shed some light on why such different performances?

We thank the reviewer for the insightful comments and agree that the trained energy model should be presented in the context of protein families. To further analyze the dependence of the energy model on protein family, we visualized the trained energy models for 24 proteins, including all proteins from the HT-SELEX dataset as well as PHO4 (PDB ID: 1A0A) and CTCF (PDB ID: 8SSQ), spanning 12 distinct protein families. To quantitatively assess similarities and differences among these energy models, we flattened each normalized energy model into an 80-dimensional vector and performed principal component analysis (PCA). As shown in Author response image 1 and Figure S11, energy models optimized from the same protein family fall within the same cluster, while those from different protein families exhibit distinct patterns. Moreover, the relative distance between energy models in PCA space reflects the degree of transferability. For example, PHO4 (PDB ID: 1A0A) is positioned close to MAX (PDB ID: 1HLO), whereas USF1 (PDB ID: 1AN4) and TCF4 (PDB ID: 6OD3) are farther away. This is consistent with the results shown in Figure 3A, where the energy model trained from PHO4 has better transferability than those from the other two systems.

We also greatly appreciate the reviewer’s suggestion to examine cases where IDEA failed to demonstrate strong discriminative power. When evaluating the model’s ability to distinguish between strong and weak binders, we used the available experimental structure most similar to the protein employed in the HT-SELEX experiments. In some instances, only the structure of the same protein from a different organism is available. For example, the HT-SELEX data for PDX1-DNA used the human PDX1 protein, but no human PDX1–DNA complex structure is available. Therefore, we used the mouse PDX1–DNA complex (PDB ID: 2H1K) for model training. The differences between species may limit the predictive accuracy of the model. A similar limitation applies to POU3F1, where an available mouse complex (PDB ID: 4Y60) was used to predict human protein–DNA interactions. Notably, DeepBind [9], a sequence-based prediction tool, also failed to distinguish strong from weak binders when using the mouse POU3F1 protein (AUC score: 0.457), but this was corrected with the human POU3F1 protein (AUC score: 0.956).

We also examined the remaining cases where IDEA did not show a clear distinction between strong and weak binders: USF1, Egr1, and PROX1. For PROX1, we initially used the structure of a protein-DNA complex (PDB ID: 4Y60) in training. However, upon closer inspection, we discovered that this structure does not include the PROX1 protein, but SOX-18, a different transcription factor. This explains the inaccurate prediction made by IDEA. Since no experimental PROX1-DNA complex structure is currently available, we have removed this case from our HT-SELEX evaluation.

IDEA also fails to fully resolve the binding preference of USF1. A closer examination of the HT-SELEX data reveals a lack of distinction among the sequences, as most sequences, including those with the lowest M-word (binding affinity) scores, contain the DNA-binding E-box sequence CACGTG. Therefore, USF1 represents a challenging example where the experimental data only consists of strong binders with limited variations in binding affinity, which likely results from differences in flanking sequences of the E-box motif.

Egr1 stands as a peculiar example. Whereas IDEA does not effectively distinguish between the strong and weak binders in the current HT-SELEX dataset, its predictions are consistent with other experimental datasets, including binding affinities measured by kMITOMI [42] (Figure S8A, B), preferred binding sequences from protein-binding microarray, an earlier HT-SELEX experiment, and bacterial one-hybrid data [43]. Therefore, further investigation of the current HT-SELEX data is needed to reconcile these differences.

We have included additional text in Section: IDEA Demonstrates Transferability across Proteins in the Same CATH Superfamily to discuss the PCA analysis and the dependence of the model’s transferability on the similarity among the learned energy models.

The revised text now reads:

“The transferability of IDEA within the same CATH superfamily can be understood from the similarities in protein-DNA binding interfaces, which determine similar learned energy models. For example, the PHO4 protein (PDB I”D: 1A0A) shares a highly similar DNA-binding interface with the MAX protein (PDB ID: 1HLO) (Figure 3B), despite sharing only a 33.41% probability of being homologous. Consequently, the energy model derived from the PHO4DNA complex (Figure 3C) exhibits a similar amino-acid-nucleotide interactive pattern as that learned from the MAX-DNA complex (Figure 2B). To further evaluate the similarity between the learned energy models and their connection to protein families, we performed principal component analysis (PCA) on the normalized energy models across 24 proteins from 12 protein families [5]. Our analysis (Figure S11) reveals that most of the energy models from the same protein family fall within the same cluster, while those from different protein families exhibit distinct patterns. Moreover, the relative distance between energy models in PCA space reflects the degree of transferability between them. For example, PHO4 (PDB ID: 1A0A) is positioned close to MAX (PDB ID: 1HLO), whereas USF1 (PDB ID: 1AN4) and TCF4 (PDB ID: 6OD3) are farther away. This is consistent with the results in Figure 3A, where the energy model trained on PHO4 has better transferability than those trained on USF1 or TCF4.”

We have also added an Appendix section titled Analysis of examples where IDEA fails to recognize strong DNA binders to discuss the examples in which IDEA did not perform well:

“We examine IDEA’s capability in identifying strong binders from the HT-SELEX dataset across 12 protein families [5]. The model successfully predicts 18 out of 22 protein-DNA systems, but the performance is reduced in 4 cases. Closer investigations revealed the source of these limitations. In some instances, only the protein from a different organism is available. For example, the PDX1 HT-SELEX data utilized the human PDX1 protein, but no human PDX1–DNA complex structure is available. Therefore, the mouse PDX1–DNA complex structure (PDB ID: 2H1K) was used for model training. Differences between model organisms may reduce predictive accuracy. A similar limitation applies to POU3F1, where an available mouse complex (PDB ID: 4Y60) was used to predict human protein–DNA interactions. Notably, DeepBind [9], a sequence-based prediction tool, also failed to distinguish strong from weak binders when using the mouse POU3F1 protein (AUC score: 0.457), but this was corrected with the human POU3F1 protein (AUC score: 0.956).

IDEA also fails to fully resolve the binding preference of USF1. A closer examination of the HT-SELEX data reveals a lack of distinction among the sequences, as most sequences, including those with the lowest M-word (binding affinity) scores, contain the DNA-binding E-box sequence CACGTG. Therefore, USF1 represents a challenging example where the experimental data only consists of strong binders with limited variations in binding affinity, which likely results from differences in flanking sequences of the E-box motif.

Egr1 stands as a peculiar example. Whereas IDEA does not effectively distinguish between the strong and weak binders in the current HT-SELEX dataset, its predictions are consistent with other experimental datasets, including binding affinities measured by k-MITOMI [42] (Figure S8A, B), preferred binding sequences from protein-binding microarray, an earlier HT-SELEX experiment, and bacterial one-hybrid data [43]. Therefore, further investigation of the current HT-SELEX data is needed to reconcile these differences.”

Comment 7: It is also not clear if IDEA’s prediction for reverse complement sequences is the same for a given sequence. If so, how is this property being modelled? Either this description is lacking or I missed it.

We thank the reviewer for the insightful comments. Given a target protein-DNA sequence, the IDEA protocol substitutes it into a known protein-DNA complex structure to evaluate the binding free energy, which can be converted into binding affinity. IDEA uses sequence identity to determine whether the forward or reverse strand of the DNA should be replaced. Only the strand most similar to the target sequence is substituted. As a result, the model treats reverse-complement sequences differently. As the orientations of test sequences are specified from 5’ to 3’ in all datasets used in this study (e.g., processed MITOMI, HT-SELEX, and ChIP-seq data), this approach ensures that the target sequences are replaced and evaluated correctly. In cases where sequence orientation is not provided (though this was not an issue in this study), we recommend replacing both the forward and reverse strands with the target sequence separately and evaluating the corresponding protein–DNA binding free energies. Since strong binders are likely to dominate the experimental signals, the higher predicted binding affinity, with stronger binding free energies, should be taken as the model’s final prediction.

We have added one section to the Methods Section titled Treatment of Complementary DNA Sequences to clarify these modeling details.

The specific text reads:

To replace the DNA sequence in the protein-DNA complex structure with a target sequence, IDEA uses sequence identity to determine whether the target sequence belongs to the forward or reverse strand of the DNA in the proteinDNA structure. The more similar strand is selected and replaced with the target sequence. As the orientations of test sequences are specified from 5’ to 3’ in all datasets used in this study (e.g., processed MITOMI, HT-SELEX, and ChIP-seq data), this approach ensures that the target sequences are replaced and evaluated correctly. In cases where sequence orientation is not provided (though this was not an issue in this study), we recommend replacing both the forward and reverse strands with the target sequence separately and evaluating the corresponding protein–DNA binding free energies. Since strong binders are likely to dominate the experimental signals, the higher predicted binding affinity, with stronger binding free energy, should be taken as the model’s final prediction.”

“Comment 8: Page 21 line 403, the E-box core should be CACGTG instead of CACGTC.

We apologize for our oversight and have corrected the relevant text.

Comment 9: The citation for DNAproDB is outdated and should be updated (PMID 39494533).

We thank the reviewer for pointing this out and have updated our citation accordingly.

Reviewer #3:

Comment 0: Summary: Protein-DNA interactions and sequence readout represent a challenging and rapidly evolving field of study. Recognizing the complexity of this task, the authors have developed a compact and elegant model. They have applied well-established approaches to address a difficult problem, effectively enhancing the information extracted from sparse contact maps by integrating artificial sequences decoy set and available experimental data. This has resulted in the creation of a practical tool that can be adapted for use with other proteins.

We appreciate the reviewer’s excellent summary of the paper, and we thank the reviewer for the insightful suggestions and comments.

Comment 1: Strengths: (1) The authors integrate sparse information with available experimental data to construct a model whose utility extends beyond the limited set of structures used for training. (2) A comprehensive methods section is included, ensuring that the work can be reproduced. Additionally, the authors have shared their model as a GitHub project, reflecting their commitment to transparency of research.

We appreciate the reviewer’s strong assessment of the strengths of this paper. In addition to sharing our model on GitHub, we have also uploaded the original data and the essential scripts required to reproduce the results presented in the manuscript. We hope this further demonstrates our commitment to transparency and reproducibility.

Comment 2: Weaknesses: (1) The coarse-graining procedure appears artificial, if not confusing, given that full-atom crystal structures provide more detailed information about residue-residue contacts. While the selection procedure for distance threshold values is explained, the overall motivation for adopting this approach remains unclear. Furthermore, since this model is later employed as an empirical potential for molecular modeling, the use of P and C5 atoms raises concerns, as the interactions in 3SPN are modeled between Cα and the nucleic base, represented by its center of mass rather than P or C5 atoms.

We appreciate the reviewer’s insightful comments. The selection of P and C5 atoms was based on different relative positions of protein and DNA across various complex structures, each with distinctive protein-DNA structural interfaces. To illustrate this, we selected two representative structures where our algorithm selected C5 and P atoms, respectively: MAX-DNA (PDB ID: 1HLO) and FOXP3 (PDB ID: 7TDW). As shown in Author response image 2, in the case of 1HLO, more C5 atoms are within the cutoff distance of 10 A from˚ the protein Cα atoms, thus capturing essential contacting interactions. In contrast, 7TDW has more P atoms within this cutoff. Importantly, several P atoms are distributed on the minor groove of the DNA, which were not captured by the C5 atoms. To maximize the inclusion of relevant structural contacts, we employed a filtering scheme that selectively chooses either P or C5 atoms based on their proximity to the protein to enhance the model prediction. We note that while this scheme is helpful, the IDEA predictions remain robust across different atom selections. To assess this robustness, we performed binding affinity predictions using only P atoms on the HT-SELEX dataset across 12 protein families [5]. Our predictions (Author response table 1) show comparable performance to that achieved using our filtering scheme.

Author response image 2.

Comparison between P and C5 atoms in proximity to the protein 3D structures of MAX–DNA (A) and FOXP-DNA (B) complexes, where P atoms (red sphere) and C5 atoms (blue sphere) that are within 10 A of Cα atoms are highlighted.

When incorporating the trained IDEA energy model into a simulation model, we acknowledge a potential mismatch between the resolution of the data-driven model (one coarse-grained site per nucleotide) and the 3SPN simulation model (three coarse-grained sites per nucleotide). The selection of nucleic base sites for molecular interactions in the 3SPN model follows our previous work [44] and its associated code implementation. While revisiting this part of the manuscript, we identified an inconsistency in the reported results in Figure 5A of our initial version: Specifically, we previously used the protein side-chain atoms, rather than only the Cα atoms, in model training. Retraining the data using the Cα atoms results in reduced prediction performance for the IDEA model (Figure 5A). Nonetheless, incorporating this updated energy model into simulations still yielded high accuracy in the predicted absolute binding free energies (Author response image 3A), demonstrating the robustness of our simulation framework in predicting absolute binding free energies against variations in atom selection during the IDEA model training. Following the reviewer’s suggestion, we also incorporated the IDEA-trained energy model as short-range van der Waals interactions between protein Cα atoms and DNA P atoms. As shown in Author response image 3B, our simulation reveals a slightly improved performance over our original implementation, with higher Pearson and Spearman correlation coefficients and a fitted slope closer to 1.0. This result suggests that a more consistent atom selection scheme between the data-driven and simulation models can improve the overall predictions. Accordingly, we have updated Figure 5 with this improved setup, using the simulation model with short-range vdW interactions implemented between protein Cα atoms and DNA P atoms (Figure 5C), ensuring consistency between the IDEA model and simulation framework.

Author response table 1.

Comparison of IDEA performance using two DNA atom selection schemes: the filtering scheme presented in the manuscript (C5 and P atoms) versus using only P atoms. Cases where the two schemes result in different atom selections are highlighted in bold.

We acknowledge that a gap still exists between the resolution of the data-driven and simulation models. To ensure a completely consistent coarse-grained level between these two models, we will work on implementing the IDEA model output for 1-bead-per-nucleotide DNA simulation models in the future.

Comment 3: (2) Although the authors use a standard set of metrics to assess model quality and predictive power, some ∆∆G predictions compared to MITOMI-derived ∆∆G values appear nonlinear, which casts doubt on the interpretation of the correlation coefficient.

Author response image 3.

Comparison of simulations using different representative atoms (A) Protein-DNA binding simulation with the IDEA-model incorporated as short-range van der Waals between protein Cα atom and nucleic base site. (B) Protein-DNA binding simulation with the IDEA-model incorporated as short-range van der Waals between protein Cα atom and DNA P atoms. The predicted free energies are robust to the choice of DNA representative atoms. The predicted binding free energies are presented in physical units, and error bars represent the standard deviation of the mean.

