Author response:

We thank the reviewers for their thorough and constructive evaluation of our manuscript titled “PSD-95 drives binocular vision maturation critical for predation”. The reviewers raised several important conceptual and technical points. Here, we address and provide additional context on the major themes and outline our planned revisions.

We acknowledge that the current prey capture task cannot directly adjudicate between PSD-95 binocular vision impairments or sensorimotor processing deficits. However, we did not observe any major impairment supporting a sensorimotor processing deficit, in contrast to a major impairment in line with binocular vision impairment. Evidence from Huang et al. (2015) [1], Favaro et al. (2018) [2] and our data with the visual water task (VWT) — thus requiring identical sensorimotor but differential visual processing—clearly demonstrated intact visual acuity but impaired orientation discrimination in PSD-95 KO mice. Therefore, we believe that a binocular integration deficit is the most likely explanation of PSD-95 KO binocular impairments. In line with this, it is unlikely that aberrations in binocular eye movements account for the observations. We appreciate that alternative explanations remain possible and merit explicit discussion. Accordingly, we intend to expand the discussion of these alternatives.

Importantly, we will provide additional experimental data demonstrating that knock-down of PSD-95 in V1 but not in superior colliculus, significantly decreases orientation discrimination analyzed with the VWT, as we had shown for PSD-95 KO mice (while control knock-down does not have this effect). We believe that this new evidence better delineates the potential neuroanatomical locus of the PSD-95-associated deficits.

Furthermore, we will provide additional head movement analyses, as suggested by Reviewer 1. Specifically, we will investigate the head angle in relation to the cricket (azimuth) in time (±1 second) around prey contact under light and dark conditions.

We will also address the potential impact of PSD-95 KO learning deficits. We agree that there are more impairments in the PSD-95 KO brain, as has been published previously. But strikingly, the binocular impairment was dominating the sensory processing. This cannot be convincingly explained by learning deficits. In fact, we have observed improved learning of PSD-95 KO mice with some tasks (e.g. cocaine conditioned place preference) [3], but no significant differences in the VWT [1,2]. Learning differences were described for another PSD-95 mouse line, expressing the N-terminus with two PDZ domains [4]. To avoid potential learning dependent confounds, we have chosen salient stimuli, like water aversion, and prey capture to reduce impacts of potential learning defects.

We agree on the strength of the random dot stereograms to isolate stereoscopic computations. However, it requires special filters in front of either eye, which renders it unsuitable for the VWT. The lengthy training with less silent stimuli of water reward, could potentially add additional confounds of PSD-95 KO deficits. Thus, we think that this would be something for future experiments to allow for integration of different visual inputs. However, the combined improved performance of WT mice with binocular vision for prey capture (depth percept) and orientation discrimination (summation) is already supporting the importance of binocular vision in mice and the dominant defect in PSD-95 KO mice.

Finally, we will address the other points raised by the reviewers through clearer exposition and reorganization of the manuscript.

Once again, we would like to thank the reviewers for their thoughtful and constructive feedback, which we believe will substantially strengthen the manuscript.

(1) Huang, X., Stodieck, S. K., Goetze, B., Cui, L., Wong, M. H., Wenzel, C., Hosang, L., Dong, Y., Löwel, S., and Schlüter, O. M. (2015). Progressive maturation of silent synapses governs the duration of a critical period. Proc. Natl. Acad. Sci. 112, E3131–E3140. https://doi.org/10.1073/pnas.1506488112.

(2) Favaro, P.D., Huang, X., Hosang, L., Stodieck, S., Cui, L., Liu, Y., Engelhardt, K.-A., Schmitz, F., Dong, Y., Löwel, S., et al. (2018). An opposing function of paralogs in balancing developmental synapse maturation. PLOS Biol. 16, e2006838. https://doi.org/10.1371/journal.pbio.2006838.

(3) Shukla, A., Beroun, A., Panopoulou, M., Neumann, P.A., Grant, S.G., Olive, M.F., Dong, Y., and Schlüter, O.M. (2017). Calcium‐permeable AMPA receptors and silent synapses in cocaine‐conditioned place preference. EMBO J. 36, 458–474. https://doi.org/10.15252/embj.201695465.

(4) Migaud, M., Charlesworth, P., Dempster, M., Webster, L.C., Watabe, A.M., Makhinson, M., He, Y., Ramsay, M.F., Morris, R.G.M., Morrison, J.H., et al. (1998). Enhanced long-term potentiation and impaired learning in mice with mutant postsynaptic density-95 protein. Nature 396, 433–439. https://doi.org/10.1038/24790.

eLife Assessment

In this valuable study, the authors present traces of bone modification on ~1.8 million-year-old proboscidean remains from Tanzania, which they infer to be the earliest evidence for stone-tool-assisted megafaunal consumption by hominins. Challenging published claims, the authors argue that persistent megafaunal exploitation roughly coincided with the earliest Acheulean tools. Notwithstanding the rich descriptive and spatial data, the behavioral inferences about hominin agency rely on traces (such as bone fracture patterns and spatial overlap) that are not unequivocal; the evidence presented to support the inferences thus remains incomplete. Given the implications of the timing and extent of hominin consumption of nutritious and energy-dense food resources, as well as of bone toolmaking, the findings of this study will be of interest to paleoanthropologists and other evolutionary biologists.

Reviewer #2 (Public review):

The revised manuscript does a good job of using less definitive language, particularly by adding "possible" qualifiers to several interpretations. This addresses the concern about overstatement.

The main issue raised in the original review, however, remains unresolved. Only two elephant bone specimens at EAK show green-bone breakage interpreted as anthropogenic, and the diagnostic basis for that interpretation is not demonstrated clearly on the EAK material itself. The manuscript discusses a suite of fracture attributes described as diagnostic of dynamic percussive breakage, but these attributes are not explicitly documented on the EAK specimens. Instead, the diagnostic traits are illustrated using material from other Olduvai contexts, and that behavior is then extrapolated to make similar claims at EAK. For a paper making a potentially important behavioral argument, the key diagnostic evidence is not clearly demonstrated at the focal assemblage.

This problem is evident in the presentation of the EAK specimens. In their response, the authors state that one EAK specimen shows "overlapping scars" and constitutes a "long bone flake"; however, these features are not clearly identifiable in the figures or captions as currently presented. The authors state that Figures S21-S23 clearly indicate human agency, including a long bone flake with overlapping scars and a view of the medullary surface, but it is unclear which specimens or surfaces these descriptions refer to. Figure S21 does appear to show green fracture and is described only as an "elephant-sized flat bone fragment with green-bone curvilinear break." Figure S22 shows the same bone and cortical surface in a different orientation, providing no additional information. In Figure S23, I cannot clearly identify a medullary surface or evidence of green-bone fracture from this image. None of these images clearly demonstrates overlapping scars, and the figures would be substantially improved by explicitly identifying the features described in the text. Even if both EAK specimens are accepted as green-broken, they do not demonstrate the co-occurrence of multiple diagnostic fracture traits such as multiple green breaks, large step fractures, hackle marks, and overlapping scars that the authors state is required to attribute dynamic percussive activity to hominins and address equifinality.

I appreciate that the authors are careful to state that spatial association between stone tools and fossils alone does not demonstrate hominin behavior, and that they treat the spatial analyses as supportive rather than decisive. While the association is intriguing, the problem is downstream: spatial association is used to strengthen an interpretation of butchery at EAK that still depends on fracture evidence that is not clearly documented at the assemblage level.

The critique concerning Nyayanga is not addressed in the revision. The manuscript proposes alternative explanations for the Nyayanga material but does not demonstrate why these are more plausible than the interpretation advanced by Plummer et al. (2023). I am not arguing that the Nyayanga material should be accepted as butchery; rather, showing that trampling is possible does not establish it as more probable than cut marks. In contrast, the EAK material is treated as evidence of butchery on the basis of evidence that, in my opinion, is more limited and less clearly demonstrated. Even if this is not the authors' intention, the uneven treatment removes an earlier megafaunal case from the comparison and strengthens the case for interpreting EAK as marking a behavioral shift toward megafaunal butchery by excluding other early cases.

While I remain concerned about how the EAK evidence is documented and interpreted, I think the manuscript is appropriate for publication and will generate useful discussion. Readers can then assess for themselves whether the available evidence supports the strength of the behavioral claims.

Author response:

The following is the authors’ response to the previous reviews

Reviewer #1 (Public review):

I am happy with the revisions the authors made, and believe that the manuscript is now stronger, representing an important contribution.

We are truly thankful to this reviewer for the very constructive comments

Reviewer #2 (Public review):

In their response, the authors state that they do not treat the EAK evidence as decisive, yet the manuscript repeatedly characterizes the assemblage in very definitive terms. For example, EAK is described as "the oldest unambiguous proboscidean butchery site at Olduvai" and as "the oldest secure proboscidean butchery evidence." These phrases communicate a high level of confidence that does not align with the more qualified position articulated in the rebuttal and extends beyond what the documented evidence securely supports.

We decided to sound less dogmatic and remove the emphasis by adding “potentially” the oldest…. We emphasize that even if we had documented cut marks, we would be on the same epistemological ground, since there is no 100% certainty that the marks identified as cut marks could be cut marks.

I appreciate the authors' clarification regarding the fracture features, and I agree that these are well-established outcomes of dynamic hammerstone percussion. At the same time, several of these traits have been documented in non-anthropogenic contexts, including helicoidal spiral fractures resulting from trampling and carnivore activity (Haynes 1983), adjacent or flake-like scars created by carnivore gnawing (Villa and Bartram 1996), hackled break surfaces produced by heavy passive breakage such as trampling or sediment pressure (Haynes 1983), and impact-related bone flakes observed in carnivore-modified assemblages (Coil et al. 2020).

We added this explanation to the final version of the article:

“This interpretation is epistemologically problematic because it does not satisfy the fundamental criteria for valid analogy as outlined by Bunge (1981), namely substantial, structural, and environmental affinity. Specifically, the cited examples involve agents, materials, and contexts that differ markedly in composition, mechanical properties, and loading regimes from those considered here. Experimental and actualistic studies demonstrate that carnivores—rather than trampling—are also capable of producing spiral fractures and overlapping bone scarring, but these observations are restricted to faunal remains of substantially smaller body size than elephants, which they can gnaw (Haynes 1983; see also Figures S30–S36). To date, no carnivore has been documented as producing comparable fracture morphologies or surface damage on elephant bones. Consequently, the proposed analogy is not supported. Moreover, Haynes (1983) provides no empirical evidence that sediment pressure or trampling can generate hackled fracture surfaces. Such features are instead associated with dynamic loading conditions, whereas passive breakage processes have not been shown to produce these types of modifications. This reasoning also applies to impact flakes on elephant bones, which can only be produced by the sole modern agent documented to dynamically fracture green proboscidean long bones: humans.”

One of the biggest issues is that there is no quantitative data or images of the bone fracture features that the authors refer to as the main diagnostic criteria at EAK. The only figures that show EAK specimens (S21, S22, S23) illustrate general green-bone fracture morphology but none of the specific traits listed in the text. In contrast, clear examples of similar features come from other Olduvai assemblages, which may be misleading to readers if they mistakenly interpret those as images from EAK. The manuscript also states that these traits "co-occur," but it is not defined whether this refers to multiple features on the same fragment or within the broader assemblage. Without images or counts that document these traits on EAK fossils, readers cannot evaluate the strength of the interpretation. Including that information would substantially strengthen the manuscript.

The arguments were addressed in the general criteria criticized by the reviewer in his/her previous review encompassing all green broken elephant bones documented. If we restrict the arguments now to EAK, then suffice to rescue the arguments from the previous reply. Images (Figs S21-23) show the EAK broken specimens and clearly indicate their human agency by two factors: a) at least one of them is a long bone flake with overlapping scars (FS 23 is showing its medullary side), and b) elephant bones impacted by carnivores (namely, hyenas) have always shown intensive gnawing and tooth-marking; lack thereof in both EAK specimens refutes a non-human carnivore agency. The former argument is interpreted as human agency because carnivores have not documented to produce such features on elephant bones.

Regarding the statement that "natural elephant long limb breaks have been documented only in pre or peri-mortem stages when an elephant breaks a leg, and only in femora (Haynes et al., 2021)," it is not entirely clear what this example is intended to illustrate in relation to the EAK assemblage. My understanding is that the authors are suggesting that naturally produced green bone fractures in elephants are very limited, perhaps occurring only in pre or peri-mortem broken leg cases, and that fractures on other elements should therefore be attributed to hominin activity. If that is not the intended argument, I would encourage clarifying this point. This appears to conflate pre-mortem injury with the broader issue of equifinality. My original comment was not referring to pre-mortem breaks but to the range of natural (i.e., non-hominin) and post-mortem processes that can generate spiral or green bone fractures similar to those described by the authors.

We elaborated such argument addressing exclusively the reviewer´s previous argument that natural limb breaking produced spiral breaks on elephant long bones, which is correctly, as Haynes describes it, the only way not involving human agency that can generate a helicoidal spiral fracture on an elephant long bone. Non-human post-mortem processes on fresh bone do not generate these features. Neither have extant carnivores documented to produce these features on elephant bones.

Finally, in considering the authors' response on the Nyayanga material, I still find the basis for their dismissal of that evidence difficult to follow and the contrasting treatment of the Nyayanga and EAK evidence raises concerns about interpretive consistency. Plummer et al. (2023) specify that bone surface modifications were examined using low-power magnification (10×-40×) and strong light sources to identify modifications and that they attributed agency (e.g., hominin, carnivore) to modifications only after excluding possible alternatives. The rebuttal does not engage with the procedures reported. The existence of newer analytical techniques does not diminish the validity of long-standing methods that have been applied across many studies. It is also unclear why abrasion is presented as a more likely explanation than stone tool cutmarks. The authors dismiss the Nyayanga images as "blurry," but this is irrelevant to the interpretation, since the analysis was based on the fossils, not the photographs. The Nyayanga dataset is dismissed without a thorough engagement, while the EAK material, despite similar uncertainties and potential for alternative explanations, is treated as definitive.

We believe the rebuttal engages with these arguments. The protocol “bone surface modifications were examined using low-power magnification (10×-40×) and strong light sources to identify modifications and that they attributed agency (e.g., hominin, carnivore) to modifications only after excluding possible alternatives” does not guarantee that any derived interpretation is correct. These methods have consistently been used for decades now in contexts in which different researchers draw different conclusions on the same marks. The underlying variables used are subjectively interpreted and tallied, and equifinal when not considering overlapping factors, such as sediment abrasion and trampling. As an example, the same marks on the Nyayanga hippo bones interpreted by the original authors as cut marks, we see them undifferentiable from trampling marks from the image evidence published.

It is clear in the final version of our article that the EAK evidence is not treated as definitive, since that would be dogmatic, and thus, non-scientific. We thank this reviewer for having given us the chance to reconsider our original phrasing.

eLife Assessment

This is an important study that identifies the developmental time window during which re-expression of TCF4 mutated in Pitt-Hopkins syndrome, can rescue phenotypic features of brain function in a TCF4 knockout mouse. The study presents compelling data using a viral transgenic intersection approach to show that TCF4 expression is required early in perinatal life. These findings have implications for the timing of possible gene therapy in people with Pitt-Hopkins-associated TCF4 mutations.

Reviewer #1 (Public review):

Summary:

This manuscript follows up previous work from this group using a conditional TCF4 mouse where Cre-expression turns "on" expression of TCF4 to investigate whether postnatal re-expression of TCF4 is effective to correct phenotypes related to Pitt-Hopkins Syndrome (PTHS) in humans. Results may inform gene therapy human PTHS gene therapy efforts on effective developmental windows for gene therapy. The authors demonstrate that re-expression of TCF4, induced by retro-orbital (RO) AAV-PHP.eB-Cre, during 2-4th postnatal week, does not rescue brain or body weight, anxiety-like or nest-building behaviors, but rescues an object location memory task, a measure of cognition. These results are novel and interesting in that they reveal distinct developmental roles for TCF4 in distinct behaviors and suggest that TCF4 plays a role in the mature brain in hippocampal and memory-related plasticity. Results may inform gene therapy design in PTHS.

Strengths:

The results are rigorous and high quality. Multiple methods are used to assess AAV-mediated re-expression of Cre, reactivation of TCF4, and the developmental time course of expression. Multiple behavioral phenotypes and molecular rescue are assessed. Most behavioral phenotypes are reproducible and robust, and it is clear whether a rescue was observed.

Weaknesses:

(1) Although the authors demonstrate the time course and spatial extent of Cre and a Cre-reporter (TdTom) in the brain with the AAV-Cre, it is unclear how many cells are transduced. Similarly, the authors do not measure TCF4 levels with immunohistochemistry or western blot. So the level of protein reactivation is unknown. A possible reason the rescue is incomplete is that the TCF4 protein is not induced in a large % of neurons in specific brain regions that mediate specific behaviors, such as the hippocampus vs. the striatum.

(2) The authors perform bulk qPCR to demonstrate a 20% increase in TCF4 RNA with Cre-mediated activation. It is unclear why the full gene reactivation is not observed. An alternative interpretation of the incomplete rescue of the phenotypes is that full TCF4 expression is required at later developmental time points.

Reviewer #2 (Public review):

Summary:

The basic helix-loop-helix transcription factor TCF4 (also known, as ITF2, SEF2, or E2-2) is a protein involved in the development and functioning of many different cell types. TCF4 plays important roles in the nervous system, both in health and disease. Its importance in the nervous system is underlined by its association with common and rare cognitive disorders. Specifically, variants of the TCF4 gene are implicated in increased susceptibility to schizophrenia, and mutations in the TCF4 gene cause Pitt-Hopkins syndrome (PTHS) or mild to moderate non-syndromic intellectual disability.

In this manuscript, the authors have studied whether reinstating TCF4 later in postnatal development in juvenile PTHS model mice could reverse behavioral phenotypes, thereby simulating gene therapy. Previous research by the same group has demonstrated that restoring TCF4 during embryonic or neonatal stages, corresponding to prenatal or neonatal periods in humans, improved phenotypes in a PTHS mouse model. In the current study, a conditional TCF4 reinstatement mouse model, Tcf4-lox-stop-lox (Tcf4-LSL), previously developed and characterized by their lab, where Cre-mediated recombination removes a floxed transcriptional stop cassette downstream of exon 17, leading to reinstatement of all TCF4 isoforms at appropriate levels in neurons, was used. The study showed that this later intervention failed to correct most phenotypes, suggesting that perinatal reinstatement of TCF4 holds the greatest potential to treat behavioral symptoms of PTHS. However, the study also suggests that some cognitive behaviors may still be responsive to TCF4 reinstatement later in life.

Strengths:

This is a very important study aimed at developing gene therapy for PTHS. The study is technically very well performed and written.

Weaknesses:

The only weakness is that a human disease is modelled in a mouse, which is evolutionarily not the closest mammal to humans. Hopefully, in the future, similar studies will also be performed in a nonhuman primate model, for example rhesus macaque.

eLife Assessment

This study proposes a cross-species transcriptomic framework to predict vaccine reactogenicity, with implications for preclinical vaccine safety assessment. The findings show that mouse muscle transcriptomic signatures capture conserved inflammatory programs and can identify highly reactogenic formulations, with supportive but limited evidence for finer discrimination among licensed human vaccines. Overall, the work establishes a valuable foundation for translational biomarkers of reactogenicity, although the strength of evidence for broad cross-species predictive performance remains incomplete and would benefit from further validation.

Reviewer #1 (Public review):

Summary:

The authors aimed to develop a translational framework for predicting vaccine reactogenicity by training a penalized ordinal regression model on mouse muscle transcriptomics and applying it across tissues and species to rank human vaccines by their inflammatory potential.

Strengths:

The study addresses an important gap in preclinical vaccine safety assessment. The identification of IL6/JAK/STAT3 signaling as a key pathway implicated in reactogenicity is biologically plausible, and the observation of coordinated changes between muscle and blood compartments supports the biological relevance of the signature. The model achieves near-perfect classification in mouse muscle tissue and successfully identifies Fluad (MF59-adjuvanted) as the most reactogenic among licensed human vaccines, consistent with clinical safety data.

Weaknesses:

The methodological foundation has several concerns. The reactogenicity class definitions rely on PC1 scores with modest variance explained, yet no sensitivity analyses demonstrate robustness to different normalization strategies, feature selection approaches, or dimensionality reduction methods. I suggest performing sensitivity analyses demonstrating that reactogenicity class definitions are robust to alternative normalization methods, feature selection criteria, and dimensionality reduction approaches.

The combined mouse analysis reveals that tissue effects dominate over vaccine-induced variation, and no explicit batch or compartment correction was reported. The authors can apply batch/compartment correction (e.g., SVA) when analyzing combined mouse muscle and blood data, then recompute PCA and downstream analyses.

The central claim regarding cross-species ranking capability is not fully supported. In human blood, the model largely distinguishes Fluad from other vaccines but shows limited separation among non-Fluad formulations, with many pairwise comparisons yielding non-significant adjusted p-values. This pattern suggests the model may be tuned to detect large inflammatory magnitudes-likely a consequence of training on extreme stimuli such as LPS and whole-cell pertussis-rather than capturing the finer gradations relevant for distinguishing licensed vaccines with moderate reactogenicity profiles. I highly suggest retraining the model, excluding extreme stimuli (LPS, Pentavac), to evaluate whether mid-range separations among licensed vaccines can be recovered.

Impact:

While the conceptual framework is promising, the current evidence does not convincingly demonstrate that the model can rank vaccines beyond identifying highly inflammatory outliers. The utility for preclinical assessment of novel vaccine candidates with moderate reactogenicity profiles remains uncertain.

Reviewer #2 (Public review):

Summary:

The authors derived a time-specific signature of reactogenicity from mouse muscle following exposure to vaccines /TLRs for capturing the reactogenicity patterns. They tested this reactogenicity signature in mouse blood, and then they applied the reactogenicity signature to human blood from subjects having received different vaccines. They identified biomarkers in mouse muscle which are also observed in mouse and human blood and could be used as a reactogenicity signature in mice, instead of CRP.

Strengths:

(1) The authors used transcriptomic response following vaccination and used common genes to human and mice for defining a reactogenic signature.

(2) As the authors used different formulations in mice, the model was trained across a broad reactogenicity spectrum, which has the advantage of being used for evaluating new vaccines/vaccine platforms.

Weaknesses:

(1) The muscle gene signature reflects local reactogenicity. Systemic reactogenicity is not specifically addressed, except where overlapping gene signatures are observed for both local and systemic reactogenicity.

(2) In the same logic, could we find additional genes in the blood which are not captured in the muscle?

(3) The peak of the reactogenicity is usually 24h; it is not certain that additional TPs have helped the findings. If they have, the authors should explain.

eLife Assessment

The study presents a valuable finding on the ubiquitin-dependent regulatory loop in which proapoptotic Bim is targeted to the E3 ubiquitin ligase Cul5-Wsb2-mediated degradation through its sequestration by BCL2 proteins. The conclusions are supported by incomplete evidence and would benefit from additional experiments addressing both the mechanistic understanding and the physiological/cancer-related significance of the study.

Reviewer #1 (Public review):

This manuscript by Toczyski and colleagues explores the role of ubiquitin-dependent degradation in the co-regulation between pro- and anti-apoptotic proteins. The binding of the pro-apoptotic sensor Bim to BCL2 anti-apoptotic proteins sequesters it into inactive complexes, inhibiting BCL2 members but also preventing Bim from activating the apoptotic executors BAX and BAK. The authors now suggest that the E3 ubiquitin ligase Cul5-Wsb2 targets Bim turnover while in complex with BLC2 members. The authors reveal the importance of WSB2 in apoptosis of neuroblastoma cell lines, highlighting the importance of Wsb2 as a cancer biomarker. In sum, this study identifies Bim as a novel Wsb2 target and suggests a novel co-receptor mechanism using BCL-2 members as bridging factors, thus adding a novel mechanistic layer to the apoptosis repressor role of Wsb2. Their experimental approach is sound, and in most cases, the conclusions are justified. However, whether Cul5-Wsb2 targets Bim via BLC2 anti-apoptotic members would require further analysis.

Major comments:

(1) They find that Wsb2 or Cul5 downregulation increases the levels of Puma and Bim isoforms, and that Wsb2 strongly interacts with all Bim isoforms. Moreover, Wsb2 regulates Bim turnover, especially visible for Bim-EL, and controls Bim-L ubiquitylation. Finally, Figure 2E suggests that Wsb2-Bim interaction is bridged by Bcl-xL, and they identify the domain in Bcl-xL/Wsb2 responsible for their binding in Figure 4A-E. However, Figure 4F shows only a mild decrease between Bim-EL and HA-Wsb2EEE, which is inconsistent with their model. This important gap should be backed up by further experimental evidence. For example, by performing (a) coIP studies between Bim and Wsb2 in the presence of Bcl-xlAAA and (b) Bim stability and ubiquitylation analysis in the presence of either Bcl-xlAAA or Wsb2EEE.

(2) The manuscript lacks quantifications and statistical analysis in most figures, which are particularly important for Figure 1D - especially regarding the upregulation of Puma and Bim isoforms upon downregulation of Cul5 and Wsb2, for Fig 3A - also including statistical analyses of Bim1 stability in presence or absence of proteasomal inhibitors, and for Figure 4D, F, especially regarding the interaction of Bim-EL- with WT and mutant Bcl-xL in 4D and with WT and mutant Wsb2 in 4F.

(3) The localization of BCL2 family members at the mitochondrial outer membrane is a crucial step in the implementation of apoptosis, and BCL2 members recruit Bim to the OM. Despite their finding suggesting that Bim insertion into the OM might be dispensable for interaction with Bim, the interaction was abolished by BH3-mimetics that disrupt Bcl-xL interaction with BIM. This suggests that Wsb2 interacts with Bim at the mitochondrial surface. Therefore, it would be interesting to investigate the sub-cellular localization Bim and WSB2 with and without ABT-263.

(4) Wsb2 mildly interacts with Bcl-xL and with Mcl1, but does not interact with Bcl-w or Bcl2. However, they show that Wsb2 recognizes Bcl-xl through a motif conserved between Bcl-xl, Bcl-w and Bcl2. Therefore, it would be helpful to precipitate Bcl-w or Bcl2 and check interaction with Wsb2.

Reviewer #2 (Public review):

Summary

This manuscript proposes an original and conceptually interesting model in which anti-apoptotic BCL-2 family proteins, particularly BCL-XL and MCL-1, not only sequester BIM but also act as adaptor "co-receptors" that recruit BIM to the CUL5-WSB2 ubiquitin ligase complex for degradation. The authors present a mechanistic framework supported by structure-guided mutagenesis, BH3 mimetic perturbations and co-immunoprecipitation assays performed in RPE1 cells. In parallel, the study shows that neuroblastoma cell lines are highly dependent on WSB2 for survival. These observations give the work both conceptual and translational relevance.

Strengths

The principal strength of the study lies in its conceptual novelty. Reframing BCL-XL and MCL-1 not only as sequestration factors but also as adaptors that facilitate substrate engagement by an E3 ligase substantially extends current models of apoptotic regulation. The mechanistic narrative developed in RPE1 cells is clear and internally consistent: the combination of AlphaFold-guided motif identification with complementary mutagenesis provides a persuasive framework for how WSB2 associates with anti-apoptotic BCL-2 family members and promotes BIM turnover. The definition of a BCL-XL/MCL-1 co-receptor mechanism for WSB2-mediated BIM degradation is therefore both intuitive and mechanistically appealing. In parallel, the authors present a distinct experimental series showing that neuroblastoma cells exhibit pronounced sensitivity to WSB2 loss, undergo apoptosis upon its depletion and display reduced competitiveness in mixed-culture assays. Although the mechanistic connection between these observations requires further clarification, the convergence of a well-defined biochemical model with a clear cancer-relevant phenotype enhances the potential biological significance of WSB2 and raises the possibility that its regulation may hold therapeutic relevance.

Weaknesses

There are several limitations that readers should consider when interpreting the study. The most fundamental issue is the disconnect between the mechanistic model established in RPE1 cells and the apoptotic phenotype observed in neuroblastoma. Although the manuscript convincingly demonstrates the WSB2-BCL-XL/MCL-1-BIM axis in RPE1 cells and independently shows that WSB2 loss compromises neuroblastoma viability, it does not examine whether BIM levels are elevated upon WSB2 depletion in neuroblastoma, nor whether apoptosis in these cells requires BIM. Without demonstrating WSB2-BCL-2-BIM complex formation or BIM dependence in the disease-relevant context, it remains unclear whether the co-receptor mechanism characterised in RPE1 cells explains the phenotype. This gap is compounded by the observation that PUMA, another potent pro-apoptotic factor, also increases following WSB2 loss, raising the possibility that multiple death pathways contribute to the outcome. The absence of a genetic rescue experiment, such as re-expression of an shRNA-resistant WSB2 restoring viability and suppressing apoptosis, further limits causal inference regarding WSB2's role in neuroblastoma.

Many central claims rely on single Western blots and pulldown assays without quantification or assessment of reproducibility. This complicates the interpretation of CHX chase experiments (where initial steady-state levels differ between samples) and limits confidence in BH3 mimetic experiments, which use a single concentration and short exposure time. Without dose-response curves, time-course analyses, caspase inhibition, or orthogonal genetic perturbation of BCL-XL or MCL-1, indirect or off-target drug effects cannot be excluded. Reduced co-IP signals in these assays could therefore reflect early apoptotic events or compound instability rather than specific disruption of protein-protein interactions.

A further limitation concerns the inference of a direct WSB2-BCL-XL interaction. The mutagenesis analyses are performed in lysates that contain endogenous or overexpressed BIM, and BH3 mimetics disrupt the WSB2 interaction only when the BCL-XL-BIM heterodimer is dismantled. The study thus cannot distinguish whether the mapped WSB2 motifs mediate direct contact with BCL-XL or whether they influence the architecture or stability of the BCL-XL-BIM complex. Because no purified protein reconstitution or biophysical binding assays are presented, the evidence for direct binding remains suggestive rather than conclusive.

The ubiquitination data also remain incomplete. Although the WSB2 mutation reduces the ubiquitin smear on BIM, the assay does not demonstrate dependence on CUL5, RBX2 or ARIH2, leaving open which ligase components are directly responsible. MLN4924 implicates CRLs more broadly, but the ubiquitination assay itself does not assign activity to the CUL5-WSB2 module.

Finally, several methodological details are insufficiently described, including the generation and validation of the doxycycline-inducible WSB2 and HA-WSB2 lines and the suitability of the WSB2-overexpressing control line used in immunoprecipitations.

Collectively, these issues do not undermine the conceptual interest of the proposed co-receptor model, but they do limit the strength of the mechanistic claims and weaken the connection between the defined mechanism and the neuroblastoma phenotype.

eLife Assessment

This important work substantially advances our understanding of the role of synaptotagmin-7 (Syt7) in short-term plasticity at cortical glutamatergic synapses. The evidence supporting the conclusions is convincing, with rigorous and elegant quantal-level iGluSnFR imaging and failure-based analyses at single boutons. The work will be of broad interest to synaptic physiologists and molecular biologists.

Reviewer #1 (Public review):

Kotzadimitriou et al. investigate how synaptotagmin-7 (syt7) contributes to short-term plasticity at cortical glutamatergic synapses. Using quantal-level iGluSnFR imaging and failure-based analyses at single boutons, the authors distinguish between synchronous and asynchronous glutamate release across boutons with differing baseline efficacy. They show that knocking out syt7 abolishes facilitation of synchronous release while leaving asynchronous facilitation largely intact, although reduced in magnitude. Furthermore, they argue that synchronous and asynchronous events arise from functionally distinct vesicle pools. The manuscript concludes that syt7 is essential for the facilitation of synchronous release, while other calcium sensors govern asynchronous release.

Strengths:

(1) The use of iGluSnFR provides a robust readout of single-synapse activity. Unlike traditional ephys methods that average the activity of thousands of synapses (which may mask the facilitation of low Pr synapses), the authors employ quantal imaging to analyze thousands of individual boutons and stratify them by efficacy. The representative images and traces in Figure 1 are of high quality, and the quantal analysis demonstrating multiple quantal peaks aligns well with previously published work (Mendonca et al., 2022; Wang et al., 2022).

(2) The failure-based analysis is thoughtfully implemented. By isolating trials in which no release occurred, the authors effectively separate facilitation from depletion, strengthening their central argument that syt7 is required for facilitation independent of vesicle depletion.

(3) The proposed model (depicted in Figure 7) is interesting and may reconcile the contradictory roles attributed to syt7, as described by others in the field. Specifically, the authors provide data to address syt7's potential function in facilitation, asynchronous release, and replenishment. However, to further support their model, which argues that "multiple Ca2+ sensors have both unique and overlapping roles in regulating synaptic plasticity," additional experiments are needed (see point 2 below).

Weaknesses:

(1) While the authors use cultures from syt7 knockout mice (and wild-type controls), there are no acute rescue experiments (e.g., syt7 viral transduction in KO cultures) or checks for compensatory changes in other proteins. Previous studies (Bacaj et al., 2013; Jackman et al., 2016) have utilized viral rescues to confirm specificity. Without such experiments, it remains theoretically possible that the chronic loss of syt7 leads to downregulation of another protein essential for facilitation. At a minimum, the authors should perform rescue experiments for at least some of their findings. Additionally, western blots for syt1 and syt7 should be conducted to confirm that their knockout is specific to syt7.

(2) The manuscript acknowledges the possible roles of Doc2a and syt3 but fails to address them experimentally. Recent work (Wu et al., 2024; Weingarten et al., 2024) has identified Doc2a as the primary sensor for asynchronous release. Even if its expression in cortical cultures remains unconfirmed (as claimed by the authors), they should, at the very least, perform Western blots for Doc2a and syt3 in both wild-type (to determine basal expression levels) and syt7 knockout cultures. Without analyzing the levels of these proteins, the mechanism/model behind the "remaining" asynchronous release remains speculative. Is it possible that these other calcium sensors are upregulated in their syt7 KO cultures and could instead explain their results?

Reviewer #2 (Public review):

Summary:

In this elegant study, the authors employ live iGluSnFR-based imaging of glutamate release from cortical boutons to dissect the distinct roles of the Ca²⁺ sensor synaptotagmin-7 (Syt7) in synaptic transmission. Although multiple functions have been attributed to Syt7 over the years, the field remains conflicted. The authors argue that one major obstacle for resolving some of these discrepancies lies in a fundamental limitation of electrophysiological recordings, which aggregate signals across all synapses to yield averaged readouts, dominated by strong, high-release-probability synapses. By using a live glutamate imaging approach combined with sensitive detection of action potential-evoked activity across different stimulation regimes, and a dedicated analysis pipeline, the authors confirm a role for Syt7 in facilitating synchronous release and in regulating the magnitude of asynchronous release. In contrast, they find no evidence that Syt7 contributes to the facilitation of asynchronous release, do not find evidence for a role for Syt7 in synaptic vesicle replenishment during AP trains, and provide evidence suggesting that the maintenance of facilitation by Syt7 may occur independently of vesicle depletion.

Strengths:

This study offers a fresh perspective on a debated issue, using a new experimental approach that the authors previously explored in the context of Synaptotagmin 1 (Mendonca et al. 2022). The authors record the response to a series of pair-pulse stimulations, followed by an AP train. By carefully quantifying individual events and by sorting events based on their efficacy, the authors extract quantitative information that they assign to different properties of synaptic function. They also devised an interesting approach for monitoring aspects of facilitation, in which they isolate PPR events where the first response did not elicit detectable release (thus regarding the release in response to the second AP as facilitating), and compare them with successful events. Together, the authors provide semi-quantitative descriptions of synchronous and asynchronous release during single, paired, and AP trains, yielding a weighted estimate of Syt7's contribution to distinct features of synaptic vesicle release that are independent of postsynaptic readouts. A major strength of the study is the confirmation of two principal proposed functions of Syt7: facilitation of synchronous release and regulation of the magnitude of asynchronous release.

Weaknesses:

The experimental approach presented here is elegant and well-executed. However, a principal limitation lies in translating electrophysiological terminology to imaging-based measurements. For instance, interpreting signals persisting beyond 10 ms as a proxy for asynchronous release relies on assumptions that would be good to experimentally justify. Could such signals arise from iGluSnFR saturation, or be affected by desensitization?. Moreover, the quantification of asynchronous release is based on very small signals that represent only a fraction of the already small synchronous release component, raising concerns about signal-to-noise limitations. A key issue is that failures to evoke glutamate release may arise from AP failures, such that the second response in a PPR does not necessarily represent facilitation. Given that many of the findings largely confirm existing literature, the study might have benefited from a different framing, for example, as an additional validation of the correspondence between electrophysiological measures and the authors' imaging-based readouts. Another point concerns the analysis of synaptic vesicle replenishment following depletion, which would ideally be addressed using alternative stimulation protocols, such as quantifying the response/success rate to single APs at varying time points after a train. Although the authors are appropriately cautious in their conclusions (e.g., with respect to Figure 5b), this limitation remains. Finally, the use of heterogeneous cortical neuronal cultures is likely to introduce substantial variability, as the authors themselves acknowledge, which may arise from the co-expression of multiple Ca²⁺ sensors across diverse cell types.

In summary, the authors were able to confirm previously-described changes in neurotransmission properties upon the loss of Syt7 using live imaging of glutamate release at the level of single boutons. They also present preliminary evidence for the interdependence of Syt7 function, synaptic vesicle replenishment, and the facilitation of asynchronous release, although these results will need to be substantiated in future studies using alternative stimulation protocols and complementary methodologies. Taken together with the group's prior work on synaptotagmin-1, this study illustrates that live imaging of glutamate release offers an alternative approach that recapitulates some elements detectable via electrophysiological analysis, while possibly revealing new insights into the function of synaptic proteins. As a whole, taking a live imaging approach may be a broadly accessible way forward to analyze synaptic function. The potential of studying synaptic proteins in diverse cell types that are difficult to access with patch-clamp electrophysiology is particularly compelling.

Reviewer #3 (Public review):

In this manuscript, the authors examine the role of Syt7 in the plasticity of synchronous and asynchronous release in cultured neurons. The experimental approach is the imaging of SF-iGluSnFR.A184V expressed in cultured neurons while delivering stimulation through whole-cell patch clamping of single neurons in the culture. In this manner, they could examine the optical signature of glutamate release in single presynaptic terminals, while separating the release events into synchronous (<10ms) and asynchronous (>10ms) while delivering both paired pulses or trains of stimuli (at 20 Hz, 50 ms between stimuli).

This manuscript employs techniques previously reported by the research group in their Mendoca et al., Nat Comms 2022 paper. This paper uses this approach to specifically examine the role of Syt7. The use of iGluSnFR in this manner provides significant rigor to the paper. The most significant weakness is that some of the events the authors discuss in this manuscript are rare, and the strength of the conclusions regarding those is somewhat unclear.

The main novel contribution of this manuscript is that single-bouton high-frequency imaging allowed them to examine paired-pulse plasticity in boutons that had not released neurotransmitter during the first pulse (failure-based analysis), thus separating between the effects of vesicle depletion and facilitation of the release machinery. This approach also allowed them to segregate their observations according to bouton-specific release efficacy. Both examinations are unavailable when performing cell-level analysis of neurotransmitter release, as is reported by most electrophysiological approaches.

The authors conclude that Syt7 contributes specifically to facilitation of synchronous release, not asynchronous release, while reducing the magnitude of the asynchronous component. Finally, the authors suggest segregation of synchronous and asynchronous release, either by differential use of calcium sensors or spatial segregation of the vesicles contributing to both modes of release.

This report contributes significantly to the discussion of the control of synaptic plasticity by different molecular players. It is not the first to examine Syt7, but its contribution to the examination of this protein is significant.

I find this report to be well executed and reasoned. In my opinion, the authors could improve the manuscript by clarifying the description of several methodological and experimental sections. Furthermore, in my opinion, some of the conclusions are overstated.

The authors state: "Because boutons along a single axon originate from the same presynaptic neuron, they are expected to share broadly similar molecular components of the vesicular release machinery and experience comparable presynaptic action potential waveforms." The authors should address the idea that presynaptic terminals from the same neuron on different postsynaptic targets can differ in the molecular components, as well as in the presynaptic side. There is ample evidence for differences between synapses onto glutamatergic and GABAergic neurons of the same neuron.

The authors used 4ms-long frames, but the stimuli are delivered at 20Hz (50ms apart). Therefore, in paired pulse stimulation, isn't there going to be a difference between the first and second stimuli regarding the timing of the frames relative to the stimulus? Isn't the deconvolution sensitive to such an offset?

A 10ms threshold for defining synchronous vs. asynchronous release full in-between frames. Doesn't this increase the chance of assigning borderline events to the wrong category?

On page 11 of the conclusion, the authors state that "Our data indicate that in our conditions during paired-pulse protocol Syt7 primarily enhances release probability rather than increasing the RRP size." While I understand the reasoning behind this statement, it should be toned down. The authors did not directly address the RRP size.

In failure-based analysis, the number of failure events in high-efficiency boutons is expected to be low. How does this affect the conclusions of the authors concerning the effects of Syt7 deletion on facilitation in high-efficiency boutons?<br /> SourceData.xlsx was not available to me, as far as I could tell.

How can the conclusions of the authors on the differential molecular composition of vesicles contributing to synchronous and asynchronous release be related to the reported effect of strontium on the nature of release? (see 10.1523/JNEUROSCI.20-12-04414.2000)

Is this the first use of failure-based analysis? If not, the authors should cite precedents. In 10.1016/s0896-6273(00)80338-4, failure of release during the 1st AP was presented, with facilitation during the 2nd, although no formal analysis was performed.

eLife Assessment

This study presents useful findings on the molecular mechanisms driving female-to-male sex reversal in the ricefield eel (Monopterus albus) during aging, which would be of interest to biologists studying sex determination. The manuscript describes an interesting mechanism potentially underlying sex differentiation in M. albus. However, the current data are incomplete and would benefit from more rigorous experimental approaches.

Reviewer #1 (Public review):

Summary:

This preprint investigates the molecular mechanism by which warm temperature induces female-to-male sex reversal in the ricefield eel (Monopterus albus), a protogynous hermaphroditic fish of significant aquacultural value in China. The study identifies Trpv4 - a temperature-sensitive Ca²⁺ channel - as a putative thermosensor linking environmental temperature to sex determination. The authors propose that Trpv4 causes Ca²⁺ influx, leading to activation of Stat3 (pStat3). pStat3 then transcriptionally upregulates the histone demethylase Kdm6b (aka Jmjd3), leading to increased dmrt1 gene expression and ovo-testes development. This work aims to bridge ecological cues with molecular and epigenetic regulators of sex change and has potential implications for sex control in aquaculture.

Strengths:

(1) This study proposes the first mechanistic pathway linking thermal cues to natural sex reversal in adult ricefield eel, extending the temperature-dependent sex determination paradigm beyond embryonic reptiles and saltwater fish

(2) The findings could have applications for aquaculture, where skewed sex ratios apparently limit breeding efficiency

Weaknesses:

Although the revised manuscript represents an improvement over the original version, substantial weaknesses remain.

Scientific Concerns

(1) Western blot normalization and exposure: The loading controls (GAPDH) in Fig. S3C appear overexposed, as do several Foxl2 blots. Because these signals are likely outside the linear range, I am not convinced that normalization is reliable. This raises concerns about the validity of the quantified results.

(2) Antibody validation and referencing (Line 776): The authors need to refer explicitly to figures demonstrating antibody validation. At present, these data are provided only as a supplementary file that is not cited in the manuscript. In addition, the Sox9a antibody appears to yield indistinguishable signals in control and RNAi conditions, suggesting that it may not recognize eel Sox9a. This issue is not addressed by the authors. Furthermore, antibody validation Western blots should be quantified.

(3) Unclear sample sizes (N values): Sample sizes remain unclear for several figures:

(a) Fig. 3F - No N value is provided. Each graph shows three data points; does this indicate that only three samples were quantified? If ten samples were collected, why were all not quantified?

(b) Fig. 4 - No N values are reported.

(c) Fig. 5A - Again, only three data points are shown per group, despite the apparent availability of twelve samples. The rationale for this discrepancy is not explained.

(4) qRT-PCR normalization: The manuscript does not specify the reference gene(s) used for qRT-PCR normalization. Although expression levels are reported as "relative," neither the identity of the reference gene(s) nor the justification for their selection is provided.

(5) Specificity of key antibodies: While the authors have made some effort to validate anti-Amh, anti-Sox9, and anti-Dmrt antibodies, the results remain incomplete. The Amh and Dmrt antibodies detect reduced protein levels following knockdown of their respective targets, which is encouraging. However, the Sox9a antibody shows no difference between control and RNAi conditions, suggesting it does not recognize eel Sox9. This is not acknowledged in the manuscript. In addition, no validation data are presented for Foxl2. Antibody validation data must be clearly referenced in the main text and presented in an interpretable and quantitative manner.

(6) Immunofluorescence data quality: The immunofluorescence images remain difficult to interpret. I strongly encourage the authors to enlarge the image panels and to present monochrome images (white signal on black background). The current presentation severely limits interpretability.

(7) Unreferenced supplementary figure: Fig. S4 is included in the submission but is not referenced anywhere in the manuscript text.

(8) Fig. 5B image resolution: The micrographs in Fig. 5B are too small to allow meaningful evaluation of the data.

(9) Unexplained data inclusion (Fig. 5E): Fig. 5E includes a pERK blot that is not mentioned in the Results section. The rationale for including these data is unclear.

(10) Poor blot quality (Fig. S3C): The blots in Fig. S3C exhibit high background and overexposure. I am concerned about the reliability of the quantification shown in panel D.

(11) Poor blot quality (Fig. S5G): The Stat3 blots in Fig. S5G contain numerous white artifacts, raising concerns about their suitability for normalization in panel H.

(12) Missing controls (Fig. 6E): Fig. 6E lacks controls for HO-3867 and Colivelin treatments alone. Without these controls, it is not possible to determine whether the reported effects are meaningful.

(13) Graphical presentation: The use of a light blue-to-pink gradient in bar graphs throughout the manuscript does not aid interpretation. I recommend using more distinct colors (e.g., red, orange, green, blue, purple, gray, black) to improve clarity. In summary, the interpretation of the study remains limited by persistent issues related to data presentation, image quality, and reagent specificity.

Author response:

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

Public Reviews: 

Reviewer #1 (Public review): 

Summary:

This study investigates the molecular mechanism by which warm temperature induces female-to-male sex reversal in the ricefield eel (Monopterus albus), a protogynous hermaphroditic fish of significant aquacultural value in China. The study identifies Trpv4 - a temperature-sensitive Ca²⁺ channel - as a putative thermosensor linking environmental temperature to sex determination. The authors propose that Trpv4 causes Ca²⁺ influx, leading to activation of Stat3 (pStat3).pStat3 then transcriptionally upregulates the histone demethylase Kdm6b (aka Jmjd3), leading to increased dmrt1 gene expression and ovo-testes development. This work aims to bridge ecological cues with molecular and epigenetic regulators of sex change and has potential implications for sex control in aquaculture.

Strengths:

(1) This study proposes the first mechanistic pathway linking thermal cues to natural sex reversal in adult ricefield eel, extending the temperature-dependent sex determination paradigm beyond embryonic reptiles and saltwater fish.

(2) The findings could have applications for aquaculture, where skewed sex ratios apparently limit breeding efficiency.

We thank you for the encouraging comments of our work, and answering your questions has greatly improved the quality of the manuscript.

Weaknesses:

(A) Scientific Concerns:

(1) There is insufficient replication and data transparency. First, the qPCR data are presented as bar graphs without individual data points, making it impossible to assess variability or replication. Please show all individual data points and clarify n (sample size) per group. Second, the Western blotting is only shown as single replicates. If repeated 2-3 times as stated, quantification and normalization (e.g., pStat3/Stat3, GAPDH loading control) are essential. The full, uncropped blots should be included in the supplementary data.

We thank you for the critical comments. Now we have remade the bar graphs with individual data points, and added the sample size per group if possible. Quantification and/or normalization of the WB data based on at least two replicates were included. The representative uncropped blots have also been loaded as the supplementary data.

(2) The biological significance of the results is not clear. Many reported fold changes (e.g., kdm6b modulation by Stat3 inhibition, sox9a in S3A) are modest (<2-fold), raising concerns about biological relevance. Can the authors define thresholds of functional relevance or confirm phenotypic outcomes in these animals?

We thank you for the inspiring comments. Most of the experiments were transient in nature, for instance, warm temperature treatment of fish for 3-4 days, the fold change of gene expression were modest.

We admit that there are some shortcomings in this work. The major one is lacking of data showing that Trpv4 inhibition/activation,or pStat3 inhibition/activation can cause a gonadal phenotype change, for instance, from ovary to ovotestis or causing females to intersex fish. We only showed that pharmacological or RNAi can lead to change in sex-biased gene expression or affect temperature-induced gene expression, but not gonadal transformation.

In natural population, the sex change of ricefield eel may take several months to one year or even longer. We propose that the magnitude and duration of temperature exposure promote sex change of ricefield eel by driving the accumulation of testicular differentiation genes in sufficient quantities. In experimental condition, to realize the gonadal phenotype change, animals may need to be under repeated pharmaceutical treatment (3 day interval treatment) for longer time to reach a threshold. However, long term treatment significantly increases the death rate of the animals, caused by stress or frequent manipulation.

Inspired by your comment, we are optimizing the experimental conditions in order to cause some phenotypic outcomes, thanks.

(3) The specificity of key antibodies is not validated. Key antibodies (Stat3, pStat3, Foxl2, Amh) were raised against mammalian proteins. Their specificity for ricefield eel proteins is unverified. Validation should include siRNA-mediated knockdown with immunoblot quantification with 3 replicates. Homemade antibodies (Sox9a, Dmrt1) also require rigorous validation.

We thank you for the comments about the specificity of the antibodies. First,when choosing the commercial antibodies, we have compared the immunogen of the animal with the corresponding amino acids of ricefield eel, making sure that it was conserved to some extent (at least> 85% similarity). Second, we have referred the published work, where the antibodies have been proven to work in zebrafish, frogs, and turtles et al. This was true for pStat3 and Stat3 antibodies (Weber et al. 2020; Ge et al., 2024). Third, the specificity for each antibody was assessed using WB, based on the predicted size of the protein and the correct control setting.

For instance, we are very confident for the specificity for Dmrt1 antibody. First, Dmrt1 protein was readily detected in testes of males but barely detected in ovaries of females (Author response image 1). Second, Dmrt1 protein was not detected in ovary of fish at cool temperature, but clearly detected in nuclei of follicles in warm temperature-treated fish (Figure 3C, 4B), in line with our qPCR results. Third, by performing IF, Dmrt1 was not detected in females reared at lower temperature. By contrast, after warm temperature treatment or Trpv4 activation, it was detected in the nuclei in specific cell types but not everywhere (Figure 3E, 6C).

Author response image 1.

Although we have carefully evaluated the antibodies before experiments as described above, in response to your concerns, we went on to validate Amh, Dmrt1, Sox9a, and Stat3 antibodies using the corresponding siRNAs (Author response image 2). The results indicated that the antibodies, although not perfect, can be used in this work, as the expected band was gone or reduced in intensity. The experiments were repeated two times, and shown were representative.

Author response image 2.

(4) Most of the imaging data (immunofluorescence) is inconclusive. Immunofluorescence panels are small and lack monochrome channels, which severely limits interpretability. Larger, better-contrasted images (showing the merge and the monochrome of important channels) and quantification would enhance the clarity of these findings.

We apologize for the poor quality of the IF images. At your suggestion, we have repeated the majority of the IF experiments, and imaging data with better quality were presented in the revised manuscript. Quantification of WB and IF was also included to enhance the clarity. Please see the revised manuscript, Thanks.

(B) Other comments about the science: 

(1) In S3A, sox9a expression is not dose-responsive to Trpv4 modulation, weakening the causal inference.

We have repeated the experiments, and new data was included for the replacement of the old one in the revised manuscript.

(2) An antibody against Kdm6b (if available) should be used to confirm protein-level changes.

We thank you for the nice suggestion. Unfortunately, current commercial antibody for Kdm6b is for mammals, which was not working in ricefield eel. At your suggestion, we are going to make one in future.

In sum, the interpretations are limited by the above concerns regarding data presentation and reagent specificity.

Reviewer #2 (Public review):

Summary:

This study presents valuable findings on the molecular mechanisms driving the female-to-male transformation in the ricefield eel (Monopterus albus) during aging. The authors explore the role of temperature-activated TRPV4 signaling in promoting testicular differentiation, proposing a TRPV4-Ca²⁺-pSTAT3-Kdm6b axis that facilitates this gonadal shift.

We thank you for the encouraging comments. Answering your questions has greatly improved our understanding of Trpv4 function in ricefield eel, and the quality of the manuscript.

Strengths:

The manuscript describes an interesting mechanism potentially underlying sex differentiation in M. albus.

Weaknesses:

The current data are insufficient to fully support the central claims, and the study would benefit from more rigorous experimental approaches.

(1) Overstated Title and Claims:

The title "TRPV4 mediates temperature-induced sex change" overstates the evidence. No histological confirmation of gonadal transformation (e.g., formation of testicular structures) is presented. Conclusions are based solely on molecular markers such as dmrt1 and sox9a, which, although suggestive, are not definitive indicators of functional sex reversal.

We thank you for pointing out this. The title has been changed to “Trpv4 links environmental temperature to testicular differentiation in hermaphroditic ricefield eel.”

(2) Temperature vs Growth Rate Confounding (Figure 1E):<br /> The conclusion that warm temperature directly induces gonadal transformation is confounded by potential growth rate effects. The authors state that body size was "comparable" between 25C and 33C groups, but fail to provide supporting data. In ectotherms, growth is intrinsically temperature-dependent. Given the known correlation between size and sex change in M. albus, growth rate-rather than temperature per se-may underlie the observed sex ratio shifts. Controlled growth-matched comparisons or inclusion of growth rate metrics are needed.

We thank you for the critical comments. We have repeated the experiments, and have carefully compared the body length and weight, and results showed that there is no big difference between 25 and 33 degree groups. Please see Figure S1D-E, and the text in the last paragraph of “Warm temperature promotes gonadal transformation” section in the Results part.

(3) TRPV4 as a Thermosensor-Insufficient Evidence:<br /> The characterisation of TRPV4 as a direct thermosensor lacks biophysical validation. The observed transcriptional upregulation of Trpv4 under heat (Figure 2) reflects downstream responses rather than primary sensor function. Functional thermosensors, including TRPV4, respond to heat via immediate ion channel activity-typically measurable within seconds-not mRNA expression over hours. No patch-clamp or electrophysiological data are provided to confirm TRPV4 activation thresholds in eel gonadal cells.

We thank you for the critical comments. The patch-clamp or electrophysiological experiments require special equipment and well-trained expert, unfortunately, our lab members and nearby collaborators have no experience in performing the kind of experiments. The Trpv4 is a well-known cation channel protein that is activated by moderate heat (> 27 degree). And a body of published work has demonstrated its role in the regulation of Ca²⁺ signals via change its configuration in response to temperature (J Physiol. 2017 Oct 25;595(22):6869–6885. doi: 10.1113/JP275052; Cell Death Dis 11, 1009 (2020). https://doi.org/10.1038/s41419-020-03181-7; Cell Death Dis 10, 497 (2019). https://doi.org/10.1038/s41419-019-1708-9; Cell calcium, https://doi.org/10.1016/j.ceca.2026.103108).

Consistently, warm temperature increased calcium influx within an hour, similar to the Trpv4 agonist treatment (Figure 2E, 5D), and addition of ion channel Trpv4 inhibitor prevents the calcium signals by war temperature treatment. Moreover, calcium signaling activity is closely linked with pStat3 activity and expression of sex-biased genes (Figures 5G, 6F). Although we did not show biophysical data, these results implied that Trpv4 is a thermosensor, and regulate the downstream pathway via the regulation of calcium signals, in line with it functions as an ion channel.

Additionally, the Ca²⁺ imaging assay (Figure 2F) lacks essential details: the timing of GSK1016790A/RN1734 administration relative to imaging is unclear, making it difficult to distinguish direct channel activity from indirect transcriptional effects.

We have added more information for Ca²⁺ imaging assay (now Figure 2E and the corresponding text in Figure 2 legend, in the revised manuscript). In particular, we added the timing of treatment to better show that it was a direct effect.

(4) Cellular Context of TRPV4 Activity Is Unclear:<br /> In situ hybridisation suggests TRPV4 expression shifts from interstitial to somatic domains under heat (Figures. 2H, S2C), implying potential cell-type-specific roles. However, the study does not clarify: (i) whether TRPV4 plays the same role across these cell types, (ii) why somatic cells show stronger signal amplification, or (iii) the cellular composition of explants used in in vitro assays. Without this resolution, conclusions from pharmacological manipulation (e.g., GSK1016790A effects) cannot be definitively linked to specific cell populations.

We thank you for the inspiring comments. We have performed IF experiments using Trpv4 specific antibodies (antibody specificity was confirmed). It was clearly shown that Trpv4 was expressed in a portion of follicle cells. To explore the identity of Trpv4-expressing somatic cells, we have performed double IF experiments using Trpv4 and Foxl2, a granulosa cell marker. The results (Figure 2H) clearly showed that Trpv4-expressing cells are a portion of Foxl2-positive granulosa cells. We propose that Trpv4-expressing granulosa cells may play an important role in sensing the temperature, and that Trpv4-expressing granulosa cells transdifferentiate into Sertoli cells by warm temperature exposure, because Dmrt1, a Sertoli cell marker, started within follicles in a typical granulosa cell location. Unfortunately, current Dmrt1/Trpv4 antibodies are both produced from rabbit. To overcome this, we are ordering mouse Dmrt1 antibodies, and in future we will perform Trpv4/Dmrt1 double IF to show if Dmrt1 positive cells co-localize with Trpv4 expressing cells. We would like to update the results to you once the antibody was available.

As our animal experiments (Figure 2H) have clearly shown the identify of Trpv4 expressing somatic cells, we did not repeat the experiments using explants, to explore the cellular composition of explants used in in vitro assays.

(5) Rapid Trpv4 mRNA Elevation and Channel Function:<br /> The authors report a dramatic increase in Trpv4 mRNA within one day of heat exposure (Figures 4D, S2B). Given that TRPV4 is a membrane channel, not a transcription factor, its rapid transcriptional sensitivity to temperature raises mechanistic questions. This finding, while intriguing, seems more correlational than functional. A clearer explanation of how TRPV4 senses temperature at the molecular level is needed.

We appreciate you for your inspiring comments. Actually, we are also wondering about how trpv4 mRNA was regulated by warm temperature. First of all, the up-regulation of trpv4 mRNA is true, as evidenced by multiple pieces of data using qPCR and ISH experiments. It appears that ovarian cells respond to the temperature changes by increasing calcium influx via Trpv4 ion channel,as well as by increasing trpv4 mRNA expression levels.

Then, how trpv4 mRNA is regulated by heat? It is well-known that gene expression can be regulated by subtle temperature change via some direct temperature sensing genes (Haltenhof et al., 2020). We hypothesized that trpv4 is a downstream target of these thermosensors, displaying a mechanism similar to mammals. Actually, we have performed some experiments, and the preliminary data were obtained, which support our hypothesis.

Because the mechanistic explanation study is undergoing and not published, we chose not to discuss it in detail in the revised manuscript. We wish to report it by the end of this year, and by then are pleased to update you with the progress.

(6) Inconclusive Evidence for the Ca²⁺-pSTAT3-Kdm6b Axis: Although the authors propose a TRPV4-Ca²⁺-pSTAT3-Kdm6b-dmrt1 pathway, intermediate steps remain poorly supported. For example, western blot data (Figures 3C, 4B) do not convincingly demonstrate significant pSTAT3 elevation at 34C. Higher-resolution and properly quantified blots are essential. The inferred signalling cascade is based largely on temporal correlation and pharmacological inhibition, which are insufficient to establish direct regulatory relationships.

We thank you for the critical comments. In response to your concerns, we have repeated experiments, and better resolution WB data with proper quantification were included in the revised manuscript. In particular, we convincingly demonstrate that 34 degree caused significant pStat3 elevation.

To directly establish regulatory relationship of the members, at your suggestion, we provided some genetic and molecular biology data to support our conclusion in the revised manuscript. For instance, we have knockdown the stat3 gene by using siRNAs, and as shown in Figure 6F, we further showed that pStat3 is functionally downstream of Trpv4. Moreover, ChIP and luciferase assays were performed to show that pStat3 directly binds and activate kdm6b (Figure 7B-C). We have also performed various pharmacological inhibition to further strength our conclusion (Figures 6B-E).

(7) Species-Specific STAT3-Kdm6b Regulation Is Unresolved:<br /> The proposed activation of Kdm6b by pSTAT3 contrasts with findings in the red-eared slider turtle (Trachemys scripta), where pSTAT3 represses Kdm6b. This divergence in regulatory direction between the two TSD species is surprising and demands further justification. Cross-species differences in binding motifs or epigenetic context should be explored. Additional evidence, such as luciferase reporter assays (using wild-type and mutant pSTAT3 binding motifs in the Kdm6b promoter) is needed to confirm direct activation.

We thank you for the inspiring comments. At your suggestion, we have performed luciferase assay using kdm6b promotor that is intact or mutated. The results were in favor of our statement. Please see Figure 7C and the related text.

A rescue experiment-testing whether Kdm6b overexpression can compensate for pSTAT3 inhibition-would also greatly strengthen the model.

We thank you for the nice suggestion. It is technically challenging to perform kdm6b overexpression or any Kdm6b gain of function experiments (we have tried to make lentivirus, however, it was not working). There is no Kdm6b-specific agonists.

Inspired by you, we are establishing constitutive kdm6b transgenic ricefield eel. Although it require at least a year to allow the fish to grow up for functional experiments, once it was established, we can directly answer some important questions.

(8) Immunofluorescence-Lack of Structural Markers: <br /> All immunofluorescence images should include structural markers to delineate gonadal boundaries. Furthermore, image descriptions in the figure legends and main text lack detail and should be significantly expanded for clarity.

We thank you for the critical comments. At your comments, we have first performed IF using beta-catenin as structural marker. However, the results were not good for some unknown reasons. Then, we used Vimentin as a structural maker, as it can label all the cells in gonads. Foxl2 was used as granulosa cell marker. Dmrt1 was used as Sertoli cell marker.

Some essential description was added in the figure legend as requested. Please see detail in the revised manuscript.

(9) Pharmacological Reagents-Mechanisms and References: <br /> The manuscript lacks proper references and mechanistic descriptions for the pharmacological agents used (e.g., GSK1016790A, RN1734, Stattic). Established literature on their specificity and usage context should be cited to support their application and interpretation in this study.

These pharmacological agents have been used by others (Ge et al., 2017; Liu et al., 2021; Weber et al., 2020; Wu et al.,2024), and they are properly cited in the manuscript.

(10) Efficiency of Experimental Interventions: <br /> The percentage of gonads exhibiting sex reversal following pharmacological or RNAi treatments should be reported in the Results. This is critical for evaluating the strength and reproducibility of the interventions.

We thank you for the critical and important comments. Actually another reviewer has asked the same question. We admit that this was the big shortcoming of the work, as we did not provide data demonstrating that Trpv4 inhibition/activation, or pStat3 inhibition/activation can cause a gonadal phenotype change, for instance, from ovary to ovotestis or causing sex reversal of fish. We only showed that pharmacological or RNAi can lead to alteration of sex-biased gene expression or affect temperature induced gene expression.

In wild population, the entire sex change of ricefield eel may take months to one year or even longer. We propose that the magnitude and duration of temperature exposure promote sex change of ricefield eel by driving the accumulation of testicular differentiation genes in sufficient quantities. In experimental condition, to realize the gonadal phenotype change, animals may need to be under repeated pharmaceutical treatment (3 day interval treatment) for longer time to reach a threshold, however, long term treatment significantly increases the death rate of the animals, caused by stress or frequent manipulation. Actually, my students have tried the experiments, unfortunately, either the number of sex-versing animals were small or the experiments lacked of repeat. So no percentage of gonadal transformation after treatment can be provided at this time, but we have indicated the number of samples when performing molecular experiments (showing expression of sex-biased genes).

Inspired by your important comment, we are optimizing the experimental conditions in order to cause some phenotypic outcomes. By then, the percentage of gonads exhibiting sex reversal following pharmacological or RNAi treatments can be calculated, showing the biological significance.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

Editorial Concerns: 

(1) The term "sex reversal" should be clearly defined upfront as female-to-male, and the developmental consequences (e.g., increase in body size post-transition) should be explicitly stated early in the introduction.

We thank our editorial for pointing out this. We have added those in the introduction Part. It reads “The species begins life as a female and then develops into a male through an intersex stage, thus displaying a female-to-male sex reversal during aging. Females are small in size (< 25 cm), and during and after sex change, there is a gradual increase in body size (> 55 cm for the majority of males).”