We thank the reviewer for the insightful comments and agree that the linear fit between our model’s prediction and the experimental data may not be the best measure of performance. The primary utility of the IDEA model is to predict high-affinity DNA-binding sequences for a given DNA-binding protein by assessing the relative binding affinities across different DNA sequences. In this regard, the ranked order of predicted sequence binding affinities serves as a better metric for evaluating the success of this model. To evaluate this, we calculated both Spearman’s rank correlation coefficient, which does not rely on linear correlation, and the Pearson correlation coefficient between our predictions and the experimental results. As shown in Figure 2, our computation shows a Spearman’s rank correlation coefficient of 0.65 for the MAX-based predictions using one MAX-DNA complex (PDB ID: 1HLO), supporting the model’s capability to effectively distinguish strong from weak binders.

As reflected in Figure 2 of the main text, although our model generally captures the relative binding affinities across different DNA sequences, its predictive accuracy diminishes for low-affinity sequences (Figure 2). This could be due to two limitations of the current modeling framework: (1) The model is residue-based and estimates binding free energy as the additive sum of contributions from individual contacting amino-acid-nucleotide pairs. This assumption does not account for cooperative effects caused by simultaneous changes at multiple nucleotide positions. One potential direction to further improve the model would be to use a finer-grained representation by incorporating more atom types within contacting residues, and to use a many-body potential to better capture cooperative effects from multiple mutations. (2) The model assumes that the target DNA adopts the same binding interface as in the reference crystal structure. However, sequencedependent DNA shape has been shown to be important in determining protein-DNA binding affinity [1]. To address this limitation, a future direction is to use deep-learningbased methods to incorporate predicted DNA shape or protein-DNA complex structures based on their sequences [2, 3] into our model prediction.

To fully evaluate the predictive power of IDEA, we have included Spearman’s rank correlation coefficient for every correlation plot in this manuscript. Across all our analyses, the Spearman’s rank correlation coefficients reveal similar predictive performance as the Pearson correlation coefficients. Additionally, we have included in our discussion the current limitations of our model and potential directions for future improvement.

We have edited our Discussion Section to include a discussion on the limitations of the current model. Specifically, the added texts are:

“Although IDEA has proved successful in many examples, it can be improved in several aspects. The model currently assumes the training and testing sequences share the same protein-DNA structure. While double-stranded DNA is generally rigid, recent studies have shown that sequence-dependent DNA shape contributes to their binding specificity [1, 2, 4]. To improve predictive accuracy, one could incorporate predicted DNA shapes or structures into the IDEA training protocol. In addition, the model is residue-based and evaluates the binding free energy as the additive sum of contributions from individual amino-acid-nucleotide contacts. This assumption does not account for cooperative effects that may arise from multiple nucleotide changes. A potential refinement could utilize a finer-grained model that includes more atom types within contacting residues and employs a many-body potential to account for such cooperative effects.”

Comment 4: (3) The discussion section lacks information about the model’s limitations and a comprehensive comparison with other models. Additionally, differences in model performance across various proteins and their respective predictive powers are not addressed.

We thank the reviewer for the insightful comments. As discussed in the response to Comment 3, the current structural model has several limitations, which may reduce predictive accuracy for weak DNA binders. We have noted these limitations in the Discussion section.

To compare the performance of IDEA with state-of-the-art protein-DNA predictive models, we examined the predictive accuracies of two additional popular computational models: ProBound [8] and DeepBind [9]. ProBound has been shown to have a better performance than several earlier predictive protein-DNA models, including JASPAR 2018 [11], HOCOMOCO [12], Jolma et al. [13], and DeepSELEX [14]. To benchmark these models’ performance, we examine each method’s capability to identify strong binders with the HT-SELEX datasets covering 22 proteins from 12 protein families [5]. As suggested by Reviewer 1, we also calculated the PRAUC score, reweighted to account for data imbalance [6], as a complementary metric for evaluating the model performance.

As shown in Figure S6, Table S1, and Table S2, IDEA ranked second among the three predictive methods. It is important to note that both ProBound and DeepBind were trained on a curated version of the HT-SELEX data [13], which overlaps with the testing data [5]. Compared with them, IDEA was trained only on the given structural and sequence information from a single protein-DNA complex, thus independent of the testing data. In order to assess how IDEA performs when incorporating knowledge from HT-SELEX data, we augmented the training by randomly including half of the HT-SELEX data (see the Methods Section Enhanced Modeling Prediction with SELEX Data). The augmented IDEA model achieved the best performance among all the models. We further benchmarked IDEA using a 10-fold cross-validation on the same HT-SELEX data [5] and found that IDEA outperformed a recent regression model that considers the shape of DNA with different sequences [5]. Overall, IDEA can be used to predict protein-DNA affinities in the absence of known binding sequence data, thereby filling a critical gap when such experimental datasets are unavailable.

In addition, we compared the performance of IDEA with both general and family-specific knowledge-based energy models. First, we incorporated a knowledge-based generic protein-DNA energy model (DBD-Hunter) learned from the protein-DNA database, reported by Skoinick and coworkers [7], into our prediction protocol. This model assigns interaction energies to different functional groups within each DNA nucleotide (e.g., phosphate (PP), sugar (SU), pyrimidine (PY), and imidazole (IM) groups). For our comparison, we averaged the energy contributions of these groups within each nucleotide and replaced the IDEA-learned energy model with this generic one to test its ability to differentiate strong binders from weak binders in the HT-SELEX dataset [5]. As shown in Figure S6, the IDEA model generally achieves better performance than the generic energy model. Additionally, we compared IDEA with rCLAMPS, a family-specific energy model developed to predict protein-DNA binding specificity in the C2H2 and homeodomain families. As shown in Table S1 and Table S2, IDEA also shows better performance than rCLAMPS in most cases across the C2H2 and homeodomain families, demonstrating that it has better predictive accuracy than both family-specific and generic knowledge-based models.

We have revised our text to include the comparison between IDEA and other predictive models. Specifically, we revised the text in Section: IDEA Generalizes across Various Protein Families.

The revised text reads:

“To examine IDEA’s predictive accuracy across different DNA-binding protein families, we applied it to calculate protein-DNA binding affinities using a comprehensive HT-SELEX dataset [5]. We focused on evaluating the capability of IDEA to distinguish strong binders from weak binders for each protein with an experimentally determined structure. We calculated the probability density distribution of the top and bottom binders identified in the SELEX experiment. A well-separated distribution indicates the successful identification of strong binders by IDEA (Figure 2D and S4). Receiver Operating Characteristic (ROC) analysis was performed to calculate the Area Under the Curve (AUC) and the precision-recall curve (PRAUC) scores for these predictions. Further details are provided in the Methods Section Evaluation of IDEA Prediction Using HT-SELEX Data. Our analysis shows that IDEA successfully differentiates strong from weak binders for 80% of the 22 proteins across 12 protein families, achieving AUC and balanced PRAUC scores greater than 0.5 (Figure 2E and S5). To benchmark IDEA’s performance against other leading methods, we compared its predictions with several popular models, including the sequence-based predictive models ProBound [8] and DeepBind [9], the familybased energy model rCLAMPS [10], and the knowledge-based energy model DBD-Hunter [7]. IDEA demonstrates performance comparable to these stateof-the-art approaches (Figure S6, Table S1, and Table S2), and incorporating sequence features further improves its prediction accuracy. We also performed 10-fold cross-validation on the binding affinities of protein–DNA pairs in this dataset and found that IDEA outperforms a recent regression model that considers the shape of DNA with different sequences [5] (Figure S7). Details are provided in Section: Comparison of IDEA predictive performance Using HT-SELEX data.”

We also added one section Comparison of IDEA predictive performance Using HT-SELEX data in the Appendix to fully explain the comparison between IDEA and other popular models.

The added texts are:

“To benchmark the performance of IDEA against state-of-the-art protein-DNA predictive models, we evaluated its ability to recognize strong binders with the HT-SELEX datasets across 22 proteins from 12 families [5]. Specifically, we compare IDEA with two widely used sequence-based models: ProBound [8] and DeepBind [9]. ProBound has demonstrated superior performance over many other predictive protein-DNA models, including JASPAR 2018 [11], HOCOMOCO [12], Jolma et al. [13], and DeepSELEX [14]. To use ProBound, we retrieved the trained binding model for each protein from motifcentral.org and used the GitHub implementation of ProBoundTools to infer the binding scores between protein and target DNA sequences. Except for POU3F1, binding models are available for all proteins. Therefore, we excluded POU3F1 and evaluated the protein-DNA binding affinities for the remaining 21 proteins. To use DeepBind, sequence-specific binding affinities were predicted directly with its web server. The Area Under the Curve (AUC) and the Precision-Recall AUC (PRAUC) scores were used as metrics for comparison. An AUC score of 1.0 indicates a perfect separation between the strong- and weak-binder distributions, while an AUC score of 0.5 indicates no separation. Because there is a significant imbalance in the number of strong and weak binders from the experimental data [5], where the strong binders are far fewer than the weak binders, we reweighted the samples to achieve a balanced evaluation, using 0.5 as the baseline for randomized prediction [6]. As summarized in Figure S6, Table S1, and Table S2, IDEA ranked second among the three predictive models. In order to assess the performance of IDEA when augmented with additional protein-DNA binding data, we augmented IDEA using randomly selected half of the HT-SELEX data (see the Methods Section Enhanced Modeling Prediction with SELEX Data). The augmented IDEA model achieved the best performance among all the models.”

“In addition, we compared the performance of IDEA with both general and family-specific knowledge-based energy models. First, we incorporated a knowledgebased generic protein-DNA energy model (DBD-Hunter) learned from the protein-DNA database, reported by Skoinick and coworkers [7], into our prediction protocol. This model assigns interaction energies to different functional groups within each DNA nucleotide, including phosphate (PP), sugar (SU), pyrimidine (PY), and imidazole (IM) groups. For our comparison, we averaged the energy contributions of these groups within each nucleotide and replaced the IDEA-learned energy model with the DBD-Hunter model to assess its ability to differentiate strong binders from weak binders in the HTSELEX dataset [5]. Additionally, we compared IDEA with rCLAMPS, a familyspecific energy model developed to predict protein-DNA binding specificity in the C2H2 and homeodomain families. rCLAMPS learns a position-dependent amino-acid-nucleotide interaction energy model. To incorporate this model into the binding free energy calculation, we averaged the energy contributions across all occurrences of each amino-acid-nucleotide pair, which resulted in a 20-by-4 residue-type-specific energy matrix. This matrix is structurally analogous to the IDEA-trained energy model and can be directly integrated into the binding free energy calculations. As shown in Figure S6, Table S1, and Table S2, the IDEA model generally outperforms DBD-Hunter and rCLAMPS, demonstrating that it can achieve better predictive accuracy than both generic and family-specific knowledge-based models.”

“We also performed 10-fold cross-validation using the same HT-SELEX datasets, following the protocol described in the Methods Section Enhanced Modeling Prediction with SELEX Data. For each protein, we divided the entire dataset into 10 equal, randomly assigned folds. In each iteration, we used randomly selected 9 of the 10 folds as the training dataset and the remaining fold as the testing dataset. This process was repeated 10 times so that each fold served as the test set once. We then reported the average R2 scores across these iterations to evaluate IDEA’s predictive performance. Our results are compared with the 1mer and 1mer+shape methods from [5], the latest regression model that considers the shape of DNA with different sequences (Figure S7). This comparative analysis shows IDEA achieved higher predictive accuracy than the state-of-the-art sequence-based protein-DNA binding predictors for proteinDNA complexes that have available experimentally resolved structures.”

“Overall, these results demonstrate that IDEA can be used to predict the proteinDNA pairs in the absence of known binding sequence data, thus filling an important gap in protein-DNA predictions when experimental binding sequence data are unavailable.”

We also greatly appreciate the reviewer’s suggestion to examine the model’s performance across different proteins. To do this, we first evaluated the dependence of IDEA prediction on the availability of experimental structures similar to the target protein-DNA complexes. To quantitatively assess similarities and differences among the IDEA-derived energy models, we flattened each normalized energy model into an 80-dimensional vector and performed principal component analysis (PCA). As shown in Author response image 1 and Figure S11, energy models optimized from the same protein family fall within the same cluster, while those from different protein families exhibit distinct patterns. Moreover, the relative distance between energy models in PCA space reflects the degree of transferability. For example, PHO4 (PDB ID: 1A0A) is positioned close to MAX (PDB ID: 1HLO), whereas USF1 (PDB ID: 1AN4) and TCF4 (PDB ID: 6OD3) are farther away. This is consistent with the results shown in Figure 3A, where the energy model trained from PHO4 has better transferability than those from the other two systems. Therefore, the availability of experimental structures from protein-DNA complexes more similar to the target can lead to better predictive performance.

We also examine cases in which the IDEA model failed to show strong discriminative power for protein-DNA complexes in the HT-SELEX datasets [5] (Figures 2E and S5). When evaluating the model’s ability to distinguish between strong and weak binders, we used the available experimental structure most similar to the protein employed in the HT-SELEX experiments. In some instances, only the structure of the same protein from a different organism is available. For example, the HT-SELEX data for PDX1-DNA used the human PDX1 protein, but no human PDX1–DNA complex structure is available. Therefore, we used the mouse PDX1–DNA complex (PDB ID: 2H1K) for model training. The differences between species may limit the predictive accuracy of the model. A similar limitation applies to POU3F1, where an available mouse complex (PDB ID: 4Y60) was used to predict human protein–DNA interactions. Notably, DeepBind [9], a sequencebased prediction tool, also failed to distinguish strong from weak binders when using the mouse POU3F1 protein (AUC score: 0.457), but this was corrected with the human POU3F1 protein (AUC score: 0.956).

We also examined the remaining cases where IDEA did not show a clear distinction between strong and weak binders: USF1, Egr1, and PROX1. For PROX1, we initially used the structure of a protein-DNA complex (PDB ID: 4Y60) in training. However, upon closer inspection, we discovered that this structure does not include the PROX1 protein, but SOX-18, a different transcription factor. This explains the inaccurate prediction made by IDEA. Since no experimental PROX1-DNA complex structure is currently available, we have removed this case from our HT-SELEX evaluation.

IDEA also fails to fully resolve the binding preference of USF1. A closer examination of the HT-SELEX data reveals a lack of distinction among the sequences, as most sequences, including those with the lowest M-word (binding affinity) scores, contain the DNA-binding E-box sequence CACGTG. Therefore, USF1 represents a challenging example where the experimental data only consists of strong binders with limited variations in binding affinity, which likely results from differences in flanking sequences of the E-box motif.