Additional information was shown in the first and second paragraph in the results Part.

(2) The manuscript references skewed sex ratios in cultured ricefield eel but fails to specify the direction (e.g., too many males or females). This should be clarified to contextualize the biological and commercial problem. 

According to your suggestion, we now added additional information, and it reads “The reproductive mode of ricefield eel, which leads to much more females than males in spawning season, severely affects the sex ratio, and decreases the productivity of broodstock. Moreover, adult females lay limited eggs (~200) due to its small size.”

(3) Define TSD (temperature-dependent sex determination) upon first use, not at the second mention.

We have checked this, and make sure it was properly done.

(4) The phrase "quality fries for aquaculture" should be reworded or defined; it is unclear to non-specialists.

We thank you for pointing out this. Now it reads “adult females lay limited eggs (~200) due to its small size, which is a limiting factor for massive production of seedling for aquaculture industry”.

(5) Several in-text citations (e.g., Weber 2020, Wu 2024) are absent from the bibliography. ]

We have double checked the reference, thanks.

(6) The inclusion of page and line numbers would facilitate peer review.

We have now shown the page and line.

(7) The discussion is written vaguely. Clarify species names when discussing comparative biology and consider breaking down complex sentences to aid comprehension for a broad audience, such as that of eLife. 

We have added the species name, and try our best to use concise expression. Thanks.

eLife Assessment

This manuscript details important findings that DNA polymerase kappa shows age-related changes in subcellular localization within different cell types in the brains of mice, from the nucleus in young cells to the cytoplasm in old cells. The authors' findings suggest that age-related alterations in POLK localization could drive mechanistic and functional changes in the aging brain. The authors provide solid evidence for their study, with data broadly supporting their claims with minor weaknesses.

Reviewer #1 (Public review):

Summary:

Abdelmageed et al. investigate age-related changes in the subcellular localization of DNA polymerase kappa (POLK) in the brains of mice. POLK has been actively investigated for its role in translesion DNA synthesis and involvement in other DNA repair pathways in proliferating cells, very little is known about POLK in a tissue-specific context or let alone in post-mitotic cells. The authors investigated POLK subcellular distribution in the brains of young, middle-aged, and old mice via immunoblotting of fractioned tissue extracts and immunofluorescence (IF). Immunoblotting revealed a progressive decrease in the abundance of nuclear POLK, while cytoplasmic POLK levels concomitantly increased. Similar findings were present when IF was performed on brain sections. Further IF studies of cingulate cortex (Cg1), motor cortex (M1, M2), and somatosensory (S1) cortical regions all showed an age-related decline in nuclear POLK. Nuclear speckles of POLK decrease in each region, meanwhile the number of cytoplasmic POLK granules decreases in all four regions, but granule size is increasing. The authors report similar findings for REV1, another Y-family DNA polymerase.

The authors then investigate the colocalization of POLK with other DNA damage response (DDR) proteins in either pyramidal neurons or inhibitory interneurons. At 18 months of age, DNA damage marker gH2AX demonstrated colocalization with nuclear POLK, while strong colocalization of POLK and 8-oxo-dG was present in geriatric mice. The authors find that cytoplasmic POLK granules colocalize with stress granule marker G3BP1, suggesting that the accumulated POLK ends up in the lysosome.

Brain regions were further stained to identify POLK patterns in NeuN+ neurons, GABAergic neurons, and other non-neuronal cell types present in the cortex. Microglia associated with pyramidal neurons or inhibitory interneurons were found to have higher abundance of cytoplasmic POLK. The authors also report that POLK localization can be regulated by neuronal activity induced by Kainic acid treatment. Lastly, the authors suggest that POLK could serve as an aging clock for brain tissue, but POLK deserves further characterization and correlation to functional changes before being considered for a biomarker.

Strengths:

Investigation of TLS polymerases in specific tissues and in post-mitotic cells is largely understudied. The potential changes in sub cellular localization of POLK and potentially other TLS polymerases opens up many questions about DNA repair and damage tolerance in the brain and how it can change with age.

Weaknesses:

The work is quite novel and interesting, and the authors do suggest some potentially interesting roles for POLK in the brain, but these are in of themselves a bit speculative. The majority of the findings of this paper draw upon findings from POLK antibody and its presumed specificity for POLK. However, this antibody has not been fully validated and would benefit from further validation of the different band sizes. More mechanistic investigation is needed before POLK could be considered as a brain aging clock but does not preclude the potential for using POLK as a biological "dating" system for the brain.

Comments on revisions:

The revised manuscript is suitably improved and addresses reviewer comments.

Reviewer #2 (Public review):

Summary:

Abdelmageed et al., demonstrate POLK expression in nervous tissue and focus mainly on neurons. Here, they describe an exciting age-dependent change in POLK subcellular localization, from the nucleus in young tissue to the cytoplasm in old tissue. They argue that the cytosolic POLK associates with stress granules. They also investigate cell-type specific expression of POLK, and quantitate expression changes induced by cell autonomous (activity) and cell nonautonomous (microglia) factors.

Comments on revisions:

Do the authors have any explanation or reason for why they weren't able to achieve a higher knockdown of POLK using siRNA in Figure 1A2? It does not seem statistically different by eye, as all values in the KD overlap with the control. This does not seem like strong evidence that their antibody works.

Author response:

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

Public Reviews:

Reviewer #1 (Public review):

Summary:

Abdelmageed et al. investigate age-related changes in the subcellular localization of DNA polymerase kappa (POLK) in the brains of mice. POLK has been actively investigated for its role in translesion DNA synthesis and involvement in other DNA repair pathways in proliferating cells, very little is known about POLK in a tissue-specific context, let alone in post-mitotic cells. The authors investigated POLK subcellular distribution in the brains of young, middle-aged, and old mice via immunoblotting of fractioned tissue extracts and immunofluorescence (IF). Immunoblotting revealed a progressive decrease in the abundance of nuclear POLK, while cytoplasmic POLK levels concomitantly increased. Similar findings were present when IF was performed on brain sections. Further, IF studies of the cingulate cortex (Cg1), the motor cortex (M1, M2), and the somatosensory (S1) cortical regions all showed an age-related decline in nuclear POLK. Nuclear speckles of POLK decrease in each region, meanwhile, the number of cytoplasmic POLK granules decreases in all four regions, but granule size is increasing. The authors report similar findings for REV1, another Y-family DNA polymerase.

The authors then investigate the colocalization of POLK with other DNA damage response (DDR) proteins in either pyramidal neurons or inhibitory interneurons. At 18 months of age, DNA damage marker gH2AX demonstrated colocalization with nuclear POLK, while strong colocalization of POLK and 8-oxo-dG was present in geriatric mice. The authors find that cytoplasmic POLK granules colocalize with stress granule marker G3BP1, suggesting that the accumulated POLK ends up in the lysosome.

Brain regions were further stained to identify POLK patterns in NeuN+ neurons, GABAergic neurons, and other non-neuronal cell types present in the cortex. Microglia associated with pyramidal neurons or inhibitory interneurons were found to have a higher abundance of cytoplasmic POLK. The authors also report that POLK localization can be regulated by neuronal activity induced by Kainic acid treatment. Lastly, the authors suggest that POLK could serve as an aging clock for brain tissue, but POLK deserves further characterization and correlation to functional changes before being considered as a biomarker.

Strengths:

Investigation of TLS polymerases in specific tissues and in post-mitotic cells is largely understudied. The potential changes in sub-cellular localization of POLK and potentially other TLS polymerases open up many questions about DNA repair and damage tolerance in the brain and how it can change with age.

Weaknesses:

The work is quite novel and interesting, and the authors do suggest some potentially interesting roles for POLK in the brain, but these are in and of themselves a bit speculative. The majority of the findings of this paper draw upon findings from POLK antibody and its presumed specificity for POLK. However, this antibody has not been fully validated and needs further work. Further validation experiments using Polk-deficient or knocked-down cells to investigate antibody specificity for both immunoblotting and immunofluorescence should be performed. More mechanistic investigation is needed before POLK could be considered as a brain aging clock.

We are thankful for the overall enthusiasm and positive comments.

(a) Concern over POLK antibody characterization in mouse:

We performed siRNA and shRNA knock downs in mouse primary cortical neurons as well as efficiently transfectable murine lines like 4T1 and Neuro-2A showing knock down of 99kDa and 120kDa bands recognized by sc-166667 anti-POLK antibody (exact figure number Figure 1 and S1). We show that in IF sc-166667 and A12052 (Figure S1G) shows similar immunostaining patterns and we used sc-166667 in all reported figures and western blots.

(b) More mechanistic investigation is needed before POLK could be considered as a brain aging clock:

We sincerely appreciate the valuable suggestion. We agree as a terminal assay POLK nucleo-cytoplasmic status is not practical for longitudinal studies. However, we believe it may serve an investigative/correlative endogenous signal for determining tissue age, that may be useful to "date" brain sections, since not many such cell biological markers exist. We have added clarification texts to address this.

Reviewer #2 (Public review):

Summary:

Abdelmageed et al., demonstrate POLK expression in nervous tissue and focus mainly on neurons. Here they describe an exciting age-dependent change in POLK subcellular localization, from the nucleus in young tissue to the cytoplasm in old tissue. They argue that the cytosolic POLK is associated with stress granules. They also investigate the cell-type specific expression of POLK, and quantitate expression changes induced by cell-autonomous (activity) and cell nonautonomous (microglia) factors.

I think it is an interesting report but requires a few more experiments to support their findings in the latter half of the paper. Additionally, a more mechanistic understanding of the pathways regulating POLK dynamics between the nucleus and cytosol, what is POLK doing in the cytosol, and what is it interacting with; would greatly increase the impact of this report. However, additional mechanistic experiments are mostly not needed to support much of the currently presented results, again, it would simply increase the impact.

(a) Concern on more mechanistic understanding of the pathways regulating POLK dynamics between the nucleus and cytosol:

We sincerely appreciate the reviewer’s enthusiasm and valuable guidance in helping us better understand the mechanism of nuclear-cytoplasmic POLK dynamics. Previously, we developed a modified aniPOND (accelerated native isolation of proteins on nascent DNA) protocol, which we termed iPoKD-MS (isolation of proteins on Pol kappa synthesized DNA followed by mass spectrometry), to capture proteins bound to nascent DNA synthesized by POLK in human cell lines (bioRxiv https://www.biorxiv.org/content/10.1101/2022.10.27.513845v3). In this dataset, we identified potential candidates that may regulate nuclear/cytoplasmic POLK dynamics. These candidates are currently undergoing validation in human cell lines, and we are preparing a manuscript on these findings. Among these, some candidates, including previously identified proteins such as exportin and importin (Temprine et al., 2020, PMID: 32345725), are being explored further as potential POLK nuclear/cytoplasmic shuttles. We are also conducting tests on these candidates in mouse cortical primary neurons to assess their role in POLK dynamics. In the revised version of the manuscript, we have included a discussion of our current understanding.

(b) Question on “… what is POLK doing in the cytosol, and what is it interacting with …”: Our data so far indicate that POLK accumulates in stress granules and lysosomes. We are very grateful for the reviewer’s insightful suggestions and will make every effort to incorporate them in the revised manuscript. We characterized POLK accumulation in the cytoplasm using six additional endo-lysosomal markers, as recommended by the reviewer. This data is now part of entirely new Figure 3.

Reviewer #3 (Public review):

Summary:

In this study, the authors show that DNA polymerase kappa POLK relocalizes in the cytoplasm as granules with age in mice. The reduction of nuclear POLK in old brains is congruent with an increase in DNA damage markers. The cytoplasmic granules colocalize with stress granules and endo-lysosome. The study proposes that protein localization of POLK could be used to determine the biological age of brain tissue sections.

Strengths:

Very few studies focus on the POLK protein in the peripheral nervous system (PNS). The microscopy approach used here is also very relevant: it allows the authors to highlight a radical change in POLK localization (nuclear versus cytoplasmic) depending on the age of the neurons. 

The conclusions of the study are strong. Several types of neurons are compared, the colocalization with several proteins from the NHEJ and BER repair pathways is tested, and microscopy images are systematically quantified.

Weaknesses:

The authors do not discuss the physical nature of POLK granules. There is a large field of research dedicated to the nature and function of condensates: in particular numerous studies have shown that some condensates but not all exhibit liquid-like properties (https://www.nature.com/articles/nrm.2017.7, https://pubmed.ncbi.nlm.nih.gov/33510441/ https://www.mdpi.com/2073-4425/13/10/1846). The change of physical properties of condensates is particularly important in cells undergoing stress and during aging. The authors should discuss this literature.

We highly appreciate the reviewer bringing up the context of biomolecular condensates. Our iPoKD-MS data referenced above suggests candidates from various biomolecular condensates that we are currently investigating. We appreciate the reviewer providing important literature cited these articles in text and potential biomolecular condensates are discussed in the revised version. 

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

The work is quite novel and interesting, and the authors do suggest some potentially interesting roles for POLK in the brain, but these are in of themselves a bit speculative. The majority of the findings of this paper rely upon the POLK antibody and its specificity for POLK, which is not fully characterized and needs further work (validation of antibodies using immunoblots of Polk KO cells or siRNA KD of POLK in murine cells) to provide confidence in the authors' findings. 

Points

siRNA knockdown of Polk in primary neurons showed a dramatic reduction in signal by IF even though qPCR analysis showed a reduction of only ~35% at the transcript level. Typically many DNA repair genes need to be knocked down by 80% or more to see discernable differences at the protein level. siRNA knockdown in a murine cell line (MEFs, neurons, or some other easily transfectable cell type) needs to be performed with immunoblotting with whole cell and fractionated (nuclear/cytoplasmic) lysates in order to better validate the anti-POLK antibodies and which bands that are visualized during immunoblotting are specific to POLK.

We performed siRNA and shRNA knock downs in mouse primary cortical neurons as well as efficiently transfectable murine lines like 4T1 and Neuro-2A showing knock down of 99kDa and 120kDa bands recognized by sc-166667 anti-POLK antibody (exact figure number Figure 1 and S1). We show that in IF sc-166667 and A12052 (Figure S1G) shows similar immunostaining patterns and we used sc-166667 in all reported figures and western blots.

Figure 1B and C, it is not clear which antibody(ies) are used for the immunoblotting of nuclear and cytoplasmic fractions and for a blot with whole tissue lysates. Please place the antibody vendor or clone next to the corresponding blot or describe it in the figure legend. Bands of varying sizes are present in 1B (and Figure S1) but only a band at 99 kDa was shown in 1C. Because there are no bands of equivalent size present in the nuclear and cytoplasmic fractions in Figure 1B, please describe or denote which bands were used for quantification purposes for nuclear and cytoplasmic POLK.

This has been clarified by using only one antibody throughout the manuscript sc-166667. We observed in whole cell lysate an intense ~99kDa and a faint ~120kDa band, which gets intense in nuclear fraction and is absent in cytoplasmic fraction. We have noted this in multiple human cell lines and hiPSC-derived neurons, which is our ongoing work. We do not know yet if the ~120kDa is a modification or isoform of POLK. We have hints from our proteomics data that it may be SUMOylated or ubiquitinylated or other post translational modifications. We added this in the discussion section.

Figure 1I, is there a quantification beyond just the representative image? There is no green staining pattern outside the cytoplasm in the 1-month-old M1 images that is present in all the other images in the panel.

Fig 1I is now Fig S1G in the revised manuscript. Since REV1 and POLH were not central to the study that focused on POLK, they were meant to be exploratory data panels and as such we did not quantify beyond the qualitative evaluation, which broadly resembled POLK’s disposition with age. We have noted there are some sample to sample variability in the background signal. In general, outside the cytoplasm as subcellularly segmented by fluorescent nissl expression, tends to be variable by brain areas but also higher in older brains

"Association with PRKDC further suggests POLK's role in the "gap-filling" step in the NHEJ repair pathway in neurons." There is no strong evidence in the literature for mammalian POLK playing a role in NHEJ. Some description of a role in HR has been described, however. The reference regarding the iPoKD-MS data set that provides evidence of POLK associating with BER and NHEJ factors is listed as Paul, 2022 but is in the reference list as Shilpi Paul 2022.

We removed this speculative statement and citation fixed.

Figure 4A, what is the age of the mouse for the representative images?

19 months and now mentioned in the figure legend

Figure 4C, Could the data from the different ages be plotted side by side to better evaluate the differences for each cell type/region?

Data is plotted side by side

Why was the one-month time point chosen as this could still represent the developing and not mature murine brain? 

Reviewer correctly noted that a 1 month brain is still developing, but mostly from the behavioral and circuit maturation standpoint. However, from cell division and neurogenesis perspective, that is considered to be complete by first postnatal month, with neuron production thereafter largely restricted to specialized adult niches in the dentate gyrus and subventricular zone–olfactory bulb pathway; these adult neurogenic stem cells are embryonically derived and are regulated in ways that are distinct from the early, expansionary developmental waves of neurogenesis. In our study we performed our measurements in the cortical areas only. (Caviness et al., 1995, PMID: 7482802; Ansorg et al., 2012, PMID: 22564330; Ming & Song, 2011, PMID: 21609825; Bond et al., 2015, PMID: 26431181; Bond et al., 2021, PMID: 33706926; Bartkowska et al., 2022, PMID: 36078144). Also, in Figure 6A it was incorrectly mentioned to be just 1month, we rechecked our metadata and noted that young brains were comprised of 1 and 2 month old brains and now it has been corrected.

Furthermore, can the authors describe which sex of mice was used in these experiments and the justification if a single sex was used? If both sexes were used, were there any dimorphic differences in POLK localization patterns?

This is an important aspect, but in the beginning to keep mice numbers within manageable limits, we were focusing more on the age component. While both males and female brains were assayed but due to uneven sample distribution between sexes, we could not estimate if there were any statistically significant sexual dimorphic differences in IN, PN and NNs. Future studies will investigate the sex component as a function of age.

The suggestion of POLK as a brain aging clock may be a bit premature as the functional and behavioral consequences of cytoplasmic POLK sequestration are not fully known. Furthermore, investigation of POLK levels in other genetic models of neurodegeneration or with gerotherapeutics would be needed to establish if the POLK brain clock is responsive to changes that shift brain aging. Lastly, this clock may be impractical and not useful for longitudinal studies due to the terminal nature of assessing POLK levels.

We agree as a terminal assay POLK nucleo-cytoplasmic status is not practical for longitudinal studies. However, we believe it may serve an investigative/correlative endogenous signal for determining tissue age, that may be useful to "date" brain sections, since not many such cell biological markers exist. We have added clarification text.

Some discussion of the Polk-null mice is warranted, as they only have a slightly shortened lifespan, and any disease phenotypes were not reported. This stands in contrast to other DNA repair-deficient mice that mimic premature aging and show behavioral and motor deficits. This calls into question the role of POLK in brain aging.

Discussion statements on Polk-null mice has been added.

Please correct the catalog number for the SCBT anti-POLK antibody to sc-166667

Typographical error has been corrected

Reviewer #2 (Recommendations for the authors):

Results:

Figure by figure 

(1) A progressive age-associated shift in subcellular localization of POLK The authors state that POLK has not been studied in nervous tissue before and they want to see if it is expressed, and if it changes subcellular location as a function of age. The authors argue age = stress like that seen in previous models using genotoxic agents and cancer cells. Indeed, POLK seems to convincingly change subcellular location from the nucleus to larger cytosolic puncta. 

(2) Nuclear POLK co-localizes with DNA damage response and repair proteins This was a difficult dataset for me to decipher. To me, it appears as though POLK colocalizes with these examined proteins in the CYTOSOL, not the nucleus. Especially, in the oldest mice.

We added in the discussion that DNA repair proteins were observed to be present in the cytoplasm and biomolecular condensates citing relevant reviews and primary references.

(3) POLK in the cytoplasm is associated with stress granules and lysosomes in old brains LAMP1 has some issues as a lysosome marker. The authors even state it can be on endosomes. It would be nice to use a marker for mature lysosomes, some fluorescent reporter that is activated only by lysosomal proteases or pH. It is also of interest if POLK is localized to the membrane or the inside of these structures. The authors have access to an airyscan which is sufficient to examine luminal vs membrane localization on larger organelles like lysosomes.

We thank the reviewer for pushing us to investigate the nature of cytoplasmic POLK in endo-lysosomal compartments. We have now added a full-page figure on the cell biological results from six different markers, subset (Cathepsin B and D) are known to present in the lumens of endo-lysosomes, in Figure 3. Further high-resolution membrane vs lumen was not pursued, which is perhaps better suited in cultured neurons rather than thick fixed tissues.

(4) Differentially altered POLK subcellular expression amongst excitatory, inhibitory, and nonneuronal cells in the cortex.

This seems fine. I don't see anything wrong with the author's statement that there is more POLK in neurons vs non-neuronal cells. 

(5) Microglia associated with IN and PN have significantly higher levels of cytoplasmic POLK I don't see really any convincing evidence of the author's claim here. They find a difference at early-old age, but not at old-old, or other ages. This is explained by "However, this effect is lost in late-old age (Figure 5D), likely due to the MG-mediated removal of the INs.". But no trend being observed, no experiment to show sufficiency, and no experiment to uncover a directional relationship; this is a tough claim to stand by.

Changes made in text to reflect speculative nature of this observation

(6) Subcellular localization of POLK is regulated by neuronal activity

Interesting and fairly difficult experiment. Can the authors talk more about what these values mean? I am confused as to why there is a decline in nuclear puncta at 80 min. Also, why are POLK counts in 6c similar at baseline between young and early-old? In Figures 5 and 6 I also worry about statistical analysis. Are all assumptions checked to use t-tests? Why not always use a test that has fewer assumptions?

We have explained in the text the artificial nature of few hour long acute slice preparations is very different and inherently a stressful environment, especially for the old brains, compared to the vascular perfused PFA fixed brain tissues tested between young and old ages.

We don’t have a proper explanation for the initial dip in nuclear puncta in both young and old brains at 80min of very similar magnitude. It could be a separate biological phenomenon that occurs at much shorter time scales that would not otherwise be captured in a fixed tissue assay and needs careful investigation using live tissue fluorescence imaging that is beyond the scope of this manuscript.

We apologize for the typographical error in the figure legend. We rechecked our R code and the tests were all Wilcoxon rank-sum (Mann–Whitney U) two-sided nonparametric.

Figure 6B & E had absurdly small p values due to large sample numbers. So, we implemented random sampling of 100 cells repeating for 200 times and presented the distribution of p values and Cohen’s d in the supplement and reported the median p value and Cohen’s in the main plot.

(7) POLK as an endogenous "aging clock" for brain tissue

Trainable model. What are the criteria for the model, and how does it work? The cutoffs it uses to classify each age group might be interesting in that the model may have identified a trait the researchers were unaware of. Otherwise, it is not especially useful. Maybe as an independent 'blind' analysis of the data?

We have added a better description of the models, assumptions and how two different unsupervised approaches converge on the same set of features with high AUROCs.

Minor questions:

The cartoons (1a, 2a-b, 5a, 6a) help a lot. However, I still had to work a bit to understand some of the graphs (e.g., 5d, 6b-e, fig 7). Is there a simpler way to present them? Maybe simply additional labelling? I'm not sure.

A more thorough discussion of statistical tests is warranted I think. I am not very clear why some were chosen (t-test vs nonparametric with fewer assumptions). Infinitesimally small p values also make me think maybe incorrect tests were done or no power analysis was performed beforehand. A fix for this is just discussing what went into the testing methods and why they were chosen.

Statistical analysis for Fig2 (using Generalized Estimating Equations), and Fig6 (with random repeated subsampling; method explained in text, figure legend updated and supplementary data on the distribution of p values and cohen’s d are added) to address the very small p values. Descriptions rewritten in relevant text.

In the absence of further mechanistic experiments, it would still be interesting to hear what the authors think is going on and what the significance of this altered subcellular location means. How do the authors think this is occurring? I think they are arguing that cytosolic localization of POLK is 100% detrimental to the neuron. ("The reduction of nuclear POLK in old brains is congruent with an increase in DNA damage markers") Do they have any idea what the 'bug' is in the POLK system then?

Statements in the discussion has been added.

Reviewer #3 (Recommendations for the authors):

POLK is detected as small " as small "speckles" inside the nucleus at a young age (1-2 months) and larger "granules" can be seen in the cytoplasm at progressively older time points (>9 months). In the nucleus, is POLK bound to DNA? In the cytoplasm, how are the POLK molecules organized: are they bound to a substrate or are they just organized as a proteins condensate without DNA?

In human U2OS cell line Dnase1 treatment leads to loss of POLK from the nucleus as well as its activity as reported in Fig5 of Paul, S. et. al. 2023 bioRxiv. While we haven’t reproduced these results in mouse primary neurons, we anticipate a similar situation which will be tested in the future. We have addressed limited aspects of the POLK in the cytoplasm in all new Fig3 with six endo-lysosomal markers, and added text.

When POLK proteins accumulate in the cytoplasm in aging cells, do they also repair condensates in the cytoplasm? What is the function of cytoplasmic POLK granules? More generally, is it known if other granules or foci, such as repair foci are found in the cytoplasms in aging cells, or in cells under stress?

Six markers for endo-lysosomes were tested to characterize the cytoplasmic granules now shown in Fig3.

While the authors quantify the number and sizes of the POLK signal, they don't discuss their physical nature. Some membrane-less condensates exhibit liquid-like properties, such as stress granules, P-bodies, or in the nucleus some repair condensates. In some diseased tissues, some condensates lose their liquid properties and become solid-like. Is it known if POLK condensates behave like liquid condensates or they are simply formed by bound molecules on DNA? Since they are larger and fewer in the cytoplasm, is it because several small puncta fused together to form a larger one? It would be worthwhile to discuss these points.

Discussion statements on the nature of condensates in context of the POLK cytoplasmic signal has been added.

eLife Assessment

This structural biology study provides insights into the assembly of the GID/CTLH E3 ligase complex. The multi-subunit complex forms unique, ring shaped assemblies and the findings presented here describe a "specificity code" regulates formation of subunit interfaces. The data supporting the conclusions are convincing, both in thoroughness and rigor. This study will be valuable to biochemists, structural biologists, and could lay foundation for novel designed protein assemblies.

Reviewer #1 (Public review):

Summary:

GID/CTLH-type RING ligases are huge multi-protein complexes that play an important role in protein ubiquitylation. The subunits of its core complex are distinct and form a defined structural arrangement, but there can be variations in subunit composition, such as exchange of RanBP9 and RanBP10. In this study, van gen Hassend and Schindelin provide new crystal structures of (parts of) key subunits and use those structures to elucidate the molecular details of the pairwise binding between those subunits. They identify key residues that mediate binding partner specificity. Using in vitro binding assays with purified protein, they show that altering those residues can switch specificity to a different binding partner.

Strengths:

This is a technically demanding study that sheds light on an interesting structural biology problem in residue-level detail. The combination of crystallization, structural modeling, and binding assays with purified mutant proteins is elegant and, in my eyes, convincing.

Weaknesses:

I mainly have some suggestions for further clarification, especially for a broad audience beyond the structural biology community.

(1) The authors establish what they call an 'engineering toolkit' for the controlled assembly of alternative compositions of the GID complex. The mutagenesis results are great for the specific questions asked in this manuscript. It would be great if they could elaborate on the more general significance of this 'toolkit' - is there anything from a technical point of view that can be generalized? Is there a biological interest in altering the ring composition for functional studies?

(2) Along the same lines, the mutagenesis required to rewire Twa1 binding was very complex (8 mutations). While this is impressive work, the 'big picture conclusion' from this part is not as clear as for the simpler RanBP9/10. It would be great if the authors could provide more context as to what this is useful for (e.g., potential for in vivo or in vitro functional studies, maybe even with clinical significance?)

(3) For many new crystal structures, the authors used truncated, fused, or otherwise modified versions of the proteins for technical reasons. It would be helpful if the authors could provide reasoning why those modifications are unlikely to change the conclusions of those experiments compared to the full-length proteins (which are challenging to work with for technical reasons). For instance, could the authors use folding prediction (AlphaFold) that incorporates information of their resolved structures and predicts the impact of the omitted parts of the proteins? The authors used AlphaFold for some aspects of the study, which could be expanded.

Reviewer #2 (Public review):

Summary:

This is a very interesting study focusing on a remarkable oligomerization domain, the LisH-CTLH-CRA module. The module is found in a diverse set of proteins across evolution. The present manuscript focuses on the extraordinary elaboration of this domain in GID/CTLH RING E3 ubiquitin ligases, which assemble into a gigantic, highly ordered, oval-shaped megadalton complex with strict subunit specificity. The arrangement of LisH-CTLH-CRA modules from several distinct subunits is required to form the oval on the outside of the assembly, allowing functional entities to recruit and modify substrates in the center. Although previous structures had shown that data revealed that CTLH-CRA dimerization interfaces share a conserved helical architecture, the molecular rules that govern subunit pairing have not been explored. This was a daunting task in protein biochemistry that was achieved in the present study, which defines this "assembly specificity code" at the structural and residue-specific level.

The authors used X-ray crystallography to solve high-resolution structures of mammalian CTLH-CRA domains, including RANBP9, RANBP10, TWA1, MAEA, and the heterodimeric complex between RANBP9 and MKLN. They further examined and characterized assemblies by quantitative methods (ITC and SEC-MALS) and qualitatively using nondenaturing gels. Some of their ITC measurements were particularly clever and involved competitive titrations and titrations of varying partners depending on protein behavior. The experiments allowed the authors to discover that affinities for interactions between partners is exceptionally tight, in the pM-nM range, and to distill the basis for specificity while also inferring that additional interactions beyond the LisH-CTLH-CRA modules likely also contribute to stability. Beyond discovering how the native pairings are achieved, the authors were able to use this new structural knowledge to reengineer interfaces to achieve different preferred partnerings.

Strengths:

Nearly everything about this work is exceptionally strong.

(1) The question is interesting for the native complexes, and even beyond that, has potential implications for the design of novel molecular machines.

(2) The experimental data and analyses are quantitative, rigorous, and thorough.

(3) The paper is a great read - scholarly and really interesting.

(4) The figures are exceptional in every possible way. They present very complex and intricate interactions with exquisite clarity. The authors are to be commended for outstanding use of color and color-coding throughout the study, including in cartoons to help track what was studied in what experiments. And the figures are also outstanding aesthetically.

Weaknesses:

There are no major weaknesses of note, but I can make a few recommendations for editing the text.

Reviewer #3 (Public review):

Summary:

Protein complexes, like the GID/CTLH-type E3 ligase, adopt a complex three-dimensional structure, which is of functional importance. Several domains are known to be involved in shaping the complexes. Structural information based on cryo-EM is available, but its resolution does not always provide detailed information on protein-protein interactions. The work by van gen Hassend and Schindelin provides additional structural data based on crystal structures.

Strengths:

The work is solid and very carefully performed. It provides high-resolution insights into the domain architecture, which helps to understand the protein-protein interactions on a detailed molecular level. They also include mutant data and can thereby draw conclusions on the specificity of the domain interactions. These data are probably very helpful for others who work on a functional level with protein complexes containing these domains.

Weaknesses:

The manuscript contains a lot of useful, very detailed information. This information is likely very helpful to investigate functional and regulatory aspects of the protein complexes, whose assembly relies on the LisH-CTLH-CRA modules. However, this goes beyond the scope of this manuscript.

eLife Assessment

This important study provides mechanistic evidence that tea-adapted two-spotted spider mite overcomes green tea catechin defenses via the horizontally transferred dioxygenase TkDOG15, supporting a two-step adaptation model, combining enzyme refinement and inducible upregulation. The evidence is convincing because multi-omics signals converge with functional validation (RNAi knockdown and recombinant enzyme assays) and well-controlled behavioral/toxicity assays to link TkDOG15 activity and expression to survival and feeding on tea.

Reviewer #1 (Public review):

Summary:

This study investigates the molecular mechanisms allowing the KSM mite to infest tea plants, a host that is toxic to the closely related TSSM mite due to high concentrations of phenolic catechins. The authors utilize a comparative approach involving tea-adapted KSM, non-adapted KSM, and TSSM to assess behavioral avoidance and physiological tolerance to catechins. The main finding is that tea-adapted KSM possesses a specific detoxification mechanism mediated by an enzyme, TkDOG15, which was acquired via horizontal gene transfer. The study demonstrates that adaptation is a two-step process: (1) structural refinement of the TkDOG15 enzyme through amino acid substitutions that enhance enzymatic efficiency against catechins, and (2) significant transcriptional upregulation of this gene in response to tea feeding. This enzymatic adaptation allows the mites to cleave and detoxify tea catechins, enabling survival on a toxic host plant.

Strengths:

A multiomics approach (transcriptomics and proteomics) provided a compelling cross-validation of its findings. Functional bioassays, such as RNAi and recombinant enzyme assays, demonstrated that the adapted mite has higher activity against catechins via TkDOG15. Other methodologies, like feeding assay using a parafilm-covered leaf disc, were effective in avoiding contact chemosensation.

Weaknesses:

Although TkDOG15 is assumed to "detoxify" catechins by ring cleavage, the study doesn't identify or characterize the breakdown metabolic products. If the metabolites are indeed non-toxic compared to the parent catechins, that would strengthen the detoxification hypothesis. Also, the transcriptomic and proteomic analyses identified other potential detoxification enzymes, such as CCEs, UGTs, and ABC (Supplementary Tables 3-1 & 3-2), which were also upregulated. The manuscript focuses almost exclusively on TkDOG15, potentially overlooking a multigenic adaptation mechanism, where these other enzymes might play synergistic roles, although it was mentioned in the discussion section.

Reviewer #2 (Public review):

Summary:

The fascinating topic of the host range of arthropods, including insects, and the detoxification of host secondary metabolites has been elucidated through studies of the host specificity of two closely related species. The discovery that key genes were acquired from fungi through horizontal gene transfer (HGT) is particularly significant.

Strengths:

(1) The discovery that the TkDOG15 enzyme, acquired through HGT from fungi, plays a key role in the detoxification of green tea catechins in the Kanzawa mite, revealing a new mechanism of plant-herbivore interactions, is highly encouraging.

(2) The verification of this finding through various experiments, including behavioral, toxicological, transcriptomic, and proteomic analyses, RNAi-based gene function analysis, and recombinant enzyme activity assays, is also highly commendable.

(3) By proposing a two-step model in which amino acid substitutions and expression regulation of a specific enzyme gene (TkDOG15) enable host adaptive evolution, this study contributes significantly to our understanding of the evolutionary mechanisms of speciation and plant defense overcoming.

Weaknesses:

While transcriptome/proteome analyses reported changes in the expression of other detoxification-related enzymes, including CCEs, UGTs, ABC transporters, DOG1, DOG4, and DOG7, it is regrettable that the contribution of each enzyme, including its interaction with TkDOG15 and the functional analysis of each enzyme within the overall catechin detoxification system, was not investigated.

eLife Assessment

This convincing study examines a novel interaction of RAB5 with VPS34 complex II. Structural data are combined with site-directed mutagenesis, sequence analysis, biochemistry, yeast mutant analysis, and prior data on RAB1-VPS34 and RAB5-VPS34 interactions to provide a new perspective on how RAB GTPases recruit related but distinct VPS34 complexes to different organelles. Although minor revisions are recommended, the judgment is that this work represents a fundamental advance in our understanding of VPS34 localization and regulation.

Reviewer #1 (Public review):

Summary:

This manuscript presents high-resolution cryoEM structures of VPS34-complex II bound to Rab5A at 3.2A resolution. The Williams group previously reported the structure of VPS34 complex II bound to Rab5A on liposomes using tomography, and therefore, the previous structure, although very informative, was at lower resolution.

The first new structure they present is of the 'REIE>AAAA' mutant complex bound to RAB5A. The structure resembles the previously determined one, except that an additional molecule of RAB5A was observed bound to the complex in a new position, interacting with the solenoid of VPS15.

Although this second binding site exhibited reduced occupancy of RAB5A in the structure, the authors determined an additional structure in which the primary binding site was mutated to prevent RAB5A binding ('REIE>ERIR'). In this structure, there is no RAB5A bound to the primary binding site on VPS34, but the RAB5A bound to VPS15 now has strong density. The authors note that the way in which RAB5A interacts with each site is distinct, though both interfaces involve the switch regions. The authors confirm the location of this additional binding site using HDX-MS.

The authors then determine multiple structures of the wild-type complex bound to RAB5A from a single sample, as they use 3D classifications to separate out versions of the complex bound to 0, 1, or 2 copies of RAB5A. Overall, the structure of VPS34-Complex II does not change between the different states, and the data indicate that both RAB5A binding sites can be occupied at the same time.

The authors then design a new mutant form of the complex (SHMIT>DDMIE) that is expected to disrupt the interaction at the secondary site between VPS15 and RAB5A. This mutation had a minor impact on the Kd for RAB5A binding, but when combined with the REIE>ERIR mutation of the primary binding site, RAB5A binding to the complex was abolished.

Comparison of sequences across species indicated that the RAB5A binding site on VPS15 was conserved in yeast, while the RAB5A binding site on VPS34 is not.

The authors tested the impact of a corresponding yeast Vps15 mutation (SHLITY>DDLIEY) predicted to disrupt interaction with yeast Rab5/Vps21, and found that this mutant Vps15 protein was mislocalized and caused defective CPY processing.

The authors then compare these structures of the RAB5A-class II complex to recently published structures from the Hurley group of the RAB1A-class I complex, and find that in both complexes the Rab protein is bound to the VPS34 binding site in a somewhat similar manner. However, a key difference is that the position of VPS34 is slightly different in the two complexes because of the unique ATL14L and UVRAG subunits in the class I and class II complexes, respectively. This difference creates a different RAB binding pocket that explains the difference in RAB specificity between the two complexes.

Finally, the higher resolution structures enable the authors to now model portions of BECLIN1 and UVRAG that were not previously modeled in the cryoET structure.

Strengths:

Overall, I found this to be an interesting and comprehensive study of the structural basis for the interaction of RAB5A with VPS34-complex II. The authors have performed experiments to validate their structural interpretations, and they present a clear and thorough comparative analysis of the Rab binding sites in the two different VPS34 complexes. The result is a much better understanding of how two different Rab GTPases specifically recruit two different, but highly similar complexes to the membrane surface.

Weaknesses:

No significant weaknesses were noted.

Reviewer #2 (Public review):

The work by Spokaite et al describes the discovery of a novel Rab5 binding site present in complex II of class III PI3K using a combination of HDX and Cryo EM. Extensive mutational and sequence analysis define this as the primordial Rab5 interface. The data presented are convincing that this is indeed a biologically relevant interface, and is important in defining mechanistically how VPS34 complexes are regulated.

This paper is a very nice expansion of their previous cryo-ET work from 2021, and is an excellent companion piece on high-resolution cryo-EM of the complex I class III complex bound to Rab1 from the Hurley lab in 2025. Overall, this work is of excellent technical quality and answers important unexplained observations on some unexpected mutational analysis from the previous work.

They used their increased affinity VPS34 mutant to determine the 3.2 ang structure of Rab5 bound to VPS34-CII. Clear density was seen for the original Rab5 interface, but an additional site was observed. Based on this structure, they mutated out the VPS34 interface, allowing for a high-resolution structure of the Rab5 bound at the VPS15 interface.

They extensively validated the VPS15 interface in the yeast variant of VPS34, showing that the Vp215-Rab5 (VPS21) interface identified is critical in controlling complex II VPS34 recruitment.

The major strengths of this paper are that the experiments appear to be done carefully and rigorously, and I have very few experimental suggestions.

Here is what I recommend based on some very minor weaknesses I observed

(1) My main concern has to do a little bit with presentation. My main issue is how the authors use mutant description. They clearly indicate the mutant sequence in the human isoform (for example, see Figure 2A, VPS15 described as 579-SHMIT-583>DDMIE); however, when they shift to the yeast version, they shift to saying VPS15 mutant, but don't define the mutant, Figure 2G). I would recommend they just include the same sequence numbering and WT to mutant replacement every time a new mutant (or species) is described. It is always easier to interpret what is being shown when the authors are jumping between species, when the exact mutant is included. This is particularly important in this paper, where we are jumping between different subunits and different species, so a clear description in the figure/figure legends makes it much easier to read for non-specialists.

(2) The HDX data very clearly shows that Rab5 is likely able to bind at both sites, which back ups the cryo EM data nicely. I am slightly confused by some of the HDX statements described in the methods.

(3) The authors state, "Only statistically significant peptides showing a difference greater than 0.25 Da and greater than 5% for at least two timepoints were kept." This seems to be confusing as to why they required multiple timepoints, and before they also describe that they required a p-value of less than 0.05. It might be clearer to state that significant differences required a 0.25 Da, 5%, and p-value of <0.05 (n=3). Also, what do they mean by kept? Does this mean that they only fully processed the peptides with differences?

(4) They show peptide traces for a selection in the supplement, but it would be ideal to include the full set of HDX data as an Excel file, including peptides with no differences, as there is a lot of additional information (deuteration levels for everything) that would be useful to share, as recommended from the Masson et al 2019 recommendations paper. This may be attached, but this reviewer could not see an example of it in the shared data dropbox folder.

Reviewer #3 (Public review):

Summary:

The manuscript of Spokaite et al. focuses on the Vps34 complex involved in PI3P production. This complex exists in two variants, one (class I) specific for autophagy, and a second one (class II) specific for the endocytic system. Both differ only in one subunit. The authors previously showed that the Vps34 complexes interact with Rab GTPases, Rab1 or Rab5 (for class II), and the identified site was found at Vps34. Now, the authors identify a conserved and overlooked Rab5 binding site in Vps15, which is required for the function of the Class II complex. In support of this, they show cryo-EM data with a second Rab5 bound to Vps15, identify the corresponding residues, and show by mutant analysis that impaired Rab5 binding also results in defects using yeast as a model system.

Overall, this is a most complete study with little to criticize. The paper shows convincingly that the two Rab5 binding sites are required for Vps34 complex II function, with the Vps15 binding site being critical for endosomal localization. The structural data is very much complete. What I am missing are a few controls that show that the mutations in Vps15 do not affect autophagy. I also found the last paragraph of the results section a bit out of place, even though this is a nice observation that the N-terminal part of BECLIN has these domains. However, what does it add to the story?

eLife Assessment

Li et al. present an important and innovative study linking developmental changes in sleep to ecological context in Drosophila mojavensis, and propose that sleep at one stage of an animal's life might anticipate needs at a future stage. The results fit well with this model, but are correlative in nature. The work is convincing, scientifically rigorous, and effectively bridges sleep biology and evolutionary ecology, opening promising new directions for the field.

Joint Public review:

Summary

This interesting work by Shuhao Li and colleagues suggests that developmental sleep and feeding behavior in larval flies is genetically programmed to prepare the animal for adult contingencies, such as in the case of flies living in harsh ecological environments, such as deserts. Thus, the work proposes that desert-dwelling flies such as Drosophila mojavensis sleep less and feed more than D. melanogaster as larvae, which allows them to feed less and sleep more as adults in the harsh desert conditions where they live. The authors argue that this is evidence for developmental sleep reallocation, which helps the adult flies survive in the desert. In general, their results support this compelling hypothesis, so this work provides a new perspective on how sleep might be differentially programmed across developmental stages according to the requirements of an ecological niche. This work is particularly innovative for several reasons. First, it extends the Drosophila sleep field beyond D. melanogaster and directly addresses questions about the evolution of sleep that remain largely unexplored. Second, it investigates the possibility that changes in sleep across development may be adaptive, rather than sleep being a static trait. Overall, this work opens new avenues of research, effectively bridges the fields of sleep biology and evolutionary ecology, and should be of broad interest to a general readership. The manuscript is scientifically sound and clearly written for a generalist audience.

There are, however, two important weaknesses that should be addressed. The first is the implicit assumption that all observed behavioral differences are adaptive; this would benefit from a more cautious framing. Second, the manuscript would be strengthened by a more detailed discussion, and potentially additional data, regarding the ecological differences experienced by D. mojavensis and D. melanogaster at distinct life-cycle stages.

Strengths:

(1) The study astutely uses desert Drosophila species as models to understand how sleep is optimized in a challenging environment. The manuscript is rigorous, experiments are well controlled, the work is very clearly presented, and the results support the main conclusions, which are quite exciting.

(2) The manuscript examines previously unexplored sleep differences in a non-melanogaster species.

(3) The study provides evidence that selective pressure can be restricted to specific developmental stages.

(4) This work offers evolutionary insights into the trade-offs between sleep and feeding across development.

Weaknesses

(1) The authors should soften interpretations so that it is not assumed that any observed difference between mojavensis and melanogaster is necessarily adaptive, or evolved due to food availability or temperature stress.

(2) The study relies on comparisons and correlations. While it seems likely that the observed differences in sleep explain the increased food consumption and energy storage in the larvae of desert flies, demonstrating this through sleep manipulation would strengthen the authors' conclusions.

(3) The question arises regarding whether transiently quiescent larvae are always really sleeping, and whether it is appropriate to treat sleep as a stochastic population-level phenomenon rather than as an individual trait.

(4) The manuscript would benefit from comparative analysis beyond mojavensis and melanogaster.

(5) A deeper discussion of the ecological differences between the 2 Drosophila species would place the results in a broader context.

(6) The feeding parameters used in adults and larvae measure different aspects of feeding, confounding comparisons.

eLife Assessment

This work presents a brain-wide atlas of vasopressin (Avp) and vasopressin receptor 1A (Avpr1a) mRNA expression in mouse brains using high-resolution RNAscope in situ hybridization. The single-transcript approach provides precise localization and identifies additional brain regions expressing Avpr1a, creating a valuable resource for the field. The revised manuscript is clearer and more impactful, with improved figures, stronger data organization, and enhanced scholarship through added context and citations. Overall, the evidence is compelling, and the atlas should be broadly of use to researchers studying vasopressin signaling and related neural circuits.

Reviewer #1 (Public review):

Summary:

Despite accumulating prior studies on the expressions of AVP and AVPR1a in the brain, a detailed, gender-specific mapping of AVP/AVPR1a neuronal nodes has been lacking. Using RNAscope, a cutting-edge technology that detects single RNA transcripts, the authors created a comprehensive neuroanatomical atlas of Avp and Avpr1a in male and female brains.

Strengths:

This well-executed study provides valuable new insights into gender differences in the distribution of Avp and Avpr1a. The atlas is an important resource for the neuroscience community.

The authors have adequately addressed all of my concerns. I have no further questions or concerns.

Reviewer #2 (Public review):

Summary:

The authors conducted a brain-wide survey of Avp (arginine vasopressin) and its Avpr1a gene expression in the mouse brain using RNAscope, a high-resolution in situ hybridization method. Overall, the findings are useful and important because they identify brain regions that express the Avpr1a transcript. A comprehensive overview of Avpr1a expression in the mouse brain could be highly informative and impactful. The authors used RNAscope (a proprietary in situ hybridization method) to assess transcript abundance of Avp and one of its receptors, Avpr1a. The finding of Avp-expressing cells outside the hypothalamus and the extended amygdala is novel and is nicely demonstrated by new photomicrographs in the revised manuscript. The Avpr1a data suggest expression in numerous brain regions. In the revised manuscript, reworked figures make the data easier to interpret.

Strengths:

A survey of Avpr1a expression in the mouse brain is an important tool for exploring vasopressin function in the mammalian brain and for developing hypotheses about cell- and circuit-level function.

Future considerations:

The work contained in the manuscript is substantial and informative. Some questions remain and would be addressed in the current manuscript. How many cells are impacted? Are transcripts spread across many cells or only present in a few cells? Is density evenly distributed through a brain region or compacted into a subfield?

Author response:

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

Reviewer #1:

We thank the reviewer for great suggestions.

(1) The X-axis labels in some panels in Figure 2C and Supplementary Figure 2B overlap, making them difficult to read. Adjusting the label spacing or formatting would improve clarity.

We thank the reviewer for the comment. All panels including Figure 2C and Supplementary Figure 2B, have now been organized the way in which X-axis labels are easily read.

(2) In the scatter dot plot bar diagrams, it appears that n=3 for most of the data. Does this represent the number of mice used or the number of tissue sections per sample? This should be clarified in the figure legends for better transparency. 

Great suggestion. In Results (page 7, lines 135-136), we now clarified that quantification was performed on every tenth section of the brain from 3 female and 3 male mice. Additionally, in the legends for scatter dot plot bar diagrams we also mentioned that n=3 represents the number of mice used.

(3) In Supplemental Figure 2B, the positive signals are not clearly visible. Providing higher-magnification images is recommended.

Great suggestion. The revised Supplemental Figure 2B, but also Figure 2A, now provide higher magnification inset images with distinctive positive signals.

Reviewer #2:

We thank the reviewer for great and critical suggestions.

(1) Introduction:

Line 58: References should be provided for this statement as it is based on a robust field of research, not on a new concept.

We thank the reviewer for the comment. We have now included relevant references as suggested (page 4, line 58).

(2) Line 100-102: This sentence seems to make new, an idea that has been well-documented since the late 1970s. Posterior pituitary hormones oxytocin and vasopressin have long been known to have multiple peripheral targets, and at least a subset of vasopressin and oxytocin neurons have robust central projections. The central targets have been the focus of study for numerous labs. Reference 34 does not relate to posterior pituitary hormones and seems mis-cited.

We have changed this sentence, excluded the reference that does not relate to posterior pituitary hormones and added 4 further references reporting other non-traditional roles of vasopressin and oxytocin (page 6, lines 100-102).

(3) Lines 102-108: While the regulation of bone is an interesting example of an under-appreciated impact of vasopressin, the example does not build to the rationale for examining central Avp and Avpr1a expression.

We mean no disrespect here, but we have recently reported neural brain-bone connections using the SNS-specific PRV152 virus (Ryu et al., 2024; PMID: 38963696) and submitted Single Transcript Level Atlas of Oxytocin and the Oxytocin Receptor in the Mouse Brain (doi: https://doi.org/10.1101/2024.02.15.580498). Surprisingly, we detected Avpr1a and Oxtr expression in certain brain areas (for example, PVH and MPOM) that connect to both bone and adipose tissue through the SNS—raising an important question regarding a central role of Avpr1a and Oxtr in bodily mass and fat regulation. 

(4) Line 111: Avp expression and Avpr1a expression have both been studied using in situ hybridization. Thus, the overall concept is less novel than hinted at in the text. Avp expression has been studied quite extensively. Avpr1a expression has not been studied in an exhaustive fashion. 

We thank the reviewer for this comment and absolutely agree that brain AVP expression has been studied extensively. As with the Avpr, we believe that RNAscope probe design and signal amplification system employed in our study allow for more specific and sensitive detection of individual RNA targets at the single transcript level with much cleaner background noise comparing to in situ hybridization method. 

(5) Results:

Line 143: RNAscope is indeed a powerful method of detecting mRNA at the single transcript level. However, using that single transcript resolution only to provide transcript per brain region analysis, losing all of the nuance of the individual transcript expression, seems like a poor use of the method potential.

This is a good point and we did notice that Avpr1a transcript expression in several brain nuclei displayed individual pattern of expression versus more ubiquitous expression in most of the other brain areas. We noted this finding in Results (page 9, lines 164-168); however, because of the word limits in Discussion, we are not sure what would be dropped to make more room and whether this is truly necessary.

(6 &7) Line 135: Sections were coded from 3 males and 3 females. I would argue that there is not enough statistical power to make inferences regarding sex differences or regional differences. In fact, the authors did not provide any statistical analysis in the manuscript at all, even though they stated they had completed statistical tests on the methods.

150-157: All statements regarding sex differences in expression are made without statistical analyses, which, if conducted, would be underpowered. Given the limitations of performing and analyzing RNAscope data en masse a low n is understandable, but it requires a much more precise description of the data and a more careful look at how the results can be interpreted.

We thank the reviewer for these comments. We mean no disrespect here, but while statistical analysis for main brain regions is relevant, it is not meaningful as far as nuclei, sub-nuclei and regions are concerned. It is noteworthy to mention that RNAscope data analysis in the whole mouse brain is an extremely drawn-out process requiring almost 2 months to conduct exhaustive manual counting of single Avpr1a transcripts in a single mouse brain—authors analyzed 6 brains. That said, statistical tests have been performed and exact P values are now shown in graphs.

(8) Line 146: I am flagging this instance, but it should be corrected everywhere it occurs. Since we cannot know the gender of a given mouse, I would recommend referring to the mouse's "sex" rather than its "gender."

Good suggestion. We made adequate changes throughout the manuscript.

(9) Line 153: The authors switch to discussing cell numbers. Why is this data relegated to the supplemental material?

Main figures in the manuscript report Avp and Avpr1a transcript density which has more important biological significance in terms of signal efficiency and cellular response dynamics. Due to the graph abundancy in the main text, we included all graphs with Avp and Avpr1a transcript counts in the supplemental material.

(10) Methods:

Line 369: "For simplicity and clarity, exact test results and exact P values are not presented." Simplicity or clarity is not a scientific rationale not to provide accurate statistics.

We now provide exact P values in the graphs and the sentence in line 369 has been corrected accordingly (page 18, lines 379-380).

(11) Line 362: The description of how data were analyzed is inadequate. More detail is needed.

Agreed. We now included a detailed description on how data was analyzed (page 18, lines 365-374).

(12) Discussion:

Line 321: "This contrasts the rudimentary attribution of a single function per brain area." While brain function is often taught in such rudimentary terms to make the information palatable to students, I do not think the scientific literature on vasopressin function published over the past 50 years would suggest that we are so naïve in interpreting the functional role of vasopressin in the brain. Clearly, vasopressin is involved in numerous brain functions that likely cross behavioral modalities.

Agreed and we removed this sentence.

(13) Line 322: "The approach of direct mapping of receptor expression in the brain and periphery provides the groundwork." On its face, this statement is true, but the present data build on the groundwork laid by others (multiple papers from Ostrowski et al. in the early 1990s).

Agreed.

(14) Figures:

Figure 1: 1B, I do not know the purpose of creating graphs with single bars (3V, ic, pir-male, and pir-female); there are no comparisons made in the graph. In the graphs with many brain regions, very little data can be effectively represented with the scale as it is. I recommend using tables to provide the count/density data and making graphs of only the most robust areas. In addition, although there is no statistical comparison, combining males and females in the same graphs might be beneficial to make a visual comparison easier. Why were cell counts only included in the supplemental material? Is that data not relevant?

We thank the reviewer for this comment. Now all figures are presented in a more effective and aesthetically pleasing way.

(15) There is a real missed opportunity to highlight some of the findings. For example, cell counts and density measures are provided for regions in the hippocampus, thalamus, and cortex that are not typically reported to contain vasopressin-expressing cells. Photomicrographs of these locations showing the RNAscope staining would be far more impactful in reporting these data. Are there cells expressing Avp, or is the Avp mRNA in these areas contained in fibers projecting to these areas from hypothalamic and forebrain sources?

Great suggestion. We now see in Figure 1D showing novel Avp transcript expression in the hippocampus, thalamus and cortex. Based on counterstained hematoxylin staining, Avp mRNA transcripts were found in somata.

(16) Figure 1C legend suggests images of the hippocampus and cortex, but all images are from the hypothalamus. Abbreviations are not defined.

Thank you for the comment. We corrected Figure 1C legend and separately included Figure 1D showing novel Avp mRNA expression in the hippocampus and cortex.

(17) Figure 2: The analysis of Avpr1a suffers from some of the same issues as the Avp analysis. In Figure 2A, the photomicrographs do not do a very good job of illustrating representative staining. The central canal image does not appear to have any obvious puncta, but the density of Avpr1a puncta suggests something different. The sex difference in the arcuate is also not clearly apparent in representative images. There is minimal visualization of the data for a project that depends so heavily on the appearance of puncta in tissue, coupled with the lack of clarity in the images provided, greatly diminished the overall enthusiasm for the data presentation. The figures in 2C would be more useful as tables with graphs used to highlight specific regions; as is, most of the data points are lost against the graph axis. Photomicrographs would also provide a better understanding of the data than graphs.

Great suggestion. The revised Figure 2A but also Supplemental Figure 2B now provide higher magnification inset images with distinctive positive signals. As with Figures 2C, we arranged all graphs in a more effective and aesthetically pleasing manner.

(18) Figure 3: Given the low number of animals and, therefore, low statistical power, I do not think that illustrating the ratios of male to female is a statistically valid comparison.

Please see response to Point 6 & Point 7.

(19) Figure 4: Pituitary is an interesting choice to analyze. However, why was only the posterior pituitary analyzed? Were Avp transcripts contained in terminals of vasopressin neuron axons or other cells? Was Avpr1a transcript present in blood vessel cells where Avp is released? A different cell type? Why not examine the anterior pituitary, which also expresses Avp receptors (although the literature suggests largely Avpr1b)?

Thank you for the great comment. We included only posterior pituitary because there were no positive Avp/Avpr1a transcripts found in the anterior pituitary. Unfortunately, we have not performed cell type-specific staining, which would have enabled greater variation in AVP and its receptor expression across various cell types.

eLife Assessment

This study provides useful insights regarding the alterations of sleep architecture in a knock-in mouse model of Alzheimer's Disease (AD). These include age-related hyperactivity that is typically associated with increased arousal, a normal homeostatic response to sleep loss, and a stronger AD-like phenotype in females. Although the analyses are robust, evidence for the proposed mechanisms underlying abnormal sleep architecture is incomplete. Overall, the study may have a focused impact on the sleep and AD fields.

Reviewer #1 (Public review):

Summary:

The manuscript titled, "Sleep-Wake Transitions Are Impaired in the AppNL-G-F Mouse Model of Early Onset Alzheimer's Disease", is about a study of sleep/wake phenomena in a knockin mouse strain carrying "three mutations in the human App gene associated with elevated risk for early onset AD". Traditional, in-depth characterization of sleep/wake states, EEG parameters, and response to sleep loss are employed to provide evidence, "supporting the use of this strain as a model to investigate interventions that mitigate AD burden during early disease stages". The sleep/wake findings of earlier studies (especially Maezono et al., 2020, as noted by the authors) were extended by several important, genotype-related observations, including age-related hyperactivity onset that is typically associated with increased arousal, a normal response to loss of sleep and to multiple sleep latency testing, and a stronger AD-like phenotype in females. The authors conclude that the AppNL-G-F mice demonstrate many of the human AD prodromal symptoms and suggest that this strain may serve as a model for prodromal AD in humans, confirming the earlier results and conclusions of Maezono et al. Finally, based on state bout frequency and duration analyses, it is suggested that the AppNL-G-F mice may develop disruptions in mechanism(s) involved in state transition.

Strengths:

The study appears to have been, technically, rigorously conducted with high quality, in-depth traditional assessment of both state and EEG characteristics, with the concordant addition of activity and temperature. The major strengths of this study derive from observations that the AppNL-G-F mice: (1) are more hyperactive in association with decreased transitions between states; (2) maintain a normal response to sleep deprivation and have normal MSLT results; and (3) display a sex specific, "stronger" insomnia-like effect of the knockin in females.

Weaknesses:

The weaknesses stem from the study's impact being limited due to its being largely confirmatory of the Maezono et al. study, with advances of importance to a potentially more focused field. Further, the authors conclude that AppNL-G-F mice have disrupted mechanism(s) responsible for state transition; however, these were not directly examined. The rationale for this conclusion is stated by the authors as based on the observations that bouts of both W and NREM tend to be longer in duration and decreased in frequency in AppNL-G-F mice. Although altered mechanism(s) of state transition (it is not clear what mechanisms are referenced here) cannot be ruled out, other explanations might be considered. For example, increased arousal in association with hyperactivity would be expected to result in increased duration of W bouts during the active phase. This would also predictably result in greater sleep pressure that is typically associated with more consolidated NREM bouts, consistent with the observations of bout duration and frequency.

Reviewer #2 (Public review):

Summary:

The authors have used a knock-in mouse model to explore late-in-life amyloid effects on sleep. This is an excellent model as the mutated genes are regulated by the endogenous promoter system. The sleep study techniques and statistical analyses are also first-rate.

The group finds an age-dependent increase in motor activity in advanced age in the NLGF homozygous knock-in mice (NLGF), with a parallel age-dependent increase in body temperature, both effects predominate in the dark period. Interestingly, the sleep patterns do not quite follow the sleep changes. Wake time is increased in NLGF mice, and there is no progression in increased wake over time. NREMS and REM sleep are both reduced, and there is no progression. Sleep-wake effects, however, show a robust light:dark effect with larger effects in the dark period. These findings support distinct effects of this mutation on activity and temperature and on sleep. This is the first description of the temporal pattern of these effects. NLGF mice show wake stability (longer bout durations in the dark period (their active period) and fewer brief arousals from sleep. Sleep homeostasis across the lights-on period is normal. Wake power spectral density is unaffected in NLGF mice at either age. Only REM power spectra are affected, with NLGF mice showing less theta and more delta. There are interesting sex differences, with females showing no gene difference in wake bout number, while males show a gene effect. Similarly, gene effects on NREM bout number seem larger in males than in females. Although there was no difference in homeostatic response, there was normalization of sleep-wake activity after sleep deprivation.

Strengths:

Approach (model extent of sleep phenotyping), analysis.

Weaknesses:

The weaknesses are summarized below and are viewed as "addressable".

(1) The term insomnia. Insomnia is defined as a subjective dissatisfaction with sleep, which cannot be ascertained in a mouse model. The findings across baseline sleep in NLGF mice support increased wake consolidation in the active period. The predominant sleep period (lights on) is largely unaffected, and the active period (lights off) shows increased activity and increased wake with longer bouts. There is a fantastic clue where NLGF effects are consistent with increased hypocretinergic (orexinergic) neuron activity in the dark period, and/or increased drive to hypocretin neurons from PVH.

(2) Sleep-wake transitions are impaired: This should not be termed an impairment. It could actually be beneficial to have greater state stability, especially wake stability in the dark or active period. There is reduced sleep in the model that can be normalized by short-term sleep loss. It is fascinating that recovery sleep normalized sleep in the NLGF in the immediate lights-on and light-off period. This is a key finding.

Reviewer #3 (Public review):

Summary:

In this study, Tisdale et al. studied the sleep/wake patterns in the biological mouse model of Alzheimer's disease. The results in this study, together with the established literature on the relationship of sleep and Alzheimer's disease progression, guided the authors to propose this mouse model for the mechanistic understanding of sleep states that translates to Alzheimer's disease patients. However, the manuscript currently suffers from a disconnect between the physiological data and the mechanistic interpretations. Specifically, the claim of "impaired transitions" is logically at odds with the observed increase in wake-state stability or possible hyperactivity. Additionally, the description of the methods, the quantification, and the figure presentation could be substantially improved. I detail some of my concerns below.

Strengths:

The selection of the knock-in model is a notable strength as it avoids the artifacts associated with APP overexpression and more closely mimics human pathology. The study utilizes continuous 14-day EEG recordings, providing a unique dataset for assessing chronic changes in arousal states. The assessment of sex as a biological variable identifies a more severe "insomniac-like" phenotype in females, which aligns with the higher prevalence and severity of Alzheimer's disease in women.

Weaknesses:

The study seems to lack a clear hypothesis-driven approach and relies mostly on explorative investigations. Moreover, lack of quantitative analytical methods as well as shaky logical conclusions, possibly not supported by data in its current form, leaves room for major improvement.

Since this paper studied sleep states, the "Methods" section is quite unclear on what specific criteria were used to classify sleep states. There is no quantitative description of classifying sleep based on clear, reproducible procedures. There are many reasonably well-characterized sleep scoring systems used in rat electrophysiological literature, which could be useful here. The authors are generally expected to describe movement speed and/or EMG and/or EEG (theta/delta/gamma) criteria used to classify these epochs. The subjective (manual) nature of this procedure provides no verifiable validation of the accuracy and interpretability of the results.

One of the bigger claims is that "state transition mechanism(s)" are impaired. However, Figure 7 shows that model mice exhibit significantly more long wake bouts (>260s) and fewer short wake bouts (<60s). Logically, an "impaired switch" (the flip-flop model, Saper et al., 2010) results in state fragmentation. The data here show the opposite: the wake state has become too stable. This suggests the primary defect is not in the transition mechanism itself, but possibly in a pathological increase in arousal drive (hyper-arousal), likely linked to the dark-phase hyperactivity shown in Figures 4 and 5. Also, a point to note is that this finding is not new.

Figure 3 heatmaps lack color bars and units. Spectral power must be quantitatively defined and methods well-explained in the Methods section. Without these, the reader cannot discern if the "reduced power" in females is a global suppression of signal or a frequency-specific shift. Additionally, the representative example used to claim shorter sleep bouts lacks the statistical weight required for a major physiological conclusion. How does a cooler color (not clear what range and what the interpretation is) mean shorter sleep bout in female mice? The authors should clearly mark the frequency ranges that support their claims. In this figure, there is a question mark following the theta/delta range. The authors should avoid speculation and state their claims based on facts. They should also add the theta and delta ranges in the plot, such that readers can draw their own conclusions.

Figure 8 and the MSLT results show that model mice are "no sleepier than WT mice" and have a functional homeostatic rebound. This presents a logical flaw in the "insomnia" narrative. True insomnia in AD patients typically involves a failure of the homeostatic process or a debilitating accumulation of sleep debt. If these mice do not show increased sleepiness (shorter latency) despite ~19% less sleep, the authors might be describing a "reduced need" for sleep or a "hyper-aroused" state, possibly not a clinical insomnia phenotype.

In Figure 9, LFP power shown and compared in percentages is problematic, as LFP power distribution is known to be skewed (follows power law). This is particularly problematic here because all the frequencies above ~20 Hz seem to be totally flattened or nonexistent, which makes this comparison of power severely limited and biased towards the relative frequency in the highly skewed portion of the LFP power spectrum, i.e., very low frequency ranges like delta, theta, and possibly beta. This ignores low, mid, and high gamma as well as ripple band frequencies. NREM sleep is known to have relatively greater ripple band (100-250 Hz) power bursts in hippocampal regions, and REM sleep is known to have synchronous theta-gamma relationships.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

The manuscript titled, "Sleep-Wake Transitions Are Impaired in the AppNL-G-F Mouse Model of Early Onset Alzheimer's Disease", is about a study of sleep/wake phenomena in a knockin mouse strain carrying "three mutations in the human App gene associated with elevated risk for early onset AD". Traditional, in-depth characterization of sleep/wake states, EEG parameters, and response to sleep loss are employed to provide evidence, "supporting the use of this strain as a model to investigate interventions that mitigate AD burden during early disease stages". The sleep/wake findings of earlier studies (especially Maezono et al., 2020, as noted by the authors) were extended by several important, genotype-related observations, including age-related hyperactivity onset that is typically associated with increased arousal, a normal response to loss of sleep and to multiple sleep latency testing, and a stronger AD-like phenotype in females. The authors conclude that the AppNL-G-F mice demonstrate many of the human AD prodromal symptoms and suggest that this strain may serve as a model for prodromal AD in humans, confirming the earlier results and conclusions of Maezono et al. Finally, based on state bout frequency and duration analyses, it is suggested that the AppNL-G-F mice may develop disruptions in mechanism(s) involved in state transition.

Strengths:

The study appears to have been, technically, rigorously conducted with high quality, in-depth traditional assessment of both state and EEG characteristics, with the concordant addition of activity and temperature. The major strengths of this study derive from observations that the AppNL-G-F mice: (1) are more hyperactive in association with decreased transitions between states; (2) maintain a normal response to sleep deprivation and have normal MSLT results; and (3) display a sex specific, "stronger" insomnia-like effect of the knockin in females.

Weaknesses:

The weaknesses stem from the study's impact being limited due to its being largely confirmatory of the Maezono et al. study, with advances of importance to a potentially more focused field. Further, the authors conclude that AppNL-G-F mice have disrupted mechanism(s) responsible for state transition; however, these were not directly examined. The rationale for this conclusion is stated by the authors as based on the observations that bouts of both W and NREM tend to be longer in duration and decreased in frequency in AppNL-G-F mice. Although altered mechanism(s) of state transition (it is not clear what mechanisms are referenced here) cannot be ruled out, other explanations might be considered. For example, increased arousal in association with hyperactivity would be expected to result in increased duration of W bouts during the active phase. This would also predictably result in greater sleep pressure that is typically associated with more consolidated NREM bouts, consistent with the observations of bout duration and frequency.

Reviewer 1 succinctly summarizes the advances of this study beyond the ground-breaking Maezono et al (2020) study of this “humanized” mouse model exhibiting amyloid deposition. Whereas Maezono et al. conducted sleep/wake studies on male App^NL-G-F mice at 6 and 12 months of age, we had the unusual opportunity to study both sexes of homozygous App^NL-G-F mice and WT littermates at 14-18 months of age and to conduct a longitudinal assessment of many of the same individuals at 18-22 months. In addition to baseline sleep/wake and EEG spectral analyses, we (1) measured subcutaneous body temperature and activity to obtain a broader picture of the physiology and behavior of this strain at advanced ages; (2) assessed baseline sleepiness in this strain using the murine version of the clinically-relevant Multiple Sleep Latency Test (MSLT); (3) evaluated the response of App^NL-G-F mice and WT littermates to a perturbation of the sleep homeostat; (4) compared the sleep/wake characteristics of male vs. female App^NL-G-F mice at 18-22 months and, (5) to assess the stability of the phenotypes, analyzed these data over a continuous 14-d recording rather than the conventional 24h recordings typical of most sleep/wake studies including Maezono et al. We found that a long wake/short sleep phenotype was characteristic of homozygous App^NL-G-F mice at these advanced ages which is also evident in the Maezono et al. (2020) study at 12 months of age (but not at 6 months), although the authors do not comment on this phenotype and instead focus on the reduced REM sleep which is particularly evident in female App^NL-G-F mice in our study. Remarkably, despite being awake ~20% longer per day, we find that App^NL-G-F mice are no sleepier than WT mice as determined by the MSLT and that their sleep homeostat is intact when challenged by 6-h sleep deprivation. At both advanced ages, the long wake/short sleep phenotype is due primarily to longer Wake bouts and shorter bouts of both NREM and REM sleep during the dark phase. Moreover, hyperactivity develops in older in App^NL-G-F mice, particularly females, which contributes to this phenotype. We agree with Reviewer 1 that “hyperactivity would be expected to result in increased duration of W bouts during the active phase” and that this could result in more consolidated NREM bouts and we will modify the manuscript to discuss this alternative. However, the suggestion of greater sleep pressure is not borne out by the MSLT studies as we did not observe the shorter sleep latencies and increased sleep during the nap opportunities on the MSLT that we have observed in other mouse strains. Moreover, due to their short sleep phenotype, App^NL-G-F mice would be entering the sleep deprivation study with a greater sleep debt than WT mice, yet we did not observe greater EEG Slow Wave Activity in this strain during recovery from sleep deprivation. Thus, we have suggested that App^NL-G-F mice are unable to transition from Wake to sleep as readily as their WT littermates. Our observations summarized above set the stage for subsequent mechanistic studies in aged App^NL-G-F mice, although realistically, mice of this age and genotype are a rare commodity.

Reviewer #2 (Public review):

Summary:

The authors have used a knock-in mouse model to explore late-in-life amyloid effects on sleep. This is an excellent model as the mutated genes are regulated by the endogenous promoter system. The sleep study techniques and statistical analyses are also first-rate.

The group finds an age-dependent increase in motor activity in advanced age in the NLGF homozygous knock-in mice (NLGF), with a parallel age-dependent increase in body temperature, both effects predominate in the dark period. Interestingly, the sleep patterns do not quite follow the sleep changes. Wake time is increased in NLGF mice, and there is no progression in increased wake over time. NREMS and REM sleep are both reduced, and there is no progression. Sleep-wake effects, however, show a robust light:dark effect with larger effects in the dark period. These findings support distinct effects of this mutation on activity and temperature and on sleep. This is the first description of the temporal pattern of these effects. NLGF mice show wake stability (longer bout durations in the dark period (their active period) and fewer brief arousals from sleep. Sleep homeostasis across the lights-on period is normal. Wake power spectral density is unaffected in NLGF mice at either age. Only REM power spectra are affected, with NLGF mice showing less theta and more delta. There are interesting sex differences, with females showing no gene difference in wake bout number, while males show a gene effect. Similarly, gene effects on NREM bout number seem larger in males than in females. Although there was no difference in homeostatic response, there was normalization of sleep-wake activity after sleep deprivation.

Strengths:

Approach (model extent of sleep phenotyping), analysis.

Weaknesses:

The weaknesses are summarized below and are viewed as "addressable".

(1) The term insomnia. Insomnia is defined as a subjective dissatisfaction with sleep, which cannot be ascertained in a mouse model. The findings across baseline sleep in NLGF mice support increased wake consolidation in the active period. The predominant sleep period (lights on) is largely unaffected, and the active period (lights off) shows increased activity and increased wake with longer bouts. There is a fantastic clue where NLGF effects are consistent with increased hypocretinergic (orexinergic) neuron activity in the dark period, and/or increased drive to hypocretin neurons from PVH.

(2) Sleep-wake transitions are impaired: This should not be termed an impairment. It could actually be beneficial to have greater state stability, especially wake stability in the dark or active period. There is reduced sleep in the model that can be normalized by short-term sleep loss. It is fascinating that recovery sleep normalized sleep in the NLGF in the immediate lights-on and light-off period. This is a key finding.

Reviewer 2 suggests a provocative hypothesis to test. Curiously, although a recent Science paper suggests that hyperexcitable hypocretin/orexin neurons in aging mice results in greater sleep/wake fragmentation, hyperexcitability of this system could result in hyperactivity and longer wake bouts in aged App^NL-G-F mice.

Reviewer #3 (Public review):

Summary:

In this study, Tisdale et al. studied the sleep/wake patterns in the biological mouse model of Alzheimer's disease. The results in this study, together with the established literature on the relationship of sleep and Alzheimer's disease progression, guided the authors to propose this mouse model for the mechanistic understanding of sleep states that translates to Alzheimer's disease patients. However, the manuscript currently suffers from a disconnect between the physiological data and the mechanistic interpretations. Specifically, the claim of "impaired transitions" is logically at odds with the observed increase in wake-state stability or possible hyperactivity. Additionally, the description of the methods, the quantification, and the figure presentation could be substantially improved. I detail some of my concerns below.

Strengths:

The selection of the knock-in model is a notable strength as it avoids the artifacts associated with APP overexpression and more closely mimics human pathology. The study utilizes continuous 14-day EEG recordings, providing a unique dataset for assessing chronic changes in arousal states. The assessment of sex as a biological variable identifies a more severe "insomniac-like" phenotype in females, which aligns with the higher prevalence and severity of Alzheimer's disease in women.

Weaknesses:

The study seems to lack a clear hypothesis-driven approach and relies mostly on explorative investigations. Moreover, lack of quantitative analytical methods as well as shaky logical conclusions, possibly not supported by data in its current form, leaves room for major improvement.

Since this paper studied sleep states, the "Methods" section is quite unclear on what specific criteria were used to classify sleep states. There is no quantitative description of classifying sleep based on clear, reproducible procedures. There are many reasonably well-characterized sleep scoring systems used in rat electrophysiological literature, which could be useful here. The authors are generally expected to describe movement speed and/or EMG and/or EEG (theta/delta/gamma) criteria used to classify these epochs. The subjective (manual) nature of this procedure provides no verifiable validation of the accuracy and interpretability of the results.

One of the bigger claims is that "state transition mechanism(s)" are impaired. However, Figure 7 shows that model mice exhibit significantly more long wake bouts (>260s) and fewer short wake bouts (<60s). Logically, an "impaired switch" (the flip-flop model, Saper et al., 2010) results in state fragmentation. The data here show the opposite: the wake state has become too stable. This suggests the primary defect is not in the transition mechanism itself, but possibly in a pathological increase in arousal drive (hyper-arousal), likely linked to the dark-phase hyperactivity shown in Figures 4 and 5. Also, a point to note is that this finding is not new.

Figure 3 heatmaps lack color bars and units. Spectral power must be quantitatively defined and methods well-explained in the Methods section. Without these, the reader cannot discern if the "reduced power" in females is a global suppression of signal or a frequency-specific shift. Additionally, the representative example used to claim shorter sleep bouts lacks the statistical weight required for a major physiological conclusion. How does a cooler color (not clear what range and what the interpretation is) mean shorter sleep bout in female mice? The authors should clearly mark the frequency ranges that support their claims. In this figure, there is a question mark following the theta/delta range. The authors should avoid speculation and state their claims based on facts. They should also add the theta and delta ranges in the plot, such that readers can draw their own conclusions.

Figure 8 and the MSLT results show that model mice are "no sleepier than WT mice" and have a functional homeostatic rebound. This presents a logical flaw in the "insomnia" narrative. True insomnia in AD patients typically involves a failure of the homeostatic process or a debilitating accumulation of sleep debt. If these mice do not show increased sleepiness (shorter latency) despite ~19% less sleep, the authors might be describing a "reduced need" for sleep or a "hyper-aroused" state, possibly not a clinical insomnia phenotype.

In Figure 9, LFP power shown and compared in percentages is problematic, as LFP power distribution is known to be skewed (follows power law). This is particularly problematic here because all the frequencies above ~20 Hz seem to be totally flattened or nonexistent, which makes this comparison of power severely limited and biased towards the relative frequency in the highly skewed portion of the LFP power spectrum, i.e., very low frequency ranges like delta, theta, and possibly beta. This ignores low, mid, and high gamma as well as ripple band frequencies. NREM sleep is known to have relatively greater ripple band (100-250 Hz) power bursts in hippocampal regions, and REM sleep is known to have synchronous theta-gamma relationships.

We agree with the reviewer that the “Classification of arousal states” section was missing the key description of how we scored the recordings into arousal states based on EEG, EMG and locomotor activity; this was an oversight as the corresponding text exists in all our previous sleep/wake studies published over several decades. Reviewer 1 also points out the alternative interpretation that “the wake state has become too stable.” However, I think we are using different words to say the same thing: that the transition from wake to sleep is impaired whether it is due to hyperarousal or to a defect in the flip/flop switch that results in greater Wake stability. We will revise Fig 3 (Reviewer 2 suggests combining with Fig 14) but note that the X-axis is labelled 0-25 Hz and that this figure was intended to be descriptive -- illustrating how unusual the female App^NL-G-F mice are relative to WT -- rather than a quantitative analysis of spectral power as in Fig. 14. Both Reviewer 2 and 3 suggest that we are using “insomnia” incorrectly, which we have simply used to describe less sleep per 24h period. Reviewer 2 states that “Insomnia is defined as a subjective dissatisfaction with sleep” and Reviewer 3 suggests a narrow definition of insomnia as due only to “a failure of the homeostatic process or a debilitating accumulation of sleep debt.” In a revised manuscript, we will define “insomnia” as an operational term to succinctly mean “less sleep”. Regarding the problem of presenting spectral power in percentages, we completely agree with the reviewer. However, we intentionally presented spectral power density, a measure of relative power, as in Figure 3A and 3B of Maezono et al. (2020). At the risk of making Fig. 9 even more busy, we will revise Fig. 9 to add labels for all Y-axes.

In addition to a revised Fig. 9, in the revised manuscript, we will reformat Tables 1-3, Figs. S1 and S2 for legibility and correct an error in Fig. 7.

eLife Assessment

This important work employed a recent functional muscle network analysis to evaluate rehabilitation outcomes in post-stroke patients. While the research direction is relevant and suggests the need for further investigation, the strength of evidence supporting the claims is incomplete. Muscle interactions can serve as biomarkers, but improvements in function are not directly demonstrated, and the method's robustness is not benchmarked against existing approaches.

Reviewer #1 (Public review):

While the revised manuscript includes additional methodological details and a supplementary comparison with conventional NMF, it would be great if the authors could add the point below as limitations in the manuscript or change the title and abstract accordingly, since core issues remain:

(1) The study claims to evaluate rehabilitation outcomes without demonstrating that patients actually improved functionally

(2) The comparison with existing methods lacks the quantitative rigor needed to establish superiority

(3) The added value of this complex framework over much simpler alternatives has not been demonstrated

The strength of evidence supporting the main claims remains incomplete. I would encourage the authors to consider discussing these points

(1) including or adding a limitation section about functional outcome measures that go beyond clinical scale scores, (2) providing/discussing quantitative benchmarks showing their method outperforms alternatives on specific, predefined metrics, and (3) clarifying the clinical pathway by which these biomarkers would inform treatment decisions.

There are specific, relatively minor points, that require attention

The authors write: "we did not focus on such complementary evidence in this study." This is a weakness for a paper claiming to provide "biomarkers of therapeutic responsiveness." The FMA-UE threshold defines responders, but there's no independent validation that patients actually functioned better in daily life. Can you please clarify?

Maybe I missed the exact point about this, but with the added NMF plot, the authors list 'lower dimensionality' among their framework's advantages, but the basis for this claim is not clear because given that 12 network components were extracted compared to 11 "conventional" synergies. Can you please clarify, as it is not clear. You claim 'lower dimensionality' as an advantage of the proposed framework (in the Supplementary Materials), yet you extracted 12 components (5 redundant + 7 synergistic networks) compared to 11 synergies from the conventional NMF approach, which does not support a clinical / outcome advantage of this method. Please clarify.

Reviewer #2 (Public review):

This study presents an important analysis of how interactions between muscles can serve as biomarkers to quantify therapeutic responses in post-stroke patients. To do so, the authors employ an information-theoretical metric (co-information) to define muscle networks and perform cluster analysis.

I thank the authors for improving the clarity of the Methods section; the newly added Figure 5 is very helpful.

One minor suggestion is that the authors should avoid overloading the notation "m" for both the EEG measurement and the matrix of II values (Eq. 1.1), which I now realise was the source of some of my initial confusion. I suggest that the authors use separate notation for these two quantities.

Author response:

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

Public Reviews:

Reviewer #1 (Public review):

Summary:

This study addresses an important clinical challenge by proposing muscle network analysis as a tool to evaluate rehabilitation outcomes. The research direction is relevant, and the findings suggest further research. The strength of evidence supporting the claims is, however, limited: the improvements in function are not directly demonstrated, the robustness of the method is not benchmarked against already published approaches, and key terminology is not clearly defined, which reduces the clarity and impact of the work.

Comments:

There are several aspects of the current work that require clarification and improvement, both from a methodological and a conceptual standpoint.

First, the actual improvements associated with the rehabilitation protocol remain unclear. While the authors report certain quantitative metrics, the study lacks more direct evidence of functional gains. Typically, rehabilitation interventions are strengthened by complementary material (e.g., videos or case examples) that clearly demonstrate improvements in activities of daily living. Including such evidence would make the findings more compelling.

We thank the reviewer for their careful consideration of our work. We agree that direct evidence for the functional gains achieved by patients is important for establishing the efficacy of a clinical intervention and that this evidence should provide comprehensive insights for clinicians, from videos to case examples as suggested. Our aim here was apply a novel computational framework to a cohort of patients undergoing rehabilitation, and in doing so, provide empirical support for its utility in standardised motor assessments. We have shown that our novel approach can identify distinct physiological responses to VR vs PT conditions across the post-stroke cohort (see Fig.2B and associated text). Hence, although the data contains virtual reality vs. conventional physical therapy experimental conditions which likely holds important insights into the clinical use case of virtual reality interventions, we did not focus on such complementary evidence in this study. In future work, research groups (including our own) investigating the important question of clinical intervention efficacy will likely gain unique and useful mechanistic insights using our approach.

Moreover, a threshold of 5 points at the FMA-UE was considered as MCID, to distinguish between responder and non-responder patients, which represents an acknowledged and applicable measure in the clinical field. The use of single cases represents low evidence of change from the perspective of expert clinicians, raising concerns on the clinical meaningful of reported results. All this given, we chose to provide stronger evidence of clinical effect (i.e. comparison between responders and non-responders) interpreted from the perspective of muscle synergies, than to support our results in single selected cases, representing a bias in terms of translation to population of people survived to a stroke.

Second, the claim that the proposed muscle network analysis is robust is not sufficiently substantiated. The method is introduced without adequate reference to, or comparison with, the extensive literature that has proposed alternative metrics. It is also not evident whether a simpler analysis (e.g., EMG amplitude) might produce similar results. To highlight the added value of the proposed method, it would be important to benchmark it against established approaches. This would help clarify its specific advantages and potential applications. Moreover, several studies have shown very good outcomes when using AI and latent manifold analyses in patients with neural lesions. Interpreting the latent space appears even easier than interpreting muscle networks, as the manifolds provide a simple encoding-decoding representation of what the patient can still perform and what they can no longer do.

To address the reviewers concerns regarding adequate evidence for the claims made about the presented framework, we have now included an application of the conventional muscle synergy analysis approach based on non-negative matrix factorisation to the post-stroke cohort (see Supplementary materials Fig.5 and associated text). We made efforts to make this comparison as fair as possible by applying the conventional approach at the population level also and clustering the activation coefficients using a similar yet more conventional approach, agglomerative clustering. Accompanying the output of this application, we have included several points of where our framework improves significantly upon conventional muscle synergy analysis:

“Comparison with conventional approaches

To more directly illustrate the advantages of the proposed framework, we carried out a standardised pre-processing of the EMG data in line with conventional muscle synergy analysis. This included rectification, low-pass filtration (cut-off: 20Hz) and smooth resampling of EMG waveforms to 50 timepoints. All data for each participant at each session was separately normalised by channel-wise variance, concatenated together and input into non-negative matrix factorisation (NMF) ('nnmf' Matlab function, 10 replications) to extract 11 muscle synergies (W1-11 of Supplementary Materials Fig.5(Left)) and their time-varying activations. The number of components to extract was determined in a conventional way as the number of components required to explain >75% of the data variance. The extracted muscle synergies included distinct shoulder- (e.g. W2), elbow (e.g. W8) and forearm-level (e.g. W1) muscle covariation patterns along with more isolated muscle contributions (e.g. UT in W3, TL in W10).

Regarding the clustering results of our framework and how they compare to conventional approaches, to facilitate this comparison we applied agglomerative clustering to the time-varying activation coefficients of all participants, trials, tasks separately for pre- and post-sessions and employed the 'evalclusters' Matlab function (Ward linkage clustering, Calinski Harabasz criterion, Klist search = 2:21) for each session. We identified two clusters both at pre-session (Criterion = 1.69) and post-session (Criterion = 1.81) as optimal fits to the population data (see Supplementary Materials Fig.5(Right)). We found no associations between pre- or post-session cluster partitions and participants FMA-UE scores. Nevertheless, we did identify significant associations between the pre-session clustering’s and S_Pre (X² = 7.08, p = 0.008) and between post-session clustering’s and conventionally-defined treatment responders (X² = 4.2, p = 0.04). These findings, along with the similar two-way clustering structure found using the NIF, highlights important commonalities between these approaches.

To summarise the main advantages of our framework over this conventional approach:

- Lower dimensionality and enhanced interpretability of extracted components.

Our framework yields a lower number of population-level components that correspond more consistently to meaningful biomechanical and physiological functions.

- Integration of pairwise muscle relationships.

By incorporating muscle-pair level analysis, our framework captures coordinated interactions between primary and stabilising muscles—relationships that conventional NMF approaches overlook.

- Separation of task-relevant and task-irrelevant activity.

The NIF isolates task-relevant coordination patterns, distinguishing them from task-irrelevant interactions driven by biomechanical or task constraints. On the other hand, task-relevant and -irrelevant muscle contributions are intermixed in conventional muscle synergy analysis.

- Ability to identify complementary functional roles.

The NIF characterises whether muscle pairs act in similar or complementary ways, providing richer insight into motor control strategies.

- Reduced dependence on variance-based optimisation.

Unlike conventional methods that rely on maximising variance explained, our framework allows detection of subtle but functionally significant interactions that contribute less to total variance.

- Improved detection of clinically relevant population structure.

The clustering component of our framework revealed distinct post-stroke subgroups with important clinical relevance, distinguishing moderately and severely impaired cohorts and treatment responders and non-responders from pre-treatment data.”

This supplementary analysis is referred to in the Methods section of the main text with reference to previous similar comparisons between our framework and conventional approaches:

“Towards finding an effective approach to clustering participants in this data based on differences in impairment severity and therapeutic (non-)responsiveness, we found that conventional clustering algorithms (e.g. agglomerative, k-means etc.) could not provide substantive outputs (see Supplementary Materials Fig.5 and associated text for a direct comparison with conventional approaches), perhaps resulting from the complex interdependencies between the modular activations.”

“To facilitate comparisons with existing approaches, we performed a conventional muscle synergy analysis on the post-stroke cohort (see Supplementary Materials Fig.5 and associated text). Further comparisons with conventional approaches can be found in our previous work (O’Reilly & Delis, 2022).”

Further, we have also referred to a previous analysis of this post-stroke dataset using the conventional approach in the discussion section, where we point out how our approach can identify salient features of post-stroke physiological responses that conventional approaches cannot:

“Further, the NIF demonstrated here an enhanced capability over traditional approaches to identify these crucial patterns, as earlier work on related versions of this dataset could not identify any differentiable fractionation events across the cohort (Pregnolato et al., 2025).”

Overall, the utility of conventional muscle synergy analysis is well recognised across the field (Hong et al 2021). Our proposed approach builds on this conventional method by addressing key limitations to further enhance this clinical utility. We also agree that manifold learning approaches are an exciting area of research that we aim to incorporate into our framework in future research. Specifically, manifold learning methods like Laplacian eigenmaps can readily be applied to the co-membership matrix produced by our clustering algorithm, exploiting the geometry of this matrix to provide a continuous rather than discrete representation of population structure. We have highlighted this possibility in the discussion section:

“Indeed, in future work, we aim to apply manifold learning approaches to the co-membership matrix derived from this clustering algorithm, providing a continuous representation of the population structure.”

Third, the terminology used throughout the manuscript is sometimes ambiguous. A key example is the distinction made between "functional" and "redundant" synergies. The abstract states: "Notably, we identified a shift from redundancy to synergy in muscle coordination as a hallmark of effective rehabilitation-a transformation supported by a more precise quantification of treatment outcomes."

However, in motor control research, redundancy is not typically seen as maladaptive. Rather, it is a fundamental property of the CNS, allowing the same motor task to be achieved through different patterns of muscle activity (e.g., alternative motor unit recruitment strategies). This redundancy provides flexibility and robustness, particularly under fatiguing conditions, where new synergies often emerge. Several studies have emphasized this adaptive role of redundancy. Thus, if the authors intend to use "redundancy" differently, it is essential to define the term explicitly and justify its use to avoid misinterpretation.

We appreciate the reviewers concerns regarding the terminology employed in this study. Indeed, we agree that redundancy is seen in the motor control literature as a positive feature of biological systems, appearing to contradict the interpretations of the redundancy-to-synergy information conversion result we have presented. We also wish to highlight that across the motor control literature and beyond, the idea of redundancy is often conflated with the related but distinct notion of degeneracy. Traditional motor control research has also recognised this difference, for example, Latash has outlined this difference in the seminal work on motor abundance (https://doi.org/10.1007/s00221-012-3000-4). A key reference discussing this conflation and these two concepts in an information-theoretic way is found here: https://doi.org/10.1093/cercor/bhaa148. To summarise what their arguments mean for our work:

- System degeneracy relates to the ability of different system components to contribute towards the same task in a context-specific way.

- System redundancy corresponds to the degree of functional overlap among system components.

Hence, conceptually speaking, informational redundancy as employed in our study (i.e. functionally-similar muscle interactions) links with system redundancy in that it quantifies the functional overlap of system components. This definition of system redundancy implies that it is an unavoidable by-product of degenerate systems (inefficient use of degrees of freedom) which should be minimised where possible. As a result of stroke, in our study and related previous work patients displayed increased informational redundancy, linking with the abnormal co-activations they typically experience for example and with previous results from traditional muscle synergy analysis showing fewer components extracted as a function of motor impairment post-stroke (i.e. higher informational redundancy) (Clark et al. 2010). Our novel contribution here is to convey how effective rehabilitation is underpinned by a redundancy-to-synergy information conversion across the muscle networks, relating in a loose sense conceptually to a reduction in system redundancy and enhancement of system degeneracy (i.e. functionally differentiated system components contributing towards task performance).

Together, and alongside the mathematical descriptions of redundant (functionally-similar) and synergistic (functionally-complementary) information in what types of functional relationships they capture, we believe the intuition behind this finding has clear links with previous research showing a) the merging of muscle synergies in response to post-stroke impairment (i.e. functional de-differentiation), b) reduction in abnormal couplings with effective rehabilitation (i.e. functional re-differentiation). To communicate this more clearly to readers, we have included the following in the corresponding discussion section:

“Previous research has shown that functional redundancy increases post-stroke (Cheung et al., 2012; Clark et al., 2010), reflecting the characteristic loss of functional specificity (i.e. functional de-differentiation) of muscle interactions post-stroke. Enhanced synergy with treatment here thus reflects the functional re-differentiation of predominantly flexor-driven muscle networks towards different, complementary task-objectives across the seven upper-limb motor tasks performed (Kim et al., 2024b), leading to improved motor function among responders.”

Finally, we have screened the updated manuscript for consistent use of terminology including functional/redundant/synergistic.

References

Clark DJ, Ting LH, Zajac FE, Neptune RR, Kautz SA. Merging of healthy motor modules predicts reduced locomotor performance and muscle coordination complexity post-stroke. Journal of neurophysiology. 2010 Feb;103(2):844-57.

Hong YN, Ballekere AN, Fregly BJ, Roh J. Are muscle synergies useful for stroke rehabilitation?. Current Opinion in Biomedical Engineering. 2021 Sep 1;19:100315.

Latash ML. The bliss (not the problem) of motor abundance (not redundancy). Experimental brain research. 2012 Mar;217(1):1-5.

O'Reilly D, Delis I. Dissecting muscle synergies in the task space. Elife. 2024 Feb 26;12:RP87651.

Sajid N, Parr T, Hope TM, Price CJ, Friston KJ. Degeneracy and redundancy in active inference. Cerebral Cortex. 2020 Nov;30(11):5750-66.

Reviewer #2 (Public review):

Summary:

This study analyzes muscle interactions in post-stroke patients undergoing rehabilitation, using information-theoretic and network analysis tools applied to sEMG signals with task performance measurements. The authors identified patterns of muscle interaction that correlate well with therapeutic measures and could potentially be used to stratify patients and better evaluate the effectiveness of rehabilitation.

However, I found that the Methods and Materials section, as it stands, lacks sufficient detail and clarity for me to fully understand and evaluate the quality of the method. Below, I outline my main points of concern, which I hope the authors will address in a revision to improve the quality of the Methods section. I would also like to note that the methods appear to be largely based on a previous paper by the authors (O'Reilly & Delis, 2024), but I was unable to resolve my questions after consulting that work.

I understand the general procedure of the method to be: (1) defining a connectivity matrix, (2) refining that matrix using network analysis methods, and (3) applying a lower-dimensional decomposition to the refined matrix, which defines the sub-component of muscle interaction. However, there are a few steps not fully explained in the text.

(1) The muscle network is defined as the connectivity matrix A. Is each entry in A defined by the co-information? Is this quantity estimated for each time point of the sEMG signal and task variable? Given that there are only 10 repetitions of the measurement for each task, I do not fully understand how this is sufficient for estimating a quantity involving mutual information.

We acknowledge the confusion caused here in how many datapoints were incorporated into the estimation of II. The number of datapoints included in each variable involved was in fact no. of timepoints x 10 repetitions. Hence for the EMGs employed in this analysis with a sampling rate of 2000Hz, the length of variables involved in this analysis could easily extend beyond 20,000 datapoints each. We have clarified this more specifically in the corresponding section of the methods:

“We carried out this application in the spatial domain (i.e. interactions between muscles across time (Ó’Reilly & Delis, 2022)) by concatenating the 10 repetitions of each task executed on a particular side (i.e. variables of length no. of timepoints x 10 trials) and quantifying II with respect to this discrete task parameter codified to describe the motor task performed at each timepoint for each trial included.”

In the previous paper (O'Reilly & Delis, 2024), the authors initially defined the co-information (Equation 1.3) but then referred to mutual information (MI) in the subsequent text, which I found confusing. In addition, while the matrix A is symmetrical, it should not be orthogonal (the authors wrote A^TA = I) unless some additional constraint was imposed?

We thank the reviewer for spotting this typo in the previous paper describing a symmetric matrix as A^TA = I which is in fact related to orthogonality instead. To clarify this error, in the current study we have correctly described the symmetric matrix as A = A^T here:

“We carried out this application in the spatial domain (i.e. interactions between muscles across time (Ó’Reilly & Delis, 2022)) by concatenating the 10 repetitions of each task executed on a particular side (i.e. variables of length no. of timepoints x 10 trials) and quantifying II with respect to this discrete task parameter codified to describe the motor task performed at each timepoint for each trial included. This computation was performed on all unique m_x and m_y pairings, generating symmetric matrices (A) (i.e. A = A^T) composed separately of non-negative redundant and synergistic values (Fig.5).”

Regarding the reviewers point about the reference to MI after equation 1.3 of the previous paper where co-Information is defined, we were referring both to the task-relevant and task-irrelevant estimates analysed there collectively in a general sense as ‘MI estimates’ as they both are derived from mutual information, task-irrelevant being the MI between two muscles conditioned on a task variable (conditional mutual information) and task-relevant being the difference between two MI values (co-I is a higher-order MI estimate). This removed the need to continuously refer to each separately throughout the paper which may in its own way cause some confusion. For clarity, in the results of that paper we also provided context for each MI estimate on how they were estimated (see beginning of “Task-irrelevant muscle couplings” and “Task-redundant muscle couplings” and “Task-synergistic muscle couplings” results sections), referring throughout the Venn diagrams depicting them (see Fig.1 of previous paper). In the present study however, for brevity and focus we did not perform an analysis on task-irrelevant muscle interactions and so decided to focus our terminology on co-I (II), a higher-order MI estimate. We acknowledge that this may have caused some confusion but highlight the efforts made to communicate each measure throughout the previous and present study. We have explicitly pointed out this specific focus on task-dependent muscle couplings in this paper at the end of the introduction of the updated manuscript:

“To do so, here we focussed our analysis on quantifying task-dependent muscle couplings (collectively referred to as II), extracting functionally-similar (i.e. redundant) and -complementary (i.e. synergistic) modules…”

(2) The authors should clarify what the following statement means: "Where a muscle interaction was determined to be net redundant/synergistic, their corresponding network edge in the other muscle network was set to zero."

We acknowledge this sentence was unclear/misleading and have now clarified this statement in the following way:

“This computation was performed on all unique m_x and m_y pairings, generating sparse symmetric matrices (A) (i.e. A = A^T) composed separately of non-negative redundant and synergistic values (Fig.5).” Additionally, we have now included an additional figure (fig.5) describing this text graphically.

(3) It should be clarified what the 'm' values are in Equation 1.1. Are these the co-information values after the sparsification and applying the Louvain algorithm to the matrix 'A'? Furthermore, since each task will yield a different co-information value, how is the information from different tasks (r) being combined here?

We thank the reviewer for their attention to detail. For clarity, at the related section of Equation 1.1, we have clarified that the input matrix is composed of co-I estimates:

“The input matrix for PNMF consisted of the sparsified A on both affected and unaffected sides from all participants at both pre- and post-sessions concatenated in their vectorised forms. More specifically, the input matrix composed of redundant or synergistic values was configured such that the set of unique muscle pairings (1 … K) on affected and unaffected sides (m_aff and m_unaff respectively)…”.

The co-I estimates in this input matrix are indeed those that survived sparsification in previous steps, however, for determining the number of modules to extract using the Louvain algorithm, this step has no direct impact or transformation on the co-I estimates and is simply employed to derive an empirical input parameter for dimensionality reduction. We refer the reviewer to the following part of this paragraph where this is described:

“The number of muscle network modules identified in this final consensus partition was used as the input parameter for dimensionality reduction, namely projective non-negative matrix factorisation (PNMF) (Fig.1(D)) (Yang & Oja, 2010). The input matrix for PNMF consisted of the sparsified A on both affected and unaffected sides from all participants at both pre- and post-sessions concatenated together in their vectorised form.”

Finally, as the reviewer has mentioned, the co-I estimates from the same muscles pairings but for different tasks, experimental sessions and participants are indeed different, reflecting their task-specific tuning, changes with rehabilitation and individual differences. To combine these representations into low-dimensional components, we employed projective non-negative matrix factorisation (PNMF). As outlined in the previous paper and earlier work on this framework (O’ Reilly & Delis, 2022), application of dimensionality reduction here can generate highly generalisable motor components, highlighting their ability to effectively represent large populations of participants, tasks and sessions, while allowing interesting individual differences mentioned by the reviewer to be buffered into the corresponding activation coefficients. These activation coefficients are for this reason the focus of the cluster analyses in the present study to characterise the post-stroke cohort. We have explicitly provided this reason in the methods section of the updated manuscript:

“We focussed on $a$ here as the extraction of population-level functional modules enabled the buffering of individual differences into the space of modular activations, making them an ideal target for identifying population structure.”

(4) In general, I recommend improving the clarity of the Methods section, particularly by being more precise in defining the quantities that are being calculated. For example, the adjacency matrix should be defined clearly using co-information at the beginning, and explain how it is changed/used throughout the rest of the section.

We thank the reviewer for their constructive advice and have gone to lengths to improve the clarity of the methods section. Firstly, we have addressed all the reviewers comments on various specific sections of the methods, including more clearly the ‘why’ and ‘how’ of what was performed. Secondly, we have now included an additional figure illustrating how co-information was quantified at the network level and separated into redundant and synergistic values (see Fig.5 of updated manuscript). Finally, we have re-structured several paragraphs of the methods section to enhance flow with additional subheadings for clarity.

(5) In the previous paper (O'Reilly & Delis, 2024), the authors applied a tensor decomposition to the interaction matrix and extracted both the spatial and temporal factors. In the current work, the authors simply concatenated the temporal signals and only chose to extract the spatial mode instead. The authors should clarify this choice.

The reviewer is correct in that a different dimensionality reduction approach was employed in the previous paper. In the present study, we instead chose to employ projective non-negative matrix factorisation, as was employed in a preliminary paper on this framework (O’Reilly & Delis, 2022). This decision was made simply based on aiming to maintain brevity and simplicity in the analysis and presentation of results as we introduce other tools to the framework (i.e. the clustering algorithm). Indeed, we could have just as easily employed the tensor decomposition to extract both spatial and temporal components, however we believed the main take away points for this paper could be more easily communicated using spatial networks only. To clarify this difference for readers we have included the following in the methods section:

“The choice of PNMF here, in contrast to the space-time tensor decomposition employed in the parent study (O’Reilly & Delis, 2024), was chosen simply to maintain brevity by focussing subsequent analyses on the spatial domain.”

References

Ó’Reilly D, Delis I. A network information theoretic framework to characterise muscle synergies in space and time. Journal of Neural Engineering. 2022 Feb 18;19(1):016031.

O'Reilly D, Delis I. Dissecting muscle synergies in the task space. Elife. 2024 Feb 26;12:RP87651.

Recommendations for the authors:

Reviewing Editor Comments:

Both reviewers are concerned with the manuscript in its current form. They questioned the relevance of the current approach in providing functional or mechanistic explanations about the rehabilitation process of post-stroke patients. Our eLife Assessment would change if you include comparisons between your current method and classical ones, in addition to improving the description of your method to strengthen the evidence of its robustness.

Reviewer #1 (Recommendations for the authors):

There is a minor typographical error in Figure 2 ("compononents" should be corrected).

This error has been rectified.

Reviewer #2 (Recommendations for the authors):

The authors should be able to address most of my concerns by providing a substantially improved version of the Methods section.

See above responses to the reviewers comments regarding the methods section.

However, I would like the authors to explain in full detail (potentially including a simulation or power analysis) the procedure for estimating the co-information quantity, and to clarify whether it is robust given the sample size used in this paper.

We refer the reviewer to our previous responses outlining with greater clarity the number of samples included in the estimation of co-I. We would also like to mention here that our framework does not make inferences on the statistical significance of individual muscle couplings (i.e. co-I estimates). Instead, these estimates are employed collectively for the sole purpose of pattern recognition. Nevertheless, to generate reliable estimates of the muscle couplings, we have employed a substantial number of samples for each co-I estimate (>20k samples in each variable) addressing the reviewers main concern her.

eLife Assessment

This important work introduces a splitGFP-based labeling tool with an analysis pipeline for the synaptic scaffold protein bruchpilot, with tests in the adult Drosophila mushroom bodies, a learning center in the Drosophila brain. The evidence supporting the conclusions is convincing.

Reviewer #1 (Public review):

Summary:

The study by Wu et al. uses endogenous bruchpilot expression in a cell-type-specific manner to assess synaptic heterogeneity in adult Drosophila melanogaster mushroom body output neurons. The authors performed genomic on locus tagging of the presynaptic scaffold protein bruchpilot (brp) with one part of splitGFP (GFP11) using the CRISPR/Cas9 methodology and co-expressed the other part of splitGFP (GFP1-10) using the GAL4/UAS system. Upon expression of both parts of splitGFP, fluorescent GFP is assembled at the C-terminus of brp, exactly where brp is endogenously expressed in active zones. For manageable analysis, a high-throughput pipeline was developed. This analysis evaluated parameters like location of brp clusters, volume of clusters, and cluster intensity as a direct measure of the relative amount of brp expression levels on site using publicly available 3D analysis tools that are integrated in Fiji. Analysis was conducted for different mushroom body cell types in different mushroom body lobes using various specific GAL4 drivers. Further validation was provided by extending analysis to R8 photoreceptors that reside in the fly medulla. To test this new method of synapse assessment, Wu et al. performed an associative learning experiment in which an odor was paired with an aversive stimulus and found that in a specific time frame after conditioning, the new analysis solidly revealed changes in brp levels at specific synapses that are associated with aversive learning. Additionally, brp levels were assessed in R8 photoreceptor terminals upon extended exposure to light.

Strengths:

Expression of splitGFP bound to brp enables intensity analysis of brp expression levels as exactly one GFP molecule is expressed per brp. This is a great tool for synapse assessment. This tool can be widely used for any synapse as long as driver lines are available to co-express the other part of splitGFP in a cell-type-specific manner. As neuropils and thus brp label can be extremely dense, the analysis pipeline developed here is very useful and important. The authors have chosen an exceptionally dense neuropil - the mushroom bodies - for their analysis and compellingly show that brp assessment can be achieved even with such densely packed active zones. The result that brp levels change upon associative learning in an experiment with odor presentation paired with punishment is likewise compelling and strongly suggests that the tool and pipeline developed here can be used in an in vivo context. Thus, the tool and its uses have the potential to fundamentally advance protein analysis not only at the synapse but especially there.

Weaknesses:

The weaknesses I perceived originally were satisfactorily explained and refuted.

Reviewer #2 (Public review):

Summary:

The authors developed a cell-type-specific fluorescence-tagging approach using a CRISPR/Cas9 induced spilt-GFP reconstitution system to visualize endogenous Bruchpilot (BRP) clusters at presynaptic active zones (AZ) in specific cell types of the mushroom body (MB) in the adult Drosophila brain. This AZ profiling approach was implemented in a high-throughput quantification process allowing to compare synapse profiles within single cells, cell-types, MB compartments and between different individuals. Aim is to in more detail analyze neuronal connectivity and circuits in this center of associative learning, notoriously difficult to investigate due to the density of cells and structures within the cells. The authors detect and characterize cell-type specific differences in BRP-dependent profiling of presynapses in different compartments of the MB, while intracellular AZ distribution was found to be stereotyped. Next to the descriptive part characterizing various AZ profiles in the MB, the authors apply an associative learning assay and Rab3 knock-down and detected consequent AZ reorganization.

Strengths:

The strength of this study lies in the outstanding resolution of synapse profiling in the extremely dense compartments of the MB. This detailed analysis will serve as an entry point for many future studies of synapse diversity in connection with functional specificity to uncover the molecular mechanisms underlying learning and memory formation and neuronal network logic. Therefore, this approach is of high importance to the scientific community and represents a valuable tool to investigate and correlate AZ architecture and synapse function in the CNS.

Weaknesses:

The results and conclusions presented in this study are conclusively and well supported by the data presented and appropriate controls. As a comment that could possibly aid and strengthen the manuscript (but not required for acceptance of the manuscript): The experiments in the study are based on spilt-GFP lines (BRP:GFP11 and UAS-GFP1-10). The authors clearly validate the new on-locus construct with a genomic GFP insertion (qPCR, confocal and STED imaging of the brain with anti-BRP (Nc82), MB morphology and memory formation). It would be important to comment on the significant overall intensity decrease of anti-BRP (Nc82) in Fig. S1B (R57C10>BRP::rGFP) and possibly a Western Blot with a correlative antibody staining against BRP might help to show that BRP protein level are not affected. Additionally, it would be important to state, at least in the Materials and Methods section, that the flies are not homozygous viable (and to offer an explanation) and to state that all experiments were performed with heterozygous flies.

Reviewer #3 (Public review):

Summary:

The authors develop a tool for marking presynaptic active zones in Drosophila brains, dependent on the GAL4 construct used to express a fragment of GFP, which will incorporate with a genome-engineered partial GFP attached to the active zone protein bruchpilot - signal will be specific to the GAL4 expressing neuronal compartment. They then use various GAL4s to examine innervation onto the mushroom bodies to dissect compartment specific differences in the size and intensity of active zones. After a description of these differences, they induce learning in flies with classic odour/electric shock pairing and observe changes after conditioning that are specific to the paired conditioning/learning paradigm.

Strengths:

The imaging and analysis appears strong. The tool is novel and exciting.

Weaknesses:

I feel that the tool could do with a little more characterisation. It is assumed that the puncta observed are AZs with no further definition or characterisation. It is not resolved if the AZs visualised here simply tagged, or are the constructs incorporated to be an active functional part of the AZ.

Comments on revisions:

Apologies, I should have thought of this in the first round of review. An experiment I would suggest (and it is not a difficult one) to address the functionality of the marker: It is mentioned that the genetically tagged half of the construct is homozygous lethal. Can this be placed in trans to a brp null, with a neuronal UAS-expression of the other half of Brp-GFP - Are the animals then 1) alive, and 2) able to fly (brp mutants can't fly, hence the name 'crashpilot') - a rescue would suggest (and that is all that would be needed here) that the reconstituted brp-GFP has function.

On another note, the paper keeps switching between different DAN-GAL4 lines. In 1H, 2Band 4A, there are informative cartoons showing the extension of the neurons for PPL1, APL and DPM neurons - could these be incorporated into figures 5, 6 and 7, and the supplementary figures to help orient the reader. Ideally they would refer to a figure (in Fig 1?) -to refer to the groups of DANs in the adult brain that are known to innervate the MBs (e.g. Fig1 in Mao and Davis, Front in Neural Circuits 2009). I suggest this because I feel that this tool will be widely used, and if non-MB aficionados can follow what's being done here I feel it will be more widely accepted.

Author response:

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

Public Reviews:

Reviewer #1 (Public review):

Summary:

The study by Wu et al. uses endogenous bruchpilot expression in a cell-type-specific manner to assess synaptic heterogeneity in adult Drosophila melanogaster mushroom body output neurons. The authors performed genomic on locus tagging of the presynaptic scaffold protein bruchpilot (BRP) with one part of splitGFP (GFP11) using the CRISPR/Cas9 methodology and co-expressed the other part of splitGFP (GFP1-10) using the GAL4/UAS system. Upon expression of both parts of splitGFP, fluorescent GFP is assembled at the N-terminus of BRP, exactly where BRP is endogenously expressed in active zones. For manageable analysis, a high-throughput pipeline was developed. This analysis evaluated parameters like location of BRP clusters, volume of clusters, and cluster intensity as a direct measure of the relative amount of BRP expression levels on site, using publicly available 3D analysis tools that are integrated in Fiji. Analysis was conducted for different mushroom body cell types in different mushroom body lobes using various specific GAL4 drivers. To test this new method of synapse assessment, Wu et al. performed an associative learning experiment in which an odor was paired with an aversive stimulus and found that, in a specific time frame after conditioning, the new analysis solidly revealed changes in BRP levels at specific synapses that are associated with aversive learning.

Strengths:

Expression of splitGFP bound to BRP enables intensity analysis of BRP expression levels as exactly one GFP molecule is expressed per BRP. This is a great tool for synapse assessment. This tool can be widely used for any synapse as long as driver lines are available to co-express the other part of splitGFP in a cell-type-specific manner. As neuropils and thus the BRP label can be extremely dense, the analysis pipeline developed here is very useful and important. The authors have chosen an exceptionally dense neuropil - the mushroom bodies - for their analysis and convincingly show that BRP assessment can be achieved with such densely packed active zones. The result that BRP levels change upon associative learning in an experiment with odor presentation paired with punishment is likewise convincing, and strongly suggests that the tool and pipeline developed here can be used in an in vivo context.

Weaknesses:

Although BRP is an important scaffold protein and its expression levels were associated with function and plasticity, I am still somewhat reluctant to accept that synapse structure profiling can be inferred from only assessing BRP expression levels and BRP cluster volume. Also, is it guaranteed that synaptic plasticity is not impaired by the large GFP fluorophore? Could the GFP10 construct that is tagged to BRP in all BRP-expressing cells, independent of GAL4, possibly hamper neuronal function? Is it certain that only active zones are labeled? I do see that plastic changes are made visible in this study after an associative learning experiment with BRP intensity and cluster volume as read-out, but I would be reassured by direct measurement of synaptic plasticity with splitGFP directly connected to BRP, maybe at a different synapse that is more accessible.

We appreciate the reviewer’s comments. In the revised manuscript, we have clarified that Brp is an important, but not the only player in the active zone. We have included new data to demonstrate that split-GFP tagging does not severely affect the localization and plasticity of Brp and the function of synapses by showing: (1) nanoscopic localization of Brp::rGFP using STED imaging; (2) colocalization between Brp::rGFP and anti-Brp signals/VGCCs; (3) activity-dependent Brp remodeling in R8 photoreceptors; (4) no defect in memory performance when labeling Brp::rGFP in KCs; These four lines of additional evidence further corroborate our approach to characterize endogenous Brp as a proxy of active zone structure.

Reviewer #2 (Public review):

Summary:

The authors developed a cell-type specific fluorescence-tagging approach using a CRISPR/Cas9 induced spilt-GFP reconstitution system to visualize endogenous Bruchpilot (BRP) clusters as presynaptic active zones (AZ) in specific cell types of the mushroom body (MB) in the adult Drosophila brain. This AZ profiling approach was implemented in a high-throughput quantification process, allowing for the comparison of synapse profiles within single cells, cell types, MB compartments, and between different individuals. The aim is to analyse in more detail neuronal connectivity and circuits in this centre of associative learning. These are notoriously difficult to investigate due to the density of cells and structures within a cell. The authors detect and characterize cell-type-specific differences in BRP-dependent profiling of presynapses in different compartments of the MB, while intracellular AZ distribution was found to be stereotyped. Next to the descriptive part characterizing various AZ profiles in the MB, the authors apply an associative learning assay and detect consequent AZ re-organisation.

Strengths:

The strength of this study lies in the outstanding resolution of synapse profiling in the extremely dense compartments of the MB. This detailed analysis will be the entry point for many future analyses of synapse diversity in connection with functional specificity to uncover the molecular mechanisms underlying learning and memory formation and neuronal network logics. Therefore, this approach is of high importance for the scientific community and a valuable tool to investigate and correlate AZ architecture and synapse function in the CNS.

Weaknesses:

The results and conclusions presented in this study are, in many aspects, well-supported by the data presented. To further support the key findings of the manuscript, additional controls, comments, and possibly broader functional analysis would be helpful. In particular:

(1) All experiments in the study are based on spilt-GFP lines (BRP:GFP11 and UAS-GFP1-10).The Materials and Methods section does not contain any cloning strategy (gRNA, primer, PCR/sequencing validation, exact position of tag insertion, etc.) and only refers to a bioRxiv publication. It might be helpful to add a Materials and Methods section (at least for the BRP:GFP11 line). Additionally, as this is an on locus insertion the in BRP-ORF, it needs a general validation of this line, including controls (Western Blot and correlative antibody staining against BRP) showing that overall BRP expression is not compromised due to the GFP insertion and localizes as BRP in wild type flies, that flies are viable, have no defects in locomotion and learning and memory formation and MB morphology is not affected compared to wild type animals.

We thank the reviewer for suggesting these important validations. We included details of the design of the construct and insertion site to the Methods section, performed several new experiments to validate the split-GFP tagging of Brp, and present the data in the revision.

First, to examine whether the transcription of the brp gene is unaffected by the insertion of GFP₁₁, we conducted qRT-PCR to compare the brp mRNA levels between brp::GFP₁₁, UAS-GFP1-10 and UAS-GFP1-10 and found no difference (Figure 1 - figure supplement 1A).

To further verify the effect of GFP₁₁ tagging at the protein level, we performed anti-Brp (nc82) immunohistochemistry of brains where GFP is reconstituted pan-neuronally. We found unaltered neuropile localization of nc82 signals (Figure 1 - figure supplement 1C). In presynaptic terminals of the mushroom body calyx, we found integration of Brp::rGFP to nc82 accumulation (Figure 1D). We performed super-resolution microscopy to verify the configuration of Brp::rGFP and confirmed the donut-shape arrangement of Brp::rGFP in the terminals of motor neurons (see Wu, Eno et al., 2025 PLOS Biology), corroborating the nanoscopic assembly of Brp::rGFP at active zones (Kittel et al., 2006 Science).

Furthermore, co-expression of RFP-tagged voltage-gated calcium channel alpha subunit Cacophony (Cac) and Brp::rGFP in PAM-γ5 dopaminergic neurons revealed strong presynaptic colocalization of their punctate clusters (Figure 1E), suggesting that rGFP tagging of Brp did not damage key protein assembly at active zones (Kawasaki et al., 2004 J Neuroscience; Kittel et al., Science).

These lines of evidence suggest that the localization of endogenous Brp is barely affected by the C-terminal GFP₁₁ insertion or GFP reconstitution therewith. This is in line with a large body of studies confirming that the N-terminal region and coiled-coil domains, but not the C-terminal, region of Brp are necessary and sufficient for active zone localization (Fouquet et al., 2009 J Cell Biol; Oswald et al., 2010 J Cell Biol; Mosca and Luo, 2014 eLife; Kiragasi et al., 2017 Cell Rep; Akbergenova et al., 2018 eLife; Nieratschker et al., 2009 PLoS Genet; Johnson et al., 2009 PLoS Biol; Hallermann et al., 2010 J Neurosci). We nevertheless report homozygous lethality and found the decreased immunoreactive signals in flies carrying the GFP₁₁ insertion (Figure 1 - figure supplement 1B).

For these reasons, we always use heterozygotes for all the experiments therefore there is no conspicuous defect in locomotion as reported in the original study (Wagh et al., 2005 Neuron). To functionally validate the heterozygotes, we measured the aversive olfactory memory performance of flies where GFP reconstitution was induced in Kenyon cells using R13F02-GAL4. We found that all these transgenes did not alter mushroom body morphology (Figure 7 - figure supplement 1) or memory performance as compared to wild-type flies (Figure 7 - figure supplement 2), suggesting the synapse function required for short-term memory formation is not affected by split-GFP tagging of Brp.

(2) Several aspects of image acquisition and high-throughput quantification data analysis would benefit from a more detailed clarification.

(a) For BRP cluster segmentation it is stated in the Materials and Methods state, that intensity threshold and noise tolerance were "set" - this setting has a large effect on the quantification, and it should be specified and setting criteria named and justified (if set manually (how and why) or automatically (to what)). Additionally, if Pyhton was used for "Nearest Neigbor" analysis, the code should be made available within this manuscript; otherwise, it is difficult to judge the quality of this quantification step.

(b) To better evaluate the quality of both the imaging analysis and image presentation, it would be important to state, if presented and analysed images are deconvolved and if so, at least one proof of principle example of a comparison of original and deconvoluted file should be shown and quantified to show the impact of deconvolution on the output quality as this is central to this study.

We thank the reviewer for suggesting these clarifications. We have included more description to the revised manuscript to clarify the setting of segmentation, which was manually adjusted to optimize the F-score (previous Figure 1D, now moved to Figure 1 -figure supplement 5). We have included the code used for analyzing nearest neighbor distance, AZ density and local Brp density in the revised manuscript (Supplementary file 1), together with a pre-processed sample data sheet (Supplementary file 2).

Regarding image deconvolution, we have clarified the differential use of deconvolved and not-deconvolved images in the revised manuscript. We have also included a quantitative evaluation of Richardson-Lucy iterative deconvolution (Figure 1 - figure supplement 4). We used 20 iterations due to only marginal FWHM improvement beyond this point (Figure 1 - figure supplement 4).

(3) The major part of this study focuses on the description and comparison of the divergent synapse parameters across cell-types in MB compartments, which is highly relevant and interesting. Yet it would be very interesting to connect this new method with functional aspects of the heterogeneous synapses. This is done in Figure 7 with an associative learning approach, which is, in part, not trivial to follow for the reader and would profit from a more comprehensive analysis.

(a) It would be important for the understanding and validation of the learning induced changes, if not (only) a ratio (of AZ density/local intensity) would be presented, but both values on their own, especially to allow a comparison to the quoted, previous AZ remodelling analysis quantifying BRP intensities (ref. 17, 18). It should be elucidated in more detail why only the ratio was presented here.

We thank the reviewer for the suggestion on the presentation of learning-induced Brp remodeling. The reported values in Figure 7C are the correlation coefficient of AZ density and local intensity in each compartment, but not the ratio. These results suggest that subcompartment-sized clusters of AZs with high Brp accumulation (Figure 6) undergo local structural remodeling upon associative learning (Figure 7). For clarity, we have included a schematic of this correlation and an example scatter plot to Figure 6. Unlike the previous studies (refs 17 and 18), we did not observe robust learning-dependent changes in the Brp intensity, possibly due to some confounding factors such as overall expression levels and conditioning protocols as described in the previous and following points, respectively.

(b) The reason why a single instead of a dual odour conditioning was performed could be clarified and discussed (would that have the same effects?).

(c) Additionally, "controls" for the unpaired values - that is, in flies receiving neither shock nor odour - it would help to evaluate the unpaired control values in the different MB compartments.

We use single odor conditioning because it is the simplest way to examine the effect of odor-shock association by comparing the paired and unpaired group. Standard differential conditioning with two odors contains unpaired odor presentation (CS-) even in the ‘paired’ group. We now show that single-odor conditioning induces memory that lasts one day as in differential conditioning (Figure 7B; Tully and Quinn, J Comp Phys A 1985).

(d) The temporal resolution of the effect is very interesting (Figure 7D), and at more time points, especially between 90 and 270 min, this might raise interesting results.

The sampling time points after training was chosen based on approximately logarithmic intervals, as the memory decay is roughly exponential (Figure 7B). This transient remodeling is consistent with the previous studies reporting that the Brp plasticity was short-lived (Zhang et al., 2018 Neuron; Turrel et al., 2022 Current Biol).

(e) Additionally, it would be very interesting and rewarding to have at least one additional assay, relating structure and function, e.g. on a molecular level by a correlative analysis of BRP and synaptic vesicles (by staining or co-expression of SV-protein markers) or calcium activity imaging or on a functional level by additional learning assays.

We thank the reviewer for raising this important point. We have performed calcium imaging of KC presynaptic terminals to correlate the structure and function in another study (see Figure 2 in Wu, Eno et al., 2025 PLOS Biology for more detail). The basal presynaptic calcium pattern along the γ compartments is strikingly similar to the compartmental heterogeneity of Brp accumulation (see also Figure 2 in this study). Considering colocalization of other active-zone components, such as Cac (Figure 1E), we propose that the learning-induced remodeling of local Brp clusters should transiently modulate synaptic properties.

As a response to other reviewers’ interest, we used Brp::rGFP to measure different forms of Brp-based structural plasticity upon constant light exposure in the photoreceptors and upon silencing rab3 in KCs. Since these experiments nicely reproduced the results of previous studies (Sugie et al., Neuron 2013; Graf et al., Neuron 2009), we believe the learning-induced plasticity of Brp clustering in KCs has a transient nature.

Reviewer #3 (Public review):

Summary:

The authors develop a tool for marking presynaptic active zones in Drosophila brains, dependent on the GAL4 construct used to express a fragment of GFP, which will incorporate with a genome-engineered partial GFP attached to the active zone protein bruchpilot - signal will be specific to the GAL4-expressing neuronal compartment. They then use various GAL4s to examine innervation onto the mushroom bodies to dissect compartment-specific differences in the size and intensity of active zones. After a description of these differences, they induce learning in flies with classic odour/electric shock pairing and observe changes after conditioning that are specific to the paired conditioning/learning paradigm.

Strengths:

The imaging and analysis appear strong. The tool is novel and exciting.

Weaknesses:

I feel that the tool could do with a little more characterisation. It is assumed that the puncta observed are AZs with no further definition or characterisation.

We performed additional validation on the tool, including (1) nanoscopic localization of Brp::rGFP using STED imaging; (2) colocalization between Brp::rGFP and anti-Brp signals/VGCCs (Figure 1D-E); 3) activity-dependent active zone remodeling in R8 photoreceptors (Figure 1F). These will be detailed in our point-by-point response below.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

(1) The authors keep stating, they profile or assess synaptic structure by analyzing BRP localization, cluster volume, and intensity. However, I do not think that BRP cluster volume and intensity warrant an educated statement about presynaptic structure as a whole. I do not challenge the usefulness of BRP cluster analysis for synapse evaluation, but as there are so many more players involved in synaptic function, BRP analysis certainly cannot explain it all. This should at least be discussed.

It is correct that Brp is not the only player in the active zone. We have included more discussion on the specific role of Brp (line 84 to 89) and other synaptic markers (line 250) and edited potentially misunderstanding text.

(2) I do see that changes in BRP expression were observed following associative learning, but is it certain, that synaptic plasticity is generally unaffected by the large GFP fluorophore? BRP is grabbing onto other proteins, both with its C- and N-termini. As the GFP is right before the stop codon, it should be at the N-terminus. How far could BRP function be hampered by this? Is there still enough space for other proteins to interact?

We thank the reviewer for sharing the concerns. We here provided three lines of evidence to demonstrate that the Brp assembly at active zones required for synaptic plasticity is unaffected by split-GFP tagging.

First, we assessed olfactory memory of flies that have Brp::rGFP labeled in Kenyon cells and found the performance comparable to wild-type (Figure 7 - figure supplement 2), suggesting the Brp function required for olfactory memory (Knapek et al., J Neurosci 2011) is unaffected by split-GFP tagging.

Second, we measured Brp remodeling in photoreceptors induced by constant light exposure (LL; Sugie et al., 2015 Neuron). Consistent with the previous study, we found that LL decreased the numbers of Brp::rGFP clusters in R8 terminals in the medulla, as compared to constant dark condition (DD). This result validates the synaptic plasticity involving dynamic Brp rearrangement in the photoreceptors. We have included this result into the revised manuscript (Figure 1F).

To further validate protein interaction of Brp::rGFP, we focused on Rab3, as it was previously shown to control Brp allocation at active zones (Graf et al., 2009 Neuron). To this end, we silenced rab3 expression in Kenyon cells using RNAi and measured the intensity of Brp::rGFP clusters in γ Kenyon cells. As previously reported in the neuromuscular junction, we found that rab3 knock-down increased Brp::rGFP accumulation to the active zones, suggesting that Brp::rGFP represents the interaction with Rab3. We have included all the new data to the revised manuscript (Figure 1 - figure supplement 3).

(3) It may well be that not only active-zone-associated BRP is labeled but possibly also BRP molecules elsewhere in the neuron. I would like to see more validation, e.g., the percentage of tagged endogenous BRP associated with other presynaptic proteins.

To answer to what extent Brp::rGFP clusters represent active zones, we double-labelled Brp::rGFP and Cac::tdTomato (Cacophony, the alpha subunit of the voltage-gated calcium channels). We found that 97% of Brp::rGFP clusters showed co-localization with Cac::tdTomato in PAM-γ5 dopamine neurons terminals (Figure 1E), suggesting most Brp::rGFP clusters represent functional AZs.

(4) Z-size is ~200 nm, while x/y pixel size is ~75 nm during acquisition. How far down does the resolution go after deconvolution?

The Z-step was 370 nm and XY pixel size was 79 nm for image acquisition. We performed 20 iterations of Richarson-Lucy deconvolution using an empirical point spread function (PSF). We found that the effect of deconvolution on the full-width at half maximum (FWHM) of Brp::rGFP clusters improves only marginally beyond 20 iterations, when the XY FWHM is around 200 nm and the XZ FWHM is around 450 nm (Figure 1 - figure supplement 4).

(5) Figure Legend 7: What is a "cytoplasm membrane marker"? Does this mean membrane-bound tdTom is sticking into the cytoplasm?

We apologize for the typo and have corrected it to “plasma membrane marker”.

(6) At the end of the introduction: "characterizing multiple structural parameters..." - which were these parameters? I was under the assumption that BRP localization, cluster volume, and intensity were assessed. I do not see how these are structural parameters. Please define what exactly is meant by "structural parameters".

We apologize for the confusion. By "structural parameters”, we indeed referred to the volume, intensity and molecular density of Brp::rGFP clusters. We have revised the sentence to “Characterizing the distinct parameters and localization of Brp::rGFP cluster.”

(7) Next to last sentence of the introduction: "Characterizing multiple structural parameters revealed a significant synaptic heterogeneity within single neurons and AZ distribution stereotypy across individuals." What do the authors mean by "significant synaptic heterogeneity"?