Egr1 stands as a peculiar example. Whereas IDEA does not effectively distinguish between the strong and weak binders in the current HT-SELEX dataset, its predictions are consistent with other experimental datasets, including binding affinities measured by kMITOMI [42] (Figure S8A, B), preferred binding sequences from protein-binding microarray, an earlier HT-SELEX experiment, and bacterial one-hybrid data [43]. Therefore, further investigation of the current HT-SELEX data is needed to reconcile these differences.

In summary, IDEA’s predictive performance depends on the availability of experimental structures closely related to the target protein-DNA complexes, both in terms of protein sequences and model organisms.

We have included additional text in Section: IDEA Demonstrates Transferability across Proteins in the Same CATH Superfamily to discuss the PCA analysis and the dependence of the model’s transferability on the similarity among the learned energy models.

The revised text now reads:

“The transferability of IDEA within the same CATH superfamily can be understood from the similarities in protein-DNA binding interfaces, which determine similar learned energy models. For example, the PHO4 protein (PDB ID: 1A0A) shares a highly similar DNA-binding interface with the MAX protein (PDB ID: 1HLO) (Figure 3B), despite sharing only a 33.41% probability of being homologous. Consequently, the energy model derived from the PHO4DNA complex (Figure 3C) exhibits a similar amino-acid-nucleotide interactive pattern as that learned from the MAX-DNA complex (Figure 2B). To further evaluate the similarity between the learned energy models and their connection to protein families, we performed principal component analysis (PCA) on the normalized energy models across 24 proteins from 12 protein families [5]. Our analysis (Figure S11) reveals that most of the energy models from the same protein family fall within the same cluster, while those from different protein families exhibit distinct patterns. Moreover, the relative distance between energy models in PCA space reflects the degree of transferability between them. For example, PHO4 (PDB ID: 1A0A) is positioned close to MAX (PDB ID: 1HLO), whereas USF1 (PDB ID: 1AN4) and TCF4 (PDB ID: 6OD3) are farther away. This is consistent with the results in Figure 3A, where the energy model trained on PHO4 has better transferability than those trained on USF1 or TCF4.”

We have also added an Appendix section titled Analysis of examples where IDEA fails to recognize strong DNA binders to discuss the examples in which IDEA did not perform well:

“We examine IDEA’s capability in identifying strong binders from the HT-SELEX dataset across 12 protein families [5]. The model successfully predicts 18 out of 22 protein-DNA systems, but the performance is reduced in 4 cases. Closer investigations revealed the source of these limitations. In some instances, only the protein from a different organism is available. For example, the PDX1 HT-SELEX data utilized the human PDX1 protein, but no human PDX1–DNA complex structure is available. Therefore, the mouse PDX1–DNA complex structure (PDB ID: 2H1K) was used for model training. Differences between model organisms may reduce predictive accuracy. A similar limitation applies to POU3F1, where an available mouse complex (PDB ID: 4Y60) was used to predict human protein–DNA interactions. Notably, DeepBind [9], a sequence-based prediction tool, also failed to distinguish strong from weak binders when using the mouse POU3F1 protein (AUC score: 0.457), but this was corrected with the human POU3F1 protein (AUC score: 0.956).

IDEA also fails to fully resolve the binding preference of USF1. A closer examination of the HT-SELEX data reveals a lack of distinction among the sequences, as most sequences, including those with the lowest M-word (binding affinity) scores, contain the DNA-binding E-box sequence CACGTG. Therefore, USF1 represents a challenging example where the experimental data only consists of strong binders with limited variations in binding affinity, which likely results from differences in flanking sequences of the E-box motif.

Egr1 stands as a peculiar example. Whereas IDEA does not effectively distinguish between the strong and weak binders in the current HT-SELEX dataset, its predictions are consistent with other experimental datasets, including binding affinities measured by k-MITOMI [42] (Figure S8A, B), preferred binding sequences from protein-binding microarray, an earlier HT-SELEX experiment, and bacterial one-hybrid data [43]. Therefore, further investigation of the current HT-SELEX data is needed to reconcile these differences.”

Comment 5: The authors provide an implementation of their model via GitHub, which is commendable. However, it unexpectedly requires the Modeller suite, despite no details about homology modeling being included in the methods section.

We thank the reviewer for the helpful comments. We did not use the homology modeling module of Modeller. Instead, we only used a single Python script, buildseq.py, from the Modeller package to extract the protein and DNA sequences from the given PDB structure. We have clarified this in the README file on our GitHub repository.

Comment 6: While the manuscript is written in clear and accessible English, some sentences are quite long and could benefit from rephrasing (e.g., lines 49-52).

Thank you for the helpful suggestion. We agree that the original sentence was overly long and have revised it by splitting it into two for improved clarity and readability.

The revised version reads:

“The very robustness of evolution [46, 47, 48, 49] provides an opportunity to extract the sequence-structure relationships embedded in existing complexes. Guided by this principle, we can learn an interpretable binding energy landscape that governs the recognition processes of DNA-binding proteins.”

Comment 7: In line 82, the citations appear out of place, as the context seems to suggest the use of the newly developed model.

Thank you for this insightful suggestion. We have rephrased the sentence to better connect with the context of this section.

The revised text now reads:

“Finally, the learned energy model can be incorporated into a simulation framework to explore the dynamics of DNA-binding processes, revealing mechanistic insights into various DNA-templated processes.”

Comment 8: Line 143 ”different structure from the bHLH TFs and thus requires a different atom” This is the first instance in the manuscript where the atom selection for distance thresholding is mentioned, making the text somewhat confusing.

We thank the reviewer for the insightful comment and agree that the atom selection scheme appears abruptly in this section. To improve clarity, we have moved the detailed atom selection scheme and its rationale to the Methods Section titled Structural Modeling of Protein and DNA.

Comment 9: Figures: Overall, the figures are visually appealing but could be further improved.

We appreciate the positive feedback regarding the visual presentation of our figures. Following the reviewer’s suggestions and to further enhance clarity, we have revised several figures to improve labeling, layout, and annotations.

Comment 10: Figure 1: The description ”highlighted in blue” considers changing to ”highlighted in blue on the structure.”.

We have revised the text based on your suggestion.

Comment 11: Figure 2: Panel B is missing a color bar legend and units, as is the case in Figure 3C. Additionally, the placement of Panel C is unconventional - it appears it should be Panel D. The color scheme for the spheres is not fully described. Panel E: There are too many colors used; consider employing different markers to improve clarity.

Thank you for the helpful suggestions.

For Figure 2B and Figure 3C, we would like to clarify that the predicted energies are presented in reduced units due to an undetermined prefactor introduced during the model optimization. This point has now been clarified in the figure captions and is also explained in the Methods section titled Training Protocol.

Additionally, we have rearranged Panels C and D to improve the figure layout and have fully described the color coding used in the structural representations.

We have updated it to read:

“Results for MAX-based predictions. (A) The binding free energies calculated by IDEA, trained using a single MAX–DNA complex (PDB ID: 1HLO), correlate well with experimentally measured MAX–DNA binding free energies [50]. ∆∆G represents the changes in binding free energy relative to that of the wild-type protein–DNA complex. (B) The heatmap, derived from the optimized energy model, illustrates key amino acid–nucleotide interactions governing MAX–DNA recognition, showing pairwise interaction energies between 20 amino acids and the four DNA bases—DA (deoxyadenosine), DT (deoxythymidine), DC (deoxycytidine), and DG (deoxyguanosine). Both the predicted binding free energies and the optimized energy model are expressed in reduced units, as explained in the Methods Section Training Protocol. Each cell represents the optimized energy contribution, where blue indicates more favorable (lower) energy values, and red indicates less favorable (higher) values. (C) The 3D structure of the MAX–DNA complex (zoomed in with different views) highlights key amino acid–nucleotide contacts at the protein–DNA interface. Notably, several DNA deoxycytidines (red spheres) form close contacts with arginines (blue spheres). Additional nucleotide color coding: adenine (yellow spheres), guanine (green spheres), thymine (pink spheres). (D) Probability density distributions of predicted binding free energies for strong (blue) and weak (red) binders of the protein ZBTB7A. The mean of each distribution is marked with a dashed line. (E) Summary of AUC scores for protein–DNA pairs across 12 protein families, calculated based on the predicted probability distributions of binding free energies.”

We fully agree that Panel E was visually overwhelming. We have revised the plot by using a combination of color and marker shapes to more clearly distinguish between different protein families, as suggested.

Comment 12: Typos:

Line 18: Gene expressions → Gene expression?

Line 28: performed → utilized ?

We really appreciate the suggestions and have corrected the text accordingly.

References

(1) Tianyin Zhou, Ning Shen, Lin Yang, Namiko Abe, John Horton, Richard S Mann, Harmen J Bussemaker, Raluca Gordan, and Remo Rohs. Quantitative modeling ofˆ transcription factor binding specificities using DNA shape. Proceedings of the National Academy of Sciences, 112(15):4654–4659, 2015.

(2) Jinsen Li, Tsu-Pei Chiu, and Remo Rohs. Predicting DNA structure using a deep learning method. Nat Commun, 15(1):1243, February 2024.

(3) Josh Abramson, Jonas Adler, Jack Dunger, Richard Evans, Tim Green, Alexander Pritzel, Olaf Ronneberger, Lindsay Willmore, Andrew J. Ballard, Joshua Bambrick, Sebastian W. Bodenstein, David A. Evans, Chia-Chun Hung, Michael O’Neill, David Reiman, Kathryn Tunyasuvunakool, Zachary Wu, Akvile˙ Zemgulytˇ e, Eirini Arvan-˙ iti, Charles Beattie, Ottavia Bertolli, Alex Bridgland, Alexey Cherepanov, Miles Congreve, Alexander I. Cowen-Rivers, Andrew Cowie, Michael Figurnov, Fabian B. Fuchs, Hannah Gladman, Rishub Jain, Yousuf A. Khan, Caroline M. R. Low, Kuba Perlin, Anna Potapenko, Pascal Savy, Sukhdeep Singh, Adrian Stecula, Ashok Thillaisundaram, Catherine Tong, Sergei Yakneen, Ellen D. Zhong, Michal Zielinski, Augustin Zˇ´ıdek, Victor Bapst, Pushmeet Kohli, Max Jaderberg, Demis Hassabis, and John M. Jumper. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature, pages 1–3, May 2024.

(4) Raktim Mitra, Jinsen Li, Jared M. Sagendorf, Yibei Jiang, Ari S. Cohen, Tsu-Pei Chiu, Cameron J. Glasscock, and Remo Rohs. Geometric deep learning of protein–DNA binding specificity. Nat Methods, 21(9):1674–1683, September 2024.

(5) Lin Yang, Yaron Orenstein, Arttu Jolma, Yimeng Yin, Jussi Taipale, Ron Shamir, and Remo Rohs. Transcription factor family-specific DNA shape readout revealed by quantitative specificity models. Mol Syst Biol, 13(2):910, February 2017.

(6) Takaya Saito and Marc Rehmsmeier. The Precision-Recall Plot Is More Informative than the ROC Plot When Evaluating Binary Classifiers on Imbalanced Datasets. PLoS ONE, 10(3):e0118432, March 2015.

(7) Mu Gao and Jeffrey Skolnick. DBD-Hunter: a knowledge-based method for the prediction of DNA-protein interactions. Nucleic Acids Res, 36(12):3978–3992, July 2008.

(8) H. Tomas Rube, Chaitanya Rastogi, Siqian Feng, Judith F. Kribelbauer, Allyson Li, Basheer Becerra, Lucas A. N. Melo, Bach Viet Do, Xiaoting Li, Hammaad H. Adam, Neel H. Shah, Richard S. Mann, and Harmen J. Bussemaker. Prediction of protein–ligand binding affinity from sequencing data with interpretable machine learning. Nat Biotechnol, 40(10):1520–1527, October 2022.

(9) Babak Alipanahi, Andrew Delong, Matthew T Weirauch, and Brendan J Frey. Predicting the sequence specificities of DNA- and RNA-binding proteins by deep learning. Nat Biotechnol, 33(8):831–838, August 2015.

(10) Joshua L. Wetzel, Kaiqian Zhang, and Mona Singh. Learning probabilistic proteinDNA recognition codes from DNA-binding specificities using structural mappings. Genome Res, 32(9):1776–1786, September 2022.

(11) Aziz Khan, Oriol Fornes, Arnaud Stigliani, Marius Gheorghe, Jaime A CastroMondragon, Robin van der Lee, Adrien Bessy, Jeanne Cheneby, Shubhada R Kulka-` rni, Ge Tan, Damir Baranasic, David J Arenillas, Albin Sandelin, Klaas Vandepoele, Boris Lenhard, Benoˆıt Ballester, Wyeth W Wasserman, Franc¸ois Parcy, and Anthony Mathelier. JASPAR 2018: update of the open-access database of transcription factor binding profiles and its web framework. Nucleic Acids Research, 46(D1):D260–D266, January 2018.

(12) Ivan V. Kulakovskiy, Ilya E. Vorontsov, Ivan S. Yevshin, Ruslan N. Sharipov, Alla D. Fedorova, Eugene I. Rumynskiy, Yulia A. Medvedeva, Arturo Magana-Mora, Vladimir B. Bajic, Dmitry A. Papatsenko, Fedor A. Kolpakov, and Vsevolod J. Makeev. HOCOMOCO: towards a complete collection of transcription factor binding models for human and mouse via large-scale ChIP-Seq analysis. Nucleic Acids Res, 46(D1):D252–D259, January 2018.

(13) Arttu Jolma, Jian Yan, Thomas Whitington, Jarkko Toivonen, Kazuhiro R. Nitta, Pasi Rastas, Ekaterina Morgunova, Martin Enge, Mikko Taipale, Gonghong Wei, Kimmo Palin, Juan M. Vaquerizas, Renaud Vincentelli, Nicholas M. Luscombe, Timothy R. Hughes, Patrick Lemaire, Esko Ukkonen, Teemu Kivioja, and Jussi Taipale. DNABinding Specificities of Human Transcription Factors. Cell, 152(1-2):327–339, January 2013.

(14) Maor Asif and Yaron Orenstein. DeepSELEX: inferring DNA-binding preferences from HT-SELEX data using multi-class CNNs. Bioinformatics, 36(Supplement 2):i634–i642, December 2020.