By “synaptic heterogeneity”, we refer to the intracellular variability of active zone cytomatrices reported by Brp clusters. For instance, the intensities of Brp::rGFP clusters within Kenyon cell subtypes were variable among compartments (Figure 2). Intracellular variability of the Brp concentration of individual active zones was higher in DPM and APL neurons than Kenyon cells (Figure 3). These variabilities demonstrate intracellular synaptic heterogeneity. We have revised the sentence to be more specific to the different characters of Brp clusters.

(8) I do not understand the last sentence of the introduction. "These cell-type-specific synapse profiles suggest that AZs are organized at multiple scales, ranging from neighboring synapses to across individuals." What do the authors mean by "ranging from neighboring synapses to across individuals"? Does this mean that even neighboring synapses in the same cell can be different?

We have revised the sentence to “These cell-type-specific synapse profiles suggest that AZs are spatially organized at multiple scales, ranging from interindividual stereotypy to neighboring synapses in the same cells.”

By “neighboring synapses", we refer to the nearest neighbor similarity in Brp levels in some cell-types (Figure 6A-C), and also the sub-compartmental dense AZ clusters with high Brp level in Kenyon cells (Figure 6D-H). By “across individuals”, we refer to the individually conserved active zone distribution patterns in some neurons (Figure 5).

(9) The title talks about cell-type-specific spatial configurations. I do not understand what is meant by "spatial configurations"? Do you mean BRP cluster volume? I think the title is a little misleading.

By “spatial configuration”, we refer to the arrangement of Brp clusters within individual mushroom body neurons. This statement is based on our findings on the intracellular synaptic heterogeneity (see also response to comment #7). We have streamlined the text description in the revised manuscript for clarity.

Reviewer #2 (Recommendations for the authors):

(1) For Figure 3A: exemplary two AZs are compared here, a histogram comparing more AZs would aid in making the point that in general, AZ of similar size have different BRP level (intensities) and how much variation exists.

We have included histograms for Brp::rGFP intensity and cluster volumes to Figure 3 in the revised manuscript.

(2) Line 52: "endogenous synapses" is a confusing term; it's probably meant that the protein levels within the synapse are endogenous and not overexpressed. 

We apologize for the confusion and have revised the term to “endogenous synaptic proteins.”

(3) It is not clear from the Materials and Methods section, whether and where deconvolved or not-deconvolved images were used for the quantification pipeline. Please comment on this. 

We have now revised the Method section to clarify how deconvolved or not-deconvolved images were differently used in the pipeline.

(4) Line 664 (C) not bold.

We have corrected the error.

(5) 725 "Files" should be Flies.

We have corrected the error.

(6) 727 two times "first".

We have corrected the error.

(7) Figure 7. All (A) etc., not bold - there should be consistent annotation. 

We want to thank the reviewer for the detailed proof and have corrected all the errors spotted.

Reviewer #3 (Recommendations for the authors):

(1) Has there been an expression of the construct in a non-neuronal cell? Astrocyte-like cell? Any glia? As some sort of control for background and activity?

As the reviewer suggested, we verified the neuronal expression specificity of Brp::rGFP. Using R86E01-GAL4 and Amon-GAL4, we compared Brp::rGFP in astrocyte-like glia and neuropeptide-releasing neurons. We found no Brp::rGFP puncta in the neuropils in astrocyte-like glia compared to neurons, suggesting Brp::rGFP is specific to neurons. We have included this new dataset to the revised manuscript (Figure 1 - figure supplement 2).

(2) Similarly, expression of the construct co-expressed with a channelrhodopsin, and induction of a 'learning'-like regime of activity, similarly in a control type of experiment, expression of an inwardly rectifying channel (e.g. Kir2.1) to show that increases in size of the BRP puncta are truly activity dependent? The NMJ may be an optimal neuron to use to see the 'donut' structures of the AZs and their increase with activity. Also, are these truly AZs we are seeing here? Perhaps try co-expressing cacophony-dsRed? If the GFP Puncta are active zones, then they should be surrounded by cacophony.

We would like to clarify that we did not find Brp::rGFP size increase upon learning. Instead, we demonstrated that associative training transiently remodelled sub-compartment-sized AZ “hot spots” in Kenyon cells, indicated by the correlation of local intensity and AZ density (Figure 6-7).

To demonstrate split-GFP tagging does not affect activity-dependent plasticity associated with Brp, we measured Brp remodeling in photoreceptors induced by constant light exposure (LL; Sugie et al., 2015 Neuron). Consistent with the previous study, we found that LL decreased the numbers of Brp::rGFP clusters in R8 terminals in the medulla, as compared to constant dark condition (DD). This result validates the synaptic plasticity involving dynamic Brp rearrangement in the photoreceptors (Figure 1F).

As the reviewer suggested, we performed the STED microscopy for the larval motor neuron and confirmed the donut-shape arrangement of Brp::rGFP (Wu, Eno et al., PLOS Biol 2025).

Also following the reviewer’s suggestion, we double-labelled Brp::rGFP and Cac::tdTomato (Cacophony, the alpha subunit of the voltage-gated calcium channels). We found that 97% Brp::rGFP clusters showed co-localization with Cac::tdTomato in PAM-γ5 dopamine neurons terminals (Figure 1E), suggesting most Brp::rGFP clusters represent functional AZs.

(3) In the introduction: Intro, a sentence about BRP - central organiser of the active zone, so a key regulator of activity.

We have included a few more sentences about the role Brp in the active zones to the revised manuscript.

(4) Figure 1 E, line 650 'cite the resource here'. 

We thank the reviewer for pointing out the error and we have corrected it.

(5) Many readers may not be MB aficionados, and to make the data more accessible, perhaps use a cartoon of an MB with the cell bodies of the neurons around the MB expressing the constructs highlighted so that the reader can have a wider idea of the anatomy in relation to the MB.

We appreciate these comments and have appended cartoons of the MB to figures to help readers understand the anatomy.

eLife Assessment

This useful study uses creative scalp EEG decoding methods to attempt to demonstrate that two forms of learned associations in a Stroop task are dissociable, despite sharing similar temporal dynamics. However, the evidence supporting the conclusions is incomplete due to concerns with the experimental design and methodology. This paper would be of interest to researchers studying cognitive control and adaptive behavior, if the concerns raised in the reviews can be addressed satisfactorily.

Reviewer #1 (Public review):

Summary:

This study focuses on characterizing the EEG correlates of item-specific proportion congruency effects. Two types of learned associations are characterized, one being associations between stimulus features and control states (SC), and the other being stimulus features and responses (SR). Decoding methods are used to identify time-resolved SC and SR correlates, which are used to test properties of their dynamics.

The conclusion is reached that SC and SR associations can independently and simultaneously guide behavior. This conclusion is based on results showing SC and SR correlates are: (1) not entirely overlapping in cross-decoding; (2) simultaneously observed on average over trials in overlapping time bins; (3) independently correlate with RT; and (4) have a positive within-trial correlation.

Strengths:

Fearless, creative use of EEG decoding to test tricky hypotheses regarding latent associations.

Nice idea to orthogonalize ISPC condition (MC/MI) from stimulus features.

Weaknesses:

I still have my concern from the first round that the decoders are overfit to temporally structured noise. As I wrote before, the SC and SR classes are highly confounded with phase (chunk of session). I do not see how the control analyses conducted in the revision adequately deal with this issue.

In the figures, there are several hints that these decoders are biased. Unfortunately, the figures are also constructed in such a way that hides or diminishes the salience of the clues of bias. This bias and lack of transparency discourage trust in the methods and results.

I have two main suggestions:

(1) Run a new experiment with a design that properly supports this question.

I don't make this suggestion lightly, and I understand that it may not be feasible to implement given constraints; but I feel that this suggestion is warranted. The desired inferences rely on successful identification of SC and SR representations. Solidly identifying SC and SR representations necessitates an experimental design wherein these variables are sufficiently orthogonalized, within-subject, from temporally structured noise. The experimental design reported in this paper unfortunately does not meet this bar, in my opinion (and the opinion of a colleague I solicited).

An adequate design would have enough phases to properly support "cross-phase" cross-validation. Deconfounding temporal noise is a basic requirement for decoding analyses of EEG and fMRI data (see e.g., leave-one-run-out CV that is effectively necessary in fMRI; in my experience, EEG is not much different, when the decoded classes are blocked in time, as here). In a journal with a typical acceptance-based review process, this would be grounds for rejection.

Please note that this issue of decoder bias would seem to weaken the rest of the downstream analyses that are based on the decoded values. For instance, if the decoders are biased, in the within-trial correlation analysis, how can we be sure that co-fluctuations along certain dimensions within their projected values are driven by signal or noise? A similar issue clouds the LMM decoding-RT correlations.

(2) Increase transparency in the reporting of results throughout main text.

Please do not truncate stimulus-aligned timecourses at time=0. Displaying the baseline period is very useful to identify bias, that is, to verify that stimulus-dependent conditions cannot be decoded pre-stimulus. Bias is most expected to be revealed in the baseline interval when the data are NOT baseline-corrected, which is why I previously asked to see the results omitting baseline correction. (But also note that if the decoders are biased, baseline-correcting would not remove this bias; instead, it would spread it across the rest of the epoch, while the baseline interval would, on average, be centered at zero.)

Please use a more standard p-value correction threshold, rather than Bonferroni-corrected p<0.001. This threshold is unusually conservative for this type of study. And yet, despite this conservativeness, stimulus-evoked information can be decoded from nearly every time bin, including at t=0. This does not encourage trust in the accuracy of these p-values. Instead, I suggest using permutation-based cluster correction, with corrected p<0.05. This is much more standard and would therefore allow for better comparison to many other studies.

I don't think these things should be done as control analyses, tucked away in the supplemental materials, but instead should be done as a part of the figures in the main text -- including decoding, RSA, cross-trial correlations, and RT correlations.

Other issues:

Regarding the analysis of the within-trial correlation of RSA betas, and "Cai 2019" bias:

The correction that authors perform in the revision -- estimating the correlation within the baseline time interval and subtracting this estimate from subsequent timepoints -- assumes that the "Cai 2019" bias is stationary. This is a fairly strong assumption, however, as this bias depends not only on the design matrix, but also on the structure of the noise (see the Cai paper), which can be non-stationary. No data were provided in support of stationarity. It seems safer and potentially more realistic to assume non-stationarity.

This analysis was included in the supplemental material. However, given that the correlation analysis presented in the Results is subject to the "Cai 2019" bias, it would seem to be more appropriate to replace that analysis, rather than supplement it.

Regardless, this seems to be a moot issue, given that the underlying decoders seem to be overfit to temporally structured noise (see point above regarding weakening of downstream analyses based on decoder bias).

Outliers and t-values:

More outliers with beta coefficients could be because the original SD estimates from the t-values are influenced more by extreme values. When you use a threshold on the median absolute deviation instead of mean +/-SD, do you still get more outliers with beta coefficients vs t-values?

Random slopes:

Were random slopes (by subject) for all within-subject variables included in the LMMs? If not, please include them, and report this in the Methods.

Reviewer #2 (Public review):

Summary:

In this EEG study, Huang et al. investigated the relative contribution of two accounts to the process of conflict control, namely the stimulus-control association (SC), which refers to the phenomenon that the ratio of congruent vs. incongruent trials affects the overall control demands, and the stimulus-response association (SR), stating that the frequency of stimulus-response pairings can also impact the level of control. The authors extended the Stroop task with novel manipulation of item congruencies across blocks in order to test whether both types of information are encoded and related to behaviour. Using decoding and RSA they showed that the SC and SR representations were concurrently present in voltage signals and they also positively co-varied. In addition, the variability in both of their strengths was predictive of reaction time. In general, the experiment has a sold design and the analyses are appropriate for the research questions.

Strength:

(1) The authors used an interesting task design that extended the classic Stroop paradigm and is effective in teasing apart the relative contribution of the two different accounts regarding item-specific proportion congruency effect.

(2) Linking the strength of RSA scores with behavioural measure is critical to demonstrating the functional significance of the task representations in question.

Weakness:

(1) The distinction between Phase 2 and Phase 1&3 behavioral results, specifically the opposite effect of MC/MI in congruent trials raises some concerns with regard to the effectiveness of the ISPC manipulation. Why do RTs and error rates under MC congruent condition in Phase 2 seem to be worse than MI congruent? Could there be other factors at play here, e.g. order effect? How does this potentially affect the neural analyses where trials from different phases were combined? Also, the manuscript does not mention whether there is counterbalancing for the color groups across participants, so far as I can tell.

Author response:

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

eLife Assessment

This useful study uses creative scalp EEG decoding methods to attempt to demonstrate that two forms of learned associations in a Stroop task are dissociable, despite sharing similar temporal dynamics. However, the evidence supporting the conclusions is incomplete due to concerns with the experimental design and methodology. This paper would be of interest to researchers studying cognitive control and adaptive behavior, if the concerns raised in the reviews can be addressed satisfactorily.

We thank the editors and the reviewers for their positive assessment of our work and for providing us with an opportunity to strengthen this manuscript. Please see below our responses to each comment raised in the reviews.

Public Reviews:

Reviewer #1 (Public review):

Summary:

This study focuses on characterizing the EEG correlates of item-specific proportion congruency effects. In particular, two types of learned associations are characterized. One being associations between stimulus features and control states (SC), and the other being stimulus features and responses (SR). Decoding methods are used to identify SC and SR correlates and to determine whether they have similar topographies and dynamics.

The results suggest SC and SR associations are simultaneously coactivated and have shared topographies, with the inference being that these associations may share a common generator.

Strengths:

Fearless, creative use of EEG decoding to test tricky hypotheses regarding latent associations. Nice idea to orthogonalize the ISPC condition (MC/MI) from stimulus features.

Thank you for acknowledging the strength in EEG decoding and design. We have addressed all your concerns raised below point by point.

Weaknesses:

(1a) I'm relatively concerned that these results may be spurious. I hope to be proven wrong, but I would suggest taking another look at a few things.

While a nice idea in principle, the ISPC manipulation seems to be quite confounded with the trial number. E.g., color-red is MI only during phase 2, and is MC primarily only during Phase 3 (since phase 1 is so sparsely represented). In my experience, EEG noise is highly structured across a session and easily exploited by decoders. Plus, behavior seems quite different between Phase 2 and Phase 3. So, it seems likely that the classes you are asking the decoder to separate are highly confounded with temporally structured noise.

I suggest thinking of how to handle this concern in a rigorous way. A compelling way to address this would be to perform "cross-phase" decoding, however I am not sure if that is possible given the design.

Thank you for raising this important issue. To test whether decoding might be confounded by temporally structured noise, we performed a control decoding analysis. As the reviewer correctly pointed out, cross-phase decoding is not possible due to the experimental design. Alternatively, to maximize temporal separation between the training and test data, we divided the EEG data in phase 2 and phase 1&3 into the first and second half chronologically. Phase 1 and 3 were combined because they share the same MC and MI assignments. We then trained the decoders on one half and tested them on the other half. Finally, we averaged the decoding results across all possible assignments of training and test data. The similar patterns (Supplementary Fig.1) observed confirmed that the decoding results are unlikely to be driven by temporally structured noise in the EEG data. The clarification has been added to page 13 of the revised manuscript.

(1b) The time courses also seem concerning. What are we to make of the SR and SC timecourses, which have aggregate decoding dynamics that look to be <1Hz?

As detailed in the response to your next comment, some new results using data without baseline correction show a narrower time window of above-chance decoding. We speculate that the remaining results of long-lasting above-chance decoding could be attributed to trials with slow responses (some responses were made near the response deadline of 1500 ms). Additionally, as shown in Figure 6a, the long-lasting above-chance decoding seems to be driven by color and congruency representations. Thus, it is also possible that the binding of color and congruency contributes to decoding. This interpretation has been added to page 17 of the revised manuscript.

(1c) Some sanity checks would be one place to start. Time courses were baselined, but this is often not necessary with decoding; it can cause bias (10.1016/j.jneumeth.2021.109080), and can mask deeper issues. What do things look like when not baselined? Can variables be decoded when they should not be decoded? What does cross-temporal decoding look like - everything stable across all times, etc.?

As the reviewer mentioned, baseline-corrected data may introduce bias to the decoding results. Thus, we cited the van Driel et al (2021) paper in the revised manuscript to justify the use of EEG data without baseline-correction in decoding analysis (Page 27 of the revised manuscript), and re-ran all decoding analysis accordingly. The new results revealed largely similar results (Fig. 2, 4, 6 and 8 in the revised manuscript) with the following exceptions: narrower time window for separatable SC subspace and SR subspace (Fig. 4b), narrower time window for concurrent representations of SC and SR (Fig. 6a-b), and wider time window for the correlations of SC/SR representations with RTs (Fig. 8).

(2) The nature of the shared features between SR and SC subspaces is unclear.

The simulation is framed in terms of the amount of overlap, revealing the number of shared dimensions between subspaces. In reality, it seems like it's closer to 'proportion of volume shared', i.e., a small number of dominant dimensions could drive a large degree of alignment between subspaces.

What features drive the similarity? What features drive the distinctions between SR and SC? Aside from the temporal confounds I mentioned above, is it possible that some low-dimensional feature, like EEG congruency effect (e.g., low-D ERPs associated with conflict), or RT dynamics, drives discriminability among these classes? It seems plausible to me - all one would need is non-homogeneity in the size of the congruency effect across different items (subject-level idiosyncracies could contribute: 10.1016/j.neuroimage.2013.03.039).

Thank you for this question. To test what dimensions are shared between SC and SR subspaces, we first identify which factors can be shared across SC and SR subspaces. For SC, the eight conditions are the four colors × ISPC. Thus, the possible shared dimensions are color and ISPC. Additionally, because the four colors and words are divided into two groups (e.g., red-blue and green-yellow, counterbalanced across subjects, see Methods), the group is a third potential shared dimension. Similarly, for SR decoders, potential shared dimensions are word, ISPC and group. Note that each class in SC and SR decoders has both congruent and incongruent trials. Thus, congruency is not decodable from SC/SR decoders and hence unlikely to be a shared dimension in our analysis. To test the effect of sharing for each of the potential dimensions, we performed RSA on decoding results of the SC decoder trained on SR subspace (SR | SC) (Supplementary Fig. 4a) and the SR decoder trained on SC subspace (SC | SR) (Supplementary Fig. 4b), where the decoders indicated the decoding accuracy of shared SC and SR representations. In the SC classes of SR | SC, word red and blue were mixed within the same class, same were word yellow and green. The similarity matrix for “Group” of SR | SC (Supplementary Fig. 4a) shows the comparison between two word groups (red & blue vs. yellow & green). The similarity matrix for “Group” of SC | SR (Supplementary Fig. 4b) shows the comparison between two color groups (red & blue vs. yellow & green).

The RSA results revealed that the contributions of group to the SC decoder (Supplementary Fig. 5a) and the SR decoder (Supplementary Fig. 5b) were significant. Meanwhile, a wider time window showed significant effect of color on the SC decoder (approximately 100 - 1100 ms post-stimulus onset, Supplementary Fig. 5a) and a narrower time window showed significant effect of word on SR decoder (approximately 100 - 500 ms post-stimulus onset, Supplementary Fig. 5b). However, we found no significant effect of ISPC on either SC or SR decoders. We also performed the same analyses on response-locked data from the time window -800 to 200 ms. The results showed shared representation of color in the SC decoder (Supplementary Fig. 5c) and group in both decoders (Supplementary Fig. 5c-d). Overall, the above results demonstrated that color, word and group information are shared between SC and SR subspaces.

Lastly, we would like to stress that our main hypothesis for the cross-subspace decoding analysis is that SR and SC subspaces are not identical. This hypothesis was supported by lower decoding accuracy for cross-subspace than within-subspace decoders and enables following analyses that treated SC and SR as separate representations.

We have added the interpretation to page 13-14 of the revised manuscript.

(3) The time-resolved within-trial correlation of RSA betas is a cool idea, but I am concerned it is biased. Estimating correlations among different coefficients from the same GLM design matrix is, in general, biased, i.e., when the regressors are non-orthogonal. This bias comes from the expected covariance of the betas and is discussed in detail here (10.1371/journal.pcbi.1006299). In short, correlations could be inflated due to a combination of the design matrix and the structure of the noise. The most established solution, to cross-validate across different GLM estimations, is unfortunately not available here. I would suggest that the authors think of ways to handle this issue.

Thank you for raising this important issue. Because the bias comes from the covariance between the regressors and the same GLM was applied to all time points in our analysis, we assume that the inflation would be similar at different time points. Therefore, we calculated the correlation of SC and SR betas ranging from -200 to 0 ms relative to stimulus onset as a baseline (i.e., no SC or SR representation is expected before the stimulus onset) and compared the post-stimulus onset correlation coefficients against this baseline. We hypothesized that if the positively within-trial correlation of SC and SR betas resulted from the simultaneous representation instead of inflation, we should observe significantly higher correlation when compared with the baseline. To examine this hypothesis, we first performed the linear discriminant analysis (Supplementary Fig. 7a) and RSA regression (Supplementary Fig. 7b) on the -200 - 0 ms window relative to stimulus onset. We then calculated the average r_baseline of SC and SR betas on that time window for each participant (group results at each time point are shown in Supplementary Fig. 7c) and computed the relative correlation at each post-stimulus onset time point using (fisher-z (r) - fisher-z (r_baseline)). Finally, we performed a simple t test at the group level on baseline-corrected correlation coefficients with Bonferroni correction. The results (Fig. 6c) showed significantly more positive correlation from 100 - 500 ms post-stimulus onset compared with baseline, supporting our hypothesis that the positive within-trial correlation of SC and SR betas arise from simultaneous representation rather than inflation. The related interpretation was added to page 17 of the revised manuscript.

(4) Are results robust to running response-locked analyses? Especially the EEG-behavior correlation. Could this be driven by different RTs across trials & trial-types? I.e., at 400 ms poststim onset, some trials would be near or at RT/action execution, while others may not be nearly as close, and so EEG features would differ & "predict" RT.

Thanks for this question. We now pair each of the stimulus-locked EEG analysis in the manuscript with response-locked analysis. To control for RT variations among trial types, when using the linear mixed model (LMM) to predict RTs from trial-wise RSA results, we included a separate intercept for each of the eight trial types in SC or SR. Furthermore, at each time point, we only included trials that have not generated a response (for stimulus-locked analysis) or already started (for response-locked analysis). All the results (Fig. 3, 5, 7, 9 in the revised manuscript) are in support of our hypothesis. We added these detailed to page 31 of the revised manuscript.

(5) I suggest providing more explanation about the logic of the subspace decoding method - what trialtypes exactly constitute the different classes, why we would expect this method to capture something useful regarding ISPC, & what this something might be. I felt that the first paragraph of the results breezes by a lot of important logic.

In general, this paper does not seem to be written for readers who are unfamiliar with this particular topic area. If authors think this is undesirable, I would suggest altering the text.

To improve clarity, we revised the first paragraph of the SC and SR association subspace analysis to list the conditions for each of the SC and SR decoders and explain more about how the concept of being separatable can be tested by cross-decoding between SC and SR subspaces. The revised paragraph now reads:

“Prior to testing whether controlled and non-controlled associations were represented simultaneously, we first tested whether the two representations were separable in the EEG data.

In other words, we reorganized the 16 experimental conditions into 8 conditions for SC (4 colors × MC/MI, while collapsing across SR levels) and SR (4 words × 2 possible responses per word, while collapsing across SC levels) associations separately. If SC and SR associations are not separable, it follows that they encode the same information, such that both SC and SR associations can be represented in the same subspace (i.e., by the same information encoded in both associations). For example, because (1) the word can be determined by the color and congruency and (2) the most-likely response can be determined by color and ISPC, the SR association (i.e., association between word and most-likely response) can in theory be represented using the same information as the SC association. On the other hand, if SC and SR associations are separable, they are expected to be represented in different subspaces (i.e., the information used to encode the two associations is different). Notably, if some, but not all, information is shared between SC and SR associations, they are still separable by the unique information encoded. In this case, the SC and SR subspaces will partially overlap but still differ in some dimensions. To summarize, whether SC and SR associations are separable is operationalized as whether the associations are represented in the same subspace of EEG data. To test this, we leveraged the subspace created by the LDA (see Methods). Briefly, to capture the subspace that best distinguishes our experimental conditions, we trained SC and SR decoders using their respective aforementioned 8 experimental conditions. We then projected the EEG data onto the decoding weights of the LDA for each of the SC and SR decoders to obtain its respective subspace. We hypothesized that if SC and SR subspaces are identical (i.e., not separable), SC/SR decoding accuracy should not differ by which subspace (SC or SR) the decoder is trained on. For example, SC decoders trained in SC subspace should show similar decoding performance as SC decoders trained in SR subspace. On the other hand, if SC and SR association representations are in different subspaces, the SC/SR subspace will not encode all information for SR/SC associations. As a result, decoding accuracy should be higher using its own subspace (e.g., decoding SC using the SC subspace) than using the other subspace (e.g., decoding SC using the SR subspace). We used cross-validation to avoid artificially higher decoding accuracy for decoders using their own subspace (see Methods).” (Page 11-12).

We also explicitly tested what information is shared between SC and SR representations (see response to comment #2). Lastly, to help the readers navigate the EEG results, we added a section “Overview of EEG analysis” to summarize the EEG analysis and their relations in the following manner:

“EEG analysis overview. We started by validating that the 16 experimental conditions (8 unique stimuli × MC/MI) were represented in the EEG data. Evidence of representation was provided by above-chance decoding of the experimental conditions (Fig. 2-3). We then examined whether the SC and SR associations were separable (i.e., whether SC and SR associations were different representations of equivalent information). As our results supported separable representations of SC and SR association (Fig. 4-5), we further estimated the temporal dynamics of each representation within a trial using RSA. This analysis revealed that the temporal dynamics of SC and SR association representations overlapped (Fig. 6a-b, Fig. 7a-b). To explore the potential reason behind the temporal overlap of the two representations, we investigated whether SC and SR associations were represented simultaneously as part of the task representation, independently from each other, or competitively/exclusively (e.g., on some trials only SC association was represented, while on other trials only SR association was represented). This was done by assessing the correlation between the strength of SC and SR representations across trials (Fig. 6c, Fig. 7c). Lastly, we tested how SC and SR representations facilitated performance (Fig.8-9).” (Page 8-9).

Minor suggestions:

(6) I'd suggest using single-trial RSA beta coefficients, not t-values, as they can be more stable (it's a t-value based on 16 observations against 9 or so regressors.... the SE can be tiny).

Thank you for your suggestion. To choose between using betas and t-values, we calculate the proportion of outliers (defined as values beyond mean ± 5 SD) for each predictor of the design matrix and each subject. We found that outliers were less frequent for t-values than for beta coefficients (t-values: mean = 0.07%, SD = 0.009%; beta-values: mean = 0.19%, SD = 0.033%). Thus, we decided to stay with t-values.

(7) Instead of prewhitening the RTs before the HLM with drift terms, try putting those in the HLM itself, to avoid two-stage regression bias.

Thank you for your suggestion. Because our current LMM included each of the eight trial types in SC or SR as separate predictors with their own intercepts (as mentioned above), adding regressors of trial number and mini blocks (1-100 blocks) introduced collinearity (as ISPC flipped during the experiment). We therefore excluded these regressors from the current LMM (Page 31).

(8) The text says classical MDS was performed on decoding *accuracy* - is this accurate?

We now clarify in the manuscript that it is the decoders’ probabilistic classification results (Page 28).

(9) At a few points, it was claimed that a negative correlation between SC and SR would be expected within single trials, if the two were temporally dissociable. Wouldn't it also be possible that they are not correlated/orthogonal?

We agree with the reviewer and revised the null hypothesis in the cross-trial correlation analysis to include no correlation as SC and SR association representations may be independent from each other (Page 17, 22).

Reviewer #2 (Public review):

Summary:

In this EEG study, Huang et al. investigated the relative contribution of two accounts to the process of conflict control, namely the stimulus-control association (SC), which refers to the phenomenon that the ratio of congruent vs. incongruent trials affects the overall control demands, and the stimulus-response association (SR), stating that the frequency of stimulusresponse pairings can also impact the level of control. The authors extended the Stroop task with novel manipulation of item congruencies across blocks in order to test whether both types of information are encoded and related to behaviour. Using decoding and RSA, they showed that the SC and SR representations were concurrently present in voltage signals, and they also positively co-varied. In addition, the variability in both of their strengths was predictive of reaction time. In general, the experiment has a solid design, but there are some confounding factors in the analyses that should be addressed to provide strong support for the conclusions.

Strengths:

(1) The authors used an interesting task design that extended the classic Stroop paradigm and is potentially effective in teasing apart the relative contribution of the two different accounts regarding item-specific proportion congruency effect, provided that some confounds are addressed.

(2) Linking the strength of RSA scores with behavioural measures is critical to demonstrating the functional significance of the task representations in question.

Thank you for your positive feedback. We hope our responses below address your concerns.

Weakness:

(1) While the use of RSA to model the decoding strength vector is a fitting choice, looking at the RDMs in Figure 7, it seems that SC, SR, ISPC, and Identity matrices are all somewhat correlated. I wouldn't be surprised if some correlations would be quite high if they were reported. Total orthogonality is, of course, impossible depending on the hypothesis, but from experience, having highly covaried predictors in a regression can lead to unexpected results, such as artificially boosting the significance of one predictor in one direction, and the other one to the opposite direction. Perhaps some efforts to address how stable the timed-resolved RSA correlations for SC and SR are with and without the other highly correlated predictors will be valuable to raising confidence in the findings.

Thank you for this important point. The results of proportion of variability explained shown in the Author response table 1 below, indicated relatively higher correlation of SC/SR with Color and Identity. We agree that it is impossible to fully orthogonalize them. To address the issue of collinearity, we performed a control RSA by removing predictors highly correlated with others. Specifically, we calculated the variance inflation factor (VIF) for each predictor. The Identity predictor had a high VIF of 5 and was removed from the RSA. All other predictors had VIFs < 4 and were kept in the RSA. The results (Supplementary Fig. 6) showed patterns similar to the results with the Identity predictor, suggesting that the findings are not significantly influenced by collinearity. We have added the interpretation to page 17 of the revised manuscript.

Author response table 1.

Proportion of variability explained (r²) of RSA predictors.

(2) In "task overview", SR is defined as the word-response pair; however, in the Methods, lines 495-496, the definition changed to "the pairing between word and ISPC" which is in accordance with the values in the RDMs (e.g., mccbb and mcirb have similarity of 1, but they are linked to different responses, so should they not be considered different in terms of SR?). This needs clarification as they have very different implications for the task design and interpretation of results, e.g., how correlated the SC and SR manipulations were.

Thank you for pointing out this important issue with how our operationalization captures the concept in questions. In the revised manuscript, we clarified the stimulus-response (SR) association is the link between the word and the most-likely response (i.e., not necessarily the actual response on the current trial). This association is likely to be encoded based on statistical learning over several trials. On each trial, the association is updated based on the stimulus and the actual response. Over multiple trials, the accumulated association will be driven towards the most-common (i.e., most-likely) response. In our ISPC manipulation, a color is presented in mostly congruent/incongruent (MC/MI) trials, which will also pair a word with a most-likely response. For example, if the color blue is MC, the color blue, which leads to the response blue, will co-occur with the word blue with high frequency. In other words, the SR association here is between the word blue and the response blue. As the actual response is not part of the SR association, in the RDM two trial types with different responses may share the same SR association, as long as they share the same word and the same ISPC manipulation, which, by the logic above, will produce the same most-likely response. These clarifications have been added to page 4 and 29 of the revised manuscript.

In the revised manuscript (Page 17), we addressed how much the correlated SC and SR predictors in the RDM could affect the correlation analysis between SC and SR association representation strength. Specifically, we conducted the RSA using the same GLM on EEG data prior to stimulus onset (Supplementary Fig. 7a-b). As no SC and SR associations are expected to be present before stimulus onset, the correlation between SC and SR representation would serve as a baseline of inflation due to correlated predictors in the GLM (Supplementary Fig. 7c, also see comment #3 of R1). The SC-SR correlation coefficients following stimulus onset was then compared to the baseline to control for potential inflation (Fig. 6c). Significantly above-baseline correlation was still observed between ~100-500 ms post-stimulus onset, providing support for the hypothesis that SC and SR are encoded in the same task representation.

Minor suggestions:

(3) Overall, I find that calling SC-controlled and SR-uncontrolled representations unwarranted. How is the level controlledness defined? Both are essentially types of statistical expectation that provide contextual information for the block of tasks. Is one really more automatic and requires less conscious processing than the other? More background/justification could be provided if the authors would like to use these terms.

Following your advice, we have added more discussion on how controlledness is conceptualized in this work and in the literature, which reads:

“We consider SC and SR as controlled and uncontrolled respectively based on the literature investigating the mechanism of ISPC effect. The SC account posits that the ISPC effect results from conflict and involves conflict adaptation, which requires the regulation of attention or control (Bugg & Hutchison, 2013; Bugg et al., 2011; Schmidt, 2018; Schmidt & Besner, 2008). On the other hand, the SR account argues that ISPC effect does not require conflict adaptation but instead reflects contingency leaning. That is, the response can be directly retrieved from the association between the stimulus and the most-likely response without top-down regulation of attention or control. As more empirical evidence emerged, researchers advocating control view began to acknowledge the role of associative learning in cognitive control regarding the ISPC effect (Abrahamse et al., 2016). SC association has been thought to include both automatic that is fast and resource saving and controlled processes that is flexible and generalizable (Chiu, 2019). Overall, we do not intend to claim that SC is entirely controlled or SR is completely automatic. We use SC-controlled and SR-uncontrolled representations to align with the original theoretical motivation and to highlight the conceptual difference between SC and SR associations.” (Page 24-25)

(4) Figures 3c and d: the figures could benefit from more explanation of what they try to show to the readers. Also for 3d, the dimensions were aligned with color sets and congruencies, but word identities were not linearly separable, at least for the first 3 axes. Shouldn't one expect that words can be decoded in the SR subspace if word-response pairs were decodable (e.g., Figure 3b)?

Thank you for the insightful observation. We now clarified that Fig. 3c and d in the original manuscript (Fig. 4c and d in the current manuscript) aim to show how each of the 8 trial types in the SC and SR subspaces are represented. The MDS approach we used for visualization tries to preserve dissimilarity between trial types when projecting from data from a high dimensional to a low dimensional space. However, such projection may also make patterns linearly separatable in high dimensional space not linearly separatable in low dimensional space. For example, if the word blue has two points (-1, -1) and (1, 1) and the word red has two points (-1, 1) and (1, -1), they are not linearly separatable in the 2D space. Yet, if they are projected from a 3D space with coordinates of (-1, -1, -0.1), (1, 1, -0.1), (-1, 1, 0.1) and (1, -1, 0.1), the two words can be linearly separatable using the 3^rd dimension. Thus, a better way to test whether word can be linearly separated in SR subspace is to perform RSA on the original high dimensional space. We performed the RSA with word (Supplementary Fig. 2) on the SR decoder trained on the SR subspace. Note that in Fig. 3c and d of the original script (Fig. 4c and d in the current manuscript) there are two pairs of words that are not linearly separable: red-blue and yellow-green. Thus, we specifically tested the separability within the two pairs using the one predictor for each pair, as shown in Supplementary Fig. 2. The results showed that within both word pairs individual words were presented above chance level (Supplementary Fig. 3). Considering that the decoders are linear, this finding indicates linear separability of the word pairs in the original SR subspace. The clarification has been added to page 13 (the end of the second paragraph) of the revised manuscript.

References

Abrahamse, E., Braem, S., Notebaert, W., & Verguts, T. (2016). Grounding cognitive control in associative learning. Psychological Bulletin, 142(7), 693-728.doi:10.1037/bul0000047.

Bugg, J. M., & Hutchison, K. A. (2013). Converging evidence for control of color-word Stroop interference at the item level. Journal of Experimental Psychology:Human Perception and Performance, 39(2), 433-449. doi:10.1037/a0029145.

Bugg, J. M., Jacoby, L. L., & Chanani, S. (2011). Why it is too early to lose control in accounts of item-specific proportion congruency effects. Journal of Experimental Psychology: Human Perception and Performance, 37(3), 844-859. doi:10.1037/a0019957.

Chiu, Y.-C. (2019). Automating adaptive control with item-specific learning. In Psychology of Learning and Motivation (Vol. 71, pp. 1-37).

Schmidt, J. R. (2018). Evidence against conflict monitoring and adaptation: An updated review. Psychonomic Bulletin & Review, 26(3), 753-771. doi:10.3758/s13423018-1520-z.

Schmidt, J. R., & Besner, D. (2008). The Stroop effect: Why proportion congruent has nothing to do with congruency and everything to do with contingency. Journal of Experimental Psychology: Learning, Memory, and Cognition, 34(3), 514-523. doi:10.1037/0278-7393.34.3.514.

eLife Assessment

Stearns and Poletti present a technically impressive study that aims to uncover a deeper understanding of microsaccade function: their role in perceptual modulation and the associated temporal dynamics. The question is useful, and advances prior work by adding temporal granularity. However, the strength of the evidence is currently incomplete. Additional analysis is needed to control for the effects of endogenous attention and to demonstrate changes in perceptual performance.

Reviewer #1 (Public review):

Summary:

Using high-precision eyetracking, the authors measure foveolar sensitivity modulations before, during, and after instructed microsaccades to a centrally cued orientation stimulus.

Strengths:

The article is clearly written, and the stimulus presentation method is sophisticated and well-established. The data provide interesting insights that will be useful for comparisons between trans-saccadic and trans-microsaccadic sensitivity modulations.

Weaknesses:

Nonetheless, I have major concerns regarding the interpretation of the measured time courses (in particular, inconsistencies in distinguishing enhancement from suppression), the attempt to disentangle these effects from endogenous attention shifts, and the overstatement of the findings' novelty.

(1) Overstatement of novelty

The authors motivate their study by stating that "the temporal dynamics of these pre-microsaccadic modulations remain unknown" (l. 55-56). However, Shelchkova & Poletti (2020) already report a microsaccade-aligned sensitivity time course. I understand that the present study uses shorter target durations and thus provides a more resolved estimate. Nonetheless, a fairer characterization of the study's novelty would be that observers' discrimination performance is continuously measured across the pre-, intra-, and post-movement interval, within the same observers and experimental design. Relatedly, the authors state that it is unclear whether pre-microsaccadic sensitivity modulations reflect "suppression at the non-foveated location, enhancement at the microsaccade target, or both" (l. 70). Guzhang et al. (2024) examined the spatial spread of pre-microsaccadic sensitivity modulations by measuring performance at the PRL, the movement target, and several other equidistant locations. They report that "whereas fine spatial vision is enhanced at the microsaccade goal location, it drops at the very center of gaze". The current authors' reasoning seems to be that performances at locations that are neither the target nor the PRL may behave differently. Why would that be the case? If my understanding is correct, I would recommend incorporating these clarifications into the motivation paragraph, so that readers less familiar with the literature do not overestimate the novelty of the findings. Moreover, and related to point 3, I am unsure if the current analyses provide decisive evidence to distinguish enhancement from suppression, as claimed by the authors.

(2) Distinction from endogenous attention

To "rule out the possible influence of covert attention" (l. 232), the authors compute a cue-aligned in addition to the movement-aligned performance time course. A difference in alignment cannot rule out the influence of a certain mechanism; it can only dilute it. Just like endogenous attention may contribute to the movement-aligned time course, movement preparation will necessarily contribute to the cue-aligned time course, since these timelines are intrinsically correlated: as the trial progresses, observers will be in later and later stages of saccade preparation. For this and several additional reasons, an effect in the cue-aligned time course is in fact expected-and, in my view, clearly present (see below). As the authors themselves note, endogenous attention has been shown to operate within the foveola and should therefore be engaged in the present experiment in addition to movement-related attentional shifts (unless the authors believe that specific design features, e.g., stimulus timing, preclude its involvement?). Regardless of the theoretical considerations, the empirical data show a pronounced, near-linear increase in performance at the target location, with d′ doubling from approximately 1 to 2. Although the interaction between condition and time does not reach significance (p = 0.09), this result should not be taken as conclusive evidence against a plausible and perhaps expected contribution of endogenous attention. I suggest an additional analysis that could more directly address these issues. In previous work (Rolfs & Carrasco, 2012; Kroell & Rolfs, 2025; see Figure 3), the relative contributions of cue-alinged influences and pre-saccadic attention were disentangled by reweighting each data point according to its position on both the cue-locked and saccade-locked timelines. Applied to the present study, the authors could compute, for each cue-to-target offset bin, its proportional contribution to each pre-movement time bin. Microsaccade-locked sensitivities could then be reweighted based on these proportions. As a result, each movement-locked time bin would contain equal contributions from all cue-locked time bins, effectively isolating the effect of microsaccade preparation.

(3) Interpretation and analysis of the time course

(3.1) Discrimination before microsaccade onset<br /> In lines 151-153, the author state "While the enhancement at the target location did not reach significance relative to baseline, the impairment at the non-target location did", suggesting that pre-movement sensitivity advantages for information presented at the target location are due to a decrease in performance at the non-target location and not an enhancement at the target location per se. After analyzing the difference between the two locations, the authors state, "These results show that approximately 100 milliseconds before microsaccade onset, discrimination rapidly improved at the intended target location while decreasing at the non-target location." (l. 159-161). How is the statement that discrimination performance rapidly improved (which is repeated throughout the manuscript) justified by the results?

More generally, the authors may benefit from applying bootstrapping or permutation-based analyses to their data. Such approaches would, for example, allow direct comparisons between congruent and incongruent conditions at every individual time point in Figure 3B and may be more sensitive to temporally confined sensitivity variations while requiring fewer assumptions than analyses based on manually segregated temporal bins and aggregate measures. If enhancement at the target location does not reach significance even in these analyses, all corresponding statements should be removed throughout the manuscript. The term "enhancement" should then be rephrased as "detection advantage" or "relative performance benefit" to emphasize the contrast to enhancement effects classically associated with pre-saccadic attention shifts.

Relatedly, the authors state that pre-microsaccadic enhancement peaks around 70 ms before microsaccade onset, which is earlier than sensitivity enhancements preceding large-scale saccades that often increase monotonically up until movement onset. The authors suggest potential reasons for this in the Discussion, yet an additional one seems conceivable based on Figure 3B. Performances at both the cue-congruent and incongruent location decrease leading up to the movement, reaching values even below their early baselines around 100 ms and 25 ms before movement onset for the incongruent and congruent location, respectively. A spatially non-specific decline that drives sensitivities toward a common absolute minimum may thus dictate the time course of detection advantages. In other words, a spatially widespread decrease in foveolar sensitivity likely contributes to both "suppression" at the non-target location and the decrease in "enhancement" at the target location. If this general decrease is due to saccadic suppression, as the authors suggest, it appears to exert a much more pronounced influence on sensitivity modulations than it does before large-scale saccades (which is interesting). Are there other findings suggesting an increased magnitude of micro-saccadic (as compared to saccadic) suppression? Another potentially related phenomenon is the decrease in pre-saccadic foveal detection performances reported twice before (Hanning & Deubel, 2022; Kroell & Rolfs, 2022). It is possible that whatever mechanism triggers this decrease is engaged by the preparation of microsaccadic and saccadic motor programs alike. In any case, I would ask the authors to acknowledge this general decrease early on to clarify that any currently significant advantage for the target location relies on varied degrees of suppression, and not on true enhancement similar to pre-saccadic attention shifts.

Moreover, in Figure 3C, the final 25 ms before microsaccade onset are excluded from the aggregate measure, presumably since including this interval substantially reduces the effect size. Since the last 25 ms before movement onset is the interval most commonly associated with saccadic suppression, I think that this choice can be justified. Nonetheless, it should be mentioned explicitly in the main text. On a minor note, the authors state that "Performance (evaluated as percent of correct responses) was averaged within a 50 millisecond sliding window, advancing in 1 ms steps (with 24 ms overlap)". Why is the overlap not 49 ms?

(3.2) Discrimination during the microsaccade:<br /> The authors state that "in the "during" trials the target must be presented during the peak speed of the microsaccade." Since the target was presented for 50 ms and the average microsaccade duration was around 60 ms, this implies that the intra-microsaccadic condition includes many trials in which the target overlapped with the pre- or post-movement fixation interval. Were there not enough trials to isolate purely intra-microsaccadic presentations? Are the results descriptively comparable?

(4) Additional analyses

Several additional analyses could strengthen the authors' conclusions. If there are enough trials in which observers erroneously saccaded to the uncued (i.e., wrong) location, these trials could experimentally isolate the influence of pre-microsaccadic attention, assuming that endogenous attention went to the cued location. In addition, the authors speculate whether differences in saccadic and microsaccadic movement latencies may underlie the differences in perceptual time courses. The latency distributions provided in the manuscript look sufficiently broad, such that intra-individual variation could be harnessed to explore this question. Do sensitivity time courses differ before microsaccades with shorter vs. longer latencies?

(5) Clarifications regarding the design

At 50 ms, the duration of the to-be discriminated stimulus, although shorter than in previous investigations, is still rather long. What is the reason for this? I would encourage the authors to state in the main text that the duration of the analyzed/plotted time bins is often shorter than the stimulus duration (i.e., there is some overlap between bins that likely introduces smoothing). In Figure 3A, it would be helpful to plot raw data points computed from non-overlapping bins on top of the moving-window estimates, to allow readers to assess the degree of smoothing and potential temporal delays introduced by this analysis. Moreover, I wonder whether the abrupt onset of the target unmasked by flickering noise masks might induce saccadic inhibition, which would manifest as a transient dip in saccade execution probability. The distributions shown in Figure 2B appear too smoothed or fitted to clearly reveal such a dip. How exactly are all distributions in the manuscript computed (e.g., binning, smoothing, fitting procedures)? Finally, on a minor note, explicitly stating on line 105 that two different orientations can be presented at the cued and non-cued location would help avoid potential confusion.

Reviewer #2 (Public review):

Summary and overall evaluation:

The authors assessed how visual discrimination of stimuli in the foveola changes before, during, and after small instructed eye movements (in the "micro" range). Consistent with (and advancing) related prior work, their main finding regards a pre-saccadic modulation of visual performance at the saccade target vs. the opposite location. This pre-saccadic modulation in foveal vision peaks ~70 ms prior to the instructed small saccade.

Strengths:

The study uses an impressive, technically advanced set-up and zooms in on peri-saccadic modulations in visual acuity at the micro scale. The findings build on related prior findings from the literature on smaller and larger eye movements and add temporal granularity over prior work from the same lab. The writing is easy to follow, and the figures are clear.

Weaknesses:

At the same time, the findings remain relatively empirical in nature and do not profoundly advance theoretical understanding beyond adding valuable granularity to existing knowledge. Relevant prior literature could be better introduced and acknowledged. In addition, there remain concerns regarding potential cue-driven attentional influences that may confound the reported effects (leaving the possibility that the reported effects may be related to cue-driven attention, rather than saccade planning/execution per se). There are also some issues regarding specific statistical inferences. I detail these points below.

Major Points:

(1) Novelty framing and introduction of relevant prior literature

At times, this study is introduced as if no prior study explored the time course of changes in visual perception surrounding small (micro) saccades. Yet, it appears that a prior study from the same lab, using a very similar task, already showed a time course (Figure 5 in Shelchkova & Poletti, 2020). While this study is discussed in the introduction, it is not mentioned that at least some pre-saccade time course was already reported there, albeit a more crude one than the one in the current article. Moreover, the 2013 study by Hafed also specifically looked at "peri-microsaccade modulation in visual perception" and also already showed a temporal modulation that peaked ~50 ms before microsaccade onset. I appreciate how the current study differs in a number of ways (focusing on visual acuity in the foveola), but I was nevertheless surprised to see the first reference to this relevant prior finding in the discussion (and without any elaboration). Though more recent, the same could be argued for the 2025 study by Bouhnik et al. on pre-microsaccade modulations in visual processing in V1, which, like the Hafed study, is first mentioned only in the discussion. Perhaps these studies could be introduced in the paragraph starting at line 48, or in the next paragraph, to do better justice to the existing literature on this topic when motivating the study. This would likely also help to better point out the major advances provided by the current study.

Relatedly, in Shelchkova & Poletti (PNAS, 2020), an apparently similar congruency effect on performance was reported >200 ms milliseconds before saccade onset, as evident from Fig 5 in that article. How should readers rhyme this with the current findings? Ideally, the authors would not only acknowledge that such a time course was already reported previously, but also discuss the discrepancies between these findings further: why may the performance effects appear much earlier in this prior study compared to in the current study, where the congruency effect emerges only ~100 ms prior to the instructed small saccade?

(2) Saccade- or cue-driven? (assumption that attention is unaltered in failed saccade trials)

Because the authors used a cue to instruct saccade direction, it remains a possibility that the reported modulations in visual performance may be driven directly by the spatial cue (cue-related attentional allocation), rather than the instructed small saccade per se. While the authors are clearly aware of this potential confound, questions remain regarding the convincingness of the presented control analyses. In my view, a more compelling control would require an additional experiment.

The central argument against a cue-locked (purely attentional) modulation is the absence of a performance modulation in so-called "failed" saccade trials. However, a key assumption here is that putative cue-driven attention was unaltered in these trials. This is never verified and, in my opinion, highly unlikely. Rather, trials with failed microsaccades could very well be the result of failing to process the cue in the first place (indeed, if the task is to make a saccade to the cue, failure to make a saccade equates failure to perform the task). In such trials, any putative cue-driven influences over spatial attention would also be expected to be substantially reduced. Accordingly, just because failed saccade trials show little performance modulation does not rule out cue-driven attention effects, because attention may also have "failed" in these failed saccade trials. The control for potential cue-driven attention effects would be more convincing if the authors included a condition with the same cues, where participants are simply not instructed to make any saccades to the cues. Unfortunately, such an experimental condition appears not to have been included here. The author may still consider adding such a control experiment.

Another argument against a cue-driven effect is that the authors found no interaction with time in the cue-locked data, whereas they did find such an interaction in the saccade-locked data. However, the lack of significance in the cue-locked data but significance in the saccade-locked data is not strong evidence against a cue-driven influence. Statistically, there is no direct comparison here, and more importantly, with longer delays, the cue-locked data may also start to show a dip (this could potentially be tested by the authors if they have enough trials available to extend their cue-locked analysis further in time). Indeed, exogenous attention, that may have been automatically evoked by the spatial cue, is known to be transient and to eventually even reverse after a brief initial facilitation (see e.g., Klein TiCS, 2000).

Finally, the authors consistently refer to "endogenous" attention (starting at line 221) when addressing potential cue-driven attention confounds. However, because the cue is not predictive, but is a spatial cue that differs in a bottom-up manner between left and right cues, "exogenous" attention is a more likely confound here in my view. Specifically, the spatial cue may automatically trigger attention in the direction of the target location it points to (and such exogenous effects would be expected even for unpredictive cues).

(3) Benefit and cost, or just cost?

Line 151 states that no statistically significant benefit for the saccade target was found compared to the neutral baseline. Yet, the claim throughout the article is distinct, such as in line 159: "These results show that approximately 100 milliseconds before microsaccade onset, discrimination rapidly improved at the intended target location". I do not question the robustness of the congruency effect, but the authors should be more careful when inferring "improved" perception at the target location because, as far as I could tell (as well as in the authors' own writing in line 151), this is not substantiated statistically when compared to the neutral baseline.

Related to this point, in Figure 3B, it would be informative to also see the average performance in the neutral cue condition (for example, as a straight line as in some other figures). This would help to better appreciate the relative benefits and/or costs compared to the neutral condition, also in the time-resolved data.

(4) Statistical inference for the comparison between failed and non-failed trials

Currently, the lack of modulation in the failed saccade trials hinges on a null effect. It would be stronger to support the claims with a significant difference in the congruency effect between failed and non-failed trials. Indeed, lack of significance in failed saccade trials does by itself not constitute valid evidence that the congruency effect is larger in saccade compared to failed saccade trials. For this, a significant interaction between saccade-trial-type (failed/non-failed) and congruency (congruent/incongruent) should be established (see e.g., Nieuwenhuis et al., Nat Neurosci, 2011).

(5) Time window justification

While the authors nicely depict their data across the full time axis, all statistics are currently performed on data extracted from specific time windows. How exactly were these time windows determined and justified? Likewise, how were the specific times picked for visualizing and statistically quantifying the data in e.g., Figures 3D and E? It would be reassuring to add justification for these specific time windows and/or to verify (using follow-up analyses) that the presented results are robust when different time windows are chosen.

(6) Microsaccade definition

Microsaccades are explicitly defined as being below half a degree. This appears rather arbitrary and rigid. Does the size of saccades not ultimately depend on the task and stimulus (e.g., Otero-Millan et al., PNAS, 2013) rather than being a fixed biological property? Perhaps this could be stated less rigidly, such as by stating how microsaccades are often observed below 0.5 degrees.

(Relatedly, one may wonder whether the type of instructed saccades that the authors studied here involves the same type of eye movements as the type of fixational microsaccades that have been the focus of ample prior studies. However, I recognize that this specific reflection may open a debate that is beyond the scope of this article.

eLife Assessment

This important study identifies a novel role for Hes5+ astrocytes in modulating the activity of descending pain-inhibitory noradrenergic neurons from the locus coeruleus during stress-induced pain facilitation. The role of glia in modulating neurological circuits including pain is poorly understood, and in that light, the role of Hes5+ astrocytes in this circuit is a key finding with broader potential impacts. This work is supported by convincing evidence, albeit somewhat limited by the indirect nature of the evidence linking adenosine to nearby neuronal modulation, and possible questions on the population specificity of the transgenic approach.

Reviewer #1 (Public review):

Review of the revised submission:

I thank the authors for their detailed consideration of my comments and for the additional data, analyses, and clarifications they have incorporated. The new behavioral experiments, quantification of targeted manipulations, and expanded methodological details strengthen the manuscript and address many of my initial concerns. While some questions remain for future work, the authors' careful responses and the additional evidence provided help resolve the main issues I raised, and I am generally satisfied with the revisions.

Review of original submission:

Summary

In this article, Kawanabe-Kobayashi et al., aim to examine the mechanisms by which stress can modulate pain in mice. They focus on the contribution of noradrenergic neurons (NA) of the locus coeruleus (LC). The authors use acute restraint stress as a stress paradigm and found that following one hour of restraint stress mice display mechanical hypersensitivity. They show that restraint stress causes the activation of LC NA neurons and the release of NA in the spinal cord dorsal horn (SDH). They then examine the spinal mechanisms by which LC→SDH NA produces mechanical hypersensitivity. The authors provide evidence that NA can act on alphaA1Rs expressed by a class of astrocytes defined by the expression of Hes (Hes+). Furthermore, they found that NA, presumably through astrocytic release of ATP following NA action on alphaA1Rs Hes+ astrocytes, can cause an adenosine-mediated inhibition of SDH inhibitory interneurons. They propose that this disinhibition mechanism could explain how restraint stress can cause the mechanical hypersensitivity they measured in their behavioral experiments.

Strengths:

(1) Significance. Stress profoundly influences pain perception; resolving the mechanisms by which stress alters nociception in rodents may explain the well-known phenomenon of stress-induced analgesia and/or facilitate the development of therapies to mitigate the negative consequences of chronic stress on chronic pain.

(2) Novelty. The authors' findings reveal a crucial contribution of Hes+ spinal astrocytes in the modulation of pain thresholds during stress.

(3) Techniques. This study combines multiple approaches to dissect circuit, cellular, and molecular mechanisms including optical recordings of neural and astrocytic Ca2+ activity in behaving mice, intersectional genetic strategies, cell ablation, optogenetics, chemogenetics, CRISPR-based gene knockdown, slice electrophysiology, and behavior.

Weaknesses:

(1) Mouse model of stress. Although chronic stress can increase sensitivity to somatosensory stimuli and contribute to hyperalgesia and anhedonia, particularly in the context of chronic pain states, acute stress is well known to produce analgesia in humans and rodents. The experimental design used by the authors consists of a single one-hour session of restraint stress followed by 30 min to one hour of habituation and measurement of cutaneous mechanical sensitivity with von Frey filaments. This acute stress behavioral paradigm corresponds to the conditions in which the clinical phenomenon of stress-induced analgesia is observed in humans, as well as in animal models. Surprisingly, however, the authors measured that this acute stressor produced hypersensitivity rather than antinociception. This discrepancy is significant and requires further investigation.

(2) Specifically, is the hypersensitivity to mechanical stimulation also observed in response to heat or cold on a hotplate or coldplate?

(3) Using other stress models, such as a forced swim, do the authors also observe acute stress-induced hypersensitivity instead of stress-induced antinociception?

(4) Measurement of stress hormones in blood would provide an objective measure of the stress of the animals.

(5) Results:

(a) Optical recordings of Ca2+ activity in behaving rodents are particularly useful to investigate the relationship between Ca2+ dynamics and the behaviors displayed by rodents.

(b) The authors report an increase in Ca2+ events in LC NA neurons during restraint stress: Did mice display specific behaviors at the time these Ca2+ events were observed such as movements to escape or orofacial behaviors including head movements or whisking?

(c) Additionally, are similar increases in Ca2+ events in LC NA neurons observed during other stressful behavioral paradigms versus non-stressful paradigms?

(d) Neuronal ablation to reveal the function of a cell population.

(e) The proportion of LC NA neurons and LC→SDH NA neurons expressing DTR-GFP and ablated should be quantified (Figures 1G and J) to validate the methods and permit interpretation of the behavioral data (Figures 1H and K). Importantly, the nocifensive responses and behavior of these mice in other pain assays in the absence of stress (e.g., hotplate) and a few standard assays (open field, rotarod, elevated plus maze) would help determine the consequences of cell ablation on processing of nociceptive information and general behavior.

(f) Confirmation of LC NA neuron function with other methods that alter neuronal excitability or neurotransmission instead of destroying the circuit investigated, such as chemogenetics or chemogenetics, would greatly strengthen the findings. Optogenetics is used in Figure 1M, N but excitation of LC→SDH NA neuron terminals is tested instead of inhibition (to mimic ablation), and in naïve mice instead of stressed mice.

(g) Alpha1Ars. The authors noted that "Adra1a mRNA is also expressed in INs in the SDH".

(h) The authors should comprehensively indicate what other cell types present in the spinal cord and neurons projecting to the spinal cord express alpha1Ars and what is the relative expression level of alpha1Ars in these different cell types.

(i) The conditional KO of alpha1Ars specifically in Hes5+ astrocytes and not in other cell types expressing alpha1Ars should be quantified and validated (Figure 2H).

(j) Depolarization of SDH inhibitory interneurons by NA (Figure 3). The authors' bath applied NA, which presumably activates all NA receptors present in the preparation.

k) The authors' model (Figure 4H) implies that NA released by LC→SDH NA neurons leads to the inhibition of SDH inhibitory interneurons by NA. In other experiments (Figure 1L, Figure 2A), the authors used optogenetics to promote the release of endogenous NA in SDH by LC→SDH NA neurons. This approach would investigate the function of NA endogenously released by LC NA neurons at presynaptic terminals in the SDH and at physiological concentrations and would test the model more convincingly compared to the bath application of NA.

(l) As for other experiments, the proportion of Hes+ astrocytes that express hM3Dq, and the absence of expression in other cells, should be quantified and validated to interpret behavioral data.

(m) Showing that the effect of CNO is dose-dependent would strengthen the authors' findings.

(n) The proportion of SG neurons for which CNO bath application resulted in a reduction in recorded sIPSCs is not clear.

(o) A1Rs. The specific expression of Cas9 and guide RNAs, and the specific KD of A1Rs, in inhibitory interneurons but not in other cell types expressing A1Rs should be quantified and validated.

(6) Methods:

It is unclear how fiber photometry is performed using "optic cannula" during restraint stress while mice are in a 50ml falcon tube (as shown in Figure 1A).

Reviewer #2 (Public review):

Summary:

This study investigates the role of spinal astrocytes in mediating stress-induced pain hypersensitivity, focusing on the LC (locus coeruleus)-to-SDH (spinal dorsal horn) circuit and its mechanisms. The authors aimed to delineate how LC activity contributes to spinal astrocytic activation under stress conditions, explore the role of noradrenaline (NA) signaling in this process, and identify the downstream astrocytic mechanisms that influence pain hypersensitivity.

The authors provide strong evidence that 1-hour restraint stress-induced pain hypersensitivity involves the LC-to-SDH circuit, where NA triggers astrocytic calcium activity via alpha1a adrenoceptors (alpha1aRs). Blockade of alpha1aRs on astrocytes-but not on Vgat-positive SDH neurons-reduced stress-induced pain hypersensitivity. These findings are rigorously supported by well-established behavioral models and advanced genetic techniques, uncovering the critical role of spinal astrocytes in modulating stress-induced pain.

However, the study's third aim-to establish a pathway from astrocyte alpha1aRs to adenosine-mediated inhibition of SDH-Vgat neurons-is less compelling. While pharmacological and behavioral evidence is intriguing, the ex vivo findings are indirect and lack a clear connection to the stress-induced pain model. Despite these limitations, the study advances our understanding of astrocyte-neuron interactions in stress-pain contexts and provides a strong foundation for future research into glial mechanisms in pain hypersensitivity.

Strengths:

The study is built on a robust experimental design using a validated 1-hour restraint stress model, providing a reliable framework to investigate stress-induced pain hypersensitivity. The authors utilized advanced genetic tools, including retrograde AAVs, optogenetics, chemogenetics, and subpopulation-specific knockouts, allowing precise manipulation and interrogation of the LC-SDH circuit and astrocytic roles in pain modulation. Clear evidence demonstrates that NA triggers astrocytic calcium activity via alpha1aRs, and blocking these receptors effectively reduces stress-induced pain hypersensitivity.

Weaknesses:

The study offers mainly indirect evidence for astrocyte-released adenosine acting on SDH-VGAT neurons. The potential contributions of astrocyte-derived D-serine and adenosine to different spinal neuron subtypes, as well as the transient "dip" in astrocytic calcium following LC optostimulation, merit further clarification in future work once appropriate tools become available.

Comments on revisions:

The authors have thoroughly addressed my previous comments, resolving most of the points I raised except those noted in the "Weaknesses" section above. I understand that some of these aspects will require future tool development.

Reviewer #3 (Public review):

Summary

This is an exciting and timely study addressing the role of descending noradrenergic systems in nocifensive responses. While it is well-established that spinally released noradrenaline (aka norepinephrine) generally acts as an inhibitory factor in spinal sensory processing, this system is highly complex. Descending projections from the A6 (locus coeruleus, LC) and the A5 regions typically modulate spinal sensory processing and reduce pain behaviours, but certain subpopulations of LC neurons have been shown to mediate pronociceptive effects, such as those projecting to the prefrontal cortex (Hirshberg et al., PMID: 29027903).

The study proposes that descending cerulean noradrenergic neurons potentiate touch sensation via alpha-1 adrenoceptors on Hes5+ spinal astrocytes, contributing to mechanical hyperalgesia. This finding is consistent with prior work from the same group (dd et al., PMID:). However, caution is needed when generalising about LC projections, as the locus coeruleus is functionally diverse, with differences in targets, neurotransmitter co-release, and behavioural effects. Specifying the subpopulations of LC neurons involved would significantly enhance the impact and interpretability of the findings.

Strengths

The study employs state-of-the-art molecular, genetic, and neurophysiological methods, including precise CRISPR and optogenetic targeting, to investigate the role of Hes5+ astrocytes. This approach is elegant and highlights the often-overlooked contribution of astrocytes in spinal sensory gating. The data convincingly support the role of Hes5+ astrocytes as regulators of touch sensation, coordinated by brain-derived noradrenaline in the spinal dorsal horn, opening new avenues for research into pain and touch modulation.

Furthermore, the data support a model in which superficial dorsal horn (SDH) Hes5+ astrocytes act as non-neuronal gating cells for brain-derived noradrenergic (NA) signalling through their interaction with substantia gelatinosa inhibitory interneurons. Locally released adenosine from NA-stimulated Hes5+ astrocytes, following acute restraint stress, may suppress the function of SDH-Vgat+ inhibitory interneurons, resulting in mechanical pain hypersensitivity. However, the spatially restricted neuron-astrocyte communication underlying this mechanism requires further investigation in future studies.

Comments on revisions:

One important point remains insufficiently resolved. In Figure S4C, two of the three visible neurons in the A5 example appear to show a white "halo" at the cell border, suggesting a merge of eGFP (green) and TH (magenta) and therefore possible transgene positivity. To draw a confident conclusion about the specificity of the approach for the A6 (LC) population, the authors are kindly asked to provide high-resolution images of several representative A5 sections, presented both as merged and as separate colour channels. Ideally, quantification across multiple rostrocaudal sections of A5, A6 and A7 should be provided. This is essential for determining whether any transgene expression occurs within the A5 nucleus, particularly given its several-millimetre rostrocaudal extent. As the behavioural phenotype arises from manipulation of only a small subset of A6 neurons, ruling out any contribution from A5 (or A7) is critical for validating pathway specificity, especially in light of prior reports showing that similar approaches can label A5 fibres.