(15) Oriol Fornes, Alberto Meseguer, Joachim Aguirre-Plans, Patrick Gohl, Patricia M Bota, Ruben Molina-Fernandez, Jaume Bonet, Altair Chinchilla-Hernandez, Ferran´ Pegenaute, Oriol Gallego, Narcis Fernandez-Fuentes, and Baldo Oliva. Structurebased learning to predict and model protein–DNA interactions and transcriptionfactor co-operativity in cis -regulatory elements. NAR Genomics and Bioinformatics, 6(2):lqae068, April 2024.

(16) Sofia Aizenshtein-Gazit and Yaron Orenstein. DeepZF: improved DNA-binding prediction of C2H2-zinc-finger proteins by deep transfer learning. Bioinformatics, 38(Suppl 2):ii62–ii67, September 2022.

(17) Stephen K Burley, Charmi Bhikadiya, Chunxiao Bi, Sebastian Bittrich, Henry Chao, Li Chen, Paul A Craig, Gregg V Crichlow, Kenneth Dalenberg, Jose M Duarte, Shuchismita Dutta, Maryam Fayazi, Zukang Feng, Justin W Flatt, Sai Ganesan, Sutapa Ghosh, David S Goodsell, Rachel Kramer Green, Vladimir Guranovic, Jeremy Henry, Brian P Hudson, Igor Khokhriakov, Catherine L Lawson, Yuhe Liang, Robert Lowe, Ezra Peisach, Irina Persikova, Dennis W Piehl, Yana Rose, Andrej Sali, Joan Segura, Monica Sekharan, Chenghua Shao, Brinda Vallat, Maria Voigt, Ben Webb, John D Westbrook, Shamara Whetstone, Jasmine Y Young, Arthur Zalevsky, and Christine Zardecki. RCSB Protein Data Bank (RCSB.org): delivery of experimentally-determined PDB structures alongside one million computed structure models of proteins from artificial intelligence/machine learning. Nucleic Acids Research, 51(D1):D488–D508, November 2022.

(18) Raktim Mitra, Ari S. Cohen, Jared M. Sagendorf, Helen M. Berman, and Remo Rohs. DNAproDB: an updated database for the automated and interactive analysis of protein-DNA complexes. Nucleic Acids Res, 53(D1):D396–D402, January 2025.

(19) Natalia Petrenko, Yi Jin, Liguo Dong, Koon Ho Wong, and Kevin Struhl. Requirements for RNA polymerase II preinitiation complex formation in vivo. eLife, 8:e43654, January 2019.

(20) Rudolf Jaenisch and Adrian Bird. Epigenetic regulation of gene expression: how the genome integrates intrinsic and environmental signals. Nat Genet, 33(3):245–254, March 2003.

(21) Claire Marchal, Jiao Sima, and David M. Gilbert. Control of DNA replication timing in the 3D genome. Nat Rev Mol Cell Biol, 20(12):721–737, December 2019.

(22) Lucia A. Hindorff, Praveen Sethupathy, Heather A. Junkins, Erin M. Ramos, Jayashri P. Mehta, Francis S. Collins, and Teri A. Manolio. Potential etiologic and functional implications of genome-wide association loci for human diseases and traits. Proceedings of the National Academy of Sciences, 106(23):9362–9367, June 2009.

(23) Tuuli Lappalainen, Alexandra J Scott, Margot Brandt, and Ira M Hall. Genomic analysis in the age of human genome sequencing. Cell, 177(1):70–84, 2019.

(24) Sonali Mukherjee, Michael F. Berger, Ghil Jona, Xun S. Wang, Dale Muzzey, Michael Snyder, Richard A. Young, and Martha L. Bulyk. Rapid analysis of the DNA-binding specificities of transcription factors with DNA microarrays. Nat Genet, 36(12):1331– 1339, December 2004.

(25) Shaoxun Liu, Pilar Gomez-Alcala, Christ Leemans, William J. Glassford, Lucas A. N. Melo, Xiang-Jun Lu, Richard S. Mann, and Harmen J. Bussemaker. Predicting the DNA binding specificity of transcription factor mutants using family-level biophysically interpretable machine learning. bioRxiv, page 2024.01.24.577115, April 2025.

(26) Tsu-Pei Chiu, Satyanarayan Rao, and Remo Rohs. Physicochemical models of protein–DNA binding with standard and modified base pairs. Proc. Natl. Acad. Sci. U.S.A., 120(4):e2205796120, January 2023.

(27) Matthew T Weirauch, Atina Cote, Raquel Norel, Matti Annala, Yue Zhao, Todd R Riley, Julio Saez-Rodriguez, Thomas Cokelaer, Anastasia Vedenko, Shaheynoor Talukder, and others. Evaluation of methods for modeling transcription factor sequence specificity. Nature biotechnology, 31(2):126–134, 2013.

(28) Chaitanya Rastogi, H. Tomas Rube, Judith F. Kribelbauer, Justin Crocker, Ryan E. Loker, Gabriella D. Martini, Oleg Laptenko, William A. Freed-Pastor, Carol Prives, David L. Stern, Richard S. Mann, and Harmen J. Bussemaker. Accurate and sensitive quantification of protein-DNA binding affinity. Proc. Natl. Acad. Sci. U.S.A., 115(16), April 2018.

(29) Rahmatullah Roche, Bernard Moussad, Md Hossain Shuvo, Sumit Tarafder, and Debswapna Bhattacharya. EquiPNAS: improved protein–nucleic acid binding site prediction using protein-language-model-informed equivariant deep graph neural networks. Nucleic Acids Research, 52(5):e27–e27, March 2024.

(30) Yufan Liu and Boxue Tian. Protein–DNA binding sites prediction based on pretrained protein language model and contrastive learning. Briefings in Bioinformatics, 25(1):bbad488, November 2023.

(31) Binh P. Nguyen, Quang H. Nguyen, Giang-Nam Doan-Ngoc, Thanh-Hoang Nguyen-Vo, and Susanto Rahardja. iProDNA-CapsNet: identifying protein-DNA binding residues using capsule neural networks. BMC Bioinformatics, 20(S23):634, December 2019.

(32) Trevor Siggers and Raluca Gordan. Protein–DNA binding: complexities and multi-ˆ protein codes. Nucleic Acids Research, 42(4):2099–2111, February 2014.

(33) Johannes Soding, Andreas Biegert, and Andrei N. Lupas. The HHpred interactive¨ server for protein homology detection and structure prediction. Nucleic Acids Research, 33(suppl 2):W244–W248, July 2005.

(34) William Humphrey, Andrew Dalke, and Klaus Schulten. VMD – Visual Molecular Dynamics. Journal of Molecular Graphics, 14:33–38, 1996.

(35) Arttu Jolma, Teemu Kivioja, Jarkko Toivonen, Lu Cheng, Gonghong Wei, Martin Enge, Mikko Taipale, Juan M Vaquerizas, Jian Yan, Mikko J Sillanpa¨a, and others.¨ Multiplexed massively parallel SELEX for characterization of human transcription factor binding specificities. Genome research, 20(6):861–873, 2010.

(36) Nobuo Ogawa and Mark D Biggin. High-throughput SELEX determination of DNA sequences bound by transcription factors in vitro. Gene Regulatory Networks: Methods and Protocols, pages 51–63, 2012.

(37) Alina Isakova, Romain Groux, Michael Imbeault, Pernille Rainer, Daniel Alpern, Riccardo Dainese, Giovanna Ambrosini, Didier Trono, Philipp Bucher, and Bart Deplancke. SMiLE-seq identifies binding motifs of single and dimeric transcription factors. Nature methods, 14(3):316–322, 2017.

(38) Paul G. Giresi, Jonghwan Kim, Ryan M. McDaniell, Vishwanath R. Iyer, and Jason D. Lieb. FAIRE (Formaldehyde-Assisted Isolation of Regulatory Elements) isolates active regulatory elements from human chromatin. Genome Res., 17(6):877–885, January 2007.

(39) Peter J Park. ChIP–seq: advantages and challenges of a maturing technology. Nature reviews genetics, 10(10):669–680, 2009.

(40) Terrence S. Furey. ChIP–seq and beyond: new and improved methodologies to detect and characterize protein–DNA interactions. Nat Rev Genet, 13(12):840–852, December 2012.

(41) Anna Bartlett, Ronan C. O’Malley, Shao-shan Carol Huang, Mary Galli, Joseph R. Nery, Andrea Gallavotti, and Joseph R. Ecker. Mapping genome-wide transcriptionfactor binding sites using DAP-seq. Nat Protoc, 12(8):1659–1672, August 2017.

(42) Marcel Geertz, David Shore, and Sebastian J Maerkl. Massively parallel measurements of molecular interaction kinetics on a microfluidic platform. Proceedings of the National Academy of Sciences, 109(41):16540–16545, 2012.

(43) Gary D. Stormo and Yue Zhao. Determining the specificity of protein–DNA interactions. Nat Rev Genet, 11(11):751–760, November 2010.

(44) Xingcheng Lin, Rachel Leicher, Shixin Liu, and Bin Zhang. Cooperative DNA looping by PRC2 complexes. Nucleic Acids Research, 49(11):6238–6248, June 2021.

(45) P. L. Privalov, A. I. Dragan, and C. Crane-Robinson. Interpreting protein/DNA interactions: distinguishing specific from non-specific and electrostatic from nonelectrostatic components. Nucleic Acids Research, 39(7):2483–2491, April 2011.

(46) J D Bryngelson and P G Wolynes. Spin glasses and the statistical mechanics of protein folding. Proc. Natl. Acad. Sci. U.S.A., 84(21):7524–7528, November 1987.

(47) J. N. Onuchic, Z. Luthey-Schulten, and P. G. Wolynes. Theory of protein folding: the energy landscape perspective. Annu Rev Phys Chem, 48:545–600, 1997.

(48) N. P. Schafer, B. L. Kim, W. Zheng, and P. G. Wolynes. Learning To Fold Proteins Using Energy Landscape Theory. Isr J Chem, 54(8-9):1311–1337, August 2014.

(49) Wen-Ting Chu, Zhiqiang Yan, Xiakun Chu, Xiliang Zheng, Zuojia Liu, Li Xu, Kun Zhang, and Jin Wang. Physics of biomolecular recognition and conformational dynamics. Rep. Prog. Phys., 84(12):126601, December 2021.

(50) Sebastian J. Maerkl and Stephen R. Quake. A Systems Approach to Measuring the Binding Energy Landscapes of Transcription Factors. Science, 315(5809):233–237, January 2007.

Author Response

The following is the authors’ response to the previous reviews.

eLife assessment

This important study reports a novel mechanism linking DHODH inhibition-mediated pyrimidine nucleotide depletion to antigen presentation. Alternative means of inducing antigen presentation provide therapeutic opportunities to augment immune checkpoint blockade for cancer treatment. While the solid mechanistic data in vitro are compelling, in vivo assessments of the functional relevance of this mechanism are still incomplete.

Public Reviews:

We thank all Reviewers for their insightful comments and excellent suggestions.

Reviewer #1 (Public Review):

The manuscript by Mullen et al. investigated the gene expression changes in cancer cells treated with the DHODH inhibitor brequinar (BQ), to explore the therapeutic vulnerabilities induced by DHODH inhibition. The study found that BQ treatment causes upregulation of antigen presentation pathway (APP) genes and cell surface MHC class I expression, mechanistically which is mediated by the CDK9/PTEFb pathway triggered by pyrimidine nucleotide depletion.

No comment from authors

The combination of BQ and immune checkpoint therapy demonstrated a synergistic (or additive) anti-cancer effect against xenografted melanoma, suggesting the potential use of BQ and immune checkpoint blockade as a combination therapy in clinical therapeutics.

No comment from authors

The interesting findings in the present study include demonstrating a novel cellular response in cancer cells induced by DHODH inhibition. However, whether the increased antigen presentation by DHODH inhibition actually contributed to the potentiation of the efficacy of immune-check blockade (ICB) is not directly examined is the limitation of the study.

No comment from authors for preceding text, comment addresses the following text

Moreover, the mechanism of the increased antigen presentation pathway by pyrimidine depletion mediated by CDK9/PTEFb was not validated by genetic KD or KO targeting by CDK9/PTEFb pathways.

We appreciate this comment, and we would like to explain why we did not pursue these approaches. According to DepMap, CRISPR/Cas9-mediated knockout of CDK9 in cancer cell lines is almost universally deleterious, scoring as “essential” in 99.8% (1093/1095) of all cell lines tested (see Author response image 1 below). This makes sense, as P-TEFb is required for productive RNA polymerase II elongation of most mammalian genes. As such, it was not feasible to generate cell lines with stable genetic knockout of CDK9 to test our hypothesis.

While knockdown of CDK9 by RNA interference could support our results, DepMap data seems to indicate that RNAi-mediated knockdown of CDK9 is generally ineffective in silencing its activity, as this perturbation scored as “essential” in only 6.2% (44/710) of tested cell lines. This suggests that incomplete depletion of CDK9 will likely not be sufficient to block APP induction downstream of nucleotide depletion. Furthermore, RNAi-mediated depletion of CDK9 may trigger transcriptional changes in the cell by virtue of its many documented protein-protein interactions, and it would be difficult to establish a consistent “time zero” at which point CDK9 protein depletion is substantial but secondary effects of this have not yet occurred to a significant degree. These factors constitute major limitations of experiments using RNAi-mediated knockdown of CDK9.

Author response image 1.

Essentiality score from CRISPR and RNAi perturbation of CDK9 in cancer cell lines https://depmap.org/portal/gene/CDK9?tab=overview&dependency=RNAi_merged

At any rate, we provide evidence that three different inhibitors of CDK9 (flavopiridol, dinaciclib, and AT7519) all inhibit our effect of interest (Fig 4B). The same results were observed using a previously validated CDK9-directed proteolysis targeting chimera (PROTAC2), and this was reversed by addition of excess pomalidomide (Fig 4C), which correlated with the presence/absence of CDK9 on western blot under the exact same conditions (Fig 4D).

It is formally possible that all CDK9 inhibitors we tested are blocking BQ-mediated APP induction by some shared off-target mechanism (or perhaps by two or more different off-target mechanisms) AND this CDK9-independent target also happens to be degraded by PROTAC2. However, this would be an extraordinarily non-parsimonious explanation for our results, and so we contend that we have provided compelling evidence for the requirement of CDK9 for BQ-mediated APP induction.

Finally, high concentrations of BQ have been reported to show off-target effects, sensitizing cancer cells to ferroptosis, and the authors should discuss whether the dose used in the in vivo study reached the ferroptotic sensitizing dose or not.

We are intrigued by the results shown to us by Reviewer #1 in the linked preprint (Mishima et al 2022, https://doi.org/10.21203/rs.3.rs-2190326/v1). We have also observed in our unpublished data that very high concentrations of BQ (>150µM) cause loss of cell viability that is not rescued by uridine supplementation and that occurs even in DHODH knockout cells. This effect of high-dose BQ must be DHODH-independent. We also agree that Mishima et al provide compelling evidence that the ferroptosis-sensitizing effect of high-dose BQ treatment is due (at least in large part) to inhibition of FSP1.

Although we showed that DHODH is strongly inhibited in tumor cells in vivo (Fig 5C), we did not directly measure the concentration of BQ in the tumor or plasma. Sykes et al (PMID: 27641501) found that the maximum plasma concentration (Cmax) for [BQ]free following a single IP administration in C57Bl6/J mice (15mg/kg) is approximately 3µM, while the Cmax for [BQ]total was around 215µM. Because polar drug molecules bound to serum proteins (predominantly albumin) are not available to bind other targets, [BQ]free is the relevant parameter.

Given a Cmax for [BQ]free of 3µM and half-life of 12.0 hours, we estimate that the steady-state [BQ]free with daily IP injections at this dose is around 4µM. Since we used an administration schedule of 10mg/kg every 24 hours, we estimate that the steady-state plasma [BQ]free in our system was 2.67µM (assuming initial Cmax of 2µM and half-life of 12.0 hours).

To derive an upper-bound estimate for the Cmax of [BQ]free over the 12-day treatment period (Fig 5A-D), we will use the observed data for 15mg/kg dose, and we will assume that 1) there is no clearance of BQ whatsoever and 2) that [BQ]free increases linearly with increasing [BQ]total. This yields a maximum free BQ concentration of 12 x 3 = 36µM.

Therefore, we consider it very unlikely that plasma concentrations of free BQ in our experiment exceeded the lower limit of the ferroptosis-sensitizing dose range reported by Mishima et al. However, without direct pharmacokinetic analysis, we cannot say for sure what the maximal [BQ]free was under our experimental conditions.

Reviewer #2 (Public Review):

In their manuscript entitled "DHODH inhibition enhances the efficacy of immune checkpoint blockade by increasing cancer cell antigen presentation", Mullen et al. describe an interesting mechanism of inducing antigen presentation. The manuscript includes a series of experiments that demonstrate that blockade of pyrimidine synthesis with DHODH inhibitors (i.e. brequinar (BQ)) stimulates the expression of genes involved in antigen presentation. The authors provide evidence that BQ mediated induction of MHC is independent of interferon signaling. A subsequent targeted chemical screen yielded evidence that CDK9 is the critical downstream mediator that induces RNA Pol II pause release on antigen presentation genes to increase expression. Finally, the authors demonstrate that BQ elicits strong anti-tumor activity in vivo in syngeneic models, and that combination of BQ with immune checkpoint blockade (ICB) results in significant lifespan extension in the B16-F10 melanoma model. Overall, the manuscript uncovers an interesting and unexpected mechanism that influences antigen presentation and provides an avenue for pharmacological manipulation of MHC genes, which is therapeutically relevant in many cancers. However, a few key experiments are needed to ensure that the proposed mechanism is indeed functional in vivo.

The combination of DHODH inhibition with ICB reflects more of an additive response instead of a synergistic combination. Moreover, the temporal separation of BQ and ICB raises the question of whether the induction of antigen presentation with BQ is persistent during the course of delayed ICB treatment. To confidently conclude that induction of antigen presentation is a fundamental component of the in vivo response to DHODH inhibition, the authors should examine whether depletion of immune cells can reduce the therapeutic efficacy of BQ in vivo.

We concur with this assessment.

Moreover, they should examine whether BQ treatment induces antigen presentation in non-malignant cells and APCs to determine the cancer specificity.

Although we showed that this occurs in HEK-293T cells, we appreciate that this cell line is not representative of human cells of any organ system in vivo. So, we agree it is important to determine if DHODH inhibition induces antigen presentation in human tissues and professional antigen presenting cells, and this is an excellent focus for future studies.

However, it should also be noted that increased antigen presentation in non-malignant host tissues would not be expected to generate an autoimmune response, because host tissues likely lack strong neoantigens, and whatever immunogenic peptides they may have would likely be presented via MHC-I at baseline (i.e. even in the absence of DHODH inhibitor treatment), since all nucleated cells express MHC-I.

This argument is strongly supported by clinical experience/data, as DHODH inhibitors (leflunomide and teriflunomide) are commonly used to treat rheumatoid arthritis and multiple sclerosis. While the pathophysiology of these autoimmune syndromes is complex, it is thought that both diseases are driven by aberrant T-cell attack on host tissues, mediated by incorrect recognition of host antigens presented via MHC-I (as well as MHC-II) as “foreign.”

If increased antigen presentation in host tissues (downstream of DHODH inhibition) could lead to a de novo autoimmune response, then administration of DHODH inhibitors would be expected to exacerbate T-cell driven autoimmune disease rather than ameliorate it. Randomized controlled trials have consistently found that treatment with DHODH inhibitors leads to improvement of rheumatoid arthritis and multiple sclerosis symptoms, which is the opposite of what one would expect if DHODH inhibitors are causing de novo autoimmune reactions in human patients.

Finally, although the authors show that DHODH inhibition induces expression of both MHC-I and MHC-II genes at the RNA level, only MHC-I is validated by flow cytometry given the importance of MHC-II expression on epithelial cancers, including melanoma, MHC-II should be validated as well.

We fully agree with this statement. We attempted to quantify cell surface MHC-II expression by FACS using the same method as for MHC-I (Figs 1G-H, 2D, and 3F). We did not detect cell surface MHC-II in any of our cancer cell lines, despite the use of high-dose interferon gamma and other stimulants (which robustly increase MHC-II mRNA in our system) in an attempt to induce expression. However, because we did not use cells known to express MHC-II as a positive control (e.g. B-cell leukemia cell lines or primary splenocytes), we do not know if our results are due to some technical failure (perhaps related to our protocol/reagents) or if they reflect a true absence of cell surface MHC-II in our cell lines.

If the latter is true, that implies that either 1) MHC-II mRNA is not translated or 2) that it is translated, but our cancer cell lines lack one or more elements of the machinery required for MHC-II antigen presentation.

In any case, it is important to determine if DHODH inhibition increases MHC-II at the cell surface of cancer cells using appropriate positive and negative controls, as this could have important implications for cancer immunotherapy.

[As a minor point, melanoma is not an epithelial cancer, as it is derived from neural crest lineage cells (melanocytes)]

Overall, the paper is clearly written and presented. With the additional experiments described above, especially in vivo, this manuscript would provide a strong contribution to the field of antigen presentation in cancer. The distinct mechanisms by which DHODH inhibition induces antigen presentation will also set the stage for future exploration into alternative methods of antigen induction.

Reviewer #3 (Public Review):

Mullen et al present an important study describing how DHODH inhibition enhances efficacy of immune checkpoint blockade by increasing cell surface expression of MHC I in cancer cells. DHODH inhibitors have been used in the clinic for many years to treat patients with rheumatoid arthritis and there has been a growing interest in repurposing these inhibitors as anti-cancer drugs. In this manuscript, the Singh group build on their previous work defining combinatorial strategies with DHODH inhibitors to improve efficacy. The authors identify an increase in expression of genes involved in the antigen presentation pathway and MHC I after BQ treatment and they narrow the mechanism to be strictly pyrimidine and CDK9/P-TEFb dependent. The authors rationalize that increased MHC I expression induced by DHODH inhibition might favor efficacy of dual immune checkpoint blockade. This combinatorial treatment prolonged survival in an immunocompetent B16F10 melanoma model.

[No comment from authors]

Previous studies have shown that DHODH inhibitors can increase expression of innate immunity-related genes but the role of DHODH and pyrimidine nucleotides in antigen presentation has not been previously reported. A strength of the manuscript is the use of multiple controls across a panel of cell lines to exclude off-target effects and to confirm that effects are exclusively dependent on pyrimidine depletion. Overall, the authors do a thorough characterization of the mechanism that mediates MHC I upregulation using multiple strategies. Furthermore, the in vivo studies provide solid evidence for combining DHODH inhibitors with immune checkpoint blockade.

No comment from authors

However, despite the use of multiple cell lines, most experiments are only performed in one cell line, and it is hard to understand why particular gene sets, cell lines or time points are selected for each experiment. It would be beneficial to standardize experimental conditions and confirm the most relevant findings in multiple cell lines.

We appreciate this comment, and we understand how the use of various cell lines may seem puzzling. We would like to explain how our cell line panel evolved over the course of the study. Our first indication that BQ caused APP upregulation came from transcriptomics experiments (Figs 1A-D, S1A) performed as part of a previous study investigating BQ resistance (Mullen et al, 2023 Cancer Letters). In that study, we used CFPAC-1 as a model for BQ sensitivity and S2-013 as a model for BQ resistance. We did RNA sequencing +/- BQ in these cell lines to look for gene expression patterns that might underlie resistance/sensitivity to BQ. When analyzing this data, we serendipitously discovered the APP/MHC phenomenon, which gave rise to the present study.

Our next step was to extend these findings to cancer cell lines of other histologies, and we prioritized cell lines derived from common cancer types for which immunotherapy (specifically ICB) are clinically approved. This is why A549 (lung adenocarcinoma), HCT116 (colorectal adenocarcinoma), A375 (cutaneous melanoma), and MDA-MB-231 (triple-negative breast cancer) cell lines were introduced.

Because PDAC is considered to have an especially “immune-cold” tumor microenvironment, we reasoned that even dramatically increasing cancer cell antigen presentation may be insufficient to elicit an effective anti-tumor immune response in vivo. So we shifted our focus towards melanoma, because a subset of melanoma patients is very responsive to ICB and loss of antigen presentation (by direct silencing or homozygous loss-of-function mutations in MHC-I components such as B2M, or by functional loss of IFN-JAK1/2-STAT signaling) has been shown to mediate ICB resistance in human melanoma patients. This is why we extended our findings to B16F10 murine melanoma cells, intending to use them for in vivo studies with syngeneic immunocompetent recipient mice.

The PDAC cell line MiaPaCa2 was introduced because a collaborator at our institution (Amar Natarajan) happened to have IKK2 knockout MiaPaCa2 cells, which allowed us to genetically validate our inhibitor results showing that IKK1 and IKK2 (crucial effectors for NF-kB signaling) are dispensable for our effect of interest.

Ultimately, realizing that our results spanned various human and murine cell lines, we chose to use HEK-293T cells to validate the general applicability of our findings to proliferating cells in 2D culture, since HEK-293T cells (compared to our cancer cell lines) have relatively few genetic idiosyncrasies and express MHC-I at baseline.

The differential in vivo survival depending on dosing schedule is interesting. However, this section could be strengthened with a more thorough evaluation of the tumors at endpoint.

Overall, this is an interesting manuscript proposing a mechanistic link between pyrimidine depletion and MHC I expression and a novel therapeutic strategy combining DHODH inhibitors with dual checkpoint blockade. These results might be relevant for the clinical development of DHODH inhibitors in the treatment of solid tumors, a setting where these inhibitors have not shown optimal efficacy yet.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

(1) The main issue is that it did not directly examine whether the increased antigen presentation by DHODH inhibition contributed to the potentiation of the efficacy of immune-check blockade (ICB). The additional effect of BQ in the xenograft tumor study was not examined to determine if it was due to increased antigen presentation toward the cancer cells or due to merely cell cycle arrest effect by pyrimidine depletion in the tumor cells. The different administration timing of ICB with BQ treatment (Fig 5E) would not be sufficient to answer this issue.

We agree with this assessment and, and we believe the experiment proposed by Reviewer #2 below (comparing the efficacy of BQ in Rag-null versus immunocompetent recipients) would address this question directly. We also think that using a more immunogenic cell line for this experiment (such as B16F10 transduced with ovalbumin or some other strong neoantigen) would be useful given the poor immunogenicity and lack of any defined strong neoantigen in B16F10 cells. An orthogonal approach would be to engraft cancer cells with or without B2M knockout into immunocompetent recipient mice (+/- BQ treatment) to further implicate MHC-I and antigen presentation. These questions will be addressed in future studies.

(2) Additionally, in the in vivo study, the increase in surface MHC1 in the protein level in by BQ treatment was not examined in the tumor samples, and it was not confirmed whether increased antigen presentation by BQ treatment actually promoted an anti-cancer immune response in immune cells. To support the story presented in the study, these data would be necessary.

We attempted to show this by immunohistochemistry, but unfortunately the anti-H2-Db antibody that we obtained for this purpose did not have satisfactory performance to assess this in our tissue samples harvested at necropsy.

(3) The mechanism of the increased antigen presentation pathway by pyrimidine depletion mediated by CDK9/PTEFb was not validated by genetic KD or KO targeting by CDK9/PTEFb pathways. In general, results only by the inhibitor assay have a limitation of off-target effects.

Please see our above reply to Reviewer #1 comment making this same point, where we spell out our rationale for not pursuing these experiments.

(4) High concentrations of BQ (> 50 uM) have been reported to show off-target effects, sensitizing cancer cells to ferroptosis, an iron-mediated lipid peroxidation-dependent cell death, independent of DHODH inhibition (https://www.researchsquare.com/article/rs-2190326/v1). It would be needed to discuss whether the dose used in the in vivo study reached the ferroptotic sensitizing dose or not.

Please see our above reply to Reviewer #1 comment making this same point, where we explain why we are very confident that the BQ dose administered in our animal experiments was far below the minimum reported BQ dose required to sensitize cancer cells to ferroptosis in vitro.

Reviewer #2 (Recommendations For The Authors):

Major Points

(1) According to the proposed model, BQ mediated induction of antigen presentation is a contributing factor to the efficacy of this therapeutic strategy. If this is true, then depletion of immune cells should reduce the therapeutic efficacy of BQ in vivo. The authors should perform the B16-F10 transplant experiments in either Rag null mice (if available) or with CD8/CD4 depletion. The expectation would be that T cell depletion (or MHC loss with genetic manipulation) should reduce the efficacy of BQ treatment. Absent this critical experiment, it is difficult to confidently conclude that induction of antigen presentation is a fundamental component of the in vivo response to DHODH inhibition.

We agree with this assessment and the proposed experiment comparing the response in Rag-null versus immunocompetent recipients. We also think that using a more immunogenic cell line for this experiment (such as B16F10 transduced with ovalbumin or some other strong neoantigen) would be useful given the poor immunogenicity and lack of any defined strong neoantigen in B16F10 cells. An orthogonal approach would be to engraft cancer cells with or without B2M knockout into immunocompetent recipient mice (+/- BQ treatment) to further implicate MHC-I and antigen presentation. These questions will be addressed in future studies.