Author response:

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

Public reviews:

Reviewer #1 (Public review):

Summary:

In this article, Kawanabe-Kobayashi et al., aim to examine the mechanisms by which stress can modulate pain in mice. They focus on the contribution of noradrenergic neurons (NA) of the locus coeruleus (LC). The authors use acute restraint stress as a stress paradigm and found that following one hour of restraint stress mice display mechanical hypersensitivity. They show that restraint stress causes the activation of LC NA neurons and the release of NA in the spinal cord dorsal horn (SDH). They then examine the spinal mechanisms by which LC→SDH NA produces mechanical hypersensitivity. The authors provide evidence that NA can act on alphaA1Rs expressed by a class of astrocytes defined by the expression of Hes (Hes+). Furthermore, they found that NA, presumably through astrocytic release of ATP following NA action on alphaA1Rs Hes+ astrocytes, can cause an adenosine-mediated inhibition of SDH inhibitory interneurons. They propose that this disinhibition mechanism could explain how restraint stress can cause the mechanical hypersensitivity they measured in their behavioral experiments.

Strengths:

(1) Significance. Stress profoundly influences pain perception; resolving the mechanisms by which stress alters nociception in rodents may explain the well-known phenomenon of stress-induced analgesia and/or facilitate the development of therapies to mitigate the negative consequences of chronic stress on chronic pain.

(2) Novelty. The authors' findings reveal a crucial contribution of Hes+ spinal astrocytes in the modulation of pain thresholds during stress.

(3) Techniques. This study combines multiple approaches to dissect circuit, cellular, and molecular mechanisms including optical recordings of neural and astrocytic Ca2+ activity in behaving mice, intersectional genetic strategies, cell ablation, optogenetics, chemogenetics, CRISPR-based gene knockdown, slice electrophysiology, and behavior.

Weaknesses:

(1) Mouse model of stress. Although chronic stress can increase sensitivity to somatosensory stimuli and contribute to hyperalgesia and anhedonia, particularly in the context of chronic pain states, acute stress is well known to produce analgesia in humans and rodents. The experimental design used by the authors consists of a single one-hour session of restraint stress followed by 30 min to one hour of habituation and measurement of cutaneous mechanical sensitivity with von Frey filaments. This acute stress behavioral paradigm corresponds to the conditions in which the clinical phenomenon of stress-induced analgesia is observed in humans, as well as in animal models. Surprisingly, however, the authors measured that this acute stressor produced hypersensitivity rather than antinociception. This discrepancy is significant and requires further investigation.

We thank the reviewer for evaluating our work and for highlighting both its strengths and weaknesses. As stated by the reviewer, numerous studies have reported acute stress-induced antinociception. However, as shown in a new additional table (Table S1) in which we have summarized previously published data using the acute restraint stress model employed in our present study, most studies reporting antinociceptive effects of acute restraint stress assessed behavioral responses to heat stimuli or formalin. This observation is consistent with the findings from our previous study (Uchiyama et al., Mol Brain, 2022 (PMID: 34980215)). The present study also confirms that acute restraint stress reduces behavioral responses to noxious heat (see also our response to Comment #2 below). In contrast to the robust and consistent antinociceptive effects observed with thermal stimuli, some studies evaluating behavioral responses to mechanical stimuli have reported stress-induced hypersensitivity (see Table S1), which aligns with our current findings. Taken together, these data support our original notion that the effects of acute stress on pain-related behaviors depend on several factors, including the nature, duration, and intensity of the stressor, as well as the sensory modality assessed in behavioral tests. We have incorporated this discussion and Table S1 into the revised manuscript (lines 344-353). Furthermore, we have slightly modified the text including the title, replacing "pain facilitation" with "mechanical pain hypersensitivity" to more accurately reflect our research focus and the conclusion of this study that LC^→SDH NAergic signaling to spinal astrocytes is required for stress-induced mechanical pain hypersensitivity. Finally, while mouse models of stress could provide valuable insights, the clinical relevance of stress-induced mechanical pain hypersensitivity remains to be elucidated and requires further investigation. We hope these clarifications address your concerns.

(2) Specifically, is the hypersensitivity to mechanical stimulation also observed in response to heat or cold on a hotplate or coldplate?

Thank you for your important comment. We have now conducted additional behavioral experiments to assess responses to heat using the hot-plate test. We found that mice subjected to restraint stress did not exhibit behavioral hypersensitivity to heat stimuli; instead, they displayed antinociceptive responses (Figure S2; lines 95-98). These results are consistent with our previous findings (Uchiyama et al., Mol Brain, 2022 (PMID: 34980215)) as well as numerous other reports (Table S1).

(3) Using other stress models, such as a forced swim, do the authors also observe acute stress-induced hypersensitivity instead of stress-induced antinociception?

As suggested by the reviewer, we conducted a forced swim test. We found that mice subjected to forced swimming, which has been reported to produce analgesic effects on thermal stimuli (Contet et al., Neuropsychopharmacology, 2006 (PMID: 16237385)), did not exhibit any changes in mechanical pain hypersensitivity (Figure S2; lines 98-99). Furthermore, a previous study demonstrated that mechanical pain sensitivity is enhanced by other stress models, such as exposure to an elevated open platform for 30 min (Kawabata et al., Neuroscience, 2023 (PMID: 37211084)). However, considering our data showing that changes in mechanosensory behavior induced by restraint stress depend on the duration of exposure (Figure S1), and that restraint stress also produced an antinociceptive effect on heat stimuli (Figure S2), stress-induced modulation of pain is a complex phenomenon influenced by multiple factors, including the stress model, intensity, and duration, as well as the sensory modality used for behavioral testing (lines 100-103).

(4) Measurement of stress hormones in blood would provide an objective measure of the stress of the animals.

A previous study has demonstrated that plasma corticosterone levels—a stress hormone—are elevated following a 1-hour exposure to restraint stress in mice (Kim et al., Sci Rep, 2018 (PMID: 30104581)), using a stress protocol similar to that employed in our current study. We have included this information with citing this paper (lines 104-105).

(5) Results:

(a) Optical recordings of Ca2+ activity in behaving rodents are particularly useful to investigate the relationship between Ca2+ dynamics and the behaviors displayed by rodents.

In the optical recordings of Ca²⁺ activity in LC neurons, we monitored mouse behavior during stress exposure. We have now included a video of this in the revised manuscript (video; lines 111-114).

(b) The authors report an increase in Ca2+ events in LC NA neurons during restraint stress: Did mice display specific behaviors at the time these Ca2+ events were observed such as movements to escape or orofacial behaviors including head movements or whisking?

By reanalyzing the temporal relationship between Ca²⁺ events and mouse behavior during stress exposure, we found that the Ca²⁺ transients and escape behaviors (struggling) occurred almost simultaneously (video). A similar temporal correlation is also observed in Ca²⁺ responses in the bed nucleus of the stria terminalis (Luchsinger et al., Nat Commun, 2021 (PMID: 34117229)). The video file has been included in the revised manuscript (video; lines 111-113, 552-553, 573-575).

Additionally, as described in the Methods section and shown in Figure S2 of the initial version (now Figure S3), non-specific signals or artifacts—such as those caused by head movements—were corrected (although such responses were minimal in our recordings).

(c) Additionally, are similar increases in Ca2+ events in LC NA neurons observed during other stressful behavioral paradigms versus non-stressful paradigms?

We appreciate the reviewer's valuable suggestion. Since the present, initial version of our manuscript focused on acute restraint stress, we did not measure Ca²⁺ events in LC-NA neurons in other stress models, but a recent study has shown an increase in Ca²⁺ responses in LC-NA neurons by social defeat stress (Seiriki et al., BioRxiv, https://www.biorxiv.org/content/10.1101/2025.03.07.641347v1).

(d) Neuronal ablation to reveal the function of a cell population.

This method has been widely used in numerous previous studies as an effective experimental approach to investigate the role of specific neuronal populations—including SDH-projecting LC-NA neurons (Ma et al., Brain Res, 2022 (PMID: 34929182); Kawanabe et al., Mol Brain, 2021 (PMID: 33971918))—in CNS function.

(e) The proportion of LC NA neurons and LC→SDH NA neurons expressing DTR-GFP and ablated should be quantified (Figures 1G and J) to validate the methods and permit interpretation of the behavioral data (Figures 1H and K). Importantly, the nocifensive responses and behavior of these mice in other pain assays in the absence of stress (e.g., hotplate) and a few standard assays (open field, rotarod, elevated plus maze) would help determine the consequences of cell ablation on processing of nociceptive information and general behavior.

As suggested, we conducted additional experiments to quantitatively analyze the number of LC^→SDH-NA neurons. We used WT mice injected with AAVretro-Cre into the SDH (L4 segment) and AAV-FLEx[DTR-EGFP] into the LC. In these mice, 4.4% of total LC-NA neurons [positive for tyrosine hydroxylase (TH)] expressed DTR-GFP, representing the LC^→SDH-NA neuronal population (Figure S4; lines 126-127). Furthermore, treatment with DTX successfully ablated the DTR-expressing LC^→SDH-NA neurons. Importantly, the neurons quantified in this analysis were specifically those projecting to the L4 segment of the SDH; therefore, the total number of SDH-projecting LC-NA neurons across all spinal segments is expected to be much higher.

We also performed the rotarod and paw-flick tests to assess motor function and thermal sensitivity following ablation of LC^→SDH-NA neurons. No significant differences were observed between the ablated and control groups (Figure S5; lines 131-134), indicating that ablation of these neurons does not produce non-specific behavioral deficits in motor function or other sensory modalities.

(f) Confirmation of LC NA neuron function with other methods that alter neuronal excitability or neurotransmission instead of destroying the circuit investigated, such as chemogenetics or chemogenetics, would greatly strengthen the findings. Optogenetics is used in Figure 1M, N but excitation of LCLC^→SDH NA neuron terminals is tested instead of inhibition (to mimic ablation), and in naïve mice instead of stressed mice.

We appreciate the reviewer’s comment. The optogenetic approach is useful for manipulating neuronal excitability; however, prolonged light illumination (> tens of seconds) can lead to undesirable tissue heating, ionic imbalance, and rebound spikes (Wiegert et al., Neuron, 2017 (PMID: 28772120)), making it difficult to apply in our experiments, in which mice are exposed to stress for 60 min. For this reason, we decided to employ the cell-ablation approach in stress experiments, as it is more suitable than optogenetic inhibition. In addition, as described in our response to weakness (1)-a) by Reviewer 3 (Public review), we have now demonstrated the specific expression of DTRs in NA neurons in the LC, but not in A5 or A7 (Figure S4; lines 127-128), confirming the specificity of LCLC^→SDH-NAergic pathway targeting in our study. Chemogenetics represent another promising approach to further strengthen our findings on the role of LCLC^→SDH-NA neurons, but this will be an important subject for future studies, as it will require extensive experiments to assess, for example, the effectiveness of chemogenetic inhibition of these neurons during 60 min of restraint stress, as well as optimization of key parameters (e.g., systemic DCZ doses).

(g) Alpha1Ars. The authors noted that "Adra1a mRNA is also expressed in INs in the SDH".

The expression of α_1ARs in inhibitory interneurons in the SDH is consistent with our previous findings (Uchiyama et al., Mol Brain, 2022 (PMID: 34980215)) as well as with scRNA-seq data (http://linnarssonlab.org/dorsalhorn/, Häring et al., Nat Neurosci, 2018 (PMID: 29686262)).

(h) The authors should comprehensively indicate what other cell types present in the spinal cord and neurons projecting to the spinal cord express alpha1Ars and what is the relative expression level of alpha1Ars in these different cell types.

According to the scRNA-seq data (https://seqseek.ninds.nih.gov/genes, Russ et al., Nat Commun, 2021 (PMID: 34588430); http://linnarssonlab.org/dorsalhorn/, Häring et al., Nat Neurosci, 2018 (PMID: 29686262)), we confirmed that α_1ARs are predominantly expressed in astrocytes and inhibitory interneurons in the spinal cord. Also, an α_1AR-expressing excitatory neuron population (Glut14) expresses Tacr1, GPR83, and Tac1 mRNAs, markers that are known to be enriched in projection neurons of the SDH. This raises the possibility that α_1A Rs may also be expressed in a subset of projection neurons, although further experiments are required to confirm this. In DRG neurons, α_1AR expression was detected to some extent, but its level seems to be much lower than in the spinal cord (http://linnarssonlab.org/drg/ Usoskin et al., Nat Neurosci, 2015 (PMID: 25420068)). Consistent with this, primary afferent glutamatergic synaptic transmission has been shown to be unaffected by α_1AR agonists (Kawasaki et al., Anesthesiology, 2003 (PMID: 12606912); Li and Eisenach, JPET, 2001 (PMID: 11714880)). This information has been incorporated into the Discussion section (lines 317-319).

(i) The conditional KO of alpha1Ars specifically in Hes5+ astrocytes and not in other cell types expressing alpha1Ars should be quantified and validated (Figure 2H).

We have previously shown a selective KO of α_1AR in Hes5⁺ astrocytes in the same mouse line (Kohro et al., Nat Neurosci, 2020 (PMID: 33020652)). This information has been included in the revised text (line 166-167).

(j) Depolarization of SDH inhibitory interneurons by NA (Figure 3). The authors' bath applied NA, which presumably activates all NA receptors present in the preparation.

We believe that the reviewer’s concern may pertain to the possibility that NA acts on non-Vgat⁺ neurons, thereby indirectly causing depolarization of Vgat⁺ neurons. As described in the Method section of the initial version, in our electrophysiological experiments, we added four antagonists for excitatory and inhibitory neurotransmitter receptors—CNQX (AMPA receptor), MK-801 (NMDA receptor), bicuculline (GABA_A receptor), and strychnine (glycine receptor)—to the artificial cerebrospinal fluid to block synaptic inputs from other neurons to the recorded Vgat⁺ neurons. Since this method is widely used for this purpose in many previous studies (Wu et al., J Neurosci, 2004 (PMID: 15140934); Liu et al., Nat Neurosci, 2010 (PMID: 20835251)), it is reasonable to conclude that NA directly acts on the recorded SDH Vgat⁺ interneurons to produce excitation (lines 193-196).

(k) The authors' model (Figure 4H) implies that NA released by LC→SDH NA neurons leads to the inhibition of SDH inhibitory interneurons by NA. In other experiments (Figure 1L, Figure 2A), the authors used optogenetics to promote the release of endogenous NA in SDH by LC→SDH NA neurons. This approach would investigate the function of NA endogenously released by LC NA neurons at presynaptic terminals in the SDH and at physiological concentrations and would test the model more convincingly compared to the bath application of NA.

We appreciate the reviewer’s valuable comment. As noted, optogenetic stimulation of LC^→SDH-NA neurons would indeed be useful to test this model. However, in our case, it is technically difficult to investigate the responses of Vgat⁺ inhibitory neurons and Hes5⁺ astrocytes to NA endogenously released from LC^→SDH-NA neurons. This would require the use of Vgat-Cre or Hes5-CreERT2 mice, but employing these lines precludes the use of NET-Cre mice, which are necessary for specific and efficient expression of ChrimsonR in LC^→SDH-NA neurons. Nevertheless, all of our experimental data consistently support the proposed model, and we believe that the reviewer will agree with this, without additional experiments that is difficult to conduct because of technical limitations (lines 382-388).

(l) As for other experiments, the proportion of Hes+ astrocytes that express hM3Dq, and the absence of expression in other cells, should be quantified and validated to interpret behavioral data.

We thank the reviewer for raising this point. In our experiments, we used an HA-tag (fused with hM3Dq) to confirm hM3Dq expression. However, it is difficult to precisely analyze individual astrocytes because, as shown in Figure 3J, the boundaries of many HA-tag⁺ astrocytes are indistinguishable. This seems to be due to the membrane localization of HA-tag, the complex morphology of astrocytes, and their tile-like distribution pattern (Baldwin et al., Trends Cell Biol, 2024 (PMID: 38180380)). Nevertheless, our previous study demonstrated that ~90% of astrocytes in the superficial laminae are Hes5⁺ (Kohro et al., Nat Neurosci, 2020 (PMID: 33020652)), and intra-SDH injection of AAV-hM3Dq labeled the majority of superficial astrocytes (Figure 3J). Thus, AAV-FLEx[hM3Dq] injection into Hes5-CreERT2 mice allows efficient expression of hM3Dq in Hes5⁺ astrocytes in the SDH. Importantly, our previous studies using Hes5-CreERT2 mice have confirmed that hM3Dq is not expressed in other cell types (neurons, oligodendrocytes, or microglia) (Kohro et al., Nat Neurosci, 2020 (PMID: 33020652); Kagiyama et al., Mol Brain, 2025 (PMID: 40289116)). This information regarding the cell-type specificity has now been briefly described in the revised version (lines 218-219).

(m) Showing that the effect of CNO is dose-dependent would strengthen the authors' findings.

Thank you for your comment. We have now demonstrated a dose-dependent effect of CNO on Ca²⁺ responses in SDH astrocytes (please see our response to Major Point (4) from Reviewer #2 (Recommendations for the Authors) (Figure S7; lines 225-228). In addition, we also confirmed that the effect of CNO is not nonspecific, as CNO application did not alter sIPSCs in spinal cord slices prepared from mice lacking hM3Dq expression in astrocytes (Figure S7; lines 225-228).

(n) The proportion of SG neurons for which CNO bath application resulted in a reduction in recorded sIPSCs is not clear.

We have included individual data points in each bar graph to more clearly illustrate the effect of CNO on each neuron (Figure 3L, N).

(o) A1Rs. The specific expression of Cas9 and guide RNAs, and the specific KD of A1Rs, in inhibitory interneurons but not in other cell types expressing A1Rs should be quantified and validated.

In addition to the data demonstrating the specific expression of SaCas9 and sgAdora1 in Vgat⁺ inhibitory neurons shown in Figure 3G of the initial version, we have now conducted the same experiments with a different sample and confirmed this specificity: SaCas9 (detected via HA-tag) and sgAdora1 (detected via mCherry) were expressed in PAX2⁺ inhibitory neurons (Author response image 1). Furthermore, as shown in Figure 3H and I in the initial version, the functional reduction of A₁Rs in inhibitory neurons was validated by electrophysiological recordings. Together, these results support the successful deletion of A₁Rs in inhibitory neurons.

Author response image 1.

Expression of HA-tag and mCherry in inhibitory neurons (a different sample from Figure 3G) SaCas9 (yellow, detected by HA-tag) and mCherry (magenta) expression in the PAX2⁺ inhibitory neurons (cyan) at 3 weeks after intra-SDH injection of AAV-FLEx[SaCas9-HA] and AAV-FLEx[mCherry]-U6-sgAdora1 in Vgat-Cre mice. Arrowheads indicate genome-editing Vgat⁺ cells. Scale bar, 25 µm.

(6) Methods:

It is unclear how fiber photometry is performed using "optic cannula" during restraint stress while mice are in a 50ml falcon tube (as shown in Figure 1A).

We apologize for the omission of this detail in the Methods section. To monitor Ca²⁺ events in LC-NA neurons during restraint stress, we created a narrow slit on the top of the conical tube, allowing mice to undergo restraint stress while connected to the optic fiber (see video). This information has now been added to the Methods section (lines 552-553).

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

(1) Scientific rigor:

It is unclear if the normal distribution of the data was determined before selecting statistical tests.

We apologize for omitting this description. For all statistical analyses in this study, we first assessed the normality of the data and then selected appropriate statistical tests accordingly. We have added this information to the revised manuscript (lines 711-712).

(2) Nomenclature:

(a) Mouse Genome Informatics (MGI) nomenclature should be used to describe mouse genotypes (i.e., gene name in italic, only first letter is capitalized, alleles in superscript).

(b) FLEx should be used instead of flex.

Thank you for the suggestion. We have corrected these terms (including FLEx) according to MGI nomenclature.

Reviewer #2 (Public review):

Summary:

This study investigates the role of spinal astrocytes in mediating stress-induced pain hypersensitivity, focusing on the LC (locus coeruleus)-to-SDH (spinal dorsal horn) circuit and its mechanisms. The authors aimed to delineate how LC activity contributes to spinal astrocytic activation under stress conditions, explore the role of noradrenaline (NA) signaling in this process, and identify the downstream astrocytic mechanisms that influence pain hypersensitivity.

The authors provide strong evidence that 1-hour restraint stress-induced pain hypersensitivity involves the LC-to-SDH circuit, where NA triggers astrocytic calcium activity via alpha1a adrenoceptors (alpha1aRs). Blockade of alpha1aRs on astrocytes - but not on Vgat-positive SDH neurons - reduced stress-induced pain hypersensitivity. These findings are rigorously supported by well-established behavioral models and advanced genetic techniques, uncovering the critical role of spinal astrocytes in modulating stress-induced pain.

However, the study's third aim - to establish a pathway from astrocyte alpha1aRs to adenosine-mediated inhibition of SDH-Vgat neurons - is less compelling. While pharmacological and behavioral evidence is intriguing, the ex vivo findings are indirect and lack a clear connection to the stress-induced pain model. Despite these limitations, the study advances our understanding of astrocyte-neuron interactions in stress-pain contexts and provides a strong foundation for future research into glial mechanisms in pain hypersensitivity.

Strengths:

The study is built on a robust experimental design using a validated 1-hour restraint stress model, providing a reliable framework to investigate stress-induced pain hypersensitivity. The authors utilized advanced genetic tools, including retrograde AAVs, optogenetics, chemogenetics, and subpopulation-specific knockouts, allowing precise manipulation and interrogation of the LC-SDH circuit and astrocytic roles in pain modulation. Clear evidence demonstrates that NA triggers astrocytic calcium activity via alpha1aRs, and blocking these receptors effectively reduces stress-induced pain hypersensitivity.

Weaknesses:

Despite its strengths, the study presents indirect evidence for the proposed NA-to-astrocyte(alpha1aRs)-to-adenosine-to-SDH-Vgat neurons pathway, as the link between astrocytic adenosine release and stress-induced pain remains unclear. The ex vivo experiments, including NA-induced depolarization of Vgat neurons and chemogenetic stimulation of astrocytes, are challenging to interpret in the stress context, with the high CNO concentration raising concerns about specificity. Additionally, the role of astrocyte-derived D-serine is tangential and lacks clarity regarding its effects on SDH Vgat neurons. The astrocyte calcium signal "dip" after LC optostimulation-induced elevation are presented without any interpretation.

We appreciate the reviewer's careful reading of our paper. According to the reviewer's comments, we have performed new additional experiments and added some discussion in the revised manuscript (please see the point-by-point responses below).

Reviewer #2 (Recommendations for the authors):

The astrocyte-mediated pathway of NA-to-astrocyte (alpha1aRs)-to-adenosine-to-SDH Vgat neurons (A1R) in the context of stress-induced pain hypersensitivity requires more direct evidence. While the data showing that the A1R agonist CPT inhibits stress-induced hypersensitivity and that stress combined with Aβ fiber stimulation increases pERK in the SDH are intriguing, these findings primarily support the involvement of A1R on Vgat neurons and are only behaviorally consistent with SDH-Vgat neuronal A1R knockdown. The role of astrocytes in this pathway in vivo remains indirect. The ex vivo chemogenetic Gq-DREADD stimulation of SDH astrocytes, which reduced sIPSCs in Vgat neurons in a CPT-dependent manner, needs revision with non-DREADD+CNO controls to validate specificity. Furthermore, the ex vivo bath application of NA causing depolarization in Vgat neurons, blocked by CPT, adds complexity to the data leaving me wondering how astrocytes are involved in such processes, and it does not directly connect to stress-induced pain hypersensitivity. These findings are potentially useful but require additional refinement to establish their relevance to the stress model.

We thank the reviewer for the insightful feedback. First, regarding the role of astrocytes in this pathway in vivo, we showed in the initial version that mechanical pain hypersensitivities induced by intrathecal NA injection and by acute restraint stress were attenuated by both pharmacological blockade and Vgat⁺ neuron-specific knockdown of A₁Rs (Figure 4A, B). Given that NA- and stress-induced pain hypersensitivity is mediated by α_1AR-dependent signaling in Hes5⁺ astrocytes (Kohro et al., Nat Neurosci, 2020 (PMID: 33020652); this study), these findings provide in vivo evidence supporting the involvement of the NA → Hes5⁺ astrocyte (via α_1ARs) → adenosine → Vgat⁺ neuron (via A₁Rs) pathway. As noted in the reviewer’s major comment (2), in vivo monitoring of adenosine dynamics in the SDH during stress exposure would further substantiate the astrocyte-to-neuron signaling pathway. However, we did not detect clear signals, potentially due to several technical limitations (see our response below). Acknowledging this limitation, we have now added a new paragraph in the end of Discussion section to address this issue. Second, the specificity of the effect of CNO has now been validated by additional experiments (see our response to major point (4)). Third, the reviewer’s concern regarding the action of NA on Vgat⁺ neurons has also been addressed (see our response to major point (3) below).

Major points:

(1) The in vivo pharmacology using DCK to antagonize D-serine signaling from alpha1a-activated astrocytes is tangential, as there is limited evidence on how Vgat neurons (among many others) respond to D-serine. This aspect requires more focused exploration to substantiate its relevance.

We propose that the site of action of D-serine in our neural circuit model is the NMDA receptors (NMDARs) on excitatory neurons, a notion supported by our previous findings (Kohro et al., Nat Neurosci, 2020 (PMID: 33020652); Kagiyama et al., Mol Brain, 2025 (PMID: 40289116)). However, we cannot exclude the possibility that D-serine also acts on NMDARs expressed by Vgat⁺ inhibitory neurons. Nevertheless, given that intrathecal injection of D-serine in naïve mice induces mechanical pain hypersensitivity (Kohro et al., Nat Neurosci, 2020 (PMID: 33020652)), it appears that the pronociceptive effect of D-serine in the SDH is primarily associated with enhanced pain processing and transmission, presumably via NMDARs on excitatory neurons. We have added this point to the Discussion section in the revised manuscript (lines 325-330).

(2) Additionally, employing GRAB-Ado sensors to monitor adenosine dynamics in SDH astrocytes during NA signaling would significantly strengthen conclusions about astrocyte-derived adenosine's role in the stress model.

We agree with the reviewer’s comment. Following this suggestion, we attempted to visualize NA-induced adenosine (and ATP) dynamics using GRAB-ATP and GRAB-Ado sensors (Wu et al., Neuron, 2022 (PMID: 34942116); Peng et al., Science, 2020 (PMID: 32883833)) in acutely isolated spinal cord slices from mice after intra-SDH injection of AAV-hSyn-GRABATP_1.0 and -GRABAdo_1.0. We confirmed expression of these sensors in the SDH (Author response image 2a) and observed increased signals after bath application of ATP (0.1 or 1 µM) or adenosine (1 µM) (Author response image 2b, c). However, we were unable to detect clear signals following NA stimulation (Author response image 2b, c). The reason for this lack of detectable changes remains unclear. If the release of adenosine from astrocytes is a highly localized phenomenon, it may be measurable using high-resolution microscopy capable of detecting adenosine levels at the synaptic level and more sensitive sensors. Further investigation will therefore be required (lines 340-341).

Author response image 2.

Ex vivo imaging of GRAB-ATP and GRAB-Ado sensors.(a) Representative images of GRAB_ATP1.0 (left, green) or GRAB_Ado1.0 (right, green) expression in the SDH at 3 weeks after SDH injection of AAV-hSyn-GRAB_Ado1.0 or AAV-hSyn-GRAB_Ado1.0 in Hes5-CreERT2 mice. Scale bar, 200 µm. (b) Left: Representative fluorescence images showing GRAB_ATP1.0 responses before and after perfusion with NA or ATP. Right: Representative traces showing responses to ATP (0.1 and 1 µM) or NA (10 µM). (c) Left: Representative fluorescence images showing GRABAdo1.0 responses before and after perfusion with NA or adenosine (Ado). Right: Representative traces showing responses to Ado (0.01, 0.1, and 1 µM), NA (10 µM), or no application (negative control).

(3) The interpretation of Figure 3D is challenging. The manuscript implies that 20 μM NA acts on Adra1a receptors on Vgat neurons to depolarize them, but this concentration should also activate Adra1a on astrocytes, leading to adenosine release and potential inhibition of depolarization. The observation of depolarization despite these opposing mechanisms requires explanation, as does the inhibition of depolarization by bath-applied A1R agonist. Of note, 20 μM NA is a high concentration for Adra1a activation, typically responsive at nanomolar levels. The discussion should reconcile this with prior studies indicating dose-dependent effects of NA on pain sensitivity (e.g., Reference 22).

Like the reviewer, we also considered that bath-applied NA could activate α_1ARs expressed on Hes5⁺ astrocytes. To clarify this point, we have performed additional patch-clamp recordings and found that knockdown of A₁Rs in Vgat⁺ neurons tended to increase the proportion of Vgat⁺ neurons with NA-induced depolarizing responses (Figure S8). Therefore, it is conceivable that NA-induced excitation of Vgat⁺ neurons may involve both a direct effect of NA activating α_1ARs in Vgat⁺ neurons and an indirect inhibitory signaling from NA-stimulated Hes5⁺ astrocytes via adenosine (lines 298-300).

The concentration of NA used in our ex vivo experiments is higher than that typically used in vitro with αR-_1Aexpressing cell lines or primary culture cells, but is comparable to concentrations used in other studies employing spinal cord slices (Kohro et al., Nat Neurosci, 2020 (PMID: 33020652); Baba et al., Anesthesiology, 2000 (PMID: 10691236); Lefton et al., Science, 2025 (PMID: 40373122)). In slice experiments, drugs must diffuse through the tissue to reach target cells, resulting in a concentration gradient. Therefore, higher drug concentrations are generally necessary in slice experiments, in contrast to cultured cell experiments, where drugs are directly applied to target cells. Importantly, we have previously shown that the pharmacological effects of 20 μM NA on Vgat⁺ neurons and Hes5⁺ astrocytes are abolished by loss of α_1ARs in these cells (Uchiyama et al., Mol Brain, 2022 (PMID: 34980215); Kohro et al., Nat Neurosci, 2020 (PMID: 33020652)), confirming the specificity of these NA actions.

Regarding the dose-dependent effect of NA on pain sensitivity, NA-induced pain hypersensitivity is abolished in Hes5⁺ astrocyte-specific α_1AR-KO mice (Kohro et al., Nat Neurosci, 2020 (PMID: 33020652)), indicating that this behavior is mediated by α_1ARs expressed on Hes5⁺ astrocytes. In contrast, the suppression of pain sensitivity by high doses of NA was unaffected in the KO mice (Kohro et al., Nat Neurosci, 2020 (PMID: 33020652)), suggesting that other adrenergic receptors may contribute to this phenomenon. Clarifying the responsible receptors will require future investigation.

(4) In Figure 3K-M, the CNO concentration used (100 μM) is unusually high compared to standard doses (1 to a few μM), raising concerns about potential off-target effects. Including non-hM3Dq controls and using lower CNO concentrations are essential to validate the specificity of the observed effects. Similarly, the study should clarify whether astrocyte hM3Dq stimulation alone (without NA) would induce hyperpolarization in Vgat neurons and how this interacts with NA-induced depolarization.

We acknowledge that the concentration of CNO used in our experiments is relatively high compared to that used in other reports. However, in our experiments, application of CNO at 1, 10, and 100 μM induced Ca²⁺ increases in GCaMP6-expressing astrocytes in spinal cord slices in a concentration-dependent manner (Figure S7). Among these, 100 μM CNO most effectively replicated the NA-induced Ca²⁺ signals in astrocytes. Based on these findings, we selected this concentration for use in both the current and previous studies (Kohro et al., Nat Neurosci., 2020 (PMID: 33020652)). Importantly, to rule out non-specific effects, we conducted control experiments using spinal cord slices from mice that did not express hM3Dq in astrocytes and confirmed that CNO had no effect on Ca²⁺ responses in astrocytes and sIPSCs in substantial gelatinosa (SG) neurons (Figure S7; lines 223-228). Thus, although the CNO concentration used is relatively high, the observed effects of CNO are not non-specific but result from the chemogenetic activation of hM3Dq-expressing astrocytes.

In this study, we used Hes5-CreERT2 and Vgat-Cre mice to manipulate gene expression in Hes5⁺ astrocytes and Vgat⁺ neurons, respectively. In order to fully address the reviewer’s comment, the use of both Cre lines is necessary. However, simultaneous and independent genetic manipulation in each cell type using Cre activity alone is not feasible with the current genetic tools. We have mentioned this as a technical limitation in the Discussion section (lines 382-388).

(5) The role of D-serine released by hM3Dq-stimulated astrocytes in (separately) modulating sub-types of neurons including excitatory neurons and Vgat positives needs more detailed discussion. If no effect of D-serine on Vgat neurons is observed, this should be explicitly stated, and the discussion should address why this might be the case.

As mentioned in our response to Major Point (1) above, we have added a discussion of this point in the revised manuscript (lines 325-330).

(6) Finally, the observed "dip" in astrocyte calcium signals below baseline following the large peaks with LC optostimulation should be discussed further, as understanding this phenomenon could provide valuable insights into astrocytic signaling dynamics in the context of single acute or repetitive chronic stress.

Thank you for your comment. We found that this phenomenon was not affected by pretreatment with the α_1AR-specific antagonist silodosin (Author response image 3), which effectively suppressed Ca²⁺ elevations evoked by stimulation of LC-NA neurons (Figure 2F). This implies that the phenomenon is independent of α_1AR signaling. Elucidating the detailed underlying mechanism remains an important direction for future investigation.

Author response image 3.

The observed "dip" in astrocyte Ca²⁺ signals was not affected by pretreatment with the α_1AR-specific antagonist silodosin. Representative traces of astrocytic GCaMP6m signals in response to optogenetic stimulation of LC-NAe^→SDHrgic axons/terminals in a spinal cord slice. Each trace shows the GCaMP6m signal before and after optogenetic stimulation (625 nm, 1 mW, 10 Hz, 5 ms pulse duration, 10 s). Slices were pretreated with silodosin (40 nM) for 5 min prior to stimulation.

Reviewer #3 (Public review):

Summary:

This is an exciting and timely study addressing the role of descending noradrenergic systems in nocifensive responses. While it is well-established that spinally released noradrenaline (aka norepinephrine) generally acts as an inhibitory factor in spinal sensory processing, this system is highly complex. Descending projections from the A6 (locus coeruleus, LC) and the A5 regions typically modulate spinal sensory processing and reduce pain behaviours, but certain subpopulations of LC neurons have been shown to mediate pronociceptive effects, such as those projecting to the prefrontal cortex (Hirshberg et al., PMID: 29027903).

The study proposes that descending cerulean noradrenergic neurons potentiate touch sensation via alpha-1 adrenoceptors on Hes5+ spinal astrocytes, contributing to mechanical hyperalgesia. This finding is consistent with prior work from the same group (dd et al., PMID:). However, caution is needed when generalising about LC projections, as the locus coeruleus is functionally diverse, with differences in targets, neurotransmitter co-release, and behavioural effects. Specifying the subpopulations of LC neurons involved would significantly enhance the impact and interpretability of the findings.

Strengths:

The study employs state-of-the-art molecular, genetic, and neurophysiological methods, including precise CRISPR and optogenetic targeting, to investigate the role of Hes5+ astrocytes. This approach is elegant and highlights the often-overlooked contribution of astrocytes in spinal sensory gating. The data convincingly support the role of Hes5+ astrocytes as regulators of touch sensation, coordinated by brain-derived noradrenaline in the spinal dorsal horn, opening new avenues for research into pain and touch modulation.

Furthermore, the data support a model in which superficial dorsal horn (SDH) Hes5+ astrocytes act as non-neuronal gating cells for brain-derived noradrenergic (NA) signalling through their interaction with substantia gelatinosa inhibitory interneurons. Locally released adenosine from NA-stimulated Hes5+ astrocytes, following acute restraint stress, may suppress the function of SDH-Vgat+ inhibitory interneurons, resulting in mechanical pain hypersensitivity. However, the spatially restricted neuron-astrocyte communication underlying this mechanism requires further investigation in future studies.

Weaknesses

(1) Specificity of the LC Pathway targeting

The main concern lies with how definitively the LC pathway was targeted. Were other descending noradrenergic nuclei, such as A5 or A7, also labelled in the experiments? The authors must convincingly demonstrate that the observed effects are mediated exclusively by LC noradrenergic terminals to substantiate their claims (i.e. "we identified a circuit, the descending LC→SDH-NA neurons").

(a) For instance, the direct vector injection into the LC likely results in unspecific effects due to the extreme heterogeneity of this nucleus and retrograde labelling of the A5 and A7 nuclei from the LC (i.e., Li et al., PMID: 26903420).

We appreciate the reviewer's valuable comments. To address this point, we performed additional experiments and demonstrated that intra-SDH injection of AAVretro-Cre followed by intra-LC injection of AAV2/9-EF1α-FLEx[DTR-EGFP] specifically results in DTR expression in NA neurons of the LC, but not of the A5 or A7 regions (Figure S4; lines 127-128). These results confirm the specificity of targeting the LC^→SDH-NAergic pathway in our study.

(b) It is difficult to believe that the intersectional approach described in the study successfully targeted LC→SDH-NA neurons using AAVrg vectors. Previous studies (e.g., PMID: 34344259 or PMID: 36625030) demonstrated that similar strategies were ineffective for spinal-LC projections. The authors should provide detailed quantification of the efficiency of retrograde labelling and specificity of transgene expression in LC neurons projecting to the SDH.

Thank you for your comment. As we described in our response to the weakness (5)-e) of Reviewer #1 (Public review), our additional analysis showed that, under our experimental conditions, expression of genes (for example DTR) was observed in 4.4% of NA (TH⁺) neurons in the LC (Figure S4; lines 126-127).

The reasons for this difference between the previous studies and our current study is unclear; however, it is likely attributed to methodological differences, including the type of viral vectors employed, species differences (mouse (PMID: 34344259, our study) vs. rat (PMID: 36625030)), the amount of AAV injected into the SDH (300 nL at three sites (PMID: 34344259), and 300 nL at a single site (our study)) and LC (500 nL at a single site (PMID: 34344259), and 300 nL at a single site (our study)), as well as the depth of AAV injection in the SDH (200–300 µm from the dorsal surface of the spinal cord (PMID: 34344259), and 120–150 µm in depth from the surface of the dorsal root entry zone (our study)).

(c) Furthermore, it is striking that the authors observed a comparably strong phenotypical change in Figure 1K despite fewer neurons being labelled, compared to Figure 1H and 1N with substantially more neurons being targeted. Interestingly, the effect in Figure 1K appears more pronounced but shorter-lasting than in the comparable experiment shown in Figure 1H. This discrepancy requires further explanation.

Although only a representative section of the LC was shown in the initial version, LC^→SDH-NA neurons are distributed rostrocaudally throughout the LC, as previously reported (Llorca-Torralba et al., Brain, 2022 (PMID: 34373893)). Our additional experiments analyzing multiple sections of the anterior and posterior regions of the LC have now revealed that approximately sixty LC^→SDH-NA neurons express DTR, and these neurons are eliminated following DTX treatment (Figure S4; lines 126-128) (it should be noted that these neurons specifically project to the L4 segment of the SDH, and the total number of LC^→SDH-NA neurons is likely much higher). Considering the specificity of LC^→SDH-NAergic pathway targeting demonstrated in our study (as described above), together with the fact that primary afferent sensory fibers from the plantar skin of the hindpaw predominantly project to the L4 segment of the SDH, these data suggest that the observed behavioral changes are attributable to the loss of these neurons and that ablation of even a relatively small number of NA neurons in the LC can have a significant impact on behavior. We have added this hypothesis in the Discussion section (lines 373-382).

Regarding the data in Figures 1H and 1K, as the reviewer pointed out, a statistically significant difference was observed at 90 min in mice with ablation of LC-NA neurons, but not in those with LC^→SDH-NA neuron ablation. This is likely due to a slightly higher threshold in the control group at this time point (Figure 1K), and it remains unclear whether there is a mechanistic difference between the two groups at this specific time point.

(d) A valuable addition would be staining for noradrenergic terminals in the spinal cord for the intersectional approach (Figure 1J), as done in Figures 1F/G. LC projections terminate preferentially in the SDH, whereas A5 projections terminate in the deep dorsal horn (DDH). Staining could clarify whether circuits beyond the LC are being ablated.

As suggested, we performed DTR immunostaining in the SDH; however, we did not detect any DTR immunofluorescence there. A similar result was also observed in the spinal terminals of DTR-expressing primary afferent fibers (our unpublished data). The reason for this is unclear, but to the best of our knowledge, no studies have clearly shown DTR expression at presynaptic terminals, which may be because the action of DTX on the neuronal cell body is necessary for cell ablation. Nevertheless, as described in our response to the weakness (5)-f) by Reviewer 1 (Public review), we have now confirmed the specific expression of DTR in the LC, but not in the A5 and A7 regions (Figure S4; lines 127-128).

(e) Furthermore, different LC neurons often mediate opposite physiological outcomes depending on their projection targets-for example, dorsal LC neurons projecting to the prefrontal cortex PFCx are pronociceptive, while ventral LC neurons projecting to the SC are antinociceptive (PMIDs: 29027903, 34344259, 36625030). Given this functional diversity, direct injection into the LC is likely to result in nonspecific effects.

To avoid behavioral outcomes resulting from a mixture of facilitatory and inhibitory effects caused by activating the entire population of LC-NA neurons, we employed a specific manipulation targeting LC^→SDH-NA neurons using AAV vectors. The specificity of this manipulation was confirmed in our previous study (Kohro et al., Nat Neurosci, 2020 (PMID: 33020652)) and in the current study (Figure S4). Using this approach, we previously demonstrated that LC neurons can exert pronociceptive effects via astrocytes in the SDH (Kohro et al., Nat Neurosci, 2020 (PMID: 33020652)). This pronociceptive role is further supported by the current study, which uses a more selective manipulation of LC^→SDH-NA neurons through a NET-Cre mouse line. In addition, intrathecal administration of relatively low doses of NA in naïve mice clearly induces mechanical pain hypersensitivity. Nevertheless, we have also acknowledged that several recent studies have reported an inhibitory role of LC^→SDH-NA neurons in spinal nociceptive signaling. The reason for these differing behavioral outcomes remains unclear, but several methodological differences may underlie the discrepancy. First, the degree of LC^→SDH-NA neuronal activity may play a role. Although direct comparisons between studies reporting pro- and anti-nociceptive effects are difficult, our previous studies demonstrated that intrathecal administration of high doses of NA in naïve mice does not induce mechanical pain hypersensitivity (Kohro et al., Nat Neurosci, 2020 (PMID: 33020652)). Second, the sensory modality used in behavioral testing may be a contributing factor as the pronociceptive effect of NA appears to be selectively observed in responses to mechanical, but not thermal, stimuli (Kohro et al., Nat Neurosci, 2020 (PMID: 33020652)). This sensory modality-selective effect is also evident in mice subjected to acute restraint stress (Table S1). Therefore, the role of LC^→SDH-NA neurons in modulating nociceptive signaling in the SDH is more complex than previously appreciated, and their contribution to pain regulation should be reconsidered in light of factors such as NA levels, sensory modality, and experimental context. In revising the manuscript, we have included some points described above in the Discussion (lines 282-291).

Conclusion on Specificity: The authors are strongly encouraged to address these limitations directly, as they significantly affect the validity of the conclusions regarding the LC pathway. Providing more robust evidence, acknowledging experimental limitations, and incorporating complementary analyses would greatly strengthen the manuscript.

We appreciate the reviewer’s comments. We fully acknowledge the limitations raised and agree that addressing them directly is important for the rigor of our conclusions on the LC pathway. To this end, we have performed additional experiments (e.g., Figure A and S4), which are now included in the revised manuscript. Furthermore, we have also newly added a new paragraph for experimental limitations in the end of Discussion section (lines 373-408). We believe these new data substantially strengthen the validity of our findings and have clarified these points in the Discussion section.

(2) Discrepancies in Data

(a) Figures 1B and 1E: The behavioural effect of stress on PWT (Figure 1E) persists for 120 minutes, whereas Ca2+ imaging changes (Figure 1B) are only observed in the first 20 minutes, with signal attenuation starting at 30 minutes. This discrepancy requires clarification, as it impacts the proposed mechanism.

Thank you for your important comment. As pointed out by the reviewer, there is a difference between the duration of behavioral responses and Ca²⁺ events, although the exact time point at which the PWT begins to decline remains undetermined (as behavioral testing cannot be conducted during stress exposure). A similar temporal difference was also observed following intraplantar injection of capsaicin (Kohro et al., Nat Neurosci, 2020 (PMID: 33020652)); while LC^→SDH-NA neuron-mediated astrocytic Ca²⁺ responses in SDH astrocytes last for 5–10 min after injection, behavioral hypersensitivity peaks around 60 min post-injection and gradually returns to baseline over the subsequent 60–120 min. These findings raise the possibility that astrocyte-mediated pain hypersensitivity in the SDH may involve a sustained alteration in spinal neural function, such as central sensitization. We have added this hypothesis to the Discussion section of the revised manuscript (lines 399-408), as it represents an important direction for future investigation.

(b) Figure 4E: The effect is barely visible, and the tissue resembles "Swiss cheese," suggesting poor staining quality. This is insufficient for such an important conclusion. Improved staining and/or complementary staining (e.g., cFOS) are needed. Additionally, no clear difference is observed between Stress+Ab stim. and Stress+Ab stim.+CPT, raising doubts about the robustness of the data.

As suggested, we performed c-FOS immunostaining and obtained clearer results (Figure 4E,F; lines 243-252). We also quantitatively analyzed the number of c-FOS⁺ cells in the superficial laminae, and the results are consistent with those obtained from the pERK experiments.

(c) Discrepancy with Existing Evidence: The claim regarding the pronociceptive effect of LC→SDH-NAergic signalling on mechanical hypersensitivity contrasts with findings by Kucharczyk et al. (PMID: 35245374), who reported no facilitation of spinal convergent (wide-dynamic range) neuron responses to tactile mechanical stimuli, but potent inhibition to noxious mechanical von Frey stimulation. This discrepancy suggests alternative mechanisms may be at play and raises the question of why noxious stimuli were not tested.

In our experiments, ChrimsonR expression was observed in the superficial and deeper laminae of the spinal cord (Figure S6). Due to the technical limitations of the optical fibers used for optogenetics, the light stimulation could only reach the superficial laminae; therefore, it may not have affected the activity of neurons (including WDR neurons) located in the deeper laminae. Furthermore, the study by Kucharczyk et al. (Brain, 2022 (PMID: 35245374)) employed a stimulation protocol that differed from ours, applying continuous stimulation over several minutes. Given that the levels of NA released from LC^→SDH-NAergic terminals in the SDH increase with the duration of terminal stimulation (as shown in Figure 2B), longer stimulation may result in higher levels of NA in the SDH. Considering also our data indicating that the pro- and anti-nociceptive effects of NA are dose dependent (Kohro et al., Nat Neurosci, 2020 (PMID: 33020652)), these differences may be related to LC^→SDH-NA neuron activity, NA levels in the SDH, and the differential responses of SDH neurons in the superficial versus deeper laminae (lines 388-395).

(3) Sole reliance on Von Frey testing

The exclusive use of von Frey as a behavioural readout for mechanical sensitisation is a significant limitation. This assay is highly variable, and without additional supporting measures, the conclusions lack robustness. Incorporating other behavioural measures, such as the adhesive tape removal test to evaluate tactile discomfort, the needle floor walk corridor to assess sensitivity to uneven or noxious surfaces, or the kinetic weight-bearing test to measure changes in limb loading during movement, could provide complementary insights. Physiological tests, such as the Randall-Selitto test for noxious pressure thresholds or CatWalk gait analysis to evaluate changes in weight distribution and gait dynamics, would further strengthen the findings and allow for a more comprehensive assessment of mechanical sensitisation.

Thank you for your suggestion. Based on our previous findings that Hes5⁺ astrocytes in the SDH selectively modulate mechanosensory signaling (Kohro et al., Nat Neurosci, 2020 (PMID: 33020652)), the present study focused on behavioral responses to mechanical stimuli using von Frey filaments. As we have not previously conducted most of the behavioral tests suggested by the reviewers, and as we currently lack the necessary equipments for these tests (e.g., Randall–Selitto test, CatWalk gait analysis, and weight-bearing test), we were unable to include them in this study. However, it will be of great interest in future research to investigate whether activation of the LC^→SDH-NA neuron-to-SDH Hes5⁺ astrocyte signaling pathway similarly sensitizes behavioral responses to other types of mechanical stimuli and also to investigate the sensory modality-selective pro- and antinociceptive role of LC^→SDH-NAergic signaling in the SDH (lines 396-399).

Overall Conclusion

This study addresses an important and complex topic with innovative methods and compelling data. However, the conclusions rely on several assumptions that require more robust evidence. Specificity of the LC pathway, experimental discrepancies, and methodological limitations (e.g., sole reliance on von Frey) must be addressed to substantiate the claims. With these issues resolved, this work could significantly advance our understanding of astrocytic and noradrenergic contributions to pain modulation.

We have made every effort to address the reviewer’s concerns through additional experiments and analyses. Based on the new control data presented, we believe that our explanation is reasonable and acceptable. Although additional data cannot be provided on some points due to methodological constraints and limitations of the techniques currently available in our laboratory, we respectfully submit that the evidence presented sufficiently supports our conclusions.

Reviewer #3 (Recommendations for the authors):

A lot of beautiful and challenging-to-collect data is presented. Sincere congratulations to all the authors on this achievement!

Notwithstanding, please carefully reconsider the conclusions regarding the LC pathway, as additional evidence is required to ensure their specificity and robustness.

We thank the reviewer for the kind comments and for raising an important point regarding the LC pathway. The reviewer’s feedback prompted us to conduct additional investigations to further strengthen the validity of our conclusions. We have incorporated these new data and analyses into the revised manuscript, and we believe that these revisions substantially enhance the robustness and reliability of our findings.

eLife Assessment

This important study provides evidence for dynamic coupling between translation initiation and elongation that can help maintain low ribosome density and translational homeostasis. The authors combine single-molecule imaging with a new approach to analyze mRNA translation kinetics using Bayesian modeling. This work is overall solid and will be of interest to those studying translational regulation.

Reviewer #1 (Public review):

Summary:

In this study, Lamberti et al. investigate how translation initiation and elongation are coordinated at the single-mRNA level in mammalian cells. The authors aim to uncover whether and how cells dynamically adjust initiation rates in response to elongation dynamics, with the overarching goal of understanding how translational homeostasis is maintained. To this end, the study combines single-molecule live-cell imaging using the SunTag system with a kinetic modeling framework grounded in the Totally Asymmetric Simple Exclusion Process (TASEP). By applying this approach to custom reporter constructs with different coding sequences, and under perturbations of the initiation/elongation factor eIF5A, the authors infer initiation and elongation rates from individual mRNAs and examine how these rates covary.

The central finding is that initiation and elongation rates are strongly correlated across a range of coding sequences, resulting in consistently low ribosome density ({less than or equal to}12% of the coding sequence occupied). This coupling is preserved under partial pharmacological inhibition of eIF5A, which slows elongation but is matched by a proportional decrease in initiation, thereby maintaining ribosome density. However, a complete genetic knockout of eIF5A disrupts this coordination, leading to reduced ribosome density, potentially due to changes in ribosome stalling resolution or degradation.

Strengths:

A key strength of this work is its methodological innovation. The authors develop and validate a TASEP-based Hidden Markov Model (HMM) to infer translation kinetics at single-mRNA resolution. This approach provides a substantial advance over previous population-level or averaged models and enables dynamic reconstruction of ribosome behavior from experimental traces. The model is carefully benchmarked against simulated data and appropriately applied. The experimental design is also strong. The authors construct matched SunTag reporters differing only in codon composition in a defined region of the coding sequence, allowing them to isolate the effects of elongation-related features while controlling for other regulatory elements. The use of both pharmacological and genetic perturbations of eIF5A adds robustness and depth to the biological conclusions. The results are compelling: across all constructs and conditions, ribosome density remains low, and initiation and elongation appear tightly coordinated, suggesting an intrinsic feedback mechanism in translational regulation. These findings challenge the classical view of translation initiation as the sole rate-limiting step and provide new insights into how cells may dynamically maintain translation efficiency and avoid ribosome collisions.

Assessment of Goals and Conclusions:

The authors successfully achieve their stated aims: they quantify translation initiation and elongation at the single-mRNA level and show that these processes are dynamically coupled to maintain low ribosome density. The modeling framework is well suited to this task, and the conclusions are supported by multiple lines of evidence, including inferred kinetic parameters, independent ribosome counts, and consistent behavior under perturbation.

Impact and Utility:

This work makes a significant conceptual and technical contribution to the field of translation biology. The modeling framework developed here opens the door to more detailed and quantitative studies of ribosome dynamics on single mRNAs and could be adapted to other imaging systems or perturbations. The discovery of initiation-elongation coupling as a general feature of translation in mammalian cells will likely influence how researchers think about translational regulation under homeostatic and stress conditions.

The data, models, and tools developed in this study will be of broad utility to the community, particularly for researchers studying translation dynamics, ribosome behavior, or the effects of codon usage and mRNA structure on protein synthesis.

Context and Interpretation:

This study contributes to a growing body of evidence that translation is not merely controlled at initiation but involves feedback between elongation and initiation. It supports the emerging view that ribosome collisions, stalling, and quality control pathways play active roles in regulating initiation rates in cis. The findings are consistent with recent studies in yeast and metazoans showing translation initiation repression following stalling events. However, the mechanistic details of this feedback remain incompletely understood and merit further investigation, particularly in physiological or stress contexts.

In summary, this is a thoughtfully executed and timely study that provides valuable insights into the dynamic regulation of translation and introduces a modeling framework with broad applicability. It will be of interest to a wide audience in molecular biology, systems biology, and quantitative imaging.

Reviewer #2 (Public review):

Summary:

This manuscript uses single-molecule run-off experiments and TASEP/HMM models to estimate biophysical parameters, i.e., ribosomal initiation and elongation rates. Combining inferred initiation and elongation rates, the authors quantify ribosomal density. TASEP modeling was used to simulate the mechanistic dynamics of ribosomal translation, and the HMM is used to link ribosomal dynamics to microscope intensity measurements. The authors' main conclusions and findings are:

- Ribosomal elongation rates and initiation rates are strongly coordinated.

- Elongation rates were estimated between 1 and 4.5 aa/sec. Initiation rates were estimated between 1 and 2 ribosomes/min. These values agree with previously reported ones.

- Ribosomal density was determined to be below 12% for all constructs and conditions.

- eIF5A-perturbations (GC7 inhibition) resulted in non-significant changes in translational bursting and ribosome density.

- eIF5A perturbations affected both elongation and initiation rates.

Strengths:

This manuscript presents an interesting scientific hypothesis to study ribosome initiation and elongation concurrently. This topic is relevant for the field. The manuscript presents a novel quantitative methodology to estimate ribosomal initiation rates from Harringtonine run-off assays. This is relevant because run-off assays have been used to estimate, exclusively, elongation rates.

Comments on revisions:

The authors have addressed my concerns. Specifically, they have expanded the discussion on unexpected eIF5A perturbation results, calculated CAI values for all constructs, and made code and data publicly available via GitHub and Zenodo. The mathematical notation is now consistent, and all variables are properly defined.

Reviewer #3 (Public review):

Disclaimer:

My expertise is in live single-molecule imaging of RNA and transcription, as well as associated data analysis and modeling. While this aligns well with the technical aspects of the manuscript, my background in translation is more limited, and I am not best positioned to assess the novelty of the biological conclusions.

Summary:

This study combines live-cell imaging of nascent proteins on single mRNAs with time-series analysis to investigate the kinetics of mRNA translation.<br /> The authors (i) used a calibration method for estimating absolute ribosome counts, and (ii) developed a new Bayesian approach to infer ribosome counts over time from run-off experiments, enabling estimation of elongation rates and ribosome density across conditions.

They report (i) translational bursting at the single-mRNA level, (ii) low ribosome density (~10% occupancy {plus minus} a few percents), (iii) that ribosome density is minimally affected by perturbations of elongation (using a drug and/or different coding sequences in the reporter), suggesting a homeostatic mechanism potentially involving a feedback of elongation onto initiation, although (iv) this coupling breaks down upon knockout of elongation factor eIF5A.

Strengths:

(1) The manuscript is well written and the conclusions are in general appropriately cautious (besides the few improvements I suggest below).

(2) The time-series inference method is interesting and promising for broader application.

(3) Simulations provide convincing support for the modeling (though some improvements are possible).

(4) The reported homeostatic effect on ribosome density is surprising and carefully validated with multiple perturbations.

(5) Imaging quality and corrections (e.g., flat-fielding, laser power measurements) are robust.

(6) Mathematical modeling is clearly described and precise; a few clarifications could improve it further.

Weaknesses:

(1) The absolute quantification of ribosome numbers (via the measurement of $i_{MP}$​) should be improved. This only affects the finding that ribosome density is low, not that it appears to be under homeostatic control. However, if $i_{MP}$​ turns out to be substantially overestimated (hence ribosome density underestimated), then "ribosomes queuing up to the initiation site and physically blocking initiation" could become a relevant hypothesis. In my first review of this work, I made recommendations, which the authors did not follow. In my view, the robustness of this particular aspect of this study remains moderate.

(2) The proposed initiation-elongation coupling is plausible, but alternative explanations such as changes in abortive elongation frequency should be considered. In their response to my previous comments, the authors indicate that this is "beyond the scope of the present work".

(3) More an opportunity for improvement than a weakness: It is unclear what the single-mRNA nature of the inference method is bringing since it is only used here to report _average_ ribosome elongation rate and density (averaged across mRNAs and across time during the run-off experiments -although the method, in principle, has the power to resolve these two aspects). In response to my previous comment, the authors note that such analyses could be incorporated in future work.

Author response:

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

Public Reviews:

Reviewer #1 (Public review): 

Summary:

In this study, Lamberti et al. investigate how translation initiation and elongation are coordinated at the single-mRNA level in mammalian cells. The authors aim to uncover whether and how cells dynamically adjust initiation rates in response to elongation dynamics, with the overarching goal of understanding how translational homeostasis is maintained. To this end, the study combines single-molecule live-cell imaging using the SunTag system with a kinetic modeling framework grounded in the Totally Asymmetric Simple Exclusion Process (TASEP). By applying this approach to custom reporter constructs with different coding sequences, and under perturbations of the initiation/elongation factor eIF5A, the authors infer initiation and elongation rates from individual mRNAs and examine how these rates covary.

The central finding is that initiation and elongation rates are strongly correlated across a range of coding sequences, resulting in consistently low ribosome density ({less than or equal to}12% of the coding sequence occupied). This coupling is preserved under partial pharmacological inhibition of eIF5A, which slows elongation but is matched by a proportional decrease in initiation, thereby maintaining ribosome density. However, a complete genetic knockout of eIF5A disrupts this coordination, leading to reduced ribosome density, potentially due to changes in ribosome stalling resolution or degradation.

Strengths:

A key strength of this work is its methodological innovation. The authors develop and validate a TASEP-based Hidden Markov Model (HMM) to infer translation kinetics at single-mRNA resolution. This approach provides a substantial advance over previous population-level or averaged models and enables dynamic reconstruction of ribosome behavior from experimental traces. The model is carefully benchmarked against simulated data and appropriately applied. The experimental design is also strong. The authors construct matched SunTag reporters differing only in codon composition in a defined region of the coding sequence, allowing them to isolate the effects of elongation-related features while controlling for other regulatory elements. The use of both pharmacological and genetic perturbations of eIF5A adds robustness and depth to the biological conclusions. The results are compelling: across all constructs and conditions, ribosome density remains low, and initiation and elongation appear tightly coordinated, suggesting an intrinsic feedback mechanism in translational regulation. These findings challenge the classical view of translation initiation as the sole rate-limiting step and provide new insights into how cells may dynamically maintain translation efficiency and avoid ribosome collisions.

We thank the reviewer for their constructive assessment of our work, and for recognizing the methodological innovation and experimental rigor of our study.

Weaknesses:

A limitation of the study is its reliance on exogenous reporter mRNAs in HeLa cells, which may not fully capture the complexity of endogenous translation regulation. While the authors acknowledge this, it remains unclear how generalizable the observed coupling is to native mRNAs or in different cellular contexts.

We agree that the use of exogenous reporters is a limitation inherent to the SunTag system, for which there is currently no simple alternative for single-mRNA translation imaging. However, we believe our findings are likely generalizable for several reasons.

As discussed in our introduction and discussion, there is growing mechanistic evidence in the literature for coupling between elongation (ribosome collisions) and initiation via pathways such as the GIGYF2-4EHP axis (Amaya et al. 2018, Hickey et al. 2020, Juszkiewicz et al. 2020), which might operate on both exogenous and endogenous mRNAs.

As already acknowledged in our limitations section, our exogenous reporters may not fully recapitulate certain aspects of endogenous translation (e.g., ER-coupled collagen processing), yet the observed initiation-elongation coupling was robust across all tested constructs and conditions.

We have now expanded the Discussion (L393-395) to cite complementary evidence from Dufourt et al. (2021), who used a CRISPR-based approach in Drosophila embryos to measure translation of endogenous genes. We also added a reference to Choi et al. 2025, who uses a ER-specific SunTag reporter to visualize translation at the ER (L395-397).

Additionally, the model assumes homogeneous elongation rates and does not explicitly account for ribosome pausing or collisions, which could affect inference accuracy, particularly in constructs designed to induce stalling. While the model is validated under low-density assumptions, more work may be needed to understand how deviations from these assumptions affect parameter estimates in real data.

We agree with the reviewer that the assumption of homogeneous elongation rates is a simplification, and that our work represents a first step towards rigorous single-trace analysis of translation dynamics. We have explicitly tested the robustness of our model to violations of the low-density assumption through simulations (Figure 2 - figure supplement 2). These show that while parameter inference remains accurate at low ribosome densities, accuracy slightly deteriorates at higher densities, as expected. In fact, our experimental data do provide evidence for heterogeneous elongation: the waiting times between termination events deviate significantly from an exponential distribution (Figure 3 - figure supplement 2C), indicating the presence of ribosome stalling and/or bursting, consistent with the reviewer's concern. We acknowledge in the Limitations section (L402-406) that extending the model to explicitly capture transcript-dependent elongation rates and ribosome interactions remains challenging. The TASEP is difficult to solve analytically under these conditions, but we note that simulation-based inference approaches, such as particle filters to replace HMMs, could provide a path forward for future work to capture this complexity at the single-trace level.

Furthermore, although the study observes translation "bursting" behavior, this is not explicitly modeled. Given the growing recognition of translational bursting as a regulatory feature, incorporating or quantifying this behavior more rigorously could strengthen the work's impact.

While we do not explicitly model the bursting dynamics in the HMM framework, we have quantified bursting behavior directly from the data. Specifically, we measure the duration of translated (ON) and untranslated (OFF) periods across all reporters and conditions (Figure 1G for control conditions and Figure 4G-H for perturbed conditions), finding that active translation typically lasts 10-15 minutes interspersed with shorter silent periods of 5-10 minutes. This empirical characterization demonstrates that bursting is a consistent feature of translation across our experimental conditions. The average duration of silent periods is similar to what was inferred by Livingston et al. 2023 for a similar SunTag reporter; while the average duration of active periods is substantially shorter (~15 min instead of ~40 min), which is consistent with the shorter trace duration in our system compared to theirs (~15 min compared to ~80 min, on average). Incorporating an explicit two-state or multi-state bursting model into the TASEP-HMM framework would indeed be computationally intensive and represents an important direction for future work, as it would enable inference of switching rates alongside initiation and elongation parameters. We have added this point to the Discussion (L415-417).

Assessment of Goals and Conclusions:

The authors successfully achieve their stated aims: they quantify translation initiation and elongation at the single-mRNA level and show that these processes are dynamically coupled to maintain low ribosome density. The modeling framework is well suited to this task, and the conclusions are supported by multiple lines of evidence, including inferred kinetic parameters, independent ribosome counts, and consistent behavior under perturbation.

Impact and Utility:

This work makes a significant conceptual and technical contribution to the field of translation biology. The modeling framework developed here opens the door to more detailed and quantitative studies of ribosome dynamics on single mRNAs and could be adapted to other imaging systems or perturbations. The discovery of initiation-elongation coupling as a general feature of translation in mammalian cells will likely influence how researchers think about translational regulation under homeostatic and stress conditions.

The data, models, and tools developed in this study will be of broad utility to the community, particularly for researchers studying translation dynamics, ribosome behavior, or the effects of codon usage and mRNA structure on protein synthesis.

Context and Interpretation:

This study contributes to a growing body of evidence that translation is not merely controlled at initiation but involves feedback between elongation and initiation. It supports the emerging view that ribosome collisions, stalling, and quality control pathways play active roles in regulating initiation rates in cis. The findings are consistent with recent studies in yeast and metazoans showing translation initiation repression following stalling events. However, the mechanistic details of this feedback remain incompletely understood and merit further investigation, particularly in physiological or stress contexts. 

In summary, this is a thoughtfully executed and timely study that provides valuable insights into the dynamic regulation of translation and introduces a modeling framework with broad applicability. It will be of interest to a wide audience in molecular biology, systems biology, and quantitative imaging.

We appreciate the reviewer's thorough and positive assessment of our work, and that they recognize both the technical innovation of our modeling framework and its potential broad utility to the translation biology community. We agree that further mechanistic investigation of initiation-elongation feedback under various physiological contexts represents an important direction for future research.

Reviewer #2 (Public review):

Summary:

This manuscript uses single-molecule run-off experiments and TASEP/HMM models to estimate biophysical parameters, i.e., ribosomal initiation and elongation rates. Combining inferred initiation and elongation rates, the authors quantify ribosomal density. TASEP modeling was used to simulate the mechanistic dynamics of ribosomal translation, and the HMM is used to link ribosomal dynamics to microscope intensity measurements. The authors' main conclusions and findings are:

(1) Ribosomal elongation rates and initiation rates are strongly coordinated.

(2) Elongation rates were estimated between 1-4.5 aa/sec. Initiation rates were estimated between 0.5-2.5 events/min. These values agree with previously reported values.

(3) Ribosomal density was determined below 12% for all constructs and conditions.

(4) eIF5A-perturbations (KO and GC7 inhibition) resulted in non-significant changes in translational bursting and ribosome density.

(5) eIF5A perturbations resulted in increases in elongation and decreases in initiation rates.

Strengths:

This manuscript presents an interesting scientific hypothesis to study ribosome initiation and elongation concurrently. This topic is highly relevant for the field. The manuscript presents a novel quantitative methodology to estimate ribosomal initiation rates from Harringtonine run-off assays. This is relevant because run-off assays have been used to estimate, exclusively, elongation rates.

We thank the reviewer for their careful evaluation of our work and for recognizing the novelty of our quantitative methodology to extract both initiation and elongation rates from harringtonine run-off assays, extending beyond the traditional use of these experiments.

Weaknesses:

The conclusion of the strong coordination between initiation and elongation rates is interesting, but some results are unexpected, and further experimental validation is needed to ensure this coordination is valid. 

We agree that some of our findings need further experimental investigation in future studies. However, we believe that the coordination between initiation and elongation is supported by multiple results in our current work: (1) the strong correlation observed across all reporters and conditions (Figure 3E), and (2) the consistent maintenance of low ribosome density despite varying elongation rates. While additional experimental validation would be valuable, we note that directly manipulating initiation or elongation independently in mammalian cells remains technically challenging. Nevertheless, our findings are consistent with emerging mechanistic understanding of collision-sensing pathways (GIGYF2-4EHP) that could mediate such coupling, as discussed in our manuscript.

(1) eIF5a perturbations resulted in a non-significant effect on the fraction of translating mRNA, translation duration, and bursting periods. Given the central role of eIF5a, I would have expected a different outcome. I would recommend that the authors expand the discussion and review more literature to justify these findings.

We appreciate this comment. This finding is indeed discussed in detail in our manuscript (Discussion, paragraphs 6-7). As we note there, while eIF5A plays a critical role in elongation, the maintenance of bursting dynamics and ribosome density upon perturbation can be explained by compensatory feedback mechanisms. Specifically, the coordinated decrease in initiation rates that counterbalances slower elongation to maintain homeostatic ribosome density. We also discuss several factors that complicate interpretation: (1) potential RQC-mediated degradation masking stronger effects in proline-rich constructs, (2) differences between GC7 treatment and genetic knockout suggesting altered stalling resolution kinetics, and (3) the limitations of using exogenous reporters that lack ER-coupled processing, which may be critical for eIF5A function in endogenous collagen translation (as suggested by Rossi et al., 2014; Mandal et al., 2016; Barba-Aliaga et al., 2021). The mechanistic complexity and tissue-specific nature of eIF5A function in mammals, which differs substantially from the better-characterized yeast system, likely contributes to the nuanced phenotype we observe. We believe our Discussion adequately addresses these points.

(2) The AAG construct leading to slow elongation is very surprising. It is the opposite of the field consensus, where codon-optimized gene sequences are expected to elongate faster. More information about each construct should be provided. I would recommend more bioinformatic analysis on this, for example, calculating CAI for all constructs, or predicting the structures of the proteins.

We agree that the slow elongation of the AAG construct is counterintuitive and indeed surprising. Following the reviewer's suggestion, we have now calculated the Codon Adaptation Index (CAI) for all constructs (Renilla 0.89, Col1a1 0.78, Col1a1 mutated 0.74). It is therefore unlikely that codon bias explains the slow translation, particularly since we designed the mutated Col1a1 construct with alanine codons selected to respect human codon usage bias, thereby minimizing changes in codon optimality. As we discuss in the manuscript, we hypothesize that the proline-to-alanine substitutions disrupted co-translational folding of the collagen-derived sequence. Prolines are critical for collagen triple-helix formation (Shoulders and Raines, 2009), and their replacement with alanines likely generates misfolded intermediates that cause ribosome stalling (Barba-Aliaga et al., 2021; Komar et al., 2024). This interpretation is supported by the high frequency (>30%) of incomplete run-off traces for AAG, suggesting persistent stalling events. Our findings thus illustrate an important potential caveat: "optimizing" a sequence based solely on codon usage can be detrimental when it disrupts functionally important structural features or co-translational folding pathways.

This highlights that elongation rates depend not only on codon optimality but also on the interplay between nascent chain properties and ribosome progression.

(3) The authors should consider using their methodology to study the effects of modifying the 5'UTR, resulting in changes in initiation rate and bursting, such as previously shown in reference Livingston et al., 2023. This may be outside of the scope of this project, but the authors could add this as a future direction and discuss if this may corroborate their conclusions. 

We thank the reviewer for this excellent suggestion. We agree that applying our methodology to 5'-UTR variants would provide a complementary test of initiation-elongation coupling, and we have now added this as a future direction in the Discussion (L417-420).

(4) The mathematical model and parameter inference routines are central to the conclusions of this manuscript. In order to support reproducibility, the computational code should be made available and well-documented, with a requirements file indicating the dependencies and their versions. 

We have added the Github link in the manuscript (https://github.com/naef-lab/suntag-analysis) and have also deposited the data (.ome.tif) on Zenodo (https://zenodo.org/records/17669332).

Reviewer #3 (Public review):

Disclaimer:

My expertise is in live single-molecule imaging of RNA and transcription, as well as associated data analysis and modeling. While this aligns well with the technical aspects of the manuscript, my background in translation is more limited, and I am not best positioned to assess the novelty of the biological conclusions.

Summary:

This study combines live-cell imaging of nascent proteins on single mRNAs with time-series analysis to investigate the kinetics of mRNA translation.

The authors (i) used a calibration method for estimating absolute ribosome counts, and (ii) developed a new Bayesian approach to infer ribosome counts over time from run-off experiments, enabling estimation of elongation rates and ribosome density across conditions.

They report (i) translational bursting at the single-mRNA level, (ii) low ribosome density (~10% occupancy

{plus minus} a few percents), (iii) that ribosome density is minimally affected by perturbations of elongation (using a drug and/or different coding sequences in the reporter), suggesting a homeostatic mechanism potentially involving a feedback of elongation onto initiation, although (iv) this coupling breaks down upon knockout of elongation factor eIF5A.

Strengths:

(1) The manuscript is well written, and the conclusions are, in general, appropriately cautious (besides the few improvements I suggest below).

(2) The time-series inference method is interesting and promising for broader applications. 

(3) Simulations provide convincing support for the modeling (though some improvements are possible). 

(4) The reported homeostatic effect on ribosome density is surprising and carefully validated with multiple perturbations.

(5) Imaging quality and corrections (e.g., flat-fielding, laser power measurements) are robust.

(6) Mathematical modeling is clearly described and precise; a few clarifications could improve it further.

We thank the reviewer for recognizing the novelty of the approach and its rigour, and for providing suggestions to improve it further.

Weaknesses:

(1) The absolute quantification of ribosome numbers (via the measurement of $i_{MP}$ ) should be improved.This only affects the finding that ribosome density is low, not that it appears to be under homeostatic control. However, if $i_{MP}$ turns out to be substantially overestimated (hence ribosome density underestimated), then "ribosomes queuing up to the initiation site and physically blocking initiation" could become a relevant hypothesis. In my detailed recommendations to the authors, I list points that need clarification in their quantifications and suggest an independent validation experiment (measuring the intensity of an object with a known number of GFP molecules, e.g., MS2-GFP MS2-GFP-labeled RNAs, or individual GEMs).

We agree with the reviewer that the estimation of the number of ribosomes is central to our finding that translation happens at low density on our reporters. This result derives from our measurement of the intensity of one mature protein (i_MP), that we have achieved by using a SunTag reporter with a RH1 domain in the C terminus of the mature protein, allowing us to stabilise mature proteins via actin-tethering. In addition, as suggested by the reviewer, we already validated this result with an independent estimate of the mature protein intensity (Figure 5 - figure supplement 2B), which was obtained by adding the mature protein intensity directly as a free parameter of the HMM. The inferred value of mature protein intensity for each construct (10-15 a.u) was remarkably close to the experimental calibration result (14 ± 2 a.u.). Therefore, we have confidence that our absolute quantification of ribosome numbers is accurate.

(2) The proposed initiation-elongation coupling is plausible, but alternative explanations, such as changes in abortive elongation frequency, should be considered more carefully. The authors mention this possibility, but should test or rule it out quantitatively. 

We thank the reviewer for the comment, but we consider that ruling out alternative explanations through new perturbation experiments is beyond the scope of the present work.

(3) The observation of translational bursting is presented as novel, but similar findings were reported by Livingston et al. (2023) using a similar SunTag-MS2 system. This prior work should be acknowledged, and the added value of the current approach clarified.

We did cite Livingston et al. (2023) in several places, but we recognized that we could add a few citations in key places, to make clear that the observation of bursting is not novel but is in agreement with previous results. We now did so in the Results and Discussion sections.

(4) It is unclear what the single-mRNA nature of the inference method is bringing since it is only used here to report _average_ ribosome elongation rate and density (averaged across mRNAs and across time during the run-off experiments - although the method, in principle, has the power to resolve these two aspects).

While decoding individual traces, our model infers shared (population-level) rates. Inferring transcript-specific parameters would be more informative, but it is highly challenging due to the uncertainty on the initial ribosome distribution on single transcripts. Pooling multiple transcripts together allows us to use some assumptions on the initial distribution and infer average elongation and initiation-rate parameters, while revealing substantial mRNA-to-mRNA variability in the posterior decoding (e.g. Figure 3 - figure Supplement 2C). Indeed, the inference still informs on the single-trace run-off time distribution (Figure 3 A) and the waiting time between termination events (Figure 3 - figure supplement 2C), suggesting the presence of stalling and bursting. In addition, the transcript-to-transcript heterogeneity is likely accounted for by our model better than previous methods (linear fit of the average run-off intensity), as suggested by their comparison (Figure 3 - figure supplement 2 A). In the future the model could be refined by introducing transcript-specific parameters, possibly in a hierarchical way, alongside shared parameters.

(5) I did not find any statement about data availability. The data should be made available. Their absence limits the ability to fully assess and reproduce the findings.

We have added the Github link in the manuscript (https://github.com/naef-lab/suntag-analysis) and have also deposited the data (.ome.tif) on Zenodo (https://zenodo.org/records/17669332).

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors): 

Major Comments:

(1) Lack of Explicit Bursting Model

Although translation "bursts" are observed, the current framework does not explicitly model initiation as a stochastic ON/OFF process. This limits insight into regulatory mechanisms controlling burst frequency or duration. The authors should either incorporate a two-state/more-state (bursting) model of initiation or perform statistical analysis (e.g., dwell-time distributions) to quantify bursting dynamics. They should clarify how bursting influences the interpretation of initiation rate estimates.

We agree with the reviewer that an explicit bursting model (e.g., a two-state telegraph model) would be the ideal theoretical framework. However, integrating such a model into the TASEP-HMM inference framework is computationally intensive and complex. As a robust first step, we have opted to quantify bursting empirically based on the decoded single-mRNA traces. As shown in Figure 1G (control) and Figure 4G (perturbed conditions), we explicitly measured the duration of "ON" (translated) and "OFF" (untranslated) periods. This statistical analysis provides a quantitative description of the bursting dynamics without relying on the specific assumptions of a telegraph model. We have clarified this in the text (L123-125) and, as suggested, added a discussion (L415-417) on the potential extensions of the model to include explicit switching kinetics in the Outlook section.

(2) Assumption of Uniform Elongation Rates

The model assumes homogeneous elongation across coding sequences, which may not hold for stalling-prone inserts (e.g., PPG). This simplification could bias inference, particularly in cases of sequence-specific pausing. Adding simulations or sensitivity analysis to assess how non-uniform elongation affects the accuracy of inferred parameters. The authors should explicitly discuss how ribosome stalling, collisions, or heterogeneity might skew model outputs (see point 4).

A strong stalling sequence that affects all ribosomes equally should not deteriorate the inference of the initiation rate, provided that the low-density assumption holds. The scenario where stalling events lead to higher density, and thus increased ribosome-ribosome interactions, is comparable to the conditions explored in Figure 2E. In those simulations, we tested the inference on data generated with varying initiation and elongation rates, resulting in ribosome densities ranging from low to high. We demonstrated that the inference remains robust at low ribosome densities (<10%). At higher densities, the accuracy of the initiation rate estimate decreases, whereas the elongation rate estimate remains comparatively robust. Additionally, the model tends to overestimate ribosome density under high-density conditions, likely because it neglects ribosome interference at the initiation site (Figure 2 figure supplement 2C). We agree that a deeper investigation into the consequences of stochastic stalling and bursting would be beneficial, and we have explicitly acknowledged this in the Limitations section.

(3) Interpretation of eIF5A Knockout Phenotype

The observation that eIF5A KO reduces initiation more than elongation, leading to decreased ribosome density, is biologically intriguing. However, the explanation invoking altered RQC kinetics is speculative and not directly tested. The authors should consider validating the RQC hypothesis by monitoring reporter mRNA stability, ribosome collision markers, or translation termination intermediates.

We thank the reviewer for the comment, but we consider that ruling out alternative explanations through new experiments is beyond the scope of the present work.

(4) To strengthen the manuscript, the authors should incorporate insights from three studies.

- Livingston et al. (PMC10330622) found that translation occurs in bursts, influenced by mRNA features and initiation factors, supporting the coupling of initiation and elongation.

- Madern et al. (PMID: 39892379) demonstrated that ribosome cooperativity enhances translational efficiency, highlighting coordinated ribosome behavior.

- Dufourt et al. (PMID: 33927056) observed that high initiation rates correlate with high elongation rates, suggesting a conserved mechanism across cell cultures and organisms.

Integrating these studies could enrich the manuscript's interpretation and stimulate new avenues of thought.

We thank the reviewer for the valuable comment. We added citations of Livingston et al. in the context of translational bursting. We already cited Madern et al. in multiple places and, although its observations of ribosome cooperativity are very compelling, they cannot be linked with our observations of a feedback between initiation and elongation, and it would be very challenging to see a similar effect on our reporters. This is why we did not expressly discuss cooperativity. We also integrated Dufourt et al. in the Discussion about the possibility of designing genetically-encoded reporter. We also added a sentence about the possibility of using an ER-specific SunTag reporter, as done recently in Choi et al., Nature (2025) (https://doi.org/10.1038/s41586-025-09718-0).

Minor Comments:

(1) Use consistent naming for SunTag reporters (e.g., "PPG" vs "proline-rich") throughout.

Thank you for the comment. However, the term proline-rich always appears together with PPG, so we believe that the naming is clear and consistent.

(2) Consider a schematic overview of the experimental design and modeling pipeline for accessibility.

Thank you for the suggestion. We consider that experimental design and modeling is now sufficiently clearly described and does not justify an additional scheme. 

(3) Clarify how incomplete run-off traces are handled in the HMM inference.

Incomplete run-off traces are treated identically to complete traces in our HMM inference. This is possible because our model relies on the probability of transitions occurring per time step to infer rates. It does not require observing the final "empty" state to estimate the kinetic parameters ɑ and λ. The loss of signal (e.g., mRNA moving out of the focal volume or photobleaching) does not invalidate the kinetic information contained in the portion of the trace that was observed. We have clarified this in the Methods section.

Reviewer #2 (Recommendations for the authors):

(1) Reproducibility:

(1.1) The authors should use a GitHub repository with a timestamp for the release version.

The code is available on GitHub (https://github.com/naef-lab/suntag-analysis).

(1.2) Make raw images and data available in a figure repository like Figshare.

The raw images (.ome.tif) are now available on Zenodo (https://zenodo.org/records/17669332).

(2) Paper reorganization and expansion of the intensity and ribosome quantification:

(2.1) Given the relevance of the initiation and elongation rates for the conclusions of this study, and the fact that the authors inferred these rates from the spot intensities. I recommend that the authors move Figure 1 Supplement 2 to the main text and expand the description of the process to relate spot intensity and number of ribosomes. Please also expand the figure caption for this image.

We agree with the importance of this validation. We have expanded the description of the calibration experiment in the main text and in the figure caption.

(2.2) I suggest the authors explicitly mention the use of HMM in the abstract.

We have now explicitly mentioned the TASEP-based HMM in the abstract.

(2.3) In line 492, please add the frame rate used to acquire the images for the run-off assays.

We have added the specific frame rate (one frame every 20 seconds) to the relevant section.

(3) Figures and captions:

(3.1) Figure 1, Supplement 2. Please add a description of the colors used in plots B, C. 

We have expanded the caption and added the color description.

(3.2) In the Figure 2 caption. It is not clear what the authors mean by "traceseLife". Please ensure it is not a typo.

Thank you for spotting this. We have corrected the typo.

(3.3) Figure 1 A, in the cartoon N(alpha)->N-1, shouldn't the transition also depend on lambda?

The transition probability was explicitly derived in the “Bayesian modeling of run-off traces” section (Eqs. 17-18), and does not depend on λ, but only on the initiation rate under the low-density assumption.

(3.4) Figure 3, Supplement 2. "presence of bursting and stalling.." has a typo.

Corrected.

(3.5) Figure 5, panel C, the y-axis label should be "run-off time (min)."

Corrected.

(3.6) For most figures, add significance bars.

(3.7) In the figure captions, please add the total number of cells used for each condition.

We have systematically indicated the number of traces (n_t) and the number of independent experiments (n_e) in the captions in this format (n_t, n_e).

(4) Mathematical Methods:

We greatly thank the reviewer for their detailed attention to the mathematical notation. We have addressed all points below.

(4.1) In lines 555, Materials and Methods, subsection, Quantification of Intensity Traces, multiple equations are not numbered. For example, after Equation (4), no numbers are provided for the rest of the equations. Please keep consistency throughout the whole document.

We have ensured that all equations are now consistently numbered throughout the document.

(4.2) In line 588, the authors mention "$X$ is a standard normal random variable with mean $\mu$ and standard deviation $s_0$". Please ensure this is correct. A standard normal random variable has a 0 mean and std 1. 

Thank you for the suggestion, we have corrected the text (L678).

(4.3) Line 546, Equation 2. The authors use mu(x,y) to describe a 2d Gaussian function. But later in line 587, the authors reuse the same variable name in equation 5 to redefine the intensity as mu = b_0 + I.

We have renamed the 2D Gaussian function to \mu_{2D}(x,y) in the spot tracking section

(4.4) For the complete document, it could be beneficial to the reader if the authors expand the definition of the relationship between the signal "y" and the spot intensity "I". Please note how the paragraph in lines 582-587 does not properly introduce "y".

We have added an explicit definition of y and its relationship to the underlying spot intensity I in the text to improve readability and clarity.

(4.5) Please ensure consistency in variable names. For example, "I" is used in line 587 for the experimental spot intensity, then line 763 redefines I(t) as the total intensity obtained from the TASEP model; please use "I_sim(t)" for simulated intensities. Please note that reusing the variable "I" for different contexts makes it hard for the reader to follow the text. 

We agree that this was confusing. We have implemented the suggestion and now distinguish simulated intensities using the notation I_S .

(4.6) Line 555 "The prior on the total intensity I is an "uninformative" prior" I ~ half_normal(1000). Please ensure it is not "I_0 ~ half_normal(1000)."? 

We confirm that “I” is the correct variable representing the total intensity in this context; we do not use an “I₀” variable here.

(4.7) In lines 595, equation 6. Ensure that the equation is correct. Shouldn't it be: s_0^2 = ln ( 1 + (sigma_meas^2 / ⟨y⟩^2) )? Please ensure that this is correct and it is not affecting the calculated values given in lines 598.

Thank you for catching this typo. We have corrected the equation in the manuscript. We confirm that the calculations performed in the code used the correct formula, so the reported values remain unchanged.

(4.8) In line 597, "the mean intensity square ^2". Please ensure it is not "the square of the temporal mean intensity."

We have corrected the text to "the square of the temporal mean intensity."

(4.9) In lines 602-619, Bayesian modeling of run-off traces, please ensure to introduce the constant "\ell". Used to define the ribosomal footprint?

We have added the explicit definition of 𝓁 as the ribosome footprint size (length of transcript occupied by one ribosome) in the "Bayesian modeling of run-off traces" section.

(4.10) Line 687 has a minor typo "[...] ribosome distribution.. Then, [...]"

We have corrected the punctuation.

(4.11) In line 678, Equation 19 introduces the constant "L_S", Please ensure that it is defined in the text.

We have added the explicit definition of L_S (the length of the SunTag) to the text surrounding Equation 19.

(4.12) In line 695, Equation 22, please consider using a subscript to differentiate the variance due to ribosome configuration. For example, instead of "sigma (...)^2" use something like "sigma_c ^2 (...)". Ensure that this change is correctly applied to Equation 24 and all other affected equations.

Thank you, we have implemented the suggestions.

(4.13) In line 696, please double-check equations 26 and 27. Specifically, the denominator ^2. Given the previous text, it is hard to follow the meaning of this variable. 

We have revised the notation in Equations 26 and 27 to ensure the denominator is consistent with the definitions provided in the text.

(4.14) In lines 726, the authors mention "[...], but for the purposes of this dissertation [...]", it should be "[...], but for the purposes of this study [...]"

Thank you for spotting this. We have replaced "dissertation" with "study."

(4.15) Equations 5, 28, 37, and the unnumbered equation between Equations 16 and 17 are similar, but in some, "y" does not explicitly depend on time. Please ensure this is correct. 

We have verified these equations and believe they are correct.

(4.16) Please review the complete document and ensure that variables and constants used in the equations are defined in the text. Please ensure that the same variable names are not reused for different concepts. To improve readability and flow in the text, please review the complete Materials and Methods sections and evaluate if the modeling section can be written more clearly and concisely. For example, Equation 28 is repeated in the text.

We have performed a comprehensive review of the Materials and Methods section. To improve conciseness and flow, we have merged the subsection “Observation model and estimation of observation parameters” with the “Bayesian modeling of run-off traces” section. This allowed us to remove redundant definitions and repeated equations (such as the previous Equation 28). We have also checked that all variables and constants are defined upon first use and that variable names remain consistent throughout the manuscript.

Reviewer #3 (Recommendations for the authors):

(1) Data Presentation

(1.1) In main Figures 1D and 4E, the traces appear to show frequent on-off-on transitions ("bursting"), but in supplementary figures (1-S1A and 4-S1A), this behavior is seen in only ~8 of 54 traces. Are the main figure examples truly representative?

We acknowledge the reviewer's point. In Figure 1D, we selected some of the longest and most illustrative traces to highlight the bursting dynamics. We agree that the term "representative" might be misleading if interpreted as "average." We have updated the text to state "we show bursting traces" to more accurately reflect the selection.

(1.2) There are 8 videos, but I could not identify which is which.

Thank you for pointing this out. We have renamed the video files to clearly correspond to the figures and conditions they represent.

(2) Data Availability:

As noted above, the data should be shared. This is in accordance with eLife's policy: "Authors must make all original data used to support the claims of the paper, or that are required to reproduce them, available in the manuscript text, tables, figures or supplementary materials, or at a trusted digital repository (the latter is recommended). [...] eLife considers works to be published when they are posted as preprints, and expects preprints we review to meet the standards outlined here." Access to the time traces would have been helpful for reviewers.

We have now added the Github link for the code (https://github.com/naef-lab/suntag-analysis) and deposited the raw data (.ome.tif files) on Zenodo (10.5281/zenodo.17669332).

(3) Model Assumptions:

(3.1) The broad range of run-off times (Figure 3A) suggests stalling, which may be incompatible with the 'low-density' assumption used on the TASEP model, which essentially assumes that ribosomes do not bump into each other. This could impact the validity of the assumptions that ribosomes behave independently, elongate at constant speed (necessary for the continuum-limit approximation), and that the rate-limiting step is the initiation. How robust are the inferences to this assumption?

We agree that the deviation of waiting times from an exponential distribution (Figure 3 - figure supplement 2C) suggests the presence of stalling, which challenges the strict low-density assumption and constant elongation speed. We explicitly explored the robustness of our model to higher ribosome densities in simulations. As shown in Figure 2 - figure supplement 2, while the model accuracy for single parameters deteriorates at very high densities (overestimating density due to neglected interference), it remains robust for estimating global rates in the regime relevant to our data. We have expanded the discussion on the limitations of the low density and homogeneous elongation rate assumptions in the text (L404-408).

(3.2) Since all constructs share the same SunTag region, elongation rates should be identical there and diverge only in the variable region. This would affect $\gamma (t)$ and hence possibly affect the results. A brief discussion would be helpful.

This is a valid point. Currently, our model infers a single average elongation rate that effectively averages the behavior over the SunTag and the variable CDS regions. Modeling distinct rates for these regions would be a valuable extension but adds significant complexity. While our current "effective rate" approach might underestimate the magnitude of differences between reporters, it captures the global kinetic trend. We have added a brief discussion acknowledging this simplification (L408-412).

(3.3) A similar point applies to the Gillespie simulations: modeling the SunTag region with a shared elongation rate would be more accurate.

We agree. Simulating distinct rates for the SunTag and CDS would increase realism, though our current homogeneous simulations serve primarily to benchmark the inference framework itself. We have noted this as a potential future improvement (L413-414).

(3.4) Equation (13) assumes that switching between bursting and non-bursting states is much slower than the elongation time. First, this should be made explicit. Second, this is not quite true (~5 min elongation time on Figure 3-s2A vs ~5-15min switching times on Figure 1). It would be useful to show the intensity distribution at t=0 and compare it to the expected mixture distribution (i.e., a Poisson distribution + some extra 'N=0' cells). 

We thank the reviewer for this insightful comment. We have added a sentence to the text explicitly stating the assumption that switching dynamics are slower than the translation time. While the timescales are indeed closer than ideal (5 min vs. 5-15 min), this assumption allows for a tractable approximation of the initial conditions for the run-off inference. Comparing the intensity distribution at t=0 to a zero-inflated Poisson distribution is an excellent suggestion for validation, which we will consider for future iterations of the model.

(4) Microscopy Quantifications:

(4.1) Figure 1-S2A shows variable scFv-GFP expression across cells. Were cells selected for uniform expression in the analysis? Or is the SunTag assumed saturated? which would then need to be demonstrated. 

All cell lines used are monoclonal, and cells were selected via FACS for consistent average cytoplasmic GFP signal. We assume the SunTag is saturated based on the established characterization of the system by Tanenbaum et al. (2014), where the high affinity of the scFv-GFP ensures saturation at expression levels similar to ours.

(4.2) As translation proceeds, free scFv-GFP may become limiting due to the accumulation of mature SunTag-containing proteins. This would be difficult to detect (since mature proteins stay in the cytoplasm) and could affect intensity measurements (newly synthesized SunTag proteins getting dimmer over time).

This effect can occur with very long induction times. To mitigate this, we optimized the Doxycycline (Dox) incubation time for our harringtonine experiments to prevent excessive accumulation of mature protein. We also monitor the cytoplasmic background for granularity, which would indicate aggregation or accumulation.

(4.3) The statements "for some traces, the mRNA signal was lost before the run-off completion" (line 195) and "we observed relatively consistent fractions of translated transcripts and trace duration distributions across reporters" (line 340) should be supported by a supplementary figure.

The first statement is supported by Figure 2 - figure supplement 1, which shows representative run-off traces for all constructs, including incomplete ones.

The second statement regarding consistency is supported by the quantitative data in Figure 1E and G, which summarize the fraction of translated transcripts and trace durations across conditions.

(4.4) Measurements of single mature protein intensity $i_{MP}$:

(4.4.1) Since puromycin is used to disassemble elongating ribosomes, calibration may be biased by incomplete translation products (likely a substantial fraction, since the Dox induction is only 20min and RNAs need several minutes to be transcribed, exported, and then fully translated).

As mentioned in the “Live-cell imaging” paragraph, the imaging takes place 40 min after the end of Dox incubation. This provides ample time for mRNA export and full translation of the synthesized proteins. Consequently, the fraction of incomplete products generated by the final puromycin addition is negligible compared to the pool of fully synthesized mature proteins accumulated during the preceding hour.

(4.4.2) Line 519: "The intensity of each spot is averaged over the 100 frames". Do I understand correctly that you are looking at immobile proteins? What immobilizes these proteins? Are these small aggregates? It would be surprising that these aggregates have really only 1, 2, or 3 proteins, as suggested by Figure 1-S2A.

We are visualizing mature proteins that are specifically tethered to the actin cytoskeleton. This is achieved using a reporter where the RH1 domain is fused directly to the C-terminus of the Renilla protein (SunTag-Renilla-RH1). The RH1 domain recruits the endogenous Myosin Va motor, which anchors the protein to actin filaments, rendering it immobile. Since each Myosin Va motor interacts with one RH1 domain (and thus one mature protein), the resulting spots represent individual immobilized proteins rather than aggregates. We have now revised the text and Methods section to make this calibration strategy and the construct design clearer (L130-140).

(4.4.3) Estimating the average intensity $i_{MP}$ of single proteins all resides in the seeing discrete modes in the histogram of Figure 1-S2B, which is not very convincing. A complementary experiment, measuring *on the same microscope* the intensity of an object with a known number of GFP molecules (e.g., MS2-GFP labeled RNAs, or individual GEMs https://doi.org/10.1016/j.cell.2018.05.042 (only requiring a single transfection)) would be reassuring to convince the reader that we are not off by an order of magnitude.

While a complementary calibration experiment would be valuable, we believe our current estimate is robust because it is independently validated by our model. When we inferred i_MP as a free parameter in the HMM (Figure 5 - figure supplement 2B), the resulting value (10-15 a.u.) was remarkably consistent with our experimental calibration (14 ± 2 a.u.). We have clarified this independent validation in the text to strengthen the confidence in our quantification (L264-272).

(4.4.4) Further on the histogram in Figure 1-S2B:

- The gap between the first two modes is unexpectedly sharp. Can you double-check? It means that we have a completely empty bin between two of the most populated bins.

We have double-checked the data; the plot is correct, though the sharp gap is likely due to the small sample size (n=29).

- I am surprised not to see 3 modes or more, given that panel A shows three levels of intensity (the three colors of the arrows).

As noted below, brighter foci exist but fall outside the displayed range of the histogram.

- It is unclear what the statistical test is and what it is supposed to demonstrate.

The Student's t-test compares the means of the two identified populations to confirm they are statistically distinct intensity groups.

- I count n = 29, not 31. (The sample is small enough that the bars of the histogram show clear discrete heights, proportional to 1, 2, 3, 4, and 5 --adding up all the counts, I get 29). Is there a mistake somewhere? Or are some points falling outside of the displayed x-range?

You are correct. Two brighter data points fell outside the displayed range. The total number of foci in the histogram is 29. We have corrected the figure caption and the text accordingly.

(5) Miscellaneous Points: 

(5.1) Panel B in Figure 2-s1 appears to be missing.

The figure contains only one panel.

(5.2) In Equation (7), $l$ is not defined (presumably ribosome footprint length?). Instead, $J$ is defined right after eq (7), as if it were used in this equation.

Thank you for pointing this out, we have corrected it.

(5.3) Line 703, did you mean to write something else than "Equation 26" (since equation 26 is defined after)?

Yes, this was a typo. We have corrected the cross-reference.

eLife Assessment

This manuscript reports important findings indicating that cell cycle progression and cytokinesis both contribute to the transition from early to late neural stem cell fates. Loss-of-function experimental evidence convincingly shows that disrupting the cell cycle or cytokinesis can alter cell fate. This work sets the stage for future investigations into the underlying mechanisms linking the cell cycle to the expression of temporal factors controlling cell fate.

Reviewer #1 (Public review):

Summary:

Drosophila larval type II neuroblasts generate diverse types of neurons by sequentially expressing different temporal identity genes during development. Previous studies have shown that transition from early temporal identity genes (such as Chinmo and Imp) to late temporal identity genes (such as Syp and Broad) depends on the activation of the expression of EcR by Seven-up (Svp) and progression through the G1/S transition of the cell cycle. In this study, Chaya and Syed examined if the expression of Syp and EcR is regulated by cell cycle and cytokinesis by knocking down CDK1 or Pav, respectively, throughout development or at specific developmental stages. They find that knocking down CDK1 or Pav either in all type II neuroblasts throughout the development or in single type neuroblast clones after larval hatching consistently leads to failure to activate late temporal identity genes Syp and EcR. To determine whether the failure of the activation of Syp and EcR is due to impaired Svp expression, they also examined Svp expression using a Svp-lacZ reporter line. They find that Svp is expressed normally in CDK1 RNAi neuroblasts. Further, knocking down CDK1 or Pav after Svp activation still leads to loss of Syp and EcR expression. Finally, they also extended their analysis to type I neuroblasts. They find that knocking down CDK1 or Pav, either at 0 hours or at 42 hours after larval hatching, also results in loss of Syp and EcR expression in type I neuroblasts. Based on these findings, the authors conclude that cycle and cytokinesis are required for the transition from early to late late temporal identity genes in both types of neuroblasts. These findings add mechanistic details to our understanding of the temporal patterning of Drosophila larval neuroblasts.

Strengths:

The data presented in the paper are solid and largely support their conclusion. Images are of high quality. The manuscript is well-written and clear.

Weaknesses:

The authors have addressed all the weaknesses in this revision.

Reviewer #2 (Public review):

Summary:

Neural stem cells produce a wide variety of neurons during development. The regulatory mechanisms of neural diversity are based on the spatial and temporal patterning of neural stem cells. Although the molecular basis of spatial patterning is well-understood, the temporal patterning mechanism remains unclear. In this manuscript, the authors focused on the roles of cell cycle progression and cytokinesis in temporal patterning and found that both are involved in this process.

Strengths:

They conducted RNAi-mediated disruption on cell cycle progression and cytokinesis. As they expected, both disruptions affected temporal patterning in NSCs.

Weaknesses:

Although the authors showed clear results, they needed to provide additional data to support their conclusion sufficiently.

For example, they can examine the effects of cell cycle acceleration on the temporal patterning.

Reviewer #3 (Public review):

Summary:

The manuscript by Chaya and Syed focuses on understanding the link between cell cycle and temporal patterning in central brain type II neural stem cells (NSCs). To investigate this, the authors perturb the progression of the cell cycle by delaying the entry into M phase and preventing cytokinesis. Their results convincingly show that temporal factor expression requires progression of the cell cycle in both Type 1 and Type 2 NSCs in the Drosophila central brain. Overall, this study establishes an important link between the two timing mechanisms of neurogenesis.

Strengths:

The authors provide solid experimental evidence for the coupling of cell cycle and temporal factor progression in Type 2 NSCs. The quantified phenotype shows an all-or-none effect of cell cycle block on the emergence of subsequent temporal factors in the NSCs, strongly suggesting that both nuclear division and cytokinesis are required for temporal progression. The authors also extend this phenotype to Type 1 NSCs in the central brain, providing a generalizable characterization of the relationship between cell cycle and temporal patterning.

Weaknesses:

One major weakness of the study is that the authors do not explore the mechanistic relationship between cell cycle and temporal factor expression. Although their results are quite convincing, they do not provide an explanation as to why Cdk1 depletion affects Syp and EcR expression but not the onset of svp. This result suggests that at least a part of the temporal cascade in NSCs is cell-cycle independent which isn't addressed or sufficiently discussed.

Author response:

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

Public Reviews:

Reviewer #1 (Public review):

Summary:

Drosophila larval type II neuroblasts generate diverse types of neurons by sequentially expressing different temporal identity genes during development. Previous studies have shown that the transition from early temporal identity genes (such as Chinmo and Imp) to late temporal identity genes (such as Syp and Broad) depends on the activation of the expression of EcR by Seven-up (Svp) and progression through the G1/S transition of the cell cycle. In this study, Chaya and Syed examined whether the expression of Syp and EcR is regulated by cell cycle and cytokinesis by knocking down CDK1 or Pav, respectively, throughout development or at specific developmental stages. They find that knocking down CDK1 or Pav either in all type II neuroblasts throughout development or in single-type neuroblast clones after larval hatching consistently leads to failure to activate late temporal identity genes Syp and EcR. To determine whether the failure of the activation of Syp and EcR is due to impaired Svp expression, they also examined Svp expression using a Svp-lacZ reporter line. They find that Svp is expressed normally in CDK1 RNAi neuroblasts. Further, knocking down CDK1 or Pav after Svp activation still leads to loss of Syp and EcR expression. Finally, they also extended their analysis to type I neuroblasts. They find that knocking down CDK1 or Pav, either at 0 hours or at 42 hours after larval hatching, also results in loss of Syp and EcR expression in type I neuroblasts. Based on these findings, the authors conclude that cycle and cytokinesis are required for the transition from early to late temporal identity genes in both types of neuroblasts. These findings add mechanistic details to our understanding of the temporal patterning of Drosophila larval neuroblasts.

Strengths:

The data presented in the paper are solid and largely support their conclusion. Images are of high quality. The manuscript is well-written and clear.

We appreciate the reviewer’s detailed summary and recognition of the study’s strengths.

Weaknesses:

The quantifications of the expression of temporal identity genes and the interpretation of some of the data could be more rigorous.

(1) Expression of temporal identity genes may not be just positive or negative. Therefore, it would be more rigorous to quantify the expression of Imp, Syp, and EcR based on the staining intensity rather than simply counting the number of neuroblasts that are positive for these genes, which can be very subjective. Or the authors should define clearly what qualifies as "positive" (e.g., a staining intensity at least 2x background).

We thank the reviewer for this helpful suggestion. In the new version, we have now clarified how positive expression was defined and added more details of our quantification strategy to the Methods section (page 11, lines 380-388; lines 426-434 in tracked changes file). Fluorescence intensity for each neuroblast was normalized to the mean intensity of neighboring wild-type neuroblasts imaged in the same field. A neuroblast was considered positive for a given factor when its normalized nuclear intensity was at least 2× the local background. This scoring criterion was applied uniformly across all genotypes and time points. All quantifications were performed on the raw LSM files in Fiji prior to assembling the figure panels.

(2) The finding that inhibiting cytokinesis without affecting nuclear divisions by knocking down Pav leads to the loss of expression of Syp and EcR does not support their conclusion that nuclear division is also essential for the early-late gene expression switch in type II NSCs (at the bottom of the left column on page 5). No experiments were done to specifically block the nuclear division in this study specifically. This conclusion should be revised.

We blocked both cell cycle progression and cytokinesis, and both these manipulations affected temporal gene transitions, suggesting that both cell cycle and cytokinesis are essential. To our knowledge, no mechanism/tool exists that selectively blocks nuclear division while leaving cell cycle progression intact. We have added more clarification on page 4, line 123 onwards (lines 126 onwards in tracked changes file).

(3) Knocking down CDK1 in single random neuroblast clones does not make the CDK1 knockdown neuroblast develop in the same environment (except still in the same brain) as wild-type neuroblast lineages. It does not help address the concern whether "type 2 NSCS with cell cycle arrest failed to undergo normal temporal progression is indirectly due to a lack of feedback signaling from their progeny", as discussed (from the bottom of the right column on page 9 to the top of the left column on page 10). The CDK1 knockdown neuroblasts do not divide to produce progeny and thus do not receive a feedback signal from their progeny as wild-type neuroblasts do. Therefore, it cannot be ruled out that the loss of Syp and EcR expression in CDK1 knockdown neuroblasts is due to the lack of the feedback signal from their progeny. This part of the discussion needs to be clarification.

Thanks to the reviewer for raising this critical point. We agree and have added more clarification of our interpretations and limitations to our studies in the revised text on page 8, line 278-282 (lines 296-300 in tracked changes file)

(4) In Figure 2I, there is a clear EcR staining signal in the clone, which contradicts the quantification data in Figure 2J that EcR is absent in Pav RNAi neuroblasts. The authors should verify that the image and quantification data are consistent and correct.

When cytokinesis is blocked using pav-RNAi, the neuroblasts become extremely large and multinucleated. In some large pav RNAi clones, we observed a weak EcR signal near the cell membrane. However, more importantly, none of the nuclear compartments showed detectable EcR staining, where EcR is typically localized. We selected a representative nuclear image for the figure panel. To clarify this observation, we have now added an explanatory note to the discussion section on page 8, lines 283-291 (lines 301-309 in tracked changes file).

Reviewer #2 (Public review):

Summary:

Neural stem cells produce a wide variety of neurons during development. The regulatory mechanisms of neural diversity are based on the spatial and temporal patterning of neural stem cells. Although the molecular basis of spatial patterning is well-understood, the temporal patterning mechanism remains unclear. In this manuscript, the authors focused on the roles of cell cycle progression and cytokinesis in temporal patterning and found that both are involved in this process.

Strengths:

They conducted RNAi-mediated disruption on cell cycle progression and cytokinesis. As they expected, both disruptions affected temporal patterning in NSCs.

We appreciate the reviewer’s positive assessment of our experimental results.

Weaknesses:

Although the authors showed clear results, they needed to provide additional data to support their conclusion sufficiently.

For example, they need to identify type II NSCs using molecular markers (Ase/Dpn).The authors are encouraged to provide a more detailed explanation of each experiment. The current version of the manuscript is difficult for non-expert readers to understand.

Thanks for your feedback. We have now included a detailed description of how we identify type II NSCs in both wild-type and mutant clones. We have also added a representative Asense staining to clearly distinguish type 1 (Ase⁺) from type 2 (Ase^-) NSCs see Figure S1. We have also added a resources table explaining the genotypes associated with each figure, which was omitted due to an error in the previous version of the manuscript. 

Reviewer #3 (Public review):

Summary:

The manuscript by Chaya and Syed focuses on understanding the link between cell cycle and temporal patterning in central brain type II neural stem cells (NSCs). To investigate this, the authors perturb the progression of the cell cycle by delaying the entry into M phase and preventing cytokinesis. Their results convincingly show that temporal factor expression requires progression of the cell cycle in both Type 1 and Type 2 NSCs in the Drosophila central brain. Overall, this study establishes an important link between the two timing mechanisms of neurogenesis.

Strengths:

The authors provide solid experimental evidence for the coupling of cell cycle and temporal factor progression in Type 2 NSCs. The quantified phenotype shows an all-ornone effect of cell cycle block on the emergence of subsequent temporal factors in the NSCs, strongly suggesting that both nuclear division and cytokinesis are required for temporal progression. The authors also extend this phenotype to Type 1 NSCs in the central brain, providing a generalizable characterization of the relationship between cell cycle and temporal patterning.

We thank the reviewer for recognizing the robustness of our data linking the cell cycle to temporal progression.

Weaknesses:

One major weakness of the study is that the authors do not explore the mechanistic relationship between the cell cycle and temporal factor expression. Although their results are quite convincing, they do not provide an explanation as to why Cdk1 depletion affects Syp and EcR expression but not the onset of svp. This result suggests that at least a part of the temporal cascade in NSCs is cell-cycle independent, which isn't addressed or sufficiently discussed.

Thank you for bringing up this important point. We are equally interested in uncovering the mechanism by which the cell cycle regulates temporal gene transitions; however, such mechanistic exploration is beyond the scope of the present study. Interestingly, while the temporal switching factor Svp is expressed independently of the cell cycle, the subsequent temporal transitions are not. We have expanded our discussion on this intriguing finding (page 9, line 307-315; lines 345-355 in tracked changes file). Specifically, we propose that svp activation marks a cell-cycle–independent phase, whereas EcR/Syp induction likely depends on cell-cycle–coupled mechanisms, such as mitosis-dependent chromatin remodeling or daughter-cell feedback. Although further dissection of this mechanism lies beyond the current study, our findings establish a foundation for future work aimed at identifying how developmental timekeeping is molecularly coupled to cell-cycle progression.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors): 

(1) Figure 1 C and D, it would be better to put a question mark to indicate that these are hypotheses to be tested. 

We appreciate this suggestion and have added question marks in Figure 1C and 1D to clearly indicate that these panels represent hypotheses under investigation clearly.

(2) Figure 2A-I, Figure 4A-I, Figure 5A-I and K-S, in addition to enlarged views of single type II neuroblasts, it would be more convincing to include zoomed-out images of the entire larval brain or at least a portion of the brain to include neighboring wild-type type II neuroblasts as internal controls. Also, it would be ideal to show EcR staining from the same neuroblasts as IMP and Syp staining. 

We thank the reviewer for this valuable input. In our imaging setup, the number of available antibody channels was limited to four (anti-Ase, anti-GFP, anti-Syp, and antiImp). Adding EcR in the same sample was therefore not technically possible, we performed EcR staining separately. 

(3) The authors cited "Syed et al., 2024" (in the middle of the right column on page 5), but this reference is missing in the "References" section and should be added. 

The missing citation has been added to the reference section.  

(4) It would be better to include Ase staining in the relevant figure to indicate neuroblast identity as type I or type II. 

We agree and now include representative Ase staining for both type 1 and type 2 NSC clones in Figure S1, along with corresponding text updates that describe these markers.

Reviewer #2 (Recommendations for the authors): 

Major comments 

(1) The present conclusion relies on the results using Cdk1 RNAi and pav RNAi. It is still possible that Cdk1 and Pav are involved in the regulation of temporal patterning independent of the regulation of cell cycle or cytokinesis, respectively. To avoid this possibility, the authors need to inhibit cell cycle progression or cytokinesis in another alternative manner. 

We thank the reviewer for raising this important point. While we cannot completely exclude gene-specific, cell-cycle-independent roles for Cdk1 or Pav, we observe consistent phenotypes across several independent manipulations that slow or block the cell cycle. Also, earlier studies using orthogonal approaches that delay G1/S (Dacapo/Rbf) or impair mitochondrial OxPhos (which lengthens G1/S; van den Ameele & Brand, 2019) produce similar temporal delays. These concordant phenotypes strongly support the interpretation that altered cell-cycle progression—rather than specific roles of a single gene—is the primary cause of the defect. While we cannot exclude additional, gene-specific effects of Cdk1 or Pav, the concordant phenotypes across independent perturbations make the cell-cycle disruption model the most parsimonious interpretation. We have clarified this reasoning in the discussion section on pages 8-9, lines 293-305 (lines 311-343 in tracked changes file).

(2) To reach the present conclusion, the authors need to address the effects of acceleration of cell cycle progression or cytokinesis on temporal patterning. 

We thank the reviewer for this insightful suggestion. To our knowledge, there are currently no established genetic tools that can specifically accelerate cell-cycle progression in Drosophila neuroblasts. However, our results demonstrate that blocking the cell cycle impairs the transition from early to late temporal gene expression. These findings suggest that proper cell-cycle progression is essential for the transition from early to late temporal identity in neuroblasts.

Minor comments 

(3) P3L2 (right), ... we blocked the NSC cell cycle...

How did they do it? 

Which fly lines were used?

Why did they use the line? 

These details are now included in the Materials and Methods and the Resource Table (pages 11-13). We used Wor-Gal4, Ase-Gal80 to drive UAS-Cdk1RNAi and UASpavRNAi in type 2 NSCs 

(4) P5L1(left), ... we used the flip-out approach...

Why did they conduct it? 

Probably, the authors have reasons other than "to further ensure." 

We have clarified in the text on page 4, lines 137-139, that the flip-out approach was used to generate random single-cell clones, enabling quantitative analysis of type 2 NSCs within an otherwise wild-type brain. 

(5) P5L8(left), ... type 2 hits were confirmed by lack of the type 1 Asense...  The authors must examine Deadpan (Dpn) expression as well. Because there are a lot of Asense (Ase) negative cells in the brain (neurons, glial cell, and neuroepithelial cells). 

Type II NSCs can be identified as Dpn+/Ase- cells.

We agree that Dpn is a helpful marker. However, we reliably distinguished type II NSCs by their lack of Ase and larger cell size relative to surrounding neurons and glia, which are smaller in size and located deeper within the clone. These differences, together with established lineage patterns, allow unambiguous identification of type 2 NSCs across all genotypes. We have now added representative type I and type 2 NSC clones to the supplemental figure S1 (E-G’) with Asense stains to demonstrate how we differentiate type I from type II NSCs. 

(6) P5L32(left), To do this, we induced... 

This sentence should be made more concise.

Please rephrase it. 

The sentence has been rewritten for clarity and concision.

(7)  P5L42(left), ...lack of EcR/Syp expression (Figure 2).  However, EcR expression is still present (Figure 2I). 

In some large pavRNAi clones, a weak EcR signal can be observed near the cell membrane; however, none of the nuclear compartments—where EcR is typically localized—show detectable staining. We selected a representative nuclear image for the figure and addressed this observation on page 8, lines 283-291 (lines 301-309 in tracked changes file).

(8) P7L29(left), ......had persistent Imp expression...

Imp expression is faint compared to that in Figure 2G.

The differences between Figures 2G and 3G should be discussed. 

We thank the reviewer for this comment. We have added a note in the Methods section clarifying that brightness and contrast were adjusted per panel for optimal visualization; thus, apparent differences in signal intensity do not reflect biological variation. Fluorescence intensity for each neuroblast was normalized to the mean intensity of neighboring wild-type neuroblasts imaged in the same field. A neuroblast was considered Imp-positive when its normalized nuclear intensity was at least 2× the local background. This scoring criterion was applied uniformly across all genotypes and time points. All quantifications were performed on the raw LSM files in Fiji prior to assembling the figure panels.

(9) P8 (Figure 5)

The Imp expression is faint compared to that in Figure 5Q.

The difference between Figure 5G and 5Q should be discussed further. 

As mentioned above, we have clarified our image processing approach in the Methods section to explain any differences in signal appearance between these figures.

(10) P10 Materials and Methods

The authors did not mention the fly lines used. This is very important for the readers. 

We thank the reviewer for bringing this oversight to our attention. The Resource Table was inadvertently omitted from the initial submission. The complete list of fly lines and reagents used in this study is now provided in the updated Resource Table.

Reviewer #3 (Recommendations for the authors): 

Major points 

(1) The authors mention that the heat-shock induction at 42ALH is well after svp temporal window and therefore the cell cycle block independently affects Syp and EcR expression. However, Figure 3 shows svp-LacZ expression at 48ALH. If svp expression is indeed transient in Type 2 NSCs, then this must be validated using an immunostaining of the svp-LacZ line with svp antibody. This is crucial as the authors claim that cell cycle block doesn't affect does affect svp expression and is required independently. 

We thank the reviewer for bringing this important issue to our attention. As noted, Svp protein is expressed transiently and stochastically in type 2 NSCs (Syed et al., 2017), making direct antibody quantification challenging upon cell cycle block. Consistent with previous work (Syed et al., 2017), we used the svp-LacZ reporter line to visualize stabilized Svp expression, which reliably captures Svp expression in type 2 NSCs (Syed et al., 2017 https://doi.org/10.7554/eLife.26287, and Dhilon et al., 2024 https://doi.org/10.1242/dev.202504).

(2) The authors have successfully slowed down the cell cycle and showed that it affects temporal progression. However, a converse experiment where the cell cycle is sped up in NSCs would be an important test for the direct coupling of temporal factor expression and cell cycle, wherein the expectation would be the precocious expression of late temporal factors in faster cycle NSCs. 

We agree that such an experiment would be ideal. However, as noted above (Reviewer #2 comment 2), to our knowledge, no suitable tools currently exist to accelerate neuroblast cell-cycle progression without pleiotropic effects.

Minor point 

The authors must include Ray and Li (https://doi.org/10.7554/eLife.75879) in the references when describing that "...cell cycle has been shown to influence temporal patterning in some systems,...".  

We thank the reviewer for this helpful suggestion. The cited reference (Ray and Li, eLife, 2022) has now been included and appropriately referenced in the revised manuscript.

eLife Assessment

This valuable study investigates the computational role of top-down feedback - a property found in biological circuits - in artificial neural network (ANN) models of the neocortex. Using hierarchical recurrent ANNs in an audiovisual integration task, the authors show that an anatomically inspired feedback motif induces a stable visual bias consistent with human perception and yields modest but meaningful benefits for learning dynamics and robustness. The strength of evidence is solid: the modeling, analyses, and controls mostly support the central claim that top-down feedback motifs impose persistent inductive biases that shape functional specialization and behavior. But the evidence for a broad, general framework that predicts behavior remains only partially supported, and the Methods would benefit from a compact, reproducible summary of hyperparameters and architectural details.

Reviewer #1 (Public review):

Summary:

Here, the authors aim to investigate the potential improvements of ANNs when used to explain brain data using top-down feedback connections found in the neocortex. To do so, they use a retinotopic and tonotopic organization to model each subregion of the ventral visual (V1, V2, V4, and IT) and ventral auditory (A1, Belt, A4) regions using Convolutional Gated Recurrent Units. The top-down feedback connections are inspired by the apical tree of pyramidal neurons, modeled either with a multiplicative effect (change of gain of the activation function) or a composite effect (change of gain and threshold of the activation function).

To assess the functional impact of the top-down connections, the authors compare three architectures: a brain-like architecture derived directly from brain data analysis, a reversed architecture where all feedforward connections become feedback connections and vice versa, and a random connectivity architecture. More specifically, in the brain-like model the visual regions provide feedforward input to all auditory areas, whereas auditory areas provide feedback to visual regions.

First, the authors found that top-down feedback influences audiovisual processing and that the brain-like model exhibits a visual bias in multimodal visual and auditory tasks. Second, they discovered that in the brain-like model, the composite integration of top-down feedback, similar to that found in the neocortex, leads to an inductive bias toward visual stimuli, which is not observed in the feedforward-only model. Furthermore, the authors found that the brain-like model learns to utilize relevant stimuli more quickly while ignoring distractors. Finally, by analyzing the activations of all hidden layers (brain regions), they found that the feedforward and feedback connectivity of a region could determine its functional specializations during the given tasks.

Strengths:

The study introduces a novel methodology for designing connectivity between regions in deep learning models. The authors also employ several tasks based on audiovisual stimuli to support their conclusions. Additionally, the model utilizes backpropagation of error as a learning algorithm, making it applicable across a range of tasks, from various supervised learning scenarios to reinforcement learning agents. Conversely, the presented framework offers a valuable tool for studying top-down feedback connections in cortical models. Thus, it is a very nice study that can also give inspiration to other fields (machine learning) to start exploring new architectures.

Reviewer #2 (Public review):

Summary:

This work addresses the question whether artificial deep neural network models of the brain could be improved by incorporating top-down feedback, inspired by the architecture of neocortex.

In line with known biological features of cortical top-down feedback, the authors model such feedback connections with both, a typical driving effect and a purely modulatory effect on the activation of units in the network.

To asses the functional impact of these top-down connections, they compare different architectures of feedforward and feedback connections in a model that mimics the ventral visual and auditory pathways in cortex on an audiovisual integration task.

Notably, one architecture is inspired by human anatomical data, where higher visual and auditory layers possess modulatory top-down connections to all lower-level layers of the same modality, and visual areas provide feedforward input to auditory layers, whereas auditory areas provide modulatory feedback to visual areas.

First, the authors find that this brain-like architecture imparts the models with a light visual bias similar to what is seen in human data, which is the opposite in a reversed architecture, where auditory areas provide feedforward drive to the visual areas.

Second, they find that, in their model, modulatory feedback should be complemented by a driving component to enable effective audiovisual integration, similar to what is observed in neural data.

Overall, the study shows some possible functional implications when adding feedback connections in a deep artificial neural network that mimic some functional aspects of visual perception in humans.

Strengths:

The study contains innovative ideas, such as incorporating an anatomically inspired architecture into a deep ANN, and comparing its impact on a relevant task to alternative architectures.

Moreover, the simplicity of the model allows it to draw conclusions on how features of the architecture and functional aspects of the top-down feedback affects performance of the network.

This could be a helpful resource for future studies of the impact of top-down connections in deep artificial neural network models of neocortex.

Weaknesses:

Some claims not yet supported.

The problem is that results are phrased quite generally in the abstract and discussion, while the actual results shown in the paper are very specific to certain implementations of top-down feedback and architectures. This could lead to misunderstanding and requires some revisions of the claims in the abstract and discussion (see below).

"Altogether our findings demonstrate that modulatory top-down feedback is a computationally relevant feature of biological brain..."

This claim is not supported, since no performance increase is demonstrated for modulatory feedback. So far, only the second half of the sentence is supported: "...and that incorporating it into ANNs affects their behavior and constrains the solutions it's likely to discover."

"This bias does not impair performance on the audiovisual tasks."

This is only true for the composite top-down feedback that combines driving and modulatory effects, whereas modulatory feedback alone can impair the performance (e.g., in the visual tasks VS1 and VS2). The fact that modulatory feedback alone is insufficient in ANNs to enable effective cross-modal integration and requires some driving component is actually very interesting, but it is not stressed enough in the abstract. This is hinted at in the following sentence, but should be made more explicitly:

"The results further suggest that different configurations of top-down feedback make otherwise identically connected models functionally distinct from each other, and from traditional feedforward and laterally recurrent models."

"Here we develop a deep neural network model that captures the core functional properties of top-down feedback in the neocortex" -> this is too strong, take out "the", because very likely there are other important properties that are not yet incorporated.

"Altogether, our results demonstrate that the distinction between feedforward and feedback inputs has clear computational implications, and that ANN models of the brain should therefore consider top-down feedback as an important biological feature."

This claim is still not substantiated by evidence provided in the paper. First, the wording is a bit imprecise, because mechanistically, it is not really the feedforward versus feedback (a purely feedforward model is not considered at all in the paper), but modulatory versus driving. Moreover, the second part of the sentence is problematic: The results imply that, computationally/functionally, driving connections are doing the job, while modulatory feedback does not really seem to improve performance (best case, it does not do any harm). It is true that it is a feature that is inspired by biology, but I don't see why the results imply that (modulatory) top-down feedback should be considered in ANN models of the brain. This would require to show that such models either improve performance, or do improve the ability to fit neural data, both which are beyond the scope of the paper.

The same argument holds for the following sentence, which is not supported by the results of the paper:

"More broadly, our work supports the conclusion that both the cellular neurophysiology and structure of feed-back inputs have critical functional implications that need to be considered by computational models of brain function."

Additional supplementary material required

Although the second version checked the influence of processing time, this was not done for the most important figure of the paper, Figure 4. A central claim in the abstract "This bias does not impair performance on the audiovisual tasks" relies on this figure, because only with composite feedback the performance is comparable between the the "drive-only" and "brain-like" models. Thus, the supplementary Figure 3 should also include the composite networks and drive only network to check the robustness of the claim with respect to process time. This robustness analysis should then also be mentioned in the text. For example, it should be mentioned whether results in these networks are robust or not with respect to process time, whether there are differences between network architectures or types of feedback in general etc.

Moreover, the current analysis for networks with modulatory feedback is a bit confusing. Why is the performance so low for the reverse model for a process time of 3 and 10? This is a very strong effect that warrants explanation. More details should be added in the caption as well. For example, are the models separately trained for the output after 3 and 10 processing steps for the comparison, or just evaluated at these times? Not training these networks separately might explain the low performance for some networks, so ideally networks are trained for each choice of processing steps.

Reviewer #3 (Public review):

Summary:

This study investigates the computational role of top-down feedback in artificial neural networks (ANNs), a feature that is prevalent in the brain but largely absent in standard ANN architectures. The authors construct hierarchical recurrent ANN models that incorporate key properties of top-down feedback in the neocortex. Using these models in an audiovisual integration task, they find that hierarchical structures introduce a mild visual bias, akin to that observed in human perception, not always compromising task performance.

Strengths:

The study investigates a relevant and current topic of considering top-down feedback in deep neural networks. In designing their brain-like model, they use neurophysiological data, such as externopyramidisation and hierarchical connectivity. Their brain-like model exhibits a visual bias that qualitatively matches human perception.

Weaknesses:

While the model is brain-inspired, it has limited bioplausibility. The model assumes a simplified and fixed hierarchy. The authors acknowledge this limitation in the discussion.

While the brain-like model showed an advantage in ignoring distracting auditory inputs, it struggled when visual information had to be ignored. This suggests that its rigid bias toward visual processing could make it less adaptive in tasks requiring flexible multimodal integration. It hence does not necessarily constitute an improvement over existing ANNs. The study does not evaluate whether the top-down feedback architecture scales well to more complex problems or larger datasets. A valuable future contribution would be to evaluate how the network's behaviour fits to human data.

Author response:

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

Public Reviews:

Reviewer #1 (Public review):

Summary:

Here, the authors aim to investigate the potential improvements of ANNs when used to explain brain data using top-down feedback connections found in the neocortex. To do so, they use a retinotopic and tonotopic organization to model each subregion of the ventral visual (V1, V2, V4, and IT) and ventral auditory (A1, Belt, A4) regions using Convolutional Gated Recurrent Units. The top-down feedback connections are inspired by the apical tree of pyramidal neurons, modeled either with a multiplicative effect (change of gain of the activation function) or a composite effect (change of gain and threshold of the activation function).

To assess the functional impact of the top-down connections, the authors compare three architectures: a brain-like architecture derived directly from brain data analysis, a reversed architecture where all feedforward connections become feedback connections and vice versa, and a random connectivity architecture. More specifically, in the brain-like model the visual regions provide feedforward input to all auditory areas, whereas auditory areas provide feedback to visual regions.

First, the authors found that top-down feedback influences audiovisual processing and that the brain-like model exhibits a visual bias in multimodal visual and auditory tasks. Second, they discovered that in the brain-like model, the composite integration of top-down feedback, similar to that found in the neocortex, leads to an inductive bias toward visual stimuli, which is not observed in the feedforward-only model. Furthermore, the authors found that the brain-like model learns to utilize relevant stimuli more quickly while ignoring distractors. Finally, by analyzing the activations of all hidden layers (brain regions), they found that the feedforward and feedback connectivity of a region could determine its functional specializations during the given tasks.

Strengths:

The study introduces a novel methodology for designing connectivity between regions in deep learning models. The authors also employ several tasks based on audiovisual stimuli to support their conclusions. Additionally, the model utilizes backpropagation of error as a learning algorithm, making it applicable across a range of tasks, from various supervised learning scenarios to reinforcement learning agents. Conversely, the presented framework offers a valuable tool for studying top-down feedback connections in cortical models. Thus, it is a very nice study that also can give inspiration to other fields (machine learning) to start exploring new architectures.

We thank the reviewer for their accurate summary of our work and their kind assessment of its strengths.

Weaknesses:

Although the study explores some novel ideas on how to study the feedback connections of the neocortex, the data presented here are not complete in order to propose a concrete theory of the role of top-down feedback inputs in such models of the brain.

(1) The gap in the literature that the paper tries to fill in the ability of DL algorithms to predict behavior: "However, there are still significant gaps in most deep neural networks' ability to predict behavior, particularly when presented with ambiguous, challenging stimuli." and "[...] to accurately model the brain."

It is unclear to me how the presented work addresses this gap, as the only facts provided are derived from a simple categorization task that could also be solved by the feedforward-only model (see Figures 4 and 5). In my opinion, this statement is somewhat far-fetched, and there is insufficient data throughout the manuscript to support this claim.

We can see now that the way the introduction was initially written led to some confusion about our goal in this study. Our goal here was not to demonstrate that top-down feedback can enable superior matches to human behaviour. Rather, our goal was to determine if top-down feedback had any real implications for processing ambiguous stimuli. The sentence that the reviewer has highlighted was intended as an explanation for why top-down feedback, and its impact on ambiguous stimuli, might be something one would want to examine for deep neural networks. But, here, we simply wanted to (1) provide an overview of the code base we have created, (2) demonstrate that top-down feedback does impact the processing of ambiguous stimuli.