(2) Does BQ treatment induce antigen presentation in non-malignant cells? APCs? If the induction of antigen presentation is not cancer specific and related to a pyrimidine depletion stress response, then there is a possibility that healthy tissues will also exhibit a similar phenotype, raising concerns about the specificity of a de novo immune response. The authors should examine antigen presentation genes in healthy tissues treated with BQ.

We agree it is important to examine if our findings regarding nucleotide depletion and antigen presentation are true of APCs and other non-transformed cells, but we are not so concerned about the possibility of raising an immune response against non-malignant host tissues, as explained above. We have reproduced the relevant section below:

“However, it should also be noted that increased antigen presentation in non-malignant host tissues would not be expected to generate an autoimmune response, because host tissues likely lack strong neoantigens, and whatever immunogenic peptides they may have would likely be presented via MHC-I at baseline, since all nucleated cells express MHC-I.

This argument is strongly supported by clinical experience/data, as DHODH inhibitors (leflunomide and teriflunomide) are commonly used to treat rheumatoid arthritis and multiple sclerosis. While the pathophysiology of these autoimmune syndromes is complex, it is thought that both diseases are driven by aberrant T-cell attack on host tissues, mediated by incorrect recognition of host antigens presented via MHC-I (as well as MHC-II) as “foreign.”

If increased antigen presentation in host tissues (downstream of DHODH inhibition) could lead to a de novo autoimmune response, then administration of DHODH inhibitors would be expected to exacerbate T-cell driven autoimmune disease rather than ameliorate it. Randomized controlled trials have consistently found that treatment with DHODH inhibitors leads to improvement of rheumatoid arthritis and multiple sclerosis symptoms, which is the opposite of what one would expect if DHODH inhibitors are causing de novo autoimmune reactions in human patients.”

(3) In the title, the authors claim that DHODH enhances the efficacy of ICB. However, the experiment shown in Figure 5D does not demonstrate this. The Kaplan Meier curves reflect more of an additive response versus a synergistic combination. Furthermore, the concurrent treatment of BQ and ICB seems to inhibit the efficacy of ICB due to BQ toxicity in immune cells. This result seems to contradict the title.

We do not agree with this assessment. Given that the effect of dual ICB alone was very marginal, while the effect of BQ monotherapy was quite marked, we cannot conclude from Fig 5 that BQ treatment inhibited ICB efficacy due to immune suppression.

(4) Related to Point 3, the temporal separation of BQ and ICB raises the question of whether the induction of antigen presentation with BQ is persistent during the course of delayed ICB treatment. One explanation for the results is that BQ treatment reduces tumor burden, and then a subsequent course of ICB also reduces tumor burden but not that the two therapies are functioning in synergy. To address this, the authors should measure the duration of BQ mediated induction of antigen presentation after stopping treatment.

We agree that the alternative explanation proposed by Reviewer #2 is possible and we appreciate the suggestion to test the stability of APP induction after stopping BQ treatment.

(5) In Figure 1, the authors show that DHODH inhibition induces expression of both MHC-I and MHC-II genes at the RNA level. However, they only validate MHC-I by flow cytometry. A simple experiment to evaluate the effect of BQ treatment on MHC-II surface expression would provide important additional mechanistic insight into the immunomodulatory effects of DHODH inhibition, especially given recent literature reinforcing the importance of MHC-II expression on epithelial cancers, including melanoma (Oliveira et al. Nature 2022).

We fully agree with this statement. We attempted to quantify cell surface MHC-II expression by FACS using the same method as for MHC-I (Figs 1G-H, 2D, and 3F). We did not detect cell surface MHC-II in any of our cancer cell lines, despite the use of high-dose interferon gamma and other stimulants (which robustly increase MHC-II mRNA in our system) in an attempt to induce expression. However, because we did not use cells known to express MHC-II as a positive control (e.g. B-cell leukemia cell lines or primary splenocytes), we do not know if our results are due to some technical failure (perhaps related to our protocol/reagents) or if they reflect a true absence of cell surface MHC-II in our cell lines.

If the latter is true, that implies that either 1) MHC-II mRNA is not translated or 2) that it is translated, but our cancer cell lines lack one or more elements of the machinery required for MHC-II antigen presentation.

In any case, it is important to determine if DHODH inhibition increases MHC-II at the cell surface of cancer cells using appropriate positive and negative controls, as this could have important implications for cancer immunotherapy.

[As a minor point, melanoma is not an epithelial cancer, as it is derived from neural crest lineage cells (melanocytes)]

Minor Points

(1) The authors show ChIP-seq tracks from Tan et al. for HLA-B. However, given the pervasive effect of Ter treatment across many HLA genes, the authors should either show tracks at additional loci, or provide a heatmap of read density across more loci. This would substantiate the mechanistic claim that RNA Pol II occupancy and activity across antigen presentation genes is the major driver of response to DHODH inhibition as opposed to mRNA stabilization/increased translation.

We appreciate this suggestion. We have changed Fig 4 by replacing the HLA-B track (old Fig 4E) with a representation of fold change (Ter/DMSO) in Pol II occupancy versus fold change (Ter/DMSO) in mRNA abundance for 23 relevant genes (new Fig 4G); both of these datasets were obtained from the Tan et al manuscript. This new figure panel (Fig 4G) also shows linear regression analysis demonstrating that Pol II occupancy and mRNA expression are significantly correlated for APP genes. While we recognize that this data in itself is not formal proof of our hypothesis, it does strongly support the notion that increased transcription is responsible for the increased mRNA abundance of APP genes that we have observed.

(2) A compelling way to demonstrate a change in antigen presentation is through mass spectrometry based immunopeptidomics. Performing immunopeptidomic analysis of BQ treated cell lines would provide substantial mechanistic insight into the outcome of BQ treatment. While this approach may be outside the scope of the current work, the authors should speculate on how this treatment may specifically alter the antigenic landscape where future directions would include empirical immunopeptidomics measurements.

We fully agree with this comment. While the abundance of cancer cell surface MHC-I is an important factor for anticancer immunity, another crucial factor is the identity of peptides that are presented. Treatments that cause presentation of more immunogenic peptides can enhance T-cell recognition even in the absence of a relative change in cell surface MHC-I abundance.

While we did not perform the immunopeptidomics experiments described, we can offer some speculation regarding this comment. As shown in Fig 1D-E, transcriptomics experiments suggest that immunoproteasome subunits (PSMB8, PSMB9, PSMB10) are upregulated upon DHODH inhibition. If this change in mRNA levels translates into greater immunoproteasome activity (which was not tested in our study), this would be expected to alter the repertoire of peptides available for presentation and could thereby change the immunopeptidome.

However, this hypothesis requires direct testing, and we hope future studies will delineate the effects of DHODH inhibition and other cancer therapies on the immunopeptidome, as this area of research will have important clinical implications.

(3) While the signaling through CDK9 seems convincing, it still does not provide a mechanistic link between depleted pyrimidines and CDK9 activity. The authors should speculate on the mechanism that signals to CDK9.

We agree with the assessment. A mechanistic link between depleted pyrimidines and CDK9 activity will be a subject of future studies.

(4) Related to minor point 2, the authors should consider a genetic approach to confirm the importance of CDK9. While the pharmacological approach, including multiple mechanistically distinct CDK9 inhibitors provides strong evidence, an additional experiment with genetic depletion of CDK9 (CRISPR KO, shRNA, etc) would provide compelling mechanistic confirmation.

Reviewer #1 raised this very same point, and we agree. Please see our reply to Reviewer #1, which details why we did not pursue this approach and argues that the evidence we present is compelling even in absence of genetic manipulation.

Additionally, please see the new Fig 4E and 4F, which is a repeat of Fig 4B using HCT116 cells. Figure 4E shows that, in this cell line, CDK9 inhibitors (flavopiridol, dinaciclib, and AT7519) block BQ-mediated APP induction, while PROTAC2 does not. Figure 4F shows that (for reasons we cannot fully explain) PROTAC2 does not lead to CDK9 degradation in HCT116 cells. This data strongly implicates CDK9, because it excludes a CDK9-degradation-independent effect of PROTAC2.

(5) Figure 2B needs a legend.

Thank you for pointing this out. We have added a legend to Fig 2B.

(6) The authors should comment in the discussion on how this strategy may be particularly useful in patients harboring genetic or epigenetic loss of interferon signaling, a known mechanism of ICB resistance. Perhaps DHODH inhibition could rescue MHC expression in cells that are deficient in interferon sensing.

Thank you for this suggestion! We have amended the Discussion section to mention this important point. Please see paragraph 2 of the revised Discussion section where we have added the following text:

“Because BQ-mediated APP induction does not require interferon signaling, this strategy may have particular relevance for clinical scenarios in which tumor antigen presentation is dampened by the loss or silencing of cancer cell interferon signaling, which has been demonstrated to confer both intrinsic and acquired ICB resistance in human melanoma patients.”

Reviewer #3 (Recommendations For The Authors):

The authors present convincing evidence of the mechanism by which pyrimidine nucleotides regulate MHC I levels and about the potential of combining DHODH inhibitors with dual immune checkpoint blockade (ICB). This is an interesting paper given the clinical relevance of DHODH inhibitors. The studies raise some questions, and some points might need clarifying as below:

• In Figure 2C, why do the authors focus on these two genes in the uridine rescue? These are important genes mediating antigen presentation, but it might be more interesting to see how H2-Db and H2-Kb expression correlate with the protein data shown in Fig 2D. Fig. 2C-2D is a relevant control, so it would be important to validate in a different cancer cell line (e.g. one of the PDAC cell lines used for the RNAseq).

We appreciate this comment. Although Fig 3C shows that BQ-induced expression of H2-Db, H2-Kb, and B2m is reversed by uridine (in B16F10 cells), we recognize that this was not the best placement for this data, as it can easily be overlooked here since uridine reversal is not the main point of Fig 3C. We have left Fig 3C as is, because we think that the uridine reversal demonstrated in that panel serves as a good internal positive control for reversal of BQ-mediated APP induction in that experiment.

We have repeated the experiments shown in the original Fig 2C and substituted the original Fig 2C with a new Fig 2C and Fig S2B, which show both Tap1 and Nlrc5 as well as H2-Db, H2-Kb, and B2m after treatment with either BQ (new Fig 2C) or teriflunomide (new Fig S2B). The original Fig S2B is now Fig S2C, and it shows that uridine has no effect on the expression of any of the genes assayed in the new Fig 2C or S2B.

The reversibility of cell surface MHC-I induction was also validated in HCT116 cells (Fig 3F). We included the uridine reversal in Fig 3F to avoid duplicating the control and BQ FACS data in multiple panels.

We have also added the qPCR data for HCT116 cells showing this same phenotype (at the mRNA level), which is the new Fig S2D.

We decided to prioritize HCT116 cells for our mechanistic studies (Figures S2D, S4A, and 4E-F) because previous reports indicate that it is diploid and therefore less genetically deranged compared to our other cancer cell lines.

• Figure 2F shows an elegant experiment to discard off-target effects related to cell death and to confirm that the increased MHC I expression is uniquely dependent on pyrimidines. DHODH has recently been involved in ferroptosis, a highly immunogenic type of cell death. What are the authors´ thoughts on BQ-induced ferroptosis as a possible contributor to the effects of ICB? Does BQ + ferroptosis inhibitor (ferrostatin) affect cell surface MHC I and/or expression of antigen processing genes?

The potential role of DHODH in ferroptosis protection (Mao et al 2021) has important implications, so we are glad that multiple reviewers raised questions concerning ferroptosis. We did not directly test the effect of ferroptosis inducing agents (with or without BQ) on MHC-I/APP expression, but that is certainly a worthwhile line of investigation.

The DHODH/ferroptosis issue is complicated by a study pointed out by Reviewer #1 that challenges the role of DHODH inhibition in BQ-mediated ferroptosis sensitization (Mishima et al, 2022). This study argues that high-dose BQ treatment causes FSP1 inhibition, and this underlies the effect of BQ on the cellular response to ferroptosis-inducing agents.

Regardless of whether BQ-induced ferroptosis-sensitization is dependent on DHODH, FSP1, or some other factor, the Mao and Mishima studies agree that a relatively high dose of BQ is required to observe these effects (100-200µM for most cell lines and >50µM even in the most ferroptosis-sensitive cell lines). As we explained above, we consider it very unlikely that the in vivo BQ exposure in our experiments (Fig 5) was high enough to cause significant ferroptosis, especially in the absence of any dedicated ferroptosis-inducing agent (which is typically required to cause ferroptosis even in the presence of high-dose BQ).

• The authors nail down the mechanism to CDK9 (Fig 4). However, all these experiments are performed in 293T cells. I would like to see a repeat of Fig. 4B in a cancer cell line (either PDAC or B16). Also, does BQ have any effect on CDK9 expression/protein levels?

We have added two figure panels that address this comment (new Fig 4E and 4F). Figure 4E (which is a repeat of Fig 4B with HCT116 cells) shows that CDK9 inhibitors (flavopiridol, AT7519, and dinaciclib) reverse BQ-mediated APP induction in HCT116 cells (this agrees with Fig S4A showing that flavopiridol reverses MHC induction by various nucleotide synthesis inhibitors in this cell line), but PROTAC2 does not. Figure 4F shows that PROTAC2 (for reasons we cannot explain) does not cause CDK9 degradation in HCT116 cells. This adds further support to our thesis that CDK9 is a critical mediator of BQ-mediated APP induction (because how else can this pattern of results be explained?). The text of the Results section has been amended to reflect this.

We chose to use HCT116 cells for this repeat experiment 1) to align with Fig S4A and 2) because, as previously mentioned, we consider HCT116 to be a good cell line for mechanistic studies because of its relative lack of idiosyncratic genetic features (compared to CFPAC-1, for example, which was derived from a patient with cystic fibrosis).

• What are the differences in tumor size for the experiment shown in Figure 5E? What about tumor cell death in the ICB vs. BQ+ICB groups?

Because this was a survival assay, direct comparisons of tumor volumes between groups was not possible at later time points, since mice that die or have to be euthanized are removed from their experimental group, which lowers the average group tumor burden at subsequent time points. Although tumor volume was the most common euthanasia criteria reached, a subset of mice were either found dead or had to be euthanized for other reasons attributed to their tumor burden (moribund state, inability to ambulate or stand, persistent bleeding from tumor ulceration, severe loss of body mass, etc.). This confounds any comparison of endpoint measurements (such as immunohistochemical quantification of tumor cell death markers, T-cell markers, etc.).