We agree with the reviewer that if our goal was to improve our ability to predict behaviour, then there was a big gap in the evidence we provided here. But, this was not our goal, and we believe that the data we provide here does convincingly show that top-down feedback has an impact on processing of ambiguous stimuli. We have updated the text in the introduction to make our goals more clear for the reader and avoid this misunderstanding of what we were trying to accomplish here. Specifically, the end of the introduction is changed to:

“To study the effect of top-down feedback on such tasks, we built a freely available code base for creating deep neural networks with an algorithmic approximation of top-down feedback. Specifically, top-down feedback was designed to modulate ongoing activity in recurrent, convolutional neural networks. We explored different architectural configurations of connectivity, including a configuration based on the human brain, where all visual areas send feedforward inputs to, and receive top-down feedback from, the auditory areas. The human brain-based model performed well on all audiovisual tasks, but displayed a unique and persistent visual bias compared to models with only driving connectivity and models with different hierarchies. This qualitatively matches the reported visual bias of humans engaged in audio-visual tasks. Our results confirm that distinct configurations of feedforward/feedback connectivity have an important functional impact on a model's behavior. Therefore, top-down feedback captures behaviors and perceptual preferences that do not manifest reliably in feedforward-only networks. Further experiments are needed to clarify whether top-down feedback helps an ANN fit better to neural data, but the results show that top-down feedback affects the processing of stimuli and is thus a relevant feature that should be considered for deep ANN models in computational neuroscience more broadly.”

(2) It is not clear what the advantages are between the brain-like model and a feedforward-only model in terms of performance in solving the task. Given Figures 4 and 5, it is evident that the feedforward-only model reaches almost the same performance as the brain-like model (when the latter uses the modulatory feedback with the composite function) on almost all tasks tested. The speed of learning is nearly the same: for some tested tasks the brain-like model learns faster, while for others it learns slower. Thus, it is hard to attribute a functional implication to the feedback connections given the presented figures and therefore the strong claims in the Discussion should be rephrased or toned down.

Again, we believe that there has been a misunderstanding regarding the goals of this study, as we are not trying to claim here that there are performance advantages conferred by top-down feedback in this case. Indeed, we share the reviewer’s assessment that the feedforward only model seems to be capable of solving this task well. To reiterate: our goal here was to demonstrate that top-down feedback alters the computations in the network and, thus, has distinct effects on behaviour that need to be considered by researchers who use deep networks to model the brain. But we make no claims of “superiority” of the brain-like model.

In-line with this, we’re not completely sure which claims in the discussion the reviewer is referring to. We note that we were quite careful in our claims. For example, in the first section of the discussion we say:

“Altogether, our results demonstrate that the distinction between feedforward and feedback inputs has clear computational implications, and that ANN models of the brain should therefore consider top-down feedback as an important biological feature.”

And later on:

“In summary, our study shows that modulatory top-down feedback and the architectural diversity enabled by it can have important functional implications for computational models of the brain. We believe that future work examining brain function with deep neural networks should therefore consider incorporating top-down modulatory feedback into model architectures when appropriate.”

If we have missed a claim in the discussion that implies superiority of the brain-like model in terms of task performance we would be happy to change it.

(3) The Methods section lacks sufficient detail. There is no explanation provided for the choice of hyperparameters nor for the structure of the networks (number of trainable parameters, number of nodes per layer, etc). Clarifying the rationale behind these decisions would enhance understanding. Moreover, since the authors draw conclusions based on the performance of the networks on specific tasks, it is unclear whether the comparisons are fair, particularly concerning the number of trainable parameters. Furthermore, it is not clear if the visual bias observed in the brain-like model is an emerging property of the network or has been created because of the asymmetries in the visual vs. auditory pathway (size of the layer, number of layers, etc).

We thank the reviewer for raising this issue, and want to provide some clarifications: First, the number of trainable parameters are roughly equal, since we were only switching the direction of connectivity (top-down versus bottom-up), not the number of connections. We confirmed the biggest difference in size is between models with composite and multiplicative feedback; models with composite feedback have roughly ~1K more parameters, and all models are within the 280K parameter range. We now state this in the methods.

Second, because superior performance was not the goal of this study, as stated above, we conducted limited hyperparameter tuning. Given the reviewer’s comment, we wondered whether this may have impacted our results. Therefore, we explored different hyperparameters for the model during the multimodal auditory tasks, which show the clearest example of the visual dominance in the brainlike model (Figure 3).

We explored different hidden state sizes, learning rates and processing times, and examined whether the core results were different. We found that extremely high learning rates (0.1) destabilize all models and that some models perform poorly under different processing times. But overall, the core results are evident across all hyperparameters where the models learn i.e the different behaviors of models with different connectivities and the visual dominance observed in the brainlike model. We now provide these results in a supplementary figure (Fig. S2, showing larger models trained with different learning rates, and Fig S3, which shows the effect of processing time on AS task performance).

Reviewer #2 (Public review):

Summary:

This work addresses the question of whether artificial deep neural network models of the brain could be improved by incorporating top-down feedback, inspired by the architecture of the neocortex.

In line with known biological features of cortical top-down feedback, the authors model such feedback connections with both, a typical driving effect and a purely modulatory effect on the activation of units in the network.

To assess the functional impact of these top-down connections, they compare different architectures of feedforward and feedback connections in a model that mimics the ventral visual and auditory pathways in the cortex on an audiovisual integration task.

Notably, one architecture is inspired by human anatomical data, where higher visual and auditory layers possess modulatory top-down connections to all lower-level layers of the same modality, and visual areas provide feedforward input to auditory layers, whereas auditory areas provide modulatory feedback to visual areas.

First, the authors find that this brain-like architecture imparts the models with a light visual bias similar to what is seen in human data, which is the opposite in a reversed architecture, where auditory areas provide a feedforward drive to the visual areas.

Second, they find that, in their model, modulatory feedback should be complemented by a driving component to enable effective audiovisual integration, similar to what is observed in neural data.

Last, they find that the brain-like architecture with modulatory feedback learns a bit faster in some audiovisual switching tasks compared to a feedforward-only model.

Overall, the study shows some possible functional implications when adding feedback connections in a deep artificial neural network that mimics some functional aspects of visual perception in humans.

Strengths:

The study contains innovative ideas, such as incorporating an anatomically inspired architecture into a deep ANN, and comparing its impact on a relevant task to alternative architectures.

Moreover, the simplicity of the model allows it to draw conclusions on how features of the architecture and functional aspects of the top-down feedback affect the performance of the network.

This could be a helpful resource for future studies of the impact of top-down connections in deep artificial neural network models of the neocortex.

We thank the reviewer for their summary and their recognition of the innovative components and helpful resources therein.

Weaknesses:

Overall, the study appears to be a bit premature, as several parts need to be worked out more to support the claims of the paper and to increase its impact.

First, the functional implication of modulatory feedback is not really clear. The "only feedforward" model (is a drive-only model meant?) attains the same performance as the composite model (with modulatory feedback) on virtually all tasks tested, it just takes a bit longer to learn for some tasks, but then is also faster at others. It even reproduces the visual bias on the audiovisual switching task. Therefore, the claims "Altogether, our results demonstrate that the distinction between feedforward and feedback inputs has clear computational implications, and that ANN models of the brain should therefore consider top-down feedback as an important biological feature." and "More broadly, our work supports the conclusion that both the cellular neurophysiology and structure of feed-back inputs have critical functional implications that need to be considered by computational models of brain function" are not sufficiently supported by the results of the study. Moreover, the latter points would require showing that this model describes neural data better, e.g., by comparing representations in the model with and without top-down feedback to recorded neural activity.

To emphasize again our specific claims, we believe that our data shows that top-down feedback has functional implications for deep neural network behaviour, not increased performance or neural alignment. Indeed, our results demonstrate that top-down feedback alters the behaviour of the networks, as shown by the differences in responses to various combinations of ambiguous stimuli. We agree with the reviewer that if our goal was to claim either superior performance on these tasks, or better fit to neural data, we would need to actually provide data supporting that claim.

Given the comments from the reviewer, we have tried to provide more clarity in the introduction and discussion regarding our claims. In particular, we now highlight that we are not trying to demonstrate that the models with top-down feedback exhibit superior performance or better fit to neural data.

As one final note, yes, the reviewer understood correctly that the “only feedforward” model is a model with only driving inputs. We have renamed the feedforward-only models to drive only models and added additional emphasis in the text to ensure that the distinction is clear for all readers.

Second, the analyses are not supported by supplementary material, hence it is difficult to evaluate parts of the claims. For example, it would be helpful to investigate the impact of the process time after which the output is taken for evaluation of the model. This is especially important because in recurrent and feedback models the convergence should be checked, and if the network does not converge, then it should be discussed why at which point in time the network is evaluated.

This is an excellent point, and we thank the reviewer for raising it. We allowed the network to process the stimuli for seven time-steps, which was enough for information from any one region to be transmitted to any other. We found in some initial investigations that if we shortened the processing time some seeds would fail to solve the task. But, based on the reviewer’s comment, we have now also run additional tests with longer processing times for the auditory tasks where we see the clearest visual bias (Figure 3). We find that different process times do not change the behavioral biases observed in our models, but may introduce difficulties ignoring visual stimuli for some models. Thus, while process time is an important hyperparameter for optimal performance of the model, the central claim of the paper remains. We include this new data in a supplementary figure S3.

Third, the descriptions of the models in the methods are hard to understand, i.e., parameters are not described and equations are explained by referring to multiple other studies. Since the implications of the results heavily rely on the model, a more detailed description of the model seems necessary.

We agree with the reviewer that the methods could have been more thorough. Therefore, we have greatly expanded the methods section. We hope the model details are now more clear.

Lastly, the discussion and testable predictions are not very well worked out and need more details. For example, the point "This represents another testable prediction flowing from our study, which could be studied in humans by examining the optical flow (Pines et al., 2023) between auditory and visual regions during an audiovisual task" needs to be made more precise to be useful as a prediction. What did the model predict in terms of "optic flow", how can modulatory from simple driving effect be distinguished, etc.

We see that the original wording of this prediction was ambiguous, thank you for pointing this out. In the study highlighted (Pines et al., 2023) the authors use an analysis technique for measuring information flow between brain regions, which is related to analysis of optical flow in images, but applied to fMRI scans. This is confusing given the current study, though. Therefore, we have changed this sentence to make clear that we are speaking of information flow here. 

Reviewer #3 (Public review):

Summary:

This study investigates the computational role of top-down feedback in artificial neural networks (ANNs), a feature that is prevalent in the brain but largely absent in standard ANN architectures. The authors construct hierarchical recurrent ANN models that incorporate key properties of top-down feedback in the neocortex. Using these models in an audiovisual integration task, they find that hierarchical structures introduce a mild visual bias, akin to that observed in human perception, not always compromising task performance.

Strengths:

The study investigates a relevant and current topic of considering top-down feedback in deep neural networks. In designing their brain-like model, they use neurophysiological data, such as externopyramidisation and hierarchical connectivity. Their brain-like model exhibits a visual bias that qualitatively matches human perception.

We thank the reviewer for their summary and evaluation of our paper’s strengths.

Weaknesses:

While the model is brain-inspired, it has limited bioplausibility. The model assumes a simplified and fixed hierarchy. In the brain with additional neuromodulation, the hierarchy could be more flexible and more task-dependent.

We agree, there are still many facets of top-down feedback that we have not captured here, and the modulation of hierarchy is an interesting example. We have added some consideration of this point to the limitations section of the discussion.

While the brain-like model showed an advantage in ignoring distracting auditory inputs, it struggled when visual information had to be ignored. This suggests that its rigid bias toward visual processing could make it less adaptive in tasks requiring flexible multimodal integration. It hence does not necessarily constitute an improvement over existing ANNs. It is unclear, whether this aspect of the model also matches human data. In general, there is no direct comparison to human data. The study does not evaluate whether the top-down feedback architecture scales well to more complex problems or larger datasets. The model is not well enough specified in the methods and some definitions are missing.

We agree with the reviewer that we have not demonstrated anything like superior performance (since the brain-like network is quite rigid, as noted) nor have we shown better match to human data with the brain-like network. This was not our intended claim. Rather, we demonstrated here simply that top-down feedback impacts behavior of the networks in response to ambiguous stimuli. We have now added statements to the introduction and discussion to make our specific claims (which are supported by our data, we believe) clear.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

I believe that the work is very nice but not so mature at this stage. Below, you can find some comments that eventually could improve your manuscript.

(1) Intro, last sentence: "Therefore, top-down feedback is a relevant feature that should be considered for deep ANN models in computational neuroscience more broadly." I don't understand what the authors refer to with this sentence. There are numerous models (deep ANNs) that have been used to model the neural activity and are much simpler than the one proposed here which contains very complex models and connectivity. Although I do agree that the top-down connections are very important there is no data to support their importance for modeling the brain.

Respectfully, we disagree with the reviewer that we don’t provide data to demonstrate the importance of top-down feedback for modelling. Indeed, we provided a great deal of data to show that top-down feedback in the networks has real functional implications for behaviour, e.g., it can induce a human-like visual bias. Thus, top-down feedback is a factor that one should care about when modelling the brain. But, we agree with the reviewer that more demonstration of the utility of using top-down feedback for achieving better fits to neural data would be an important next step. 

(2) I suggest adding some extra supplementary simulations where, for example, the number of data for visual and auditory pathways is equal in size (i.e., the same number of examples), the number of layers is identical (3 per pathway), and also the number of parameters. Doing this would help strengthen the claims presented in the paper.

In fact, all of the hyperparameters the reviewer mentions here were identical for the different networks, so the experiments the reviewer is requesting here were already part of the paper. We now clarify this in the text.

(3) Results: I suggest adding Tables with quantifications of the presented results. For example, best performance, epochs to converge, etc. As it is now, it is very hard to follow the evidence shown in Figures.

This is a good suggestion, we have now added this table to the start of the supplemental figures.

(4) Figure 2e, 3e: Although VS3, and AS3 have been used only for testing, the plot shows alignments with respect to training epochs. The authors should clarify in the Methods if they tested the network with all intermediate weights during VS1/VS2 or AS1/AS2 training.

Testing scenarios in this context meant that the model was never shown the scenario/task during training, but the models were indeed evaluated on the VS3 and AS3 after each training epoch. We have added clarifications to the figure legends.

(5) Methods: It would be beneficial to discuss how specific hyperparameters were selected based on prior research, empirical testing, or theoretical considerations. Also, it is not clear how the alignment (visual or audio) is calculated. Do the authors use the examples that have been classified correctly for both stimuli or do they exclude those from the analysis (maybe I have missed it).

As noted above, because superior performance was not the goal of this study, we conducted limited hyperparameter tuning. But we have extended the results with additional hyperparameter tuning in a supplementary figure, and describe the hyperparameter choices more thoroughly in the methods. As well, all data includes all model responses, regardless of whether they were correct or not. We now clarify this in the methods.

(6) Code: The code repository lacks straightforward examples demonstrating how to utilize the modeling approach. Given that it is referred to as a "framework", one would expect it to facilitate easy integration into various models and tasks. Including detailed instructions or clear examples would significantly improve usability and help users effectively apply the proposed methodology.

We agree with the reviewer, this would be beneficial. We have revised the README of the codebase to explain the model and its usage more clearly and included an interactive jupyter notebook with example training on MNIST.

Some minor comments are given below. Generally speaking, the Figures need to be more carefully checked for consistent labels, colors, etc.

(1) Page 4, 1st paragraph - grammar correction: "a larger infragranular layer" or "larger infragranular layers"

Thank you for catching this, we have fixed the text.

(2) Page 4, 2nd para - rephrase: "In three additional control ANNs" → "In the third additional control ANN"

In fact, we did mean three additional control ANNs, each one representing a different randomized connectivity profile. We now clarify this in the text and provide the connectivity of the two other random graphs in the supplemental figures.

(3) Page 4, VAE acronym needs to be defined before its first use

The variational autoencoder is introduced by its full name in the text now.

(4) Page 4: Fig. 2c reference should be Fig. 2b, Fig. 2d should be Fig. 2c, Fig. 2b should be Fig. 2d, VS4; Fig. 2b, bottom should be VS4; Fig. 2f, Fig. 2f to Fig. 2g. Double check the Figure references in the text. Here is very confusing for the reader.

We have now fixed this, thank you for catching it.

(5) Page 5, 1st para: "Altogether, our results demonstrated both" → "Altogether, our results demonstrated that both"

This has been updated.

(6) Figure 2: In the e and g panels the x label is missing.

This was actually because the x-axis were the same across the panels, but we see how this was unclear, so we have updated the figure.

(7) Figure 3: There is no panel g (the title is missing); In panels b, c, e, and g the y label is missing, and in panels e and g the x label is missing. Also, the Feedforward model is shown in panel g but it is introduced later in the text. Please remove it from Figure 3. Also in legend: "AV Reverse graph" → "Reverse graph". Also, "Accuracy" and "Alignment" should be presented as percentages (as in Figure 2).

This has been corrected.

(8) Figure 4; x labels are missing.

As with point (6), this was actually because the x-axis were the same across the panels, but we see how this was unclear, so we have updated the figure.

(9) Page 7; I can’t find the cited Figure S1.

Apologies, we have added the supplemental figure (now as S4). It shows the results of models with multiplicative feedback on the task in Fig 5 (as opposed to models with composite feedback shown in the main figure).

Reviewer #2 (Recommendations for the authors):

(1) Discussion Section 3.1 is only a literature review, and does not really add any value.

Respectfully, we think it is important to relate our work to other computational work on the role of top-down feedback, and to make clear what our specific contribution is. But, we have updated the text to try to place additional emphasis on our study’s contribution, so that this section is more than just a literature review.

“Our study adds to this previous work by incorporating modulatory top-down feedback into deep, convolutional, recurrent networks that can be matched to real brain anatomy. Importantly, using this framework we could demonstrate that the specific architecture of top-down feedback in a neural network has important computational implications, endowing networks with different inductive biases.”

(2) Including ipython notebooks and some examples would be great to make it easier to use the code.

We now provide a demo of how to use the code base in a jupyter notebook.

(3) The description of the model is hard to comprehend. Please name and describe all parameters. Also, a figure would be great to understand the different model equations.

We have added definitions of all model terms and parameters.

(4) The terminology is not really clear to me. For example "The results further suggest that different configurations of top-down feedback make otherwise identically connected models functionally distinct from each other and from traditional feedforward only recurrent models." The feedforward and only recurrent seem to contradict each other. Would maybe driving and modulatory be a better term here? I also saw in the code that you differentiate between three types of inputs, modulatory, threshold offset and basal (like feedforward). How about you only classify connections based on these three type? I was also confused about the feedforward only model, because I was unsure whether it is still feedback connections but with "basal" quality, or whether feedback connections between modalities and higher-to-lower level layers were omitted altogether.

We take the reviewer’s point here. To clarify this, we have updated the text to refer to “driving only” rather than “feedforward only”, to make it obvious that what we change in these models is simply whether the connection has any modulatory impact on the activity. 

(5) "incorporating it into ANNs can affect their behavior and help determine the solutions that the network can discover." -> Do you mean constrain? Overall, I did not really get this point.

Yes, we mean that it constrains the solutions that the network is likely to discover.

(6) "ignore the auditory inputs when they visual inputs were unambiguous" -> the not they

This has been fixed. Thank you for catching it.

(7) xlabel in Figure 4 is missing.

This has been fixed, thank you for catching it.

Reviewer #3 (Recommendations for the authors):

Major:

(1) How alignment is computed is not defined. In addition to a proper definition in the methods section, it would be nice to briefly define it when it first appears in the results section.

We’ve added an explicit definition of how alignment is calculated in the methods and emphasized the calculation when its first explained in the results

(2) A connectivity matrix for the feedforward-only model is missing and could be added.

We have added this to Figure 1.

(3) The connectivity matrix for each random model should also be shown.

We’ve shown each of the random model configurations in the new supplemental figure S1.

(4) Initial parameters are not defined, such as W, b etc. A table with all model parameters would be great.

We have added a table to the methods listing all of the parameters.

(5) Would be nice to show the t-sne plots (not just the NH score) for each model and each task in the appendix.

We can provide these figures on request. They massively increase the file size of the paper pdf, as there’s 49 of them for each task and each model, 980 in total. An example t-SNE plot is provided in figure 6.

Minor:

(1) Page 4:

"we refer to this as Visual-dominant Stimulus case 1, or VS1; Fig. 1a, top)." This should be Fig. 2a.

(2) "In stimulus condition VS1, all of the models were able to learn to use the auditory clues to disambiguate the images (Fig. 2c)."

This should be Fig. 2b.

(3) "In comparison, in VS2, we found that the brainlike model learned to ignore distracting audio inputs quickly and consistently compared to the random models, and a bit more rapidly than the auditory information (Fig 2d)."

This should be Fig. 2c.

(4) "VS3; Fig. 2b, top"

This should be Fig. 2d

(5) "while all other models had to learn to do so further along in training (Fig. 2e)."

It is not stated explicitly, but this suggests that the image-aligned target was considered correct, and that weight updates were happening.

(6) "VS4; Fig. 2b, bottom"

This should be Fig. 2f

(7) "adept at learning (Fig. 2f)."

This should be Fig. 2g

(8) Figure 3:b,c,e y-labels are missing

3f: both x and y labels are missing

(9) Figure labeling in the text is not consistent (Fig. 1A versus Fig. 2a)

(10) Doubled "the" in ""This shows that the inductive bias towards vision in the brainlike model depended on the presence of the multiplicative component of the the feedback"

(11) Page 9 Figure 6: The caption says b shows the latent spaces for the VS2 task, whereas the main text refers to 6b as showing the latent space for the AS2 task. Please correct which task it is.

(12) Methods 4.1 page 13

"which is derived from the feedback input (h_{l−1})"

This should be h_{l+1}

(13) r_l, u_l, u and c are not defined to which aspects of the model they refer to

Even though this is based on a previous model, the methods section should completely describe the model.

Equations 1,2,3: the notation [x;y] is unclear and should be defined.

Equation 5: u should probably be u_l.

(14) Page 14 typo: externopyrmidisation.

(15) It is confusing to use different names for the same thing: the all-feedforward model, the all feedforward network, the feedforward network, and the feedforward-only model are probably all the same? Consistent naming would help here.

Thank you for the detailed comments! We’ve fixed the minor errors and renamed the feedforward models to drive-only models.

eLife Assessment

This study investigates the temporal dynamics of neural activity preceding self-initiated movements and makes a valuable contribution to this field. The authors identify key methodological and analytical limitations in previous work and introduce a novel approach to overcome the shortcomings in assessing how predictive neural activity is of an upcoming event. Applying generally solid methods and analyses, they show that a late-stage neural event, ~100 ms before movement execution, is most predictive of upcoming movements, whereas earlier neural activity is less informative. Although interesting, additional analyses are needed to strengthen confidence in this central claim.

Reviewer #1 (Public review):

Summary:

Jeay-Bizot and colleagues investigate the neural correlates of the preparation of, and commitment to, a self-initiated motor action. In their introduction, they differentiate between theoretical proposals relating to the timing of such neural correlates relative to the time of a recorded motor action (e.g., a keypress). These are categorised into 'early' and 'late' timing accounts. The authors advocate for 'late' accounts based on several arguments that align well with contemporary models of decision-making in other domains (for example, evidence accumulation models applied to perceptual decisions). They also clearly describe prevalent methodological issues related to the measurement of event-related potentials (ERPs) and time-frequency power to gauge the timing of the commitment to making a motor action. These methodological insights are communicated clearly and denote potentially important limitations on the inferences that can be drawn from a large body of existing work.

To attempt to account for such methodological concerns, the authors devise an innovative experiment that includes an experimental condition whereby participants make a motor action (a right-hand keypress) to make an image disappear. They also include a condition whereby the stimulus presentation program automatically proceeds at a set time that is matched to the response timing in a previous trial. In this latter condition, no motor action is required by the participant. The authors then attempt to determine the times at which they can differentiate between these two conditions (motor action vs no motor action) based on EEG and MEG data, using event-related potential analyses, time-frequency analyses, and multivariate classifiers. They also apply analysis techniques based on comparing M/EEG amplitudes at different time windows (as used in previous work) to compare these results to those of their key analyses.

When using multivariate classifiers to discriminate between conditions, they observed very high classification performance at around -100ms from the time of the motor response or computer-initiated image transition, but lower classification performance and a lack of statistically significant effects across analyses for earlier time points. Based on this, they make the key claim that measured M/EEG responses at the earlier time points (i.e., earlier than around -100ms from the motor action) do not reliably correlate with the execution of a motor action (as opposed to no such action being prepared or made). This is argued to favour 'late' accounts of motor action commitment, aligning with the well-made theoretical arguments in favour of these accounts in the introduction. Although the exact time window related to 'late' accounts is not concretely specified, an effect that occurs around -100ms from response onset is assumed here to fall within that window.

Importantly, this claim relies on accepting the null hypothesis of zero effect for the time points preceding around -100ms based on a somewhat small sample of n=15 and some additional analyses of individual participant datasets. Although the authors argue that their classifiers are sensitive to detecting relevant effects, and the study appears well-powered to detect the (likely to be large magnitude) M/EEG signal differences occurring around the time of the response or computer-initiated image transition, there is no guarantee that the study is adequately sensitive to detect earlier differences in M/EEG signals. These earlier effects are likely to be more subtle and exhibit lower signal-to-noise ratios, but would still be relevant to the 'early' vs 'late' debate framed in the manuscript. This, along with some observed patterns in the data, may substantially reduce the confidence one may have in the key claim about the onset timing of M/EEG signal differences.

Notably, there is some indication of above-chance (above 0.5 AUC) classification performance at time points earlier than -100ms from the response, as visible in Figure 3A for the task-based EEG analyses (EEG OC dataset, blue line). While this was not statistically significantly above chance for their n=15 sample, these results do not appear to be clear evidence in favour of a zero-effect null-hypothesis. In Figures 2A-B, there are also visible differences in the ERPs across conditions, from around the time that motor action-related components have been previously observed (around -500ms from the response). The plotted standard errors in the data are large enough to indicate that the study may not have been adequately powered to differentiate between the conditions.

Although the authors acknowledge this limitation in the discussion section of their manuscript, their counter-argument is that the classifiers could reliably differentiate between conditions at time points very close to the motor response, and in the time-based analyses where substantive confounds are likely to be present, as demonstrated in a set of analyses. Based on this data, the authors imply that the study is sufficiently powered to detect effects across the range of time points used in the analyses. While it's commendable that these extra analyses were run, they do not provide convincing evidence that the study is necessarily sensitive to detecting more subtle effects that may occur at earlier time points. In other words, the ability of classifiers (or other analysis methods) to detect what are likely to be very prominent, large effects around the time of the motor response does not guarantee that such analyses will detect smaller magnitude effects at other time points.

In summary, the authors develop some very important lines of argument for why existing work may have misestimated the timing of neural signals that precede motor actions. This in itself is an important contribution to the field. However, their attempt to better estimate the timing of such signals is limited by a reliance on accepting the null hypothesis based on non-statistically significant results, and arguably a limited degree of sensitivity to detect subtle but meaningful effects.

Strengths:

This manuscript provides compelling reasons why existing studies may have misestimated the timing of the neural correlates of motor action preparation and execution. They provide additional analyses as evidence of the relevant confounds and provide simulations to back up their claims. This will be important to consider for many in the field. They also endeavoured to collect large numbers of trials per participant to also examine effects in individuals, which is commendable and arguably better aligned with contemporary theory (which pertains to how individuals make decisions to act, rather than groups of people).

The innovative control condition in their experiment may also be very useful for providing complementary evidence that can better characterise the neural correlates of motor action preparation and commitment. The method for matching image durations across active and passive conditions is particularly well thought-out and provides a nice control for a range of potential confounding factors.

Weaknesses:

There is a mismatch between the stated theoretical phenomenon of interest (commitment to making a motor action) and what is actually tested in the study (differences in neural responses when an action is prepared and made compared to when no action is required). The assumed link between these concepts could be made more explicit for readers, particularly because it is argued in the manuscript that neural correlates of motor action preparation are not necessarily correlates of motor action commitment.

As mentioned in the summary, the main issue is the strong reliance on accepting the null hypothesis of no differences between motor action and computer initiation conditions based on a lack of statistically significant results from the modest (n=15) sample. Although a larger sample will increase measurement precision at the group level, there are some EEG data processing changes that could increase the signal-to-noise ratio of the analysed data and produce more precise estimates of effects, which may improve the ability to detect more subtle effects, or at least provide more confidence in the claims of null effects.

First, it is stated in the EEG acquisition and preprocessing section that the 64-channel Biosemi EEG data were recorded with a common average reference applied. Unless some non-standard acquisition software was used (of which we are not aware exists), Biosemi systems do not actually apply this reference at recording (it is for display purposes only, but often mistaken to be the actual reference applied). As stated in the Biosemi online documentation, a reference should be subsequently applied offline; otherwise, there is a substantial decrease in the signal-to-noise ratio of the EEG data, and a large portion of ambient alternating current noise is retained in the recordings. This can be easily fixed by applying a referencing scheme (e.g., the common average reference) offline as one of the first steps of data processing. If this was, in fact, done offline, it should be clearly communicated in the manuscript.

In addition, the data is downsampled using a non-integer divisor of the original sampling rate (a 2,048 Hz dataset is downsampled to 500 Hz rather than 512 Hz). Downsampling using a non-integer divisor is not recommended and can lead to substantial artefacts in raw data as a result, as personally observed by this Reviewer in Biosemi data. Finally, although a 30 Hz low-pass filter is applied for visualisation purposes of ERPs, no such filter is applied prior to analyses, and no method is used to account for alternating current noise that is likely to be in the data. As noted above, much of the alternating current noise will be retained when an offline reference is not applied, and this is likely to further degrade the quality of the data and reduce one's ability to identify subtle patterns in EEG signals. Changes in data processing to address these issues would likely lead to more precise estimates of EEG signals (and by extension differences across conditions).

With regard to possible effects extending hundreds of milliseconds before the response, it would be helpful for the authors to more precisely clarify the time windows associated with 'early' and 'late' theories in this case. The EEG data that would be required to support 'early' theories is also not made sufficiently clear. For example, even quite early neural correlates of motor actions in this task (e.g., around -500ms from the response, or earlier) could still be taken as evidence for the 'late' theories if these correlates simply reflect the accumulation of evidence toward making a decision and associated motor action, as implied by the Leaky Stochastic Accumulator model described by the authors. In other words, even observations of neural correlates of motor action preparation that occur much earlier than the response would not constitute clear evidence against the 'late' account if this neural activity represents an antecedent to a decision and action (rather than commitment to the action), as the authors point out in the introduction.

In addition, there is some discrepancy regarding the data that is used by the classifiers to differentiate between the conditions in the EEG data and the claims about the timing of neural responses that differentiate between conditions. Unless we reviewers are mistaken, the Sliding Window section of the methods states that the AUC scores in Figure 3 are based on windows of EEG data that extend from the plotted time point until 0.5 seconds into the past. In other words, an AUC value at -100ms from the response is based on classifiers applied to data ranging from -600 to -100 milliseconds relative to the response. In this case, the range of data used by the classifiers extends much earlier than the time points indicated by Figure 3, and it is difficult to know whether the data at these earlier time points may have contributed (even in subtle ways) to the success of the classifiers. This may undermine the claim that neural responses only become differentiable from around -100ms from response onset. The spans of these windows used for classification could be made more explicit in Figure 3, and classification windows that are narrower could be included in a subset of analyses to ensure that classifiers only using data in a narrow window around the response show the high degree of classification performance in the dataset. If we are mistaken, then perhaps these details could be clarified in the method and results sections.

Reviewer #2 (Public review):

Summary:

The authors set out to investigate how well the onset of a self-initiated movement could be predicted at different times prior to action onset. To do so, they collected EEG and MEG data across 15 human participants who watched natural landscape images on a screen. These participants performed active self-initiated movements or observed passive actions to have a new image appear. By comparing the neural activity prior to active and time-matched passive actions, the authors found that even though a build-up of neural activity is visible close to 1s prior to action, action onset could only be reliably predicted around 100ms prior to action. These results confirm what was already suggested in previous literature: the commitment to action is only clear from the late stages in the visible neural ramp-up to action onset.

Strengths:

(1) The paper presents a well-thought-out methodology to assess the predictive value of neural activity prior to a self-initiated movement and passively observed action, while keeping all other experimental factors identical. This methodology can be applied outside the specific scope of this paper as well, in efforts to assess the correspondence of a neural signature with an observed behavior.

(2) The results are a strong confirmation of what was suggested less clearly in previous research (Trevena & Miller, 2010, Consciousness & Cognition; Schmidt et al., 2016, Neuroscience & Biobehavioral Reviews; Travers et al., 2020, NeuroImage).

Weaknesses:

(1) Although the authors conducted a solid confirmatory study, the importance of this confirmation is less clear to me. How do the current results change our interpretation of the relation between conscious intention and neural preparation for action? Do these results affect our interpretation of free will? Why does it matter at all whether we see neural preparatory activity prior to the report of a conscious intention to act, or prior to action observation? This study does not clarify the relationship between the observed neural phenomenon, the action or the experienced intention. It does not explain whether this relation is causal, correlational or something else.

(2) Whereas Derchi et al. (2023, Scientific Reports) were able to keep the entire experimental context similar across intended and unintended conditions, Jeay-Bizot et al. have one big difference between their passive and active conditions: the presence of a movement. Therefore, the present results explain the presence or absence of a movement rather than the presence or absence of an intention to act.

eLife Assessment

This fundamental study reports the effects of the psychedelic drug psilocin on iPSC-derived human cortical neurons, analyzing different aspects of structural and functional neuronal plasticity. The evidence is convincing and supports the value of using iPSC-derived human cortical neurons for testing the potentially translational effects of psilocin and other psychedelic-related compounds.

Reviewer #1 (Public review):

Summary:

This study reports the effects of psilocin on iPSC-derived human cortical neurons.

Strengths:

The characterization was comprehensive, involving immunohistochemistry of various markers, 5-HT2A receptors, BDNF, and TrkB, transcriptomics analyses, morphological determination, electrophysiology, and finally synaptic protein measurements. The results are in close agreement with prior work (PMID 29898390) on rat cultured cortical neurons. Nevertheless, there is value in confirming those earlier findings and furthermore to demonstrate the effects in human neurons, which are important for translation. The genetic, proteomics, and cell structure analyses used in this paper are its major strength. The study supports the value of using iPSC-derived human cortical neurons for drug development involving psychedelics-related compounds.

Weaknesses:

(1) Line 140: 5-HT2A receptor expression was found via immunocytochemistry to reside in the somatodendritic and axonal compartments. However, prior work from ex vivo tissue using electron microscopy has found predominantly 5-HT2A receptor expression in the somatodendritic compartment (PMID: 12535944). Was this antibody validated to be 5-HT2A receptor-specific? Can the authors reason why the discrepancy may arise, and if the axonal expression is specific to the cultured neurons?

(2) Line 143: It would be helpful to specify the dose of psilocin tested, and describe how this dose was chosen.

(3) Figure 1: The interpretation is that the differential internalization in the axonal and somatodendritic compartments is time-dependent. However, given that only one dose is tested, it is also possible that this reflects dose dependence, with the longer time exposure leading to higher dose exposure, so these variables are related. That is, if a higher dose is given, internalization may also be observed after 10 minutes in the dendritic compartment.

(4) Figure 3 & 4: What is the 'control' here? A more appropriate control for the 24 hours after psilocin application would be 24 hours after vehicle application. Here the authors are looking at before and after, but the factor of time elapsed and perturbation via application is not controlled for.

(5) The sample size was not clearly described. In the figure legend, N = the number of neurites is provided, but it is unclear how many cells have been analyzed, and then how many of those cells belong to the same culture. These are important sample size information that should be provided. Relatedly, statistical analyses should consider that the neurites from the same cells are not independent. If the neurites indeed come from the same cells, then the sample size is much smaller and a statistical analysis considering the nested nature of the data should be used.

Comments on revisions:

The authors performed substantial experiments to check validity of the HTR2A antibody for the revision. Briefly, they found that western blot shows a single band, abolished by a blocking peptide, in neural progenitors and iPSC-derived neurons, suggesting positive results. However, they also detected immunofluorescence signals in HEK293 and HeLa cells, which do not express 5-HT2A receptors as scRNAseq analysis of these cells show complete absence of the transcript. Therefore the antibody has epitope-selective binding but also has some non-specific binding, precluding its use. The authors rightfully removed the data related to the antibody in the revised manuscript. The account is repeated here to highlight to anyone who may find the information helpful. Overall, the additional results added rigor to the study.

Reviewer #2 (Public review):

In this article, Schmidt et al use iPSC-derived human cortical neurons to test the effects the psychedelic psilocin in different models of neuroplasticity.

Using human iPSC-derived cortical neurons, the authors test the expression of 5-HT2A and subcellular distribution, as well as the effect of different times of exposure to psilocin on 5-HT2A expression. The authors evaluated the effect of the 5-HT2 antagonist ketanserin, as well as the inhibition of dynamin-dependent endocytic pathways with dynasore. Gene expression and plasticity (structural and functional) was also evaluated after different times of exposure to psilocin.

In general, results are interesting since they use the iPSC to evaluate the potentially translationally relevant effects of psilocin (the active metabolite of the psychedelic psilocybin).

Comments on revisions:

The authors have addressed all of my previous concerns. A particular strength of the rebuttal is that the authors corroborated the lack of selectivity/specificity of the anti-5-HT2A antibody used in earlier versions of the manuscript.

Author response:

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

Reviewer #1:

Comment 1: 5-HT2A Antibody Specificity

Was this antibody validated to be 5-HT2A receptor-specific? Can the authors reason why the discrepancy may arise, and if the axonal expression is specific to the cultured neurons?

We performed extensive validation of the anti-5-HT2A receptor antibody (Alomone #ASR-033), which is summarized in the accompanying Author response images:

Positive findings (Author response image 1c-e, Author response image 2a): (1) Western blot showed a single band at the expected molecular weight (~50 kDa) in neural progenitors and iPSCderived neurons. (2) The blocking peptide (#BLP-SR033) abolished Western blot bands and markedly reduced immunofluorescence signals in neurons, confirming epitope-specific binding.

Negative findings (Author response image 1a-b, Author response image 2a-b, Author response image 3): (1) We detected positive immunofluorescence signals in HEK293 and HeLa cells (Author response image 1a-b), which do not express 5-HT2AR. (2) Western blot also showed bands in HEK293 and HeLa cells (Author response image 2a-b). (3) Single-cell RNA-seq analysis of HEK293T cells confirmed complete absence of HTR2A expression (Author response image 3a). (4) qPCR showed no detectable HTR2A transcripts in iPSCs or HeLa cells (Ct > 36), while neural progenitors and neurons showed clear expression (Author response image 3b). (5) siRNA knockdown experiments failed to produce a corresponding decrease in immunofluorescence or Western blot signals, despite reduced HTR2A transcript levels (data not shown).

BLAST analysis: Protein BLAST analysis of the 13-amino acid immunogenic peptide sequence identified the human 5-HT2A receptor as the top hit (9/13 amino acids overlap). However, shorter sequence similarities were also found with other proteins, including APPBP1 (6/9 amino acids), Immunoglobulin Heavy Chain (6/7 amino acids), and Interleukin31 receptor (6/8 amino acids). While these partial homologies do not provide a definitive mechanistic explanation for the observed off-target binding, they illustrate that the epitope sequence is not entirely unique to the 5-HT2A receptor.

Conclusion: While our validation confirmed epitope-specific binding (blocking peptide effective in neurons), the antibody clearly detects something in cells that demonstrably lack HTR2A gene expression. This indicates off-target binding to other proteins sharing the epitope sequence. We have therefore removed all antibody-based 5-HT2A receptor experiments from the revised manuscript. This includes the receptor internalization data from Figure 1. The remaining findings (BDNF upregulation, gene expression changes, morphological effects, electrophysiology) are supported by independent methods including pharmacological blockade with ketanserin.

Comment 2: Psilocin Dose Selection

It would be helpful to specify the dose of psilocin tested, and describe how this dose was chosen.

We used 10 µM psilocin based on: (1) The seminal study by Ly et al. (2018), which demonstrated neuroplasticity effects at this concentration in rat cortical neurons. (2) Our own dose-response experiments (Figure S2B) showing maximal BDNF increase at 10 µM compared to lower concentrations (10 nM, 100 nM, 1 µM). We have clarified this in the revised Methods section.

Comment 3: Dose vs. Time Dependence

Given that only one dose is tested, it is also possible that this reflects dose dependence, with the longer time exposure leading to higher dose exposure.

We agree that dose dependence cannot be excluded with our current experimental design. This point is now moot as we have removed the 5-HT2A receptor internalization experiments from the manuscript. Future studies in our group will address dose-dependent effects on other readouts.

Comment 4: Control Conditions

What is the 'control' here? A more appropriate control would be 24 hours after vehicle application.

The control condition is indeed a vehicle (DMSO) control collected at the same time point as the experimental condition (i.e., 24 hrs post-treatment). We have clarified this in the revised figure legends and Methods section to avoid confusion.

Comment 5: Sample Size Description

The sample size was not clearly described. Statistical analyses should consider that neurites from the same cells are not independent.

We have expanded the sample size descriptions in the figure legends. Analyses were performed using 5-10 microscope images per condition, with 15 ROIs per image, across at least two independent differentiations from two genetic backgrounds. Regarding independence: each neurite segment exists within a distinct microenvironment and can be considered an independent measurement unit, consistent with established practices in the field (Paul et al., 2021, CNS Neurosci Ther). We acknowledge this increases statistical power and have noted this in the Methods.

Reviewer #2:

Comment 1: 5-HT2A Antibody Validation

Without validation (using for example knockdown techniques to decrease expression of 5HT2A), the experiments using this antibody should be excluded from the manuscript.

We agree with this assessment. As detailed in our response to Reviewer 1 (Comment 1) and documented in the Response to Reviewer Figure, our extensive validation attempts—including siRNA knockdown—could not conclusively demonstrate antibody specificity. We have removed all antibody-based 5-HT2A receptor experiments from the revised manuscript.

Comment 2: Serotonin in Cell Media

Did the authors evaluate whether 5-HT is present in the cell media?

The cell culture media used in our experiments does not contain serotonin. We have explicitly stated this in the revised Methods section.

Comment 3: Statistical Analysis of Figure S1F

Some of the datasets are not statistically analyzed, such as Figure S1F.

Figure S1F related to the 5-HT2A receptor experiments and has been removed from the revised manuscript along with the associated data.

Comment 4: Translational Validity of Prolonged Exposure

The authors continuously exposed cells to psilocin for hours or days. Since this is not the model of what occurs in vivo, the findings lack translational validity.

We acknowledge this limitation. Most experiments (BDNF, gene expression, branching) were conducted 24–48 hrs after a brief 10-minute exposure, which better reflects the in vivo situation. Prolonged exposures (96 hrs) were used specifically for synaptogenesis experiments based on literature showing that repeated LSD administration enhances spine density (Inserra et al., 2022; De Gregorio et al., 2022). Our in vitro system lacks metabolizing enzymes and glial cells, which may introduce temporal biases. We have added a discussion of these limitations in the revised manuscript.

Comment 5: Ketanserin Effect on BDNF

In Figure 2E, ketanserin by itself seems to reduce BDNF density. How do the authors conclude that ketanserin blocks psi-induced effects?

We identified that one cell line (Ctrl 1) with inherently higher BDNF density was inadvertently excluded from the ketanserin-only condition. After removing Ctrl 1 from all conditions and reanalyzing, the difference between Ctrl and Ket alone is no longer significant. The significant difference between Psi+Ket and Ket alone demonstrate that psilocin exerts effects that ketanserin can block, consistent with 5-HT2A receptor mediation. The revised figure and statistical analysis are included in the updated manuscript.

Comment 6: mCherry Localization mCherry (Fig 4A) seems to be retained in the nucleus.

The CamKII promoter drives expression of cytoplasmic mCherry, which fills the entire neuron including soma, dendrites, and axons. The apparent nuclear signal reflects mCherry accumulation in the soma, which surrounds the nucleus. The images clearly show mCherry extending into neurites, which was essential for our Sholl analysis of neuronal complexity.

Comment 7: Reference 36

Reference 36 is a review article that does not mention psilocin.

Our statement refers broadly to serotonergic psychedelics increasing neurotrophic factors. Reference 36 (Colaço et al., 2020) examines ayahuasca, which contains the serotonergic psychedelic DMT. We have revised the text to clarify this point.

Summary of Major Revisions

(1) Removed all 5-HT2A receptor antibody-based experiments from Figure 1 and supplementary figures due to inconclusive specificity validation. An Author response image documenting our validation attempts is provided.

(2) Clarified control conditions (vehicle controls at matched time points) in figure legends.

(3) Expanded sample size descriptions in Methods and figure legends.

(4) Re-analyzed ketanserin experiments with consistent cell line inclusion.

(5) Added discussion of translational limitations.

(6) Added new Figure S5 summarizing proposed signaling pathways.

(7) Expanded discussion on the relevance of iPSC-derived neurons for drug development.

Author response image 1.

Immunostaining for 5-HT2A receptor across cell types and peptide-blocking control. (a) HEK293 cells display a positive immunofluorescent signal despite not endogenously expressing 5-HT2AR, indicating nonspecific antibody reactivity. (b) HeLa cells also exhibit a positive signal despite lacking endogenous 5-HT2AR expression, further demonstrating nonspecific antibody binding in non-expressing cell types. (c) Neural progenitor cells show clear positive 5-HT2AR staining. (d) iPSC-derived neurons exhibit robust and well-defined 5-HT2AR staining. (e) Application of the Alomone 5-HT2AR blocking peptide (#BLP-SR033) markedly reduces neuronal signal intensity, supporting epitope-specific binding.

Author response image 2.

Western blot analysis of 5-HT2A receptor abundance and peptide-blocking control. (a-b) In line with the immunofluorescence a single band is detected in iPSCs, HEK cells, neural progenitors, iPSC-derived neurons and (b) HeLa cells. (a) Preincubation of the primary antibody with the corresponding blocking peptide abolishes this band across all samples, consistent with specific binding of the antibody to its intended epitope.

Author response image 3.

Lack of detectable 5-HT2AR expression in HEK and HeLa cells. (a) Analysis of a human-only HEK293T single-cell RNA-seq dataset (10x Genomics; https://www.10xgenomics.com/datasets/293-t-cells-1-standard-1-1-0, accessed 2025-11-25) shows no meaningful HTR2A expression, whereas other genes such as GAPDH, TP53, MYC, and ACTB are robustly detected. Consistently, evaluation of a “Barnyard” dataset - an equal mixture of human HEK293T and mouse NIH3T3 cells (10x Genomics; https://www.10xgenomics.com/datasets/20-k-1-1mixture-of-human-hek-293-t-and-mouse-nih-3-t-3-cells-3-ht-v-3-1-3-1-high-6-1-0, accessed 2025-1125) reveals only ~4 of ~10,000 droplets with minimal HTR2A signal, confirming the absence of meaningful expression.(b) (b) qPCR analysis further demonstrates no detectable HTR2A transcripts in iPSCs or HeLa cells (Ct > 36), while neural progenitors and iPSC-derived cortical neurons show expression when normalized to housekeeping genes GAPDH and TBP.

eLife Assessment

This study provides valuable insight into stress biology by showing that yeast populations can rapidly evolve a trehalose producing resting state that substantially improves survival and rapid regrowth after freeze-thaw. This finding is consistent with the role of trehalose metabolism as a biophysical adaptation that is broadly relevant to the community working on environmental resilience and dormancy. The evidence is convincing: the authors integrate experimental evolution, cell-level biophysical measurements, and modelling in a mutually reinforcing manner.

Reviewer #1 (Public review):

Summary:

This manuscript presents findings on the adaptation mechanisms of Saccharomyces cerevisiae under extreme stress conditions. The authors try to generalize this to adaptation to stress tolerance. A major finding is that S. cerevisiae evolves a quiescence-like state with high trehalose to adapt to freeze-thaw tolerance independent of their genetic background. The manuscript is comprehensive, and each of the conclusions is well supported by careful experiments.

Strengths:

This is excellent interdisciplinary work.

I have commented on the response of the authors, in-line, below. This is to maintain the conversation thread with the authors.

Comment 1:

Earlier papers have shown that loss of ribosomal proteins, that slow growth, leads to better stress tolerance in S. cerevisiae. Given this, isn't it expected that any adaptation that slows down growth would, overall, increase stress tolerance? Even for other systems, it has been shown that slowing down growth (by spore formation in yeast or bacteria/or dauer formation in C. elegans) is an effective strategy to combat stress and hence is a likely route to adaptation. The authors stress this as one of the primary findings. I would like the authors to explain their position, detailing how their findings are unexpected in the context of the literature.

Response:

We agree that the link between slower growth and higher stress tolerance has been well stud-ied. What is distinctive here is that repeated, near-lethal freeze-thaw selected not only for a tolerant/quiescent-like state but also for a shorter lag on re-entry. In this regime of freeze-thaw-regrowth, cells that are tolerant but slow to restart would be outcompeted by naive fast growers. Our quiescence-based selection simulations reproduce exactly this constraint. We have added this explanation to the Results to make clear that the novelty is the co-evolution of a tolerant, trehalose-rich state together with rapid regrowth under an alternating regime.

Comment to Response: I get the point. I believe that the outcome is highly dependent on how selection pressure is administered. So, generalizing this over all stresses (as done in the abstract) may not be accurate.

Comment 2:

Convergent evolution of traits: I find the results unsurprising. When selecting for a trait, if there is a major mode to adapt to that stress, most of the strains would adapt to that mode, independent of the route. According to me, finding out this major route was the objective of many of the previous reports on adaptive evolution. The surprising part in the previous papers (on adaptive evolution of bacteria or yeast) was the resampling of genes that acquired mutations in multiple replicates of an evolution experiments, providing a handle to understand the major genetic route or the molecular mechanism that guides the adaptation (for example in this case it would be - what guides the over-accumulation of trehalose). I fail to understand why the authors find the results surprising, and I would be happy to understand that from the authors. I may have missed something important.

Response:

Our surprise was precisely that we did not see the classical pattern of "phenotypic convergence + repeated mutations in the same locus/module." All independently evolved lines converged on a trehalose-rich, mechanically reinforced, quiescence-like phenotype, but population sequencing across lines did not reveal a single repeatedly hit gene or small shared pathway, even when we increased selection stringency (1-3 freeze-thaw cycles per round). We have now stated in the manuscript that this decoupling (strong phenotypic convergence, non-overlapping genetic routes) is the central inference: selection is acting on a physiologically defined state that multiple genotypes can reach.

Comment to Response: You indeed saw a case of phenotypic convergence. Converging towards trehalose-rich, mechanically reinforced, quiescent like - are phenotypes that have converged. This is what prevented lysis. The same locus need not be mutated over and over again, if the trehalose pathway is controlled by many processes (it is, and many are still unknown as I point in the next comment), many different mutations on different loci can result in the same regulation! I do not see the decoupling between phenotypic convergence and decoupling of genetic mutations as surprising or novel; molecular and cellular biology is replete with such examples where deletion(mutation) of hundreds of different genes can have the same phenotypic outcome (yeast deletion library screening, indirect effects etc). If this was a specific question unsolved in evolutionary biology, then the matter is different.

A minor point: Here I would also like to point out that the three phenotypes you measure may be linked to each other, so their independent evolution may just be a cause-effect relationship. For example Trehalose accumulation may drive the other two. This has not been deconvoluted in this manuscript.

Comment 3:

Adaptive evolution would work on phenotype, as all of selective evolution is supposed to. So, given that one of the phenotypes well-known in literature to allow free-tolerance is trehalose accumulation, I think it is not surprising that this trait is selected. For me, this is not a case of "non-genetic" adaptation as the authors point out: it is likely because perturbation of many genes can individually result in the same outcome - up-regulation of trehalose accumulation. Thereby, although the adaptation is genetic, it is not homogeneous across the evolving lines - the end result is. Do the authors check that the trait is actually a non-genetic adaptation, i.e., if they regrow the cells for a few generations without the stress, the cells fall back to being similarly only partially fit to freeze-thaw cycles? Additionally, the inability to identify a network that is conserved in the sequencing does not mean that there is no regulatory pathway. A large number of cryptic pathways may exist to alter cellular metabolic states.<br /> This is a point in continuation of point #2, and I would like to understand what I have missed.

Response:

We agree, and we have removed the wording "non-genetic adaptation." The evolved populations retain high survival even after regrowth for {greater than or equal to}25 generations without freeze-thaw, so the adaptation is clearly genetically maintained. What our data show is that there is no single genetic route to the shared phenotype; different mutations can all drive cells into the same trehalose-rich, quiescence-like, mechanochemically reinforced state. We now describe this as "genetic diversification with phenotypic convergence."

Comment to Response: While the last term does explain what is going on, isn't it an outcome that is routine in cell biology (as pointed out in my previous comment to your response)? I apologize for not understanding the punchline that is provided in the last few sentences of the abstract.

Comment 4:

To propose the convergent nature, it would be important to check for independently evolved lines and most probably more than 2 lines. It is not clear from their results section if they have multiple lines that have evolved independently.

Response:

We indeed evolved four independent lines and maintained two independent controls. We have added this information at the start of the Results so that the level of replication is immediately clear.

Comment to Response: Previous large scale studies have done hundreds of sequencing to oversample the pathway and figure out reproducible loci. With pooled sequencing (as mentioned below) and only 4 sample evolution, I am not sure that you would have the power in your study to conclude in the loci are sampled or not! If there were 10 gene LOFs that control Trehalose levels (which you can find from the published deletion screening experiment), then four of the experiments are likely to go through one of these routes; what is the likely event that you would identify the same route in two pools? It is unlikely, and therefore, sequencing of 4 pools cannot tell you if the mutation path is repeatedly sampled or not.

Comment 5:

For the genomic studies, it is not clear if the authors sequenced a pool or a single colony from the evolved strains. This is an important point, since an average sequence will miss out on many mutations and only focus on the mutations inherited from a common ancestral cell. It is also not clear from the section.

Response:

We sequenced population samples from the evolved lines. Our specific question was whether independently evolved lines would show the same high-frequency genetic solution, as is often seen in parallel evolution. Pool sequencing may under-sample rare/private variants, but it is appropriate for detecting such shared, high-frequency routes - and we do not find any. We have clarified this rationale in the Methods/Results.

Comment to Response: Please provide the average sequencing depth of each sequencing run. It is essential to understand the power of this study in identifying mutations. What coverage was used in Xgenome size?

Reviewer #2 (Public review):

Summary:

The authors used experimental evolution, repeatedly subjecting Saccharomyces cerevisiae populations to rapid liquid-nitrogen freeze-thaw cycles, while tracking survival, cellular biophysics, metabolite levels, and whole-genome sequence changes. Within 25 cycles, viability rose from ~2 % to ~70 % in all independent lines, demonstrating rapid and highly convergent adaptation despite distinct starting genotypes. Evolved cells accumulated about three-fold more intracellular trehalose, adopted a quiescence-like phenotype (smaller, denser, non-budding cells), showed cytoplasmic stiffening and reduced membrane damage, and re-entered growth with shorter lags-traits that together protected them from ice-induced injury. Whole-genome indicated that multiple genetic routes can yield the same mechano-chemical survival strategy. A population model in which trehalose controls quiescence entry, growth rate, lag, and freeze-thaw survival reproduced the empirical dynamics, implicating physiological state transitions rather than specific mutations as the primary adaptive driver. The study therefore concludes that extreme-stress tolerance can evolve quickly through a convergent, trehalose-rich quiescence-like state that reinforces membrane integrity and cytoplasmic structure.

Strengths:

Experimental design, data presentation and interpretation, writing

Weaknesses:

None

Comments on revisions:

The revised manuscript is improved and addresses the reviews concerns adequately.

Author response:

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

We thank the editor and the reviewers for the detailed and constructive comments. In revising the manuscript we have: (i) clarified what is new relative to prior stress tolerance work, (ii) made explicit that we observe phenotypic convergence without a shared genetic route, (iii) stated upfront that we evolved four independent lines plus two controls, and (iv) corrected figure legends, statistics, and the missing citations. Below we respond point-by-point.

Public Reviews:

Reviewer #1 (Public review):

Summary:

This manuscript presents findings on the adaptation mechanisms of Saccharomyces cerevisiae under extreme stress conditions. The authors try to generalize this to adaptation to stress tolerance. A major finding is that S. cerevisiae evolves a quiescence-like state with high trehalose to adapt to freeze-thaw tolerance independent of their genetic background. The manuscript is comprehensive, and each of the conclusions is well supported by careful experiments.

Strengths:

This is excellent interdisciplinary work.

Weaknesses:

I have questions regarding the overall novelty of the proposal, which I would like the authors to explain.

(1) Earlier papers have shown that loss of ribosomal proteins, that slow growth, leads to better stress tolerance in S. cerevisiae. Given this, isn’t it expected that any adaptation that slows down growth would, overall, increase stress tolerance? Even for other systems, it has been shown that slowing down growth (by spore formation in yeast or bacteria/or dauer formation in C. elegans) is an effective strategy to combat stress and hence is a likely route to adaptation. The authors stress this as one of the primary findings. I would like the authors to explain their position, detailing how their findings are unexpected in the context of the literature.

We agree that the link between slower growth and higher stress tolerance has been well studied. What is distinctive here is that repeated, near-lethal freeze–thaw selected not only for a tolerant/quiescent-like state but also for a shorter lag on re-entry. In this regime of freeze–thaw–regrowth, cells that are tolerant but slow to restart would be outcompeted by naive fast growers. Our quiescence-based selection simulations reproduce exactly this constraint. We have added this explanation to the Results to make clear that the novelty is the co-evolution of a tolerant, trehaloserich state together with rapid regrowth under an alternating regime.

(2) Convergent evolution of traits: I find the results unsurprising. When selecting for a trait, if there is a major mode to adapt to that stress, most of the strains would adapt to that mode, independent of the route. According to me, finding out this major route was the objective of many of the previous reports on adaptive evolution. The surprising part in the previous papers (on adaptive evolution of bacteria or yeast) was the resampling of genes that acquired mutations in multiple replicates of an evolution experiments, providing a handle to understand the major genetic route or the molecular mechanism that guides the adaptation (for example in this case it would be - what guides the overaccumulation of trehalose). I fail to understand why the authors find the results surprising, and I would be happy to understand that from the authors. I may have missed something important.

Our surprise was precisely that we did not see the classical pattern of “phenotypic convergence + repeated mutations in the same locus/module.” All independently evolved lines converged on a trehalose-rich, mechanically reinforced, quiescence-like phenotype, but population sequencing across lines did not reveal a single repeatedly hit gene or small shared pathway, even when we increased selection stringency (1–3 freeze–thaw cycles per round). We have now stated in the manuscript that this decoupling (strong phenotypic convergence, non-overlapping genetic routes) is the central inference: selection is acting on a physiologically defined state that multiple genotypes can reach.

(3) Adaptive evolution would work on phenotype, as all of selective evolution is supposed to. So, given that one of the phenotypes well-known in literature to allow free-tolerance is trehalose accumulation, I think it is not surprising that this trait is selected. For me, this is not a case of ”non-genetic” adaptation as the authors point out: it is likely because perturbation of many genes can individually result in the same outcome - up-regulation of trehalose accumulation. Thereby, although the adaptation is genetic, it is not homogeneous across the evolving lines - the end result is. Do the authors check that the trait is actually a non-genetic adaptation, i.e., if they regrow the cells for a few generations without the stress, the cells fall back to being similarly only partially fit to freeze-thaw cycles? Additionally, the inability to identify a network that is conserved in the sequencing does not mean that there is no regulatory pathway. A large number of cryptic pathways may exist to alter cellular metabolic states.

This is a point in continuation of point #2, and I would like to understand what I have missed.

We agree, and we have removed the wording “non-genetic adaptation.” The evolved populations retain high survival even after regrowth for ≥25 generations without freeze–thaw, so the adaptation is clearly genetically maintained. What our data show is that there is no single genetic route to the shared phenotype; different mutations can all drive cells into the same trehalose-rich, quiescencelike, mechanochemically reinforced state. We now describe this as “genetic diversification with phenotypic convergence.”

(4) To propose the convergent nature, it would be important to check for independently evolved lines and most probably more than 2 lines. It is not clear from their results section if they have multiple lines that have evolved independently.

We indeed evolved four independent lines and maintained two independent controls. We have added this information at the start of the Results so that the level of replication is immediately clear.

(5) For the genomic studies, it is not clear if the authors sequenced a pool or a single colony from the evolved strains. This is an important point, since an average sequence will miss out on many mutations and only focus on the mutations inherited from a common ancestral cell. It is also not clear from the section.

We sequenced population samples from the evolved lines. Our specific question was whether independently evolved lines would show the same high-frequency genetic solution, as is often seen in parallel evolution. Pool sequencing may under-sample rare/private variants, but it is appropriate for detecting such shared, high-frequency routes — and we do not find any. We have clarified this rationale in the Methods/Results.

Reviewer #2 (Public review):

Summary:

The authors used experimental evolution, repeatedly subjecting Saccharomyces cerevisiae populations to rapid liquid-nitrogen freeze-thaw cycles while tracking survival, cellular biophysics, metabolite levels, and whole-genome sequence changes. Within 25 cycles, viability rose from ~2 % to ~70 % in all independent lines, demonstrating rapid and highly convergent adaptation despite distinct starting genotypes. Evolved cells accumulated about threefold more intracellular trehalose, adopted a quiescence-like phenotype (smaller, denser, non-budding cells), showed cytoplasmic stiffening and reduced membrane damage, and re-entered growth with shorter lag traits that together protected them from ice-induced injury. Whole-genome sequencing indicated that multiple genetic routes can yield the same mechano-chemical survival strategy. A population model in which trehalose controls quiescence entry, growth rate, lag, and freeze-thaw survival reproduced the empirical dynamics, implicating physiological state transitions rather than specific mutations as the primary adaptive driver. The study therefore concludes that extreme-stress tolerance can evolve quickly through a convergent, trehalose-rich quiescence-like state that reinforces membrane integrity and cytoplasmic structure.

Strengths:

The strengths of the paper are the experimental design, data presentation and interpretation, and that it is well-written.

(1) While the phenotyping is thorough, a few more growth curves would be quite revealing to determine the extent of cross-stress protection. For example, comparing growth rates under YPD vs. YPEG (EtOH/glycerol), and measuring growth at 37ºC or in the presence of 0.8 M KCl.

We thank the referee for the interesting suggestions. However, growth rates alone may be difficult to interpret since WT strains also show different growth rates under these conditions. Therefore, comparing the relative fitness or survival of the evolved strains versus the WT under these stresses would be more informative. In the present study we limited growth/survival measurements to what was needed to parameterize the adaptation model in YPD under the freeze–thaw regime. We have now added a statement in the Discussion that, given the shared trehalose/mechanical mechanism, such cross-stress assays are an expected and straightforward follow-up.

(2) Is GEMS integrated prior to evolution? Are the evolved cells transformable?

Yes. GEMs were integrated prior to evolution, because the non-integrated evolved population showed low transformation efficiency, likely due to altered cell-wall properties.

(3) From the table, it looks like strains either have mutations in Ras1/2 or Vac8. Given the known requirements of Ras/PKA signaling for the G1/S checkpoint (to make sure there are enough nutrients for S phase), this seems like a pathway worth mentioning and referencing. Regarding Vac8, its emerging roles in NVJ and autophagy suggest another nutrient checkpoint, perhaps through TORC1. The common theme is rewired metabolism, which is probably influencing the carbon shuttling to trehalose synthesis.

We appreciate the reviewer’s suggestion to consider pathways like Ras/PKA (linked to Ras1/2) and autophagy/TORC1 (linked to Vac8) as potential upstream modulators. While these pathways are involved in nutrient sensing and metabolic regulation, we choose not to emphasize them specifically. This is because (i) some evolved lines lack Ras1/2 or Vac8 variants, and (ii) none of the variants lies directly in trehalose synthesis/degradation pathways. Furthermore, direct links to trehalose accumulation are not well established for these specific variants in this context, and pathways like Ras are global regulators with broad effects. Together with the strongly convergent phenotype, this supports our main inference that multiple genetic/metabolic routes can feed into the same trehalose-rich, mechanochemically reinforced, quiescence-like state. We have added a note in the discussion regarding metabolic rewiring and trehalose.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

Generally, the results sections should have more details. The figures should be corrected, and the legends should be checked for correctness. The manuscript seems to have been assembled in haste?

We have expanded the relevant Results subsections with one-sentence motivations (why each measurement was performed) and we have corrected the figure legends for ordering and consistency.

Figure 3: It will be good to have the correct p-values on the figure itself. P-values are typically less than 1, unless there is some special method (here the values presented are , etc). Please explain how the P-values were obtained in the figure legend itself.

Figure 3 now shows the actual p-values. The legend specifies the details and the sample sizes used.

Figure 5: It is not clear what the error bars show in 5B, E (different evolved population/ clones/ cells?). All the figure legends are mixed up, please correct them. It is difficult to follow the paper.