• The different response in the concurrent vs delayed treatment is very interesting. The authors suggest two possible mechanisms to explain this: "1) Concurrent BQ dampens the initial anticancer immune response generated by dual ICB, or b) cancer cell MHC-I and related genes are not maximally upregulated at the time of ICB administration with concurrent treatment". However, and despite the caveat of comparing the in vitro to the in vivo setting, Fig 2D shows upregulation of MHC I already at 24h of treatment in B16 cells. Have the authors checked T cell infiltration in the concurrent and delayed treatment setting?

For the same reasons described in response to the preceding comment, tumors harvested upon mouse death/euthanasia from our survival experiment were not suitable for cross-cohort comparison of tumor endpoint measurements. An additional experiment in which mice are necropsied at a prespecified time point (before any mice have died or reached euthanasia criteria, as in the experiment for Fig 5A-D) would be required to answer this question.

• Page 5, line 181 -do the authors mean "nucleotide salvage inhibitors" instead of "synthesis"?

We believe the reviewer is referring to the following sentence:

“The other drugs screened included nucleotide synthesis inhibitors (5-fluorouracil, methotrexate, gemcitabine, and hydroxyurea), DNA damage inducers (oxaliplatin, irinotecan, and cytarabine), a microtubule targeting drug (paclitaxel), a DNA methylation inhibitor (azacytidine), and other small molecule inhibitors (Fig 2F).”

In this context, we believe our use of “synthesis” instead of “salvage” is correct, because methotrexate and 5-FU inhibit thymidylate synthase (which mediates de novo dTTP synthesis), while gemcitabine and hydroxyurea inhibit ribonucleotide reductase (which mediates de novo synthesis of all dNTPs).

Author response:

The following is the authors’ response to the original reviews

Public Reviews:

Reviewer #1 (Public review):

This work employs both in vitro and in vivo/transplant methods to investigate the contribution of BDNF/TrkB signaling to enhancing differentiation and dentin-repair capabilities of dental pulp stem cells in the context of exposure to a variety of inflammatory cytokines. A particular emphasis of the approach is the employment of dental pulp stem cells in which BDNF expression has been enhanced using CRISPR technology. Transplantation of such cells is said to improve dentin regeneration in a mouse model of tooth decay.

The study provides several interesting findings, including demonstrating that exposure to several cytokines/inflammatory agents increases the quantity of (activated) phospho-Trk B in dental pulp stem cells.

However, a variety of technical issues weaken support for the major conclusions offered by the authors. These technical issues include the following:

Thank you for your keen observation and evaluation, which helped us significantly improve our manuscript. We have addressed the concerns and comments point by point in detail and substantially revised the manuscript and Figures. We hope that the manuscript is acceptable in the current improvised version.

Detailed response to your comments/concerns is as follows:

(1) It remains unclear exactly how the cytokines tested affect BDNF/TrkB signaling. For example, in Figure 1C, TNF-alpha increases TrkB and phospho-TrkB immunoreactivity to the same degree, suggesting that the cytokine promotes TrkB abundance without stimulating pathways that activate TrkB, whereas in Figure 2D, TNF-alpha has little effect on the abundance of TrkB, while increasing phospho-TrkB, suggesting that it affects TrkB activation and not TrkB abundance.

Thank you for your kind concern. Recently, we have demonstrated the effect and interaction of TNF-alpha and Ca2+/calmodulin-dependent protein kinase II on the regulation of the inflammatory hDPSCs dentino-differentiation via BDNF/TrkB receptor signaling using TrkB inhibitor (Ref. below, and Figure 9). Moreover, we agree with your concern, and we have re-analyzed our replicates and found a better trend and significant abundance of TrkB as well (please refer to revised Figure 2D).

Ref.: Kim, Ji Hyun, et al. (2025) "Ca 2+/calmodulin-dependent protein kinase II regulates the inflammatory hDPSCs dentino-differentiation via BDNF/TrkB receptor signaling." Frontiers in Cell and Developmental Biology 13: 1558736.

(2) I find the histological images in Figure 3 to be difficult to interpret. I would have imagined that DAPI nuclear stains would reveal the odontoblast layer, but this is not apparent. An adjacent section labeled with conventional histological stains would be helpful here. Others have described Stro-1 as a stem cell marker that is expressed on a minority of cells associated with vasculature in the dental pulp, but in the images in Figure 3, Stro-l label is essentially co-distributed with DAPI, in both control and injured teeth, indicating that it is expressed in nearly all cells. Although the authors state that the Stro-1-positive cells are associated with vasculature, but I see no evidence that is true.

Thank you for your concern. STRO-1 is a mesenchymal stem cell marker also expressed in dental pulp stem cells; both populations are distributed in the pulp. DPSCs can contribute to tissue repair and regeneration in inflamed pulp by differentiating into odontoblasts and forming reparative dentin. Moreover, in the case of carious and inflamed pulp, they are disorganized depending on the extent of infection/injury. Our purpose here was to point out DPSCs presence, not vasculature, which will differentiate into odontoblasts in such a scenario. We have revised Figure 3 by adding magnified images and dotted lines to indicate the boundary between the pulp and dentin.

Ref. Volponi A. A., Pang Y., Sharpe P. T. Stem cell-based biological tooth repair and regeneration. Trends in Cell Biology. 2010;20(12):715–722.

(3) The data presented convincingly demonstrate that they have elevated BDNF expression in their dental pulp stem cells using a CRISPR-based approach I have a number of questions about these findings. Firstly, nowhere in the paper do they describe the nature of the CRISPR plasmid they are transiently transfecting. Some published methods delete segments of the BDNF 3'-UTR while others use an inactivated Cas9 to position an active transactivator to sequences in the BDNF promoter. If it is the latter approach, transient transfection will yield transient increases in BDNF expression. Also, as BDNF employs multiple promoters, it would be helpful to know which promoter sequence is targeted, and finally, knowing the identity of the guide RNAs would allow assessment for the potential of off-target effects I am guessing that the investigators employ a commercially obtained system from Santa Cruz, but nowhere is this mentioned. Please provide this information.

Dear Reviewer, yes, you are right. We have used a commercially obtained system from Santa Cruz, i.e., BDNF CRISPR Activation Plasmid (h): sc-400029-ACT and UltraCruz® Transfection Reagent (sc-395739), and they have been mentioned in Chemicals and Reagents section of Materials and Methods as follows.

“BDNF CRISPR Activation Plasmid (h) is a synergistic activation mediator (SAM) transcription activation system designed to upregulate gene expression specifically BDNF CRISPR Activation Plasmid (h) consists of three plasmids at a 1:1:1 mass ratio: a plasmid encoding the deactivated Cas9 (dCas9) nuclease (D10A and N863A) fused to the transactivation domain VP64, and a blasticidin resistance gene; a plasmid encoding the MS2-p65-HSF1 fusion protein, and a hygromycin resistance gene; a plasmid encoding a target-specific 20 nt guide RNA fused to two MS2 RNA aptamers, and a puromycin resistance gene.”

The resulting SAM complex binds to a site-specific region approximately 200-250 nt upstream of the transcriptional start site and provides robust recruitment of transcription factors for highly efficient gene activation

Following transfection, gene activation efficiency could be assayed by WB, IF, or IHC using antibody: pro-BDNF Antibody (5H8): sc-65514

Author response image 1.

(4) Another question left unresolved is whether their approach elevated BDNF, proBDNF, or both. Their 28 kDa western blot band apparently represents proBDNF exclusively, with no mature BDNF apparent, yet only mature BDNF effectively activates TrkB receptors. On the other hand, proBDNF preferentially activates p75NTR receptors. The present paper never mentions p75NTR, which is a significant omission, since other investigators have demonstrated that p75NTR controls odontoblast differentiation.

Dear reviewer, thank you for your noticing the error.

Pro-BDNF is produced as a 32-kDa precursor that undergoes N-glycosylation and glycosulfation on residues located within the pro-domain of the precursor. N-terminal cleavage of the precursor generates mature BDNF as well as a minor truncated form of the precursor (28 kDa) that arises by a different processing mechanism than mature BDNF. The precursor undergoes N-terminal cleavage within the trans-Golgi network and/or immature secretory vesicles to generate mature BDNF (14 kDa).

We checked our data and band size, and it shows a little mistake (Thank you for your keen observation and pointing out). The CRISPR protocol required verification of gene activation by checking pro-BDNF, as mentioned in the methodology. The labeling has been revised in the figure as pro-BDNF, and the actual blot with a ladder has been shown below for clarification.

(5) In any case, no evidence is presented to support the conclusion that the artificially elevated BDNF expression has any effect on the capability of the dental pulp stem cells to promote dentin regeneration. The results shown in Figures 4 and 5 compare dentin regeneration with BDNF-over-expressing stem cells with results lacking any stem cell transplantation. A suitable control is required to allow any conclusion about the benefit of over-expressing BDNF.

We have tested the presence of BDNF overexpressing cells by the higher expression of GFP here. Moreover, a significant increment in the dentin mineralization volume indicates the advantage of BDNF-over-expressing stem cells. Recently, we published the in vitro effects of BDNF/TrkB on DPSCs odontoblastic differentiation strongly supporting our in vivo data. Currently, we are in a difficult position to conduct the animal study within a short period of time. We would definitely consider using positive control in our future studies.

Ref.: Kim, Ji Hyun, et al. (2025) "Ca 2+/calmodulin-dependent protein kinase II regulates the inflammatory hDPSCs dentino-differentiation via BDNF/TrkB receptor signaling." Frontiers in Cell and Developmental Biology 13: 1558736.

(6) Whether increased BDNF expression is beneficial or not, the evidence that the BDNF-overexpressing dental pulp stem cells promote dentin regeneration is somewhat weak. The data presented indicate that the cells increase dentin density by only 6%. The text and figure legend disagree on whether the p-value for this effect is 0.05 or 0.01. In either case, nowhere is the value of N for this statistic mentioned, leaving uncertainty about whether the effect is real.

A significant increment in the dentin mineralization volume by about 7.76% indicates the advantage of BDNF-over-expressing stem cells, and we believe this could be a breakthrough to advance stem cell engineering and therapy further to get this percentage higher in the future. The text in the result section shows that the p-value for this effect is 0.05. While N was 3 previously, we analyzed two more samples by CT scan and revised results, taking N = 5, which improved the results a little more to about 8.53%. Thank you for noticing; the figure legend has been corrected to 0.05.

Similarly, our in vitro data in the current study supports the notion that it adds up to mineralization and odontoblastic differentiation. We recently published that BDNF/TrkB significantly enhances calcium deposits and mineralization using a battery of in vitro experiments.

Ref.: Kim, Ji Hyun, et al. (2025) "Ca 2+/calmodulin-dependent protein kinase II regulates the inflammatory hDPSCs dentino-differentiation via BDNF/TrkB receptor signaling." Frontiers in Cell and Developmental Biology 13: 1558736.

(7) The final set of experiments applies transcriptomic analysis to address the mechanisms mediating function differences in dental pulp stem cell behavior. Unfortunately, while the Abstract indicates " we conducted transcriptomic profiling of TNFα-treated DPSCs, both with and without TrkB antagonist CTX-B" that does not describe the experiment described, which compared the transcriptome of control cells with cells simultaneously exposed to TNF-alpha and CTX-B. Since CTX-B blocks the functional response of cells to TNF-alpha, I don't understand how any useful interpretation can be attached to the data without controls for the effect of TNF alone and CTX-B alone.

Dear reviewer, yes, we did it alone and together as well. Earlier, we showed only the combined results and mentioned the interaction between TNFα and TrkB. We have included the results from TNFα alone and combined them with CTX-B for better comparison (Please refer to Figure 8). Figure 8C1 clearly shows the reversal of certain factors with the treatment of TrkB inhibitor compared to figure 8C with TNFα alone treated group.

Reviewer #2 (Public review):

Summary:

In this manuscript, the authors investigate the potential for overexpressing BDNF in dental pulp stem cells to enhance dentin regeneration. They suggest that in the inflammatory environment of injured teeth, there is increased signaling of TrkB in response to elevated levels of inflammatory molecules.

Strengths:

The potential application to dentin regeneration is interesting.

Weaknesses:

There are a number of concerns with this manuscript to be addressed.

Thank you for your compliments, keen observation, and evaluation, which helped us significantly improve our manuscript. We have addressed the concerns and comments point by point in detail and substantially revised the manuscript and Figures. We hope that the manuscript is acceptable in the current improvised version.

Detailed response to your comments/concerns is as follows:

(1) Insufficient citation of the literature. There is a vast literature on BDNF-TrkB regulating survival, development, and function of neurons, yet there is only one citation (Zhang et al 2012) which is on Alzheimer's disease.

More references have been cited accordingly.

(2) There are several incorrect statements. For example, in the introduction (line 80) TrkA is not a BDNF receptor.

Thank you for noticing the typo; the sentence has been corrected.

(3) Most important - Specific antibodies must be identified by their RRID numbers. To state that "Various antibodies were procured:... from BioLegend" is unacceptable, and calls into question the entire analysis. Specifically, their Western blot in Figure 4B indicates a band at 28 kDa that they say is BDNF, however the size of BDNF is 14 kDa, and the size of proBDNF is 32 and 37 kDa, therefore it is not clear what they are indicating at 28 kDa. The validation is critical to their analysis of BDNF-expressing cells.

Dear reviewer, thank you for your kind concern. Sorry for the inconvenience; we have added RRID numbers of antibodies.

Pro-BDNF is produced as a 32-kDa precursor that undergoes N-glycosylation and glycosulfation on residues located within the pro-domain of the precursor. N-terminal cleavage of the precursor generates mature BDNF as well as a minor truncated form of the precursor (28 kDa) that arises by a different processing mechanism than mature BDNF. The precursor undergoes N-terminal cleavage within the trans-Golgi network and/or immature secretory vesicles to generate mature BDNF (14 kDa).

We checked our data and band size, and it shows a mistake in recognizing ladder size. It is actually a 14kDa band which has been shown. The labeling has been revised in the figure, and the actual blot with a ladder has been shown below for clarification. Similarly, our data focused on the fact that the observed cellular effects are more consistent with BDNF/TrkB-mediated pathways, which are known to promote survival and differentiation.

(4) Figure 2 indicates increased expression of TrkB and TrkA, as well as their phosphorylated forms in response to inflammatory stimuli. Do these treatments elicit increased secretion of the ligands for these receptors, BDNF and NGF, respectively, to activate their phosphorylation? Or are they suggesting that the inflammatory molecules directly activate the Trk receptors? If so, further validation is necessary to demonstrate that.

Thank you for your kind concern. TNF-α increases the number of TrkB receptors. The enhanced TrkB activation may result from a greater number of receptors and/or increased activation of individual receptors. In either case, inflammatory agents enhance the TrkB receptor signaling pathway.

Recently, we have demonstrated the effect and interaction of TNF-alpha and Ca2+/calmodulin-dependent protein kinase II on the regulation of the inflammatory hDPSCs dentino-differentiation via BDNF/TrkB receptor signaling using TrkB inhibitor (Ref. below, and Figure 9). For now, we have added figure 9 for the proposed mechanism of action based on our recent and current study.

Ref.: Kim, Ji Hyun, et al. (2025) "Ca 2+/calmodulin-dependent protein kinase II regulates the inflammatory hDPSCs dentino-differentiation via BDNF/TrkB receptor signaling." Frontiers in Cell and Developmental Biology 13: 1558736.

(5) Figure 7 - RNA-Seq data, what is the rationale for treatment with TNF+ CTX-B? How does this identify any role for TrkB signaling? They never define their abbreviations, but if CTX-B refers to cholera toxin subunit B, which is what it usually refers to, then it is certainly not a TrkB antagonist.

Thank you for your concern. Cyclotraxin-B (CTX-B) is a TrkB antagonist (mentioned in the revised manuscript). In order to identify the underlying mechanism, we ought to locate certain transcriptional factors interacting with the TrkB/BDNF signaling, leading to differentiation and dentinogenesis. Therefore, we treated it with a TrkB inhibitor.

Earlier, we showed only the combined results and mentioned the interaction between TNFα and TrkB. We have included the results from TNFα alone and combined them with CTX-B for better comparison (Please refer to Figure 8). Figure 8C1 clearly shows the reversal of certain factors with the treatment of TrkB inhibitor compared to figure 8C with TNFα alone treated group. We agree that the precise role of CTX-B in modulating TrkB signaling requires further clarification and have now included this point in the revised discussion while we are currently working on this aspect.

Reviewer #3 (Public review):

In general, although the authors interpret their results as pointing towards a possible role of BDNF in dentin regeneration, the results are over-interpreted due to the lack of proper controls and focus on TrkB expression, but not its isoforms in inflammatory processes. Surprisingly, the authors do not study the possible role of p75 in this process, which could be one of the mechanisms intervening under inflammatory conditions.

Thank you for your compliments, keen observation, and evaluation, which helped us significantly improve our manuscript. We have addressed the concerns and comments point by point in detail and substantially revised the manuscript and Figures. We hope that the manuscript is acceptable in the current improvised version.

Detailed response to your comments/concerns is as follows:

(1) The authors claim that there are two Trk receptors for BDNF, TrkA and TrkB. To date, I am unaware of any evidence that BDNF binds to TrkA to activate it. It is true that two receptors have been described in the literature, TrkB and p75 or NGFR, but the latter is not TrkA despite its name and capacity to bind NGF along with other neurotrophins. It is crucial for the authors to provide a reference stating that TrkA is a receptor for BDNF or, alternatively, to correct this paragraph.

Dear reviewer, we apologize for the inconvenience; it was an error. BDNF binds to TrkB, and the sentence has been corrected.

(2) The authors discuss BDNF/TrkB in inflammation. Is there any possibility of p75 involvement in this process?

Mature BDNF binds to the high-affinity receptor tyrosine kinase B (TrkB), activating signaling cascades, while pro-BDNF binds to the p75 neurotrophin receptor (p75NTR). So, we don’t think there’s a possibility, as our data shows mature BDNF production. Here, we initially screened the TrkA and TrkB involvement in dentinogenesis and chose to work with BDNF and its receptor TrkB. Future studies can be directed to elucidate its mechanism of action in the context of dentinogenesis.

(3) The authors present immunofluorescence (IF) images against TrkB and pTrkB in the first figure. While they mention in the materials and methods section that these antibodies were generated for this study, there is no proof of their specificity. It should be noted that most commercial antibodies labeled as anti-TrkB recognize the extracellular domain of all TrkB isoforms. There are indications in the literature that pathological and excitotoxic conditions change the expression levels of TrkB-Fl and TrkB-T1. Therefore, it is necessary to demonstrate which isoform of TrkB the authors are showing as increased under their conditions. Similarly, it is essential to prove that the new anti-p-TrkB antibody is specific to this Trk receptor and, unlike other commercial antibodies, does not act as an anti-phospho-pan-Trk antibody.

Thank you for your kind concern.

Human TrkB has 7 isoforms and predicted Mw ranges from 35 to 93kDa. It has 11 potential N-glycosylation sites. The given antibody (isotype: Mouse IgG2a, κ) has been shown to interact with SHC1, PLCG1 and/or PLCG2, SH2B1 and SH2B2, NGFR, SH2D1A, SQSTM1 and KIDINS220, FRS2.

And, sorry for the misunderstanding and text mistake. We procured all the antibodies from the market using proven products, and didn’t check any specific isoform. We have mentioned the details of antibodies and reagents in the chemicals section of the methodology.

(4) I believe this initial conclusion could be significantly strengthened, without opening up other interpretations of the results, by demonstrating the specificity of the antibodies via Western blot (WB), both in the presence and absence of BDNF and other neurotrophins, NGF, and NT-3. Additionally, using WB could help reinforce the quantification of fluorescence intensity presented by the authors in Figure 1. It's worth noting that the authors fixed the cells with 4% PFA for 2 hours, which can significantly increase cellular autofluorescence due to the extended fixation time, favoring PFA autofluorescence. They have not performed negative controls without primary antibodies to determine the level of autofluorescence and nonspecific background. Nor have they indicated optimizing the concentration of primary antibodies to find the optimal point where the signal is strong without a significant increase in background. The authors also do not mention using reference markers to normalize specific fluorescence or indicating that they normalized fluorescence intensity against a standard control, which can indeed be done using specific signal quantification techniques in immunocytochemistry with a slide graded in black-and-white intensity controls. From my experience, I recommend caution with interpretations from fluorescence quantification assays without considering the aforementioned controls.

Thank you for your insightful comments. We have now included a negative control image in the revised Figures. This control confirms that the observed fluorescence signal is specific and not due to autofluorescence or nonspecific background. In our lab, we have been using these antibodies and already optimized the concentration to use in certain cell types. Additionally, we followed the manufacturer’s recommended antibody concentration and protocol throughout our experiments to ensure an optimal signal-to-noise ratio.

We agree that extended fixation with 4% PFA may increase autofluorescence; however, including negative controls helps account for this effect. We also ensured consistent imaging parameters and applied the same exposure settings across all samples to allow for a valid comparison of fluorescence intensity. We appreciate your emphasis on careful quantification and have clarified these methodological details in the revised Methods section.

(5) In Figure 2, the authors determine the expression levels of TrkA and TrkB using qPCR. Although they specify the primers used for GAPDH as a control in materials and methods, they do not indicate which primers they used to detect TrkA and TrkB transcripts, which is essential for determining which isoform of these receptors they are detecting under different stimulations. Similarly, I recommend following the MIQE guidelines (Minimum Information for Publication of Quantitative Real-Time PCR experiments), so they should indicate the amplification efficiency of their primers, the use of negative and positive controls to validate both the primer concentration used, and the reaction, the use of several stable reference genes, not just one.

We appreciate the reviewer’s suggestion regarding the specificity of primers and the amplification efficiency. In response, we have now included the primer sequences used for detecting TrkA and TrkB transcripts in the revised Materials and Methods section (Quantitative real-time PCR analysis of odontogenic differentiation marker gene expression in dental pulp stem cells). This ensures clarity on which isoforms of these receptors were assessed under different conditions. We also acknowledge the importance of following MIQE guidelines, and we got the primer provided by Integrated DNA Technologies with standard desalting purification and guaranteed yield.

(6) Moreover, the authors claim they are using the same amounts of cDNA for qPCRs since they have quantified the amounts using a Nanodrop. Given that dNTPs are used during cDNA synthesis, and high levels remain after cDNA synthesis from mRNA, it is not possible to accurately measure cDNA levels without first cleaning it from the residual dNTPs. Therefore, I recommend that the authors clarify this point to determine how they actually performed the qPCRs. I also recommend using two other reference genes like 18S and TATA Binding Protein alongside GAPDH, calculating the geometric mean of the three to correctly apply the 2^-ΔΔCt formula.

Thank you for your kind concern. We agree that residual dNTPs from cDNA synthesis could impact the accuracy of cDNA quantification. To address this, we have used the commercially available and guaranteed kit. The kit used is mentioned in Materials and Methods. We will definitely consider using 18S and TATA Binding Protein alongside GAPDH in our future studies. For now, we request you consider the results generated against GAPDH control.

(7) Similarly, given that the newly generated antibodies have not been validated, I recommend introducing appropriate controls for the validation of in-cell Western assays.

We apologize for the text mistake. Antibodies were procured commercially and not generated. We have corrected the sentence.

(8) The authors' conclusion that TrkB levels are minimal (Figure 2E) raises questions about what they are actually detecting in the previous experiments might not be the TrkB-Fl form. Therefore, it is essential to demonstrate beyond any doubt that both the antibodies used to detect TrkB and the primers used for qPCR are correct, and in the latter case, specify at which cycle (Ct) the basal detection of TrkB transcripts occurs. Treatment with TNF-alpha for 14 days could lead to increased cell proliferation or differentiation, potentially increasing overall TrkB transcript levels due to the number of cells in culture, not necessarily an increase in TrkB transcripts per cell.

Thank you for your comments. We appreciate your kind concerns. Here, we are trying to demonstrate that TrkB gets activated in inflammatory conditions. We have also provided the details on primers and antibodies. We have used commercial antibodies and qPCR primers, and they have been extensively validated with previous publications. The efficiency and validation of qPCR primers were provided by a company.

Moreover, we used the minimal concentration of TNF-alpha twice a week, and before using it, we did preliminary experiments to determine whether it affected any experimental condition.

(9) Overall, there are reasonable doubts about whether the authors are actually detecting TrkB in the first three images, as well as the phosphorylation levels and localization of this receptor in the cells. For example, in Figure 3 A to J, it is not clear where TrkB is expressed, necessitating better resolution images and a magnified image to show in which cellular structure TrkB is expressed.

Thank you for your comment. Here, we aimed to show the expression of TrkB receptors in inflamed/infected pulp, especially in minority-distributed DPSCs. TrkB is present on the cell membrane and perinuclear region. We have provided a single-cell (magnified) image in the figure for better clarification.

(10) In Figure 4, the authors indicate they have generated cells overexpressing BDNF after recombination using CRISPR technology. However, the WB they show in Figure 4B, performed under denaturing conditions, displays a band at approximately 28kDa. This WB is absolutely incorrect with all published data on BDNF detection via this technique. I believe the authors should demonstrate BDNF presence by showing a WB with appropriate controls and BDNF appearing at 14kDa to assume they are indeed detecting BDNF and that the cells are producing and secreting it. What antibodies have been used by the authors to detect BDNF? Have the authors validated it? There are some studies reporting the lack of specificity of certain commercial BDNF antibodies, therefore it is necessary to show that the authors are convincingly detecting BDNF.

Dear reviewer, thank you for your kind concern. Firstly, we apologize for the inconvenience.

Pro-BDNF is produced as a 32-kDa precursor that undergoes N-glycosylation and glycosulfation on residues located within the pro-domain of the precursor. N-terminal cleavage of the precursor generates mature BDNF and a minor truncated form of the precursor (28 kDa) that arises by a different processing mechanism than mature BDNF. The precursor undergoes N-terminal cleavage within the trans-Golgi network and/or immature secretory vesicles to generate mature BDNF (14 kDa).

We checked our data and band size, and it shows a mistake in recognizing ladder size. It is actually a 14kDa band which has been shown. The labeling has been revised in the figure, and the actual blot with a ladder has been shown below for clarification. Similarly, our data focused on the fact that the observed cellular effects are more consistent with BDNF/TrkB-mediated pathways, which are known to promote survival and differentiation.

(11) While the RNA sequencing data indicate changes in gene expression in cells treated with TNFalpha+CTX-B compared to control, the authors do not show a direct relationship between these genetic modifications with the rest of their manuscript's argument. I believe the results from these RNA sequencing assays should be put into the context of BDNF and TrkB, indicating which genes in this signaling pathway are or are not regulated, and their importance in this context.

Thank you for your concern. In order to identify the underlying mechanism, we ought to locate certain transcriptional factors interacting with the TrkB/BDNF signaling, leading to differentiation and dentinogenesis. Therefore, we treated it with a TrkB inhibitor.

Earlier, we showed only the combined results and mentioned the interaction between TNFα and TrkB. We have included the results from TNFα alone and combined them with CTX-B for better comparison (Please refer to Figure 8). Figure 8C1 clearly shows the reversal of certain factors with the treatment of TrkB inhibitor compared to figure 8C with TNFα alone treated group. We agree that the precise role of CTX-B in modulating TrkB signaling requires further clarification. We have now included this point in the revised discussion while working on this aspect. In a parallel study, we are trying to dig deep, especially the TCF family, as they have been documented to interact indirectly with BDNF and TrkB.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

Some minor textual issues

Line 120: It is obvious that TNFα stimulation caused significant phosphorylation of TrkB (p < 0.01) compared to TrkA (p < 0.05).

Thank you for noticing the typo. The sentence has been corrected.

The authors should consider rewording this sentence - I do not understand the intended meaning.

Line 126: pronounced peak at 10 ng/mL. I am not convinced there is a peak. Looks like a plateau to me. To call it a peak one would have to show that the values at 10 ng/ml and 20 ng/ml are statistically different.

We meant here the peak compared to 0.1 and 1ng/mL concentration and not compared to 20 ng/mL. The sentence has been elaborated accordingly.

Reviewer #3 (Recommendations for the authors):

The authors should show how they have validated the specificity of all the used antibodies as well as the efficiency and specificity of their qPCR data.

We procured the commercially available antibodies (all of them have been extensively validated with previous publications) and also performed negative controls (provided in revised figures). We frequently used Western blot and validate it with band size. Primer sequences are also provided in the revised manuscript. We checked its specificity with R² of Standard Curve ≥ 0.98 and the single peak of melting curves. We edited accordingly in line 263.

Once again, we thank all of you for your efforts in evaluating our study. It really helped us improve the quality of the manuscript. We hope all the queries have been answered and the revised manuscript is acceptable.