Figure 5 legends now state clearly what the error bars represent (biological replicates) and which panels are from single-cell measurements. We have checked the panel lettering and legend order for consistency with the flow of the main text.

Reviewer #3 (Recommendations for the authors):

Overall, the paper is outstanding, well-written, and insightful.

A point to address is that there are missing citations on lines 60, 91.

We have added the missing citations at both locations. We apologize for the omission, which was due to a compilation error. This error has been fixed, and the bibliography has been corrected (now containing 74 references).

eLife Assessment

The authors present an important set of data implicating ETFDH as an epigenetically suppressed gene in cancer with tumor suppressive functions. The evidence is convincing, with the authors demonstrating that suppression of ETFDH activity results in accumulation of amino acids that impact metabolism via hyperactive mTORC1.

Reviewer #1 (Public review):

In their manuscript, Papadopoli et al explore the role of ETFDH in transformation. They note that ETFDH protein levels are decreased in cancer, and that deletion of ETFDH in cancer cell lines results in increased tumorigenesis, elevated OXPHOS and glycolysis, and a reduction in lipid and amino acid oxidation. The authors attribute these effects to increased amino acid levels stimulating mTORC1 signaling and driving alterations in BCL6 and EIF4EBP1. They conclude that ETFDH1 is epigenetically silenced in a proportion of neoplasms, suggesting a tumor-suppressive function. Overall, the authors logically present clear data and perform appropriate experiments to support their hypotheses.

Reviewer #2 (Public review):

Summary:

The altered metabolism of tumors enables their growth and survival. Classically, tumor metabolism often involves increased activity of a given pathway in intermediary metabolism to provide energy or substrates needed for growth. Papadopoli et al. investigate the converse - the role of mitochondrial electron transfer flavoprotein dehydrogenase (ETFDH) in cancer metabolism and growth. The authors present compelling evidence that ETFDH insufficiency, which is detrimental in non-malignant tissues, paradoxically enhances bioenergetic capacity and accelerates neoplastic growth in cancer cells in spite of the decreased metabolic fuel flexibility that this affords tumor cells. This is achieved through the retrograde activation of the mTORC1/BCL-6/4E-BP1 axis, leading to metabolic and signaling reprogramming that favors tumor progression.

Strengths:

This review focuses primarily on the cancer metabolism aspects of the manuscript.

The study provides robust evidence linking ETFDH insufficiency to enhanced cancer cell bioenergetics and tumor growth.

The use of multiple cancer cell lines and in vivo models strengthens the generalizability of the findings.

The mechanistic insights into the mTORC1/BCL-6/4E-BP1 axis and its role in metabolic reprogramming are of general interest within and outside the immediate field of tumor metabolism.

Conclusion:

This manuscript provides significant insights into the role of ETFDH insufficiency in cancer metabolism and growth. The findings highlight the potential of targeting the mTORC1/BCL-6/4E-BP1 axis in ETFDH-deficient cancers. The compelling data support the conclusions presented in the manuscript, which will be valuable to the cancer metabolism community.

[Editors' note: The authors have addressed each of the two weaknesses previously listed in the public review, providing new experimental data on nucleotides and showing that the catalytic activity is required via the suggested addback experiment.]

Author response:

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

Public Reviews:

Reviewer #1 (Public review):

Authors state, "we identified ETF dehydrogenase (ETFDH) as one of the most dispensable metabolic genes in neoplasia." Surely there are thousands of genes that are dispensable for neoplasia. Perhaps the authors can revise this sentence and similar sentiments in the text.

We agree with the reviewer and have corrected the text accordingly. Specifically, we rephrased the sentence: “Surprisingly, we observed that in contrast to muscle, ETFDH is one of the most non-essential metabolic genes in cancer cells.” to “Surprisingly, we observed that in contrast to muscle, ETFDH is a non-essential gene in acute lymphoblastic leukemia NALM-6 cells”

Authors state, "These findings show that ETFDH loss elevates glutamine utilization in the CAC to support mitochondrial metabolism." While elevated glutamine to CAC flux is consistent with the statement that increased glutamine, the authors have not measured the effect of restoring glutamine utilization to baseline on mitochondrial metabolism. Thus, the causality implied by the authors can only be inferred based on the data presented. Indeed, the increased glutamine consumption may be linked to the increase in ROS, as glutamate efflux via system xCT is a major determinant of glutamine catabolism in vitro.

Indeed. We changed the statement "These findings show that ETFDH loss elevates glutamine utilization in the CAC to support mitochondrial metabolism." to "Collectively, these data demonstrate that ETF insufficiency in cancer cells remodels mitochondrial metabolism and increases the glutamine consumption and anaplerosis."

Authors state that the mechanism described is an example of "retrograde signaling". However, the mechanism seems to be related to a reduction in BCAA catabolism, suggesting that the observed effects may be a consequence of altered metabolic flux rather than a direct signaling pathway. The data presented do not delineate whether the observed effects stem from disrupted mitochondrial communication or from shifts in nutrient availability and metabolic regulation.

Notwithstanding that the term “retrograde” was used to refer to signaling from mitochondria to mTORC1, rather than from mTORC1 to mitochondria [1], we have removed the term “retrograde signaling” throughout the manuscript.

The authors should discuss which amino acids that are ETFDH substrates might affect mTORC1 activity or consider whether other ETFDH substrates might also affect mTORC1 in their discussion. Along these lines, the authors might consider discussing why amino acids that are not ETFDH substrates are increased upon ETFDH loss.

Based on the literature, we expect that branched chain amino acids that are ETFDH substrates (e.g., leucine) are likely to play a major role in activating mTORC1 upon ETFDH abrogation. As expected, the aforementioned amino acids are among those that are the most highly upregulated in ETFDH deficient cells (Fig 3A). We have, however, never formally tested the role of branched chain amino acid in activating mTORC1 in the context of ETFDH disruption. The increase in amino acids that are not metabolized via ETFDH, is likely to stem from global metabolic rewiring of ETFDH-deficient cells and observed alterations in amino acid uptake (e.g., glutamine; Fig 2F). We discuss this in the revised version of the paper as follows:

“Several metabolites can be sensed via signaling partners upstream of mTORC1, including leucine, arginine, methionine/SAM, and threonine [2]. Branched-chain amino acids (leucine, isoleucine, and valine), which are among the highest upregulated metabolites in ETFDH deficient cells (Fig 3A) serve as ETFDH substrates, and have been described to display strong activation capabilities towards mTORC1 in the literature [3,4]. Glutamine can also activate mTORC1 through Arf family of GTPases [5]. Indeed, glutamine can supplement the non-essential amino acid (NEAA) pool through transamination [6] and amino acid uptake [7]. Accordingly, the maintenance of NEAA that are non-ETFDH substrates may be supported by the global metabolic rewiring fueled by enhanced glutamine metabolism in ETFDH-deficient cells. Deciphering the mechanisms leading to accumulation of specific amino acids and their role in ETFDH-dependent mTORC1 modulation is warranted.”

Reviewer #2 (Public review):

The authors would strengthen the paper considerably by adding back catalytically inactive ETFDH to show that the activity of this enzyme is responsible for the increased growth phenotypes and changes in labeling that they observe.

Based on the Reviewers’ suggestions we performed these experiments. Herein, we took advantage of Y304A/G306E ETFDH mutant that impairs electron transfer from ETF and cannot substitute for the wild type (WT) gene function in ETFDH-deficient myoblasts [8]. We expressed WT and Y304A/G306E ETFDH mutant in ETFDH KO HCT116 colorectal cancer cells and confirmed that they are expressed to a comparable level (Supplementary Figure 6C). Re-expression of WT decreased proliferation, while suppressing mTORC1 signaling and increasing 4E-BP1 levels relative to control (vector infected) ETFDH KO EV HCT116 cells (Supplementary Figure 6D). In contrast, proliferation rates, mTORC1 signaling and 4E-BP1 levels remained largely unchanged upon Y304A/G306E ETFDH mutant expression in ETFDH KO HCT116 cells (Supplementary Figure 6D). Similarly, re-expression of WT ETFDH disrupted the bioenergetic phenotype associated with ETFDH loss, in contrast to re-expression of Y304A/G306E ETFDH mutant, which exhibited similar bioenergetic profiles as ETFDH KO control (Supplementary Figure 6E-F). Collectively these findings argue that the ETFDH activity is required for its tumor suppressive effects.

If nucleotide pool and labeling data are available, or can be obtained readily, this would significantly strengthen the tracing data already obtained.

We followed Reviewer’s suggestion and measured nucleotide levels. This revealed that loss of ETFDH results in increase in steady-state nucleotide pools (Supplementary Figure 2K), consistent with increased aspartate labelling and accelerated tumor growth.

References

(1) Morita, M. et al. mTORC1 controls mitochondrial activity and biogenesis through 4EBP-dependent translational regulation. Cell Metab 18, 698-711 (2013). https://doi.org/10.1016/j.cmet.2013.10.001

(2) Valenstein, M. L. et al. Structural basis for the dynamic regulation of mTORC1 by amino acids. Nature 646, 493-500 (2025). https://doi.org/10.1038/s41586-025-09428-7

(3) Appuhamy, J. A., Knoebel, N. A., Nayananjalie, W. A., Escobar, J., & Hanigan, M. D. Isoleucine and leucine independently regulate mTOR signaling and protein synthesis in MAC-T cells and bovine mammary tissue slices. J Nutr 142, 484-491 (2012). https://doi.org/10.3945/jn.111.152595

(4) Herningtyas, E. H. et al. Branched-chain amino acids and arginine suppress MaFbx/atrogin-1 mRNA expression via mTOR pathway in C2C12 cell line. Biochim Biophys Acta 1780, 1115-1120 (2008). https://doi.org/10.1016/j.bbagen.2008.06.004

(5) Jewell, J. L. et al. Metabolism. Differential regulation of mTORC1 by leucine and glutamine. Science 347, 194-198 (2015). https://doi.org/10.1126/science.1259472

(6) Tan, H. W. S., Sim, A. Y. L. & Long, Y. C. Glutamine metabolism regulates autophagy-dependent mTORC1 reactivation during amino acid starvation. Nat Commun 8, 338 (2017). https://doi.org/10.1038/s41467-017-00369-y

(7) Chen, R. et al. The general amino acid control pathway regulates mTOR and autophagy during serum/glutamine starvation. J Cell Biol 206, 173-182 (2014).https://doi.org/10.1083/jcb.201403009

(8) Herrero Martin, J. C. et al. An ETFDH-driven metabolon supports OXPHOS efficiency in skeletal muscle by regulating coenzyme Q homeostasis. Nat Metab 6, 209-225 (2024). https://doi.org/10.1038/s42255-023-00956-y

eLife Assessment

This manuscript investigates the extremely interesting and important claim that the human hippocampus represents interactions with multiple social interaction partners on two relatively abstract social dimensions - and that this ability correlates with the social network size of the participant. This research potentially demonstrates the intricate role of the hippocampus in navigating our social world. While most of the results are solid, the paper requires some further clarification.

Reviewer #1 (Public review):

Schafer et al. tested whether the hippocampus tracks social interactions as sequences of neural states within an abstract social space defined by the dimensions of affiliation and power, using a narrative-based task in which participants engaged in dynamic social interactions. The study showed that individual social relationships were represented as distinct trajectories of hippocampal activity patterns. These neural trajectories systematically reflected trial-by-trial changes in affiliation and power between the participant and each character, suggesting that the hippocampus encodes sequences of socially relevant events and their relational structure, extending its well-established role beyond spatial representations.

A major strength of this study is the use of a richly structured, narrative-based task that allows social relationships to evolve dynamically over time. The use of representational similarity analysis provides a principled framework for linking behavioral trajectories in social space to neural pattern dynamics.

One potential limitation concerns temporal autocorrelation in the neural data, as nearby trials are inherently related both behaviorally and temporally within a continuous narrative. Although the authors carefully attempted to control for temporal distance and related confounds, fully disentangling representational similarity driven by social structure from similarity driven by temporal proximity remains challenging within a single-session task design.

While the findings of a two-dimensional representational structure is an important contribution, it remains an open question whether such a representation reflects an inherent property of how the human brain encodes social relationships, or whether it is partly driven by task constraints in which social interactions were limited to changes along two (affiliation and power) dimensions. Future studies that allow social relationships to vary along richer or higher-dimensional feature spaces will be necessary to determine the generality of low dimensional representations.

Reviewer #2 (Public review):

The substantially revised paper has increased in clarity and is much more accessibe and straightforward than the first version. The analyses are now clearer and support the conclusions better. There are however some remaining methodological weakness, which in my mind still renders the evidence to not be entirely convincing.

(1) The temporal autocorrelation concern is not fully convincingly addressed. The temporal autocorrelation curves supplied in the supplements are really helpful, but linearly regressing out the temporal distance from the neural distance clearly does not work, as one can see from the right panel of supplementary Figure 1. If the method had worked correctly the line should have been flat. The analysis however shows that decision trials with a lag > 2 are basically independent - so a simple way to address this is to restrict the RSA analysis to trials with a decision lag of > 2. This analysis would strengthen the paper a lot.

(2) In the final analysis, the authors use all the trials to make the claim that the hippocampus represents the characters in a shared social space. However, as within-character distances are still included in the analysis, this result could still be driven by the effects of within-character representations that are not shared across characters. A simple way of addressing this concern would be to only include between-character distances in this analysis, making it truly complementary to the previous within-character analysis. It would also be very interesting to compare the the within- and between-character analyses in the hippocampus directly.

(3) Overall, the correction for multiple comparisons in the fMRI and the resulting corrected p-values are not sufficiently explained and documented in the paper. What was exactly permuted in the tests? Was correction applied in a voxel-wise or cluster-wise fashion? If cluster-wise, the cluster-wise p-values need to be reported.

Author response:

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

Public reviews:

Reviewer #1 (Public review):

Summary:

Schafer et al. tested whether the hippocampus tracks social interactions as sequences of neural states within an abstract social space defined by dimensions of affiliation and power, using a task in which participants engaged in narrative-based social interactions. The findings of this study revealed that individual social relationships are represented by unique sequences of hippocampal activity patterns. These neural trajectories corresponded to the history of trial-to-trial affiliation and power dynamics between participants and each character, suggesting an extended role of the hippocampus in encoding sequences of events beyond spatial relationships.

The current version has limited information on details in decoding and clustering analyses which can be improved in the future revision.

Strengths:

(1) Robust Analysis: The research combined representational similarity analysis with manifold analyses, enhancing the robustness of the findings and the interpretation of the hippocampus's role in social cognition.

(2) Replicability: The study included two independent samples, which strengthens the generalizability and reliability of the results.

Weaknesses:

I appreciate the authors for utilizing contemporary machine-learning techniques to analyze neuroimaging data and examine the intricacies of human cognition. However, the manuscript would benefit from a more detailed explanation of the rationale behind the selection of each method and a thorough description of the validation procedures. Such clarifications are essential to understand the true impact of the research. Moreover, refining these areas will broaden the manuscript's accessibility to a diverse audience.

We thank the reviewer for these comments and have addressed them in various ways.

First, we removed the spline-based decoding and spectral clustering analyses. As we detail in our response to the recommendations, these approaches were complex and raised legitimate interpretational concerns, making it unclear how they supported our core claims. The revised manuscript now focuses on a set of representational similarity analyses to show representations consistent with social dimension similarity (affiliation vs. power decision trials) and social location similarity (trajectory/map-like coding based on participant choices).

Second, we expanded the Methods and Results to more clearly explain the analyses, the questions they address, and associated controls and robustness tests. The dimension similarity analysis tests whether hippocampal patterns differentiate affiliation and power decisions in a way consistent with an abstract dimension representation. The location similarity RSAs test whether within-character neural pattern distances scale with Euclidean distance in social space (relationship-specific trajectories), and whether pattern distances across all characters scale with location distances when distances are globally standardized, consistent with a shared map-like coordinate system.

Third, we emphasize new controls. For the dimension similarity RSA, we test for potential confounds such as word count, text sentiment, and reaction time differences between affiliation and power trials. For the location similarity RSA, we control for temporal distance between trials and show (in the Supplement) that the reported effects cannot be explained by temporal autocorrelation in the fMRI data or by the relationship between temporal distance and behavioral location distance.

We believe that these changes address the reviewer’s request for clearer rationale and validation.

Reviewer #2 (Public review):

Summary:

Using an innovative task design and analysis approach, the authors set out to show that the activity patterns in the hippocampus related to the development of social relationships with multiple partners in a virtual game. While I found the paper highly interesting (and would be thrilled if the claims made in the paper turned out to be true), I found many of the analyses presented either unconvincing or slightly unconnected to the claims that they were supposed to support. I very much hope the authors can alleviate these concerns in a revision of the paper.

Strengths & Weaknesses:

(1) The innovative task design and analyses, and the two independent samples of participants are clear strengths of the paper.

We thank the reviewer for this comment.

(2) The RSA analysis is not what I expected after I read the abstract and tile of the result section "The hippocampus represents abstract dimensions of affiliation and power". To me, the title suggests that the hippocampus has voxel patterns, which could be read out by a downstream area to infer the affiliation and power value, independent of the exact identity of the character in the current trial. The presented RSA analysis however presents something entirely different - namely that the affiliation trials and power trials elicit different activity patterns in the area indicated in Figure 3. What is the meaning of this analysis? It is not clear to me what is being "decoded" here and alternative explanations have not been considered. How do affiliation and power trials differ in terms of the length of sentences, complexity of the statements, and reaction time? Can the subsequent decision be decoded from these areas? I hope in the revision the authors can test these ideas - and also explain how the current RSA analysis relates to a representation of the "dimensions of affiliation and power".

We agree that this analysis needed to be better justified and explained. We have revised the text to clarify that by “represents the interaction decision trials along abstract social dimensions” we mean that hippocampal multivoxel patterns differentiate affiliation and power decisions in a way consistent with the conceptual framework of underlying latent dimensions. The analysis tests one simple prediction of this view – that on average these trial types are separable in the neural patterns. We have added details to the Methods, showing how the affiliation and power trials do not differ in word count or in sentiment, but do differ in their semantics, as assessed by a Large Language Model, as we expect from our task assumptions. Thanks to the reviewer’s comment, we also tested for and found a reaction time difference between affiliation and power trials, that we now control for.

(3) Overall, I found that the paper was missing some more fundamental and simpler RSA analyses that would provide a necessary backdrop for the more complicated analyses that followed. Can you decode character identity from the regions in question? If you trained a simple decoder for power and affiliation values (using the LLE, but without consideration of the sequential position as used in the spline analysis), could you predict left-out trials? Are affiliation and power represented in a way that is consistent across participants - i.e. could you train a model that predicts affiliation and power from N-1 subjects and then predict the Nth subject? Even if the answer to these questions is "no", I believe that they are important to report for the reader to get a full understanding of the nature of the neural representations in these areas. If the claim is that the hippocampus represents an "abstract" relationship space, then I think it is important to show that these representations hold across relationships. Otherwise, the claim needs to be adjusted to say that it is a representation of a relationship-specific trajectory, but not an abstract social space.

We appreciate this comment and agree on the value of clear, conceptually simple analyses. To address this concern, we have simplified our main analysis significantly by removing the spline-based analysis and substituting it with a multiple regression representational similarity analysis approach. We test whether within-character neural pattern distances scale with distance in social space (relationship-specific trajectories), and whether pattern distances across all characters scale with location distances when distances are globally standardized. We find evidence for both, consistent with a shared map-like coordinate system.

We agree that decoding character identity and an across-participant decoding approach could be informative. However, our current task is not well designed for such analyses and as such would complicate the paper. Although we agree that these questions are interesting, they would test questions that are outside the scope of this paper. 

(4) To determine that the location of a specific character can be decoded from the hippocampal activity patterns, the authors use a sequential analysis in a lowdimensional space (using local linear embedding). In essence, each trial is decoded by finding the pair of two temporally sequential trials that is closest to this pattern, and then interpolating the power/affiliation values linearly between these two points. The obvious problem with this analysis is that fMRI pattern will have temporal autocorrelation and the power and affiliation values have temporal autocorrelation. Successful decoding could just reflect this smoothness in both time series. The authors present a series of control analyses, but I found most of them to not be incisive or convincing and I believe that they (and their explanation of their rationale) need to be improved. For example, the circular shifting of the patterns preserves some of the autocorrelation of the time series - but not entirely. In the shifted patterns, the first and last items are considered to be neighboring and used in the evaluation, which alone could explain the poor performance. The simplest way that I can see is to also connect the first and last item in a circular fashion, even when evaluating the veridical ordering. The only really convincing control condition I found was the generation of new sequences for every character by shuffling the sequence of choices and re-creating new artificial trajectories with the same start and endpoint. This analysis performs much better than chance (circular shuffling), suggesting to me that a lot of the observed decoding accuracy is indeed simply caused by the temporal smoothness of both time series.

We thank the reviewer for emphasizing this important concern; we agree that we did not sufficiently address this in the initial submission. This concern is one main reason we removed the spline-based analysis and now use regression-based representational similarity analyses in its place. In the revision, we report autocorrelation-related analyses in the supplement, and via controls and additional analysis show that temporal distance (or its square) cannot explain the location-like effects. This substantially improves our ability to interpret the findings.

(5) Overall, I found the analysis of the brain-behavior correlation presented in Figure 5 unconvincing. First, the correlation is mostly driven by one individual with a large network size and a 6.5 cluster. I suspect that the exclusion of this individual would lead to the correlation losing significance. Secondly, the neural measure used for this analysis (determining the number of optimal clusters that maximize the overlap between neural clustering and behavioral clustering) is new, non-validated, and disconnected from all the analyses that had been reported previously. The authors need to forgive me for saying so, but at this point of the paper, would it not be much more obvious to use the decoding accuracy for power and affiliation from the main model used in the paper thus far? Does this correlate? Another obvious candidate would be the decoding accuracy for character identity or the size of the region that encodes affiliation and power. Given the plethora of candidate neural measures, I would appreciate if the authors reported the other neural measures that were tried (and that did not correlate). One way to address this would have been to select the method on the initial sample and then test it on the validation sample - unfortunately, the measure was not pre-registered before the validation sample was collected. It seems that the correlation was only found and reported on the validation sample?

We agree that this analysis was too complicated and under constrained, and thus not convincing. We think that removing this cluster-based analysis is the most conservative response to the reviewer’s concerns and have removed it from the revised paper.

Recommendations to the authors:

Reviewer #1 (Recommendations for the authors):

The manuscript's description of the shuffling analysis performed during decoding is currently ambiguous, particularly concerning the control variables. This ambiguity is present only in the Figure 4 legends and requires a more detailed explanation within the methods section. It is essential to clarify whether the permutation process was conducted within each character's data set or across multiple characters' data sets. If permutations were confined to within-character data, the conclusion would be that the hippocampus encodes context-specific information rather than providing a twodimensional common space.

We thank the reviewer for this comment. We have now removed the spline analysis due to these and other problems and have replaced it with representational similarity analyses that are both more rigorous and easier to interpret. We think these analyses allow us to make the claim that the characters are represented in a common space. 

In the methods, we explain the analyses (page 23-24, lines 475-500):

“We also expected the hippocampus to represent the different characters’ changing social locations, which are implicit in the participant’s choices. We used multiple regression searchlight RSA to test whether hippocampal pattern dissimilarity increases with social location distance, based on participant-specific trial-wise beta images where boxcar regressors spanned each trial’s reaction time.”

“We ran two complementary regression analyses to address two related questions. First, we asked whether the hippocampus represents how a specific relationship changes over time. For this analysis, for each participant and each searchlight, we computed character-specific (i.e., only for same character trial pairs) correlation distances between trial-wise beta patterns and Euclidean distances between the social location behavioral coordinates. Distances were zscored within character trial pairs to isolate character-specific changes. The second analysis asked whether the there is a common map-like representation, where all trials, regardless of relationship, are represented in a shared coordinate system. Here, we included all trial pairs and z-scored the distances globally. For both regression analyses, we included control distances to control for possible confounds. To account for generic time-related changes, we controlled for absolute scan-time difference, as this correlated with location distance across participants (see Temporal autocorrelation of hippocampal beta patterns in the supplement). Although the square of this temporal distance did not explain any additional variance in behavioral distances, we ran a robustness analysis including both temporal distance and its square and saw qualitatively the same clusters with similar effect sizes. As such, we report the main analysis only. We included binary dimension difference (0 = trial pairs of different dimension, 1 = trials pairs of the same dimension), to ensure effects could not be explained by dimension-related effects. In the group-level model, we controlled for sample and the average reaction time between affiliation and power decisions.”

In the results, we describe the results and our interpretation (pages 11-12, lines 185208):

“We have shown that the left hippocampus represents the affiliation and power trials differently, consistent with an abstract dimensional representation. Does it also represent the changing social coordinates of each character? To test this, we multiple-regression RSA searchlight to test whether left hippocampus patterns represent the characters’ changing social locations across interactions (see Figure 3). We restricted the distances to those from trial pairs from the same character and standardized the distances within character (see Figure 3BD). We controlled for temporal distance to ensure the effect was not explainable by the time between trials, and for whether the trials shared the same underlying dimension (affiliation or power; see Location similarity searchlight analyses for more details). At the group level, we controlled for sample and the average reaction time difference between affiliation and power trials. Using the same testing logic as the dimensionality similarity analysis, we first tested our hypothesis in the bilateral hippocampus and found widespread effects in both the left (peak voxel MNI x/y/z = -35/-22/-15, cluster extent = 1470 voxels) and right (peak voxel MNI x/y/z = 37/-19/-14, cluster extent = 1953 voxels) hemispheres. The whole-brain searchlight analysis revealed additional clusters in the left putamen (-27/-3/14, cluster extent = 131 voxels) and left posterior cingulate cortex (-10/-28/41, cluster extent = 304 voxels).”

“We then asked a second, complementary question: does the hippocampus represent all interactions, across characters, within a shared map? To test for this map-like structure, we repeated the analysis but now included all trial pairs, z-scoring distances globally rather than within character (Figure 3E-F). The remainder of the procedure followed the same logic as the preceding analysis. The hippocampus analysis revealed an extensive right hippocampal cluster (27/27/-14, cluster extent = 1667 voxels). The whole-brain analysis did not show any significant clusters.”

We also describe the results in the discussion (page 12, lines 220-226): 

“Then, we show that the hippocampus tracks the changing social locations (affiliation and power coordinates), above and beyond the effects of dimension or time; the hippocampus seemed to reflect both the changing within-character locations, tracking their locations over time, and locations across characters, as if in a shared map. Thus, these results suggest that the hippocampus does not just encode static character-related representations but rather tracks relationship changes in terms of underlying affiliation and power.”

The manuscript's description of the decoding analysis is unclear regarding the variability of the decoded positions. The authors appear to decode the position of a character along a spline, which raises the question of whether this position correlates with time, since characters are more likely to be located further from the center in later trials. There is a concern that the decoded position may not solely reflect the hippocampal encoding of spatial location, but could also be influenced by an inherent temporal association. Given that a character's position at time t is likely to be similar to its positions at t−1 and t+1, it is crucial that the authors clearly articulate their approach to separating spatial representation from temporal autocorrelation. While this issue may have been addressed in the construction of the test set, the manuscript does not seem to adequately explain how such biases were mitigated in the training set.

We agree that temporal confounding needs to be better accounted for, as our claims depend on space-like signals being separable from time-like ones. We address this in several ways in the revised manuscript.

First, we emphasize that this is a narrative-based task, where temporal structure is relevant. As such, our analyses aim to demonstrate that effects go beyond simple temporal confounds, like trial order or time elapsed.

Despite the temporal structure to the task, the decisions for the same character are spaced in time, and interleaved with other characters’ decisions, reducing the chance that a simple temporal confound could explain trajectory-related effects. We now describe the task better in the revised methods (page 16, lines 314-318):

“All six characters’ decision trials are interleaved with one another and with narrative slides. On average, after a decision trial for a given character, participants view ~11 narrative slides and complete ~3 decisions for other characters before returning to that same character, such that each character’s choices are separated by an average of ~20 seconds (range 12 seconds to 10 min).”

To address temporal autocorrelation in the fMRI time series, we used SPM’s FAST algorithm. Briefly, FAST models temporal autocorrelation as a weighted combination of candidate correlation functions, using the best estimate to remove autocorrelated signal.

We also now report the temporal autocorrelation profile of the hippocampal beta series in the supplement, including (pages 29-31, lines 593-656):

“The Social Navigation Task is a narrative-based task, where the relationships with characters evolve over time; trial pairs that are close in time may have more similar fMRI patterns for reasons unrelated to social mapping (e.g., slow drift). It is important to account for the role of time in our analyses, to ensure effects go beyond simple temporal confounds, like the time between decision trials. To aid in this, we quantified how fMRI signals change over time using a pattern autocorrelation function across decision trial lags. We defined the left and right hippocampus and the left and right intracalcarine cortex using the HarvardOxford atlas and thresholded them at 50% probability. We chose intracalcarine corex as an early visual control region that largely corresponds to primary visual cortex (V1), as it is likely to be driven by the visually presented narrative. We used the same trial-wise beta images as in the location similarity RSA (boxcar regressors spanning each decision trial’s reaction time). For each participant and region-of-interest (ROI), we extracted the decision trial-by-voxel beta matrix and quantified three kinds of temporal dependence: beta autocorrelation, multivoxel pattern correlation and multivoxel pattern correlation after regressing out temporal distance.”

“To estimate the temporal autocorrelation of the trial-wise beta values, we treated each voxel’s beta values as a time series across trials and measured how much a voxel’s response on one trial correlated (Pearson) with its response on previous trials. We averaged these voxel wise autocorrelations within each ROI. At one trial apart (lag 1), both the hippocampus and V1 showed small positive autocorrelations, indicating modest trial-to-trial carryover in response amplitude (see Supplemental figure 1) that by three trials apart was approximately 0.”

“Because our representational similarity analyses depend on trial-by-trial pattern similarity, we also estimated how multivoxel patterns were autocorrelated over time. For each lag, we computed the Pearson correlation between each trial’s voxelwise pattern and the pattern from the trial that many trials earlier, then averaged those correlations to obtain a single autocorrelation value for that lag. At one trial apart, both regions showed positive autocorrelation, with V1 having greater autocorrelation than the hippocampus; pattern correlations between trials 3 or 4 trials apart reduced across participants, settling into low but positive values. Then, for each participant and ROI, we regressed out the effect of absolute trial onset differences from all pairwise pattern correlations, to mirror the effects of controlling for these temporal distances in regressions. After removing this temporal distance component, the short lag pattern autocorrelation dropped substantially in both regions. The similarity in autocorrelation profiles between the two regions suggests that significant similarity effects in the hippocampus are unlikely to be driven by generic temporal autocorrelation.”

“Relationship between behavioral location distance and temporal distance “

“We also quantified how temporal distances between trials relates to their behavioral location distances, participant by participant. Our dimension similarity analysis controls for temporal distance between trials by design (see Social dimension similarity searchlight analysis), but our location similarity analysis does not. To decide on covariates to include in the analysis, we tested whether temporal distances can explain behavioral location distances. For each participant, we computed the correlations between trial pairs’ Euclidean distances in social locations and their linear temporal distances (“linear”) and the temporal distances squared (“quadratic”), to test for nonlinear effects. We then summarized the correlations using one-sample t-tests. The linear relationship was statistically significant (t₄₉ = 12.24, p < 0.001), whereas the quadratic relationship was not (t₄₉ = -0.55, p = 0.586). Similarly, in participant specific regressions with both linear and quadratic temporal distances, the linear effect was significant (t₄₉ = 5.69, p < 0.001) whereas the quadratic effect was not (t₄₉ = 0.20, p = 0.84). Based on this, we included linear temporal distances as a covariate in our location similarity analyses (see Location similarity searchlight analyses), and verified that adding a quadratic temporal distance covariate does not alter the results. Thus, the reported location-related pattern similarity effects go beyond what can be explained by temporal distance alone.”

How the free parameter of spectral clustering was determined, if there is any?

The interpretation of the number of hippocampal activity clusters is ambiguous. It is suggested that this number could fluctuate due to unique activity patterns or the fit to behaviorally defined trajectories. A lower number of clusters might indicate either a noisier or less distinct representation, raising the question of the necessity and interpretability of such a complex analysis. This concern is compounded by the potential sensitivity of the clustering to the variance in Euclidean distances of each trial's position relative to the center. If a character's position is consistently near the center, this could artificially reduce the perceived number of clusters. Furthermore, the manuscript should address whether there is any correlation between the number of clusters and behavioral performance. Specifically, what are the implications if participants are able to perform the task adequately with a smaller number of distinct hippocampal representation states?

The rationale for conducting both cluster analysis and position decoding as separate analyses remains unclear. While cluster analysis can corroborate the findings of position decoding, it is not apparent why the authors chose to include trials across characters for cluster analysis but not for decoding analysis. An explanation of the reasoning behind this methodological divergence would help in understanding the distinct contributions of each analysis to the study's findings.

The paper by Cohen et al. (1997), which provides the questionnaire for measuring the social network index, is not cited in the references. Upon reviewing the questionnaire that the author may have used, it appears that the term "social network size" does not refer to the actual size but to a score or index derived from the questionnaire responses. It may be more appropriate to replace the term "size" with a different term to more accurately reflect this distinction.

Thank you for seeking these clarifications. Given the complexity of this analysis, we have decided to drop it to focus instead on our dimension and location representational similarity analysis results.

Reviewer #2 (Recommendations for the authors):

How did the participants' decisions on previous trials influence the future trials that the subjects saw? If the different participants were faced with different decision trials, then how did you compare their decision? If two participants made the same decisions, would they have seen exactly the same sequence of trials (see point X on how the trial sequence was randomized).

All participants experience the same narrative, with the same decisions (i.e., the same available options); their choices (i.e., the options they select) are what implicitly shape each character’s affiliation and power locations, and thus each character’s trajectory. In other words, the narrative is fixed; what changes is the social coordinates assigned to each trial’s outcome depending on the participant’s choice of how to interact from the two narrative options. This means that we can meaningfully compare participants' neural patterns, given that every participant received the same text and images throughout.

We have now added details on the narrative structure, replacing more ambiguous statements with a clearer description (page 16, lines 309-318):

“The sequence of trials, including both narrative and decision trials, were fixed across participants; all that differs are the choices that the participants make. Narrative trials varied in duration, depending on the content (range 2-10 seconds), but were identical across participants. Decision trials always lasted 12 seconds, with two options presented until the participant made a choice, after which a blank screen was presented for the remainder of the duration. All six characters’ decision trials are interleaved with one another, and with the narrative slides. On average, after a decision trial for a given character, participants view ~11 narrative slides and complete ~3 decisions for other characters before returning to another decision with the same character, such that each character’s choices are separated by an average of ~20 seconds (ranging from 12 seconds to 10 min).”

Figure 2B: I assume that "count" is "count of participants"? It would be good to indicate this on the axis/caption.

Thank you for noting this. We have now removed this figure to improve the clarity of our figures. 

We have shown that the hippocampus represents the interaction decision trials along abstract social dimensions, but does it track each relationship's unique sequence of abstract social coordinates?". Please clarify what you mean by "represents the interaction decision trials”.

By “represents the interaction decision trials along abstract social dimensions”, we mean that when the participant makes a choice during the social interactions the hippocampal patterns represent the current social dimension of the choice (affiliation vs power). In other words, the hippocampal BOLD patterns differentiate affiliation and power decisions, consistent with our hypothesis of abstract social dimension representation in the hippocampus. We have clarified this (page 11, lines 185-187):

“We have shown that the left hippocampus represents the affiliation and power trials differently, consistent with an abstract dimensional representation.”

Page 8: "Hippocampal sequences are ordered like trajectories": It is not entirely clear to me what is meant by the split midpoint. Is this the midpoint of the piece-wise linear interpolation between two points, or simply the mean of all piecewise splines from one character? If the latter, is the null model the same as simply predicting the mean affiliation and power value for this character? If yes, please clarify and simplify this for the reader.

Page 8: "Hippocampal sequences track relationship-specific paths". First, I was misled by the "relationship-specific". I first understood this to mean that you wanted to test whether two relationships (i.e. the identity of the partner) had different representations in Hippocampus, even if the power/affiliation trajectories are the same. I suggest changing the title of this section.

The analysis in this section also breaks any temporal autocorrelation of measured patterns - so I am not sure if this is a strong analysis that should be interpreted at all. This analysis seems to not address the claim and conclusion that is drawn from it. I assume that the random trajectories have different choices and different affiliation/power values than the true trajectories. So the fact that the true trajectories can be better decoded simply shows that either choices or affiliation and power (or both) are represented in the neural code - but not necessarily anything beyond this.

Page 9: "Neural trajectories reflect social locations, not just choices". The motivation of this analysis is not clear to me. As I understand this analysis, both social location and choices are changed from the real trajectories. How can it then show that it reflects social locations, not just the choices?

Figure 4 caption: "on the -based approximation" Is there a missing "point"-[based] here?

We agree with the reviewer that this analysis is hard to interpret and does not adequately address concerns regarding temporal autocorrelation, and as such we have removed it from the manuscript. We describe the new results that include controlling for temporal distance between trials (pages 11-12, lines 185-208):

“We have shown that the left hippocampus represents the affiliation and power trials differently, consistent with an abstract dimensional representation. Does it also represent the changing social coordinates of each character? To test this, we multiple-regression RSA searchlight to test whether left hippocampus patterns represent the characters’ changing social locations across interactions (see Figure 3). We restricted the distances to those from trial pairs from the same character and standardized the distances within character (see Figure 3BD). We controlled for temporal distance to ensure the effect was not explainable by the time between trials, and for whether the trials shared the same underlying dimension (affiliation or power; see Location similarity searchlight analyses for more details). At the group level, we controlled for sample and the average reaction time difference between affiliation and power trials. Using the same testing logic as the dimensionality similarity analysis, we first tested our hypothesis in the bilateral hippocampus and found widespread effects in both the left (peak voxel MNI x/y/z = -35/-22/-15, cluster extent = 1470 voxels) and right (peak voxel MNI x/y/z = 37/-19/-14, cluster extent = 1953 voxels) hemispheres. The whole-brain searchlight analysis revealed additional clusters in the left putamen (-27/-3/14, cluster extent = 131 voxels) and left posterior cingulate cortex (-10/-28/41, cluster extent = 304 voxels).”

“We then asked a second, complementary question: does the hippocampus represent all interactions, across characters, within a shared map? To test for this map-like structure, we repeated the analysis but now included all trial pairs, z-scoring distances globally rather than within character (Figure 3E-F). The remainder of the procedure followed the same logic as the preceding analysis. The hippocampus analysis revealed an extensive right hippocampal cluster (27/27/-14, cluster extent = 1667 voxels). The whole-brain analysis did not show any significant clusters.”

We emphasize that the results are robust to the inclusion of temporal distance squared, in the methods (pages 23-24, lines 493-496):

“Although the square of this temporal distance did not explain any additional variance in behavioral distances, we ran a robustness analysis including both temporal distance and its square and saw qualitatively the same clusters with similar effect sizes.”

Page 8: last paragraph: The text sounds like you have already shown that you can decode character identity from the patterns - but I do not believe you have it this point. I would consider this would be an interesting addition to the paper, though.

This section has been removed, and we have been careful to not imply this in the current version of the manuscript. While we agree a character identity decoding would enrich our argument, we do not believe our task is well-suited to capture a character identity effect. Each character only has 12 decision trials, and these trials are partially clustered in time - this is one problem of temporal autocorrelation that we thank the reviewers for pushing us to consider in more detail. Dimension and location patterns, on the other hand, are more natural to analyze in our task, especially in representational similarity analyses that test whether the relevant differences scale with neural distances.

Page 14ff: Why is "Analysis section" not part of "Materials and Methods"? I believe adding the analysis after a careful description of the methods would improve the clarity of this section.

We agree with the reviewer and have now consolidated these two sections.

Two or three examples of Affiliation and Power decision trials should be provided, so the reader can form a more thorough understanding of how these dimensions were operationalized. For the RSA analysis, it is important to consider other differences between these two types of trials.

We agree that adding examples will clarify the operationalization of these dimensions. We now include example affiliation and power trials in a table (page 17-18).

We thank the reviewer for noting the need to rule out alternative hypotheses; we have added several such tests. Affiliation and power trials were not different in word count (page 17, lines 329-332):

“To ensure that any observed neural or behavioral differences were not confounded by trivial features of the text, we tested for differences between the affiliation and power trials (where the two options are concatenated). There were no differences in word count (affiliation average = 26.6, power average = 25.6; t-test p = 0.56).”

They were also not different in their sentiment, as assessed by a Large Language Model (LLM) analysis (page 17, lines 332-335): 

“The text’s sentiment also did not differ between these trial types (t-test p = 0.72), as quantified by comparing sentiment compound scores (from most negative, −1, to most positive, +1), using a Large Language Model (LLM) specialized for sentiment analysis [26]. “

The affiliation and power trials were different in terms of semantic content, consistent with our assumptions (page 17, lines 337-347):

“Our framework assumes that affiliation and power trials differ in their semantic content–that is, in the conceptual meaning of the text, beyond word count or sentiment. To test this assumption, we used an LLM-based semantic embedding analysis. Each decision trial was embedded into a semantic vector. We then measured the cosine similarity between pairs of trials and calculated the difference between average within-dimension similarity (affiliation-affiliation and power-power comparisons) and average between-dimension similarity (affiliationpower comparisons) and assessed its statistical significance with permutation testing (1,000 shuffles of trial labels). As expected, decision trials of the same dimension were more similar to each other than trials of different dimension, across multiple LLMs (OpenAI’s text-embedding-3-small [27]: similarity difference = 0.041, p < 0.001; all-MiniLM-L12-v2 [28]: similarity difference = 0.032, p < 0.001).”

The affiliation and power trials were different in average reaction time. To control for this difference in the dimension RSA analysis, we added each participant’s absolute value reaction time difference between the trial types as a covariate. The results were nearly identical to what they were before. We updated the text to reflect this new control (page 23, lines 471-474):

“However, there was a significant difference in the average reaction time between affiliation and power decisions across participants (t₄₉ = 6.92, p < 0.001; affiliation mean = 4.92 seconds (s), power mean = 4.51 s), so we controlled for this in the group-level analysis.”

The exact implementation and timing of the behavioral tasks should be described better. How many narrative trials were intermixed with the decision trials? Which characters were they assigned to? How was the sequence of trials determined? Was it fixed across participants, or randomized?

We agree that additional details are helpful. In the Methods, we now describe this with more detail (page 16, lines 301-318):

“There are two types of trials: “narrative” trials where background information is provided or characters talk or take actions (a total of 154 trials), and “decision” trials where the participant makes decisions in one-on-one interactions with a character that can change the relationship with that character (a total of 63 trials). On each decision, participants used a button response box to select between the two options. The options (1 or 2, assigned to the index and middle fingers) choice directions (+/-1 arbitrary unit on the current dimension) were counterbalanced.”

“The sequence of trials, including both narrative and decision trials, were fixed across participants; all that differs are the choices that the participants make. Narrative trials varied in duration, depending on the content (range 2-10 seconds), but were identical across participants. Decision trials always lasted 12 seconds, with two options presented until the participant made a choice, after which a blank screen was presented for the remainder of the duration. All six characters’ decision trials are interleaved with one another, and with the narrative slides. On average, after a decision trial for a given character, participants view ~11 narrative slides and complete ~3 decisions for other characters before returning to another decision with the same character, such that each character’s choices are separated by an average of ~20 seconds (ranging from 12 seconds to 10 min).”

What is the exact timing of trials during fMRI acquisition - i.e. how long were the trials, what was the ITI, were there long phases of rest to determine the resting baseline? These are all important factors that will determine the covariance between regressors and should be reported carefully. Ideally, I would like to see the trial-by-trial temporal auto-correlation structure across beta-weights to be reported.

We thank the reviewer for asking for this clarification. We have added the following text to clarify the trial timing (page 16, lines 314-318):

“All six characters’ decision trials are interleaved with one another and with narrative slides. On average, after a decision trial for a given character, participants view ~11 narrative slides and complete ~3 decisions for other characters before returning to that same character, such that each character’s choices are separated by an average of ~20 seconds (range 12 seconds to 10 min).”

We now describe the temporal autocorrelation patterns in the supplement, including how we decided on how to control for temporal distance in representational similarity analyses (pages 29-31, lines 593-656):

“The Social Navigation Task is a narrative-based task, where the relationships with characters evolve over time; trial pairs that are close in time may have more similar fMRI patterns for reasons unrelated to social mapping (e.g., slow drift). It is important to account for the role of time in our analyses, to ensure effects go beyond simple temporal confounds, like the time between decision trials. To aid in this, we quantified how fMRI signals change over time using a pattern autocorrelation function across decision trial lags. We defined the left and right hippocampus and the left and right intracalcarine cortex using the HarvardOxford atlas and thresholded them at 50% probability. We chose intracalcarine corex as an early visual control region that largely corresponds to primary visual cortex (V1), as it is likely to be driven by the visually presented narrative. We used the same trial-wise beta images as in the location similarity RSA (boxcar regressors spanning each decision trial’s reaction time). For each participant and region-of-interest (ROI), we extracted the decision trial-by-voxel beta matrix and quantified three kinds of temporal dependence: beta autocorrelation, multivoxel pattern correlation and multivoxel pattern correlation after regressing out temporal distance.”

“To estimate the temporal autocorrelation of the trial-wise beta values, we treated each voxel’s beta values as a time series across trials and measured how much a voxel’s response on one trial correlated (Pearson) with its response on previous trials. We averaged these voxel wise autocorrelations within each ROI. At one trial apart (lag 1), both the hippocampus and V1 showed small positive autocorrelations, indicating modest trial-to-trial carryover in response amplitude (see Supplemental figure 1) that by three trials apart was approximately 0.”

“Because our representational similarity analyses depend on trial-by-trial pattern similarity, we also estimated how multivoxel patterns were autocorrelated over time. For each lag, we computed the Pearson correlation between each trial’s voxelwise pattern and the pattern from the trial that many trials earlier, then averaged those correlations to obtain a single autocorrelation value for that lag. At one trial apart, both regions showed positive autocorrelation, with V1 having greater autocorrelation than the hippocampus; pattern correlations between trials 3 or 4 trials apart reduced across participants, settling into low but positive values. Then, for each participant and ROI, we regressed out the effect of absolute trial onset differences from all pairwise pattern correlations, to mirror the effects of controlling for these temporal distances in regressions. After removing this temporal distance component, the short lag pattern autocorrelation dropped substantially in both regions. The similarity in autocorrelation profiles between the two regions suggests that significant similarity effects in the hippocampus are unlikely to be driven by generic temporal autocorrelation.”

“Relationship between behavioral location distance and temporal distance “

“We also quantified how temporal distances between trials relates to their behavioral location distances, participant by participant. Our dimension similarity analysis controls for temporal distance between trials by design (see Social dimension similarity searchlight analysis), but our location similarity analysis does not. To decide on covariates to include in the analysis, we tested whether temporal distances can explain behavioral location distances. For each participant, we computed the correlations between trial pairs’ Euclidean distances in social locations and their linear temporal distances (“linear”) and the temporal distances squared (“quadratic”), to test for nonlinear effects. We then summarized the correlations using one-sample t-tests. The linear relationship was statistically significant (t₄₉ = 12.24, p < 0.001), whereas the quadratic relationship was not (t₄₉ = -0.55, p = 0.586). Similarly, in participant specific regressions with both linear and quadratic temporal distances, the linear effect was significant (t₄₉ = 5.69, p < 0.001) whereas the quadratic effect was not (t₄₉ = 0.20, p = 0.84). Based on this, we included linear temporal distances as a covariate in our location similarity analyses (see Location similarity searchlight analyses), and verified that adding a quadratic temporal distance covariate does not alter the results. Thus, the reported location-related pattern similarity effects go beyond what can be explained by temporal distance alone.”

eLife Assessment

This study presents a potentially valuable exploration of the role of thalamic nuclei in language processing. The results will be of interest to researchers interested in the neurobiology of language. However, the evidence is incomplete to support robust conclusions at this point.

Reviewer #1 (Public review):

Summary:

The manuscript by Mengxing et al., reports an assessment of three first-order thalamic nuclei (auditory, visual, somatosensory) in a 3 x 2 factorial design to test for specificity of responses in first-order thalamic nuclei to linguistic processing particularly in the left hemisphere. The conditions are reading, speech production, and speech comprehension and their respective control conditions. The authors report the following results:

(1) BOLD-response analyses: left MGB linguistic vs non-linguistic significant; left LGN linguistic vs non-linguistic significant. There is no hemisphere x stimulus interaction.

(2) MVPA: left MGB linguistic vs. non-linguistic significant; bilateral VLN linguistic vs. non-linguistic significant; significant lateralisation in MGB (left MGB responses better classified linguistic vs. non-linguistic in contrast to right).

(3) Functional connectivity: there is, in general, connectivity between the thalamic ROIs and the respective primary cortices independent of linguistics.

Strengths:

The study has a clear and comprehensive design and addresses a timely topic. First-order thalamic nuclei and their interaction with the respective cerebral cortex area are likely key to understanding how perception works in a world where one has to compute highly dynamic stimuli often in an instant. Speech is a prime example of an ecologically important, extremely dynamic, and complex stimulus. The field of the contribution of cerebral cortex-thalamic loops is wide open, and the study presents a solid approach to address their role in different speech modalities (i.e., reading, comprehension, production).

Weaknesses:

I see two major overall weaknesses in the manuscript in its current form:

(1) Statistics:

Unfortunately, I have doubts about the solidity of the statistics. In the analyses of the BOLD responses, the authors do not find significant hemisphere x stimulus interactions. In my view, such results would pre-empt doing a post-hoc t-test. Nevertheless, the authors motivate their post-hoc t-test by 'trends' in the interaction and prior hypotheses. I see two difficulties with that. First, the origin of the prior hypotheses is somewhat unclear (see also the comment below on hypotheses), and the post-hoc t-test is not corrected for multiple comparisons. I find that it is a pity that the authors did not derive more specific hypotheses grounded in the literature to guide the statistical testing, as I think these would have been available, and the response properties of the MGB and LGN also make sense in light of them. In addition, I was wondering whether the MVPA results would also need to be corrected for the three tests, i.e., the three ROIs.

Hypotheses:

In my view, it is relatively unclear where the hypotheses precisely come from. For example, the paragraph on the hypotheses in the introduction (p. 6-7) is devoid of references. I also have the impression that the hypotheses are partly not taking into account previous reports on first-order thalamic nuclei involvement in linguistic vs. non-linguistic processing. For example, the authors test for lateralisation of linguistic vs. non-linguistic responses in all nuclei. However, from previous literature, one could derive the hypothesis that the lateralisation in MGB for speech might be there - previous work shows, for example, that speech recognition abilities consistently correlate with left MGB only (von Kriegstein et al., 2008 Curr Biol; Mihai et al., 2019 eLife). In addition, the involvement of the MGB in speech in noise processing is present in the left MGB (Mihai et al., 2021, J Neuroscience). Developmental dyslexia, which is supposed to be based on imprecise phonological processing (Ramus et al., 2004 TiCS), has alterations in left MGB (Diaz et al., 2012 PNAS; Galaburda et al., 1994 PNAS) and left MGB connections to planum temporale (Tschentscher et al., 2019 J Neurosci) as well as altered lateralisation (Müller-Axt et al., 2025 Brain). Conversely, in the LGN, I'm not aware of any studies showing lateralisation for speech. See, for example, Diaz et al., 2018, Neuroimage, where there are correlations of LGN task-dependent modulation with visual speech recognition behaviour in both LGNs. Thus, based on this literature, one could have predicted the result pattern displayed, for example, in Figure 3A at least for MGB and LGN.

In summary, the motivation for the different hypotheses needs to be carved out more and couched into previous literature that is directly relevant to the topic. The above paragraph is, of course, my view on the topic, but currently, the paper lacks different literature as references to fully understand where the hypotheses are derived from.

Reviewer #2 (Public review):

Summary:

This study investigates the involvement of first-order thalamic nuclei in language-related tasks using task-based fMRI in a 3 × 2 design contrasting linguistic and non-linguistic versions of reading, speech comprehension, and speech production. By focusing on the LGN, MGN, and VLN and combining activation, connectivity, lateralization, and multivariate pattern analyses, the authors aim to characterize modality-specific and language-related thalamic contributions.

Strength:

A major strength of the work is its hypothesis-driven and multimodal analytical approach, and the modality-specific engagement of first-order thalamic nuclei is robust and consistent with known thalamocortical organization. This is a very sound study overall.

Weaknesses:

However, several conceptual issues complicate the interpretation of the results as evidence for linguistic modulation per se. A central concern relates to the operationalization of the linguistic versus non-linguistic contrast. In the present design, linguistic and non-linguistic stimuli differ along multiple dimensions beyond linguistic content. For example, written words and scrambled images differ in spatial frequency structure, edge composition, contrast regularities, and familiarity, while intelligible speech and acoustically scrambled sounds differ substantially in temporal and spectral statistics. This is particularly relevant given that first-order thalamic nuclei such as the LGN are known to be highly sensitive to low-level sensory properties. As a result, observed differences in thalamic responses may reflect sensitivity to stimulus properties rather than linguistic processing per se, and this limits the specificity of claims regarding linguistic modulation.

Relatedly, although the manuscript frequently refers to effects "depending on the linguistic nature of the stimuli," the statistical evidence for linguistic versus non-linguistic modulation is uneven across analyses. Whole-brain contrasts collapse across stimulus type and primarily test modality effects. Similarly, the primary ROI analyses of activation amplitude are collapsed across linguistic and non-linguistic conditions and convincingly demonstrate modality-specific engagement of thalamic nuclei, but do not in themselves provide evidence for linguistic modulation. Linguistic effects emerge only in later, more targeted analyses focusing on hemispheric lateralization and multivariate pattern classification, and these effects are nucleus-, modality-, and analysis-specific rather than general. Taken together, these results suggest that linguistic modulation constitutes a secondary and selective finding, whereas modality-specific task engagement represents the primary and most robust outcome of the study.

An additional interpretational issue concerns task engagement and attention. The tasks differ substantially in cognitive demands (e.g., passive reading and listening versus overt speech production), and linguistic and non-linguistic blocks may differ systematically in salience or engagement. This is particularly important given prior evidence, cited by the authors, that LGN and MGN activity can be modulated by task demands and attention. In the absence of behavioral measures indexing task engagement or compliance, it is difficult to determine whether differences between linguistic and non-linguistic conditions reflect linguistic processing per se or are mediated by attentional factors.

Finally, while the manuscript emphasizes the novelty of evaluating thalamic involvement in language, thalamic contributions to language have been documented previously in both lesion and functional imaging studies. The contribution of the present work, therefore, lies less in establishing thalamic involvement in language per se, and more in its focus on specific first-order nuclei, its multimodal design, and its combination of univariate, connectivity, and multivariate analyses. Moderating claims of novelty would help place the findings more clearly within the existing literature.

Author response:

We acknowledge the concerns raised by both reviewers and plan to address them in our revision:

Regarding Reviewer #1's comments: We will strengthen the statistical framework and address the concerns about multiple comparison corrections. We will also expand our literature review to better motivate our hypotheses, particularly incorporating the work on lateralization patterns in MGN/LGN and the existing evidence on first-order thalamic nuclei in linguistic processing.

Regarding Reviewer #2's comments: We acknowledge the valid concern that linguistic and non-linguistic stimuli differ beyond linguistic content, including some low-level sensory properties. We will elaborate on the creation and properties of these stimuli in the Methods section and upload stimuli examples to an online repository to provide transparency about differences. We will also add a discussion of this limitation in the Discussion section, acknowledging that disentangling effects of linguistic processing from low-level stimulus properties will require further testing in future research. Additionally, we will moderate part of our claims and reorganize the presentation of results as suggested, and clarify our contribution relative to existing literature.

eLife Assessment

This manuscript reports high-resolution cryo-EM structures of a trimethylamine N-oxide demethylase and advances the intriguing hypothesis that the enzyme is bifunctional, coupling TMAO demethylation to formaldehyde capture at a distal tetrahydrofolate-binding site via an enclosed intramolecular tunnel. Supported by biochemical assays and molecular dynamics simulations, the structural findings are valuable and potentially of broad interest, particularly the unusual oligomeric architecture and the proposed conduit for a reactive intermediate. However, the mechanistic framework is considered incomplete, raising substantial concerns regarding the proposed catalytic mechanism, metal/cofactor requirements, and the interpretation of biochemical data supporting formaldehyde channelling.

Reviewer #1 (Public review):

Summary:

Thach et al. report on the structure and function of trimethylamine N-oxide demethylase (TDM). They identify a novel complex assembly composed of multiple TDM monomers and obtain high-resolution structural information for the catalytic site, including an analysis of its metal composition, which leads them to propose a mechanism for the catalytic reaction.

In addition, the authors describe a novel substrate channel within the TDM complex that connects the N-terminal Zn²-dependent TMAO demethylation domain with the C-terminal tetrahydrofolate (THF)-binding domain. This continuous intramolecular tunnel appears highly optimized for shuttling formaldehyde (HCHO), based on its negative electrostatic properties and restricted width. The authors propose that this channel facilitates the safe transfer of HCHO, enabling its efficient conversion to methylenetetrahydrofolate (MTHF) at the C-terminal domain as a microbial detoxification strategy.

Strengths:

The authors provide convincing high-resolution cryo-EM structural evidence (up to 2 Å) revealing an intriguing complex composed of two full monomers and two half-domains. They further present evidence for the metal ion bound at the active site and articulate a plausible hypothesis for the catalytic cycle. Substantial effort is devoted to optimizing and characterizing enzyme activity, including detailed kinetic analyses across a range of pH values, temperatures, and substrate concentrations. Furthermore, the authors validate their structural insights through functional analysis of active-site point mutants.

In addition, the authors identify a continuous channel for formaldehyde (HCHO) passage within the structure and support this interpretation through molecular dynamics simulations. These analyses suggest an exciting mechanism of specific, dynamic, and gated channeling of HCHO. This finding is particularly appealing, as it implies the existence of a unique, completely enclosed conduit that may be of broad interest, including potential applications in bioengineering.

Weaknesses:

Although the idea of an enclosed channel for HCHO is compelling, the experimental evidence supporting enzymatic assistance in the reaction of HCHO with THF is less convincing. The linear regression analysis shown in Figure 1C demonstrates a THF concentration-dependent decrease in HCHO, but the concentrations used for THF greatly exceed its reported KD (enzyme concentration used in this assay is not reported). It has previously been shown that HCHO and THF can couple spontaneously in a non-enzymatic manner, raising the possibility that the observed effect does not require enzymatic channeling. An additional control that can rule out this possibility would help to strengthen the evidence. For example, mutating the THF binding site to prevent THF binding to the protein complex could clarify whether the observed decrease in HCHO depends on enzyme-mediated proximity effects. A mutation which would specifically disable channeling could be even more convincing (maybe at the narrowest bottleneck).

Another concern is that the observed decrease in HCHO could alternatively arise from a reduced production of HCHO due to a negative allosteric effect of THF binding on the active site. From this perspective, the interpretation would be more convincing if a clear coupled effect could be demonstrated, specifically, that removal of the product (HCHO) from the reaction equilibrium leads to an increase in the catalytic efficiency of the demethylation reaction.

While the enzyme kinetics appear to have been performed thoroughly, the description of the kinetic assays in the Methods section is very brief. Important details such as reaction buffer composition, cofactor identity and concentration (Zn²⁺), enzyme concentration, defined temperature, and precise pH are not clearly stated. Moreover, a detailed methodological description could not be found in the cited reference (6), if I am not mistaken.

The composition of the complex is intriguing but raises some questions. Based on SDS-PAGE analysis, the purified protein appears to be predominantly full-length TDM, and size-exclusion chromatography suggests an apparent molecular weight below 100 kDa. However, the cryo-EM structure reveals a substantially larger complex composed of two full-length monomers and two half-domains.

Given the lack of clear evidence for proteolytic fragments on the SDS-PAGE gel, it is unclear how the observed stoichiometry arises. This raises the possibility of higher-order assemblies or alternative oligomeric states. Did the authors attempt to pick or analyze larger particles during cryo-EM processing? Additional biophysical characterization of particle size distribution - for example, using interferometric scattering microscopy (iSCAT)-could help clarify the oligomeric state of the complex in solution.

The authors mention strict symmetry in the complex, yet C2 symmetry was enforced during refinement. While this is reasonable as an initial approach, it would strengthen the structural interpretation to relax the symmetry to C1 using the C2-refined map as a reference. This could reveal subtle asymmetries or domain-specific differences without sacrificing the overall quality of the reconstruction.

In this context, the proposed catalytic role of Zn²⁺ raises additional questions. Why is a 2:1 enzyme-to-metal stoichiometry observed, and how does this reconcile with previous reports? This point warrants discussion. Does this imply asymmetric catalysis within the complex? Would the stoichiometry change under Zn²⁺-saturating conditions, as no Zn²⁺ appears to be added to the buffers? It would be helpful to clarify whether Zn²⁺ occupancy is equivalent in both active sites when symmetry is not imposed, or whether partial occupancy is observed.

The divalent ion Zn2+ is suggested to activate water for the catalytic reaction. I am not sure if there is a need for a water molecule to explain this catalytic mechanism. Can you please elaborate on this more? As one aspect, it might be helpful to explain in more detail how Zn-OH and D220 are recovered in the last step before a new water molecule comes in.

Overall, the authors were successful in advancing our structural and functional understanding of the TDM complex. They suggest an interesting oligomeric complex composition which should be investigated with additional biophysical techniques.

Additionally, they provide an intriguing hypothesis for a new type of substrate channeling. Additional kinetic experiments focusing on HCHO and THF turnover by enzymatic proximity effects would strengthen this potentially fundamental finding. If this channeling mechanism can be supported by stronger experimental evidence, it would substantially advance our understanding and knowledge of biologic conduits and enable future efforts in the design of artificial cascade catalysis systems with high conversion rate and efficiency, as well as detoxification pathways.

Reviewer #2 (Public review):

Summary:

The manuscript reports a cryo-EM structure of TMAO demethylase from Paracoccus sp. This is an important enzyme in the metabolism of trimethylamine oxide (TMAO) and trimethylamine (TMA) in human gut microbiota, so new information about this enzyme would certainly be of interest.

Strengths:

The cryo-EM structure for this enzyme is new and provides new insights into the function of the different protein domains, and a channel for formaldehyde between the two domains.

Weaknesses:

(1) The proposed catalytic mechanism in this manuscript does not make sense. Previous mechanistic studies on the Methylocella silvestris TMAO demethylase (FEBS Journal 2016, 283, 3979-3993, reference 7) reported that, as well as a Zn2+ cofactor, there was a dependence upon non-heme Fe2+, and proposed a catalytic mechanism involving deoxygenation to form TMA and an iron(IV)-oxo species, followed by oxidative demethylation to form DMA and formaldehyde.

In this work, the authors do not mention the previously proposed mechanism, but instead say that elemental analysis "excluded iron". This is alarming, since the previous work has a key role for non-heme iron in the mechanism. The elemental analysis here gives a Zn content of about 0.5 mol/mol protein (and no Fe), whereas the Methylocella TMAO demethylase was reported to contain 0.97 mol Zn/mol protein, and 0.35-0.38 mol Fe/mol protein. It does, therefore, appear that their enzyme is depleted in Zn, and the absence of Fe impacts the mechanism, as explained below.

The proposed catalytic mechanism in this manuscript, I am sorry to say, does not make sense to me, for several reasons:

(i) Demethylation to form formaldehyde is not a hydrolytic process; it is an oxidative process (normally accomplished by either cytochrome P450 or non-heme iron-dependent oxygenase). The authors propose that a zinc (II) hydroxide attacks the methyl group, which is unprecedented, and even if it were possible, would generate methanol, not formaldehyde.

(ii) The amine oxide is then proposed to deoxygenate, with hydroxide appearing on the Zn - unfortunately, amine oxide deoxygenation is a reductive process, for which a reducing agent is needed, and Zn2+ is not a redox-active metal ion;

(iii) The authors say "forming a tetrahedral intermediate, as described for metalloproteinase", but zinc metalloproteases attack an amide carbonyl to form an oxyanion intermediate, whereas in this mechanism, there is no carbonyl to attack, so this statement is just wrong.

So on several counts, the proposed mechanism cannot be correct. Some redox cofactor is needed in order to carry out amine oxide deoxygenation, and Zn2+ cannot fulfil that role. Fe2+ could do, which is why the previously proposed mechanism involving an iron(IV)-oxo intermediate is feasible. But the authors claim that their enzyme has no Fe. If so, then there must be some other redox cofactor present. Therefore, the authors need to re-analyse their enzyme carefully and look either for Fe or for some other redox-active metal ion, and then provide convincing experimental evidence for a feasible catalytic mechanism. As it stands, the proposed catalytic mechanism is unacceptable.

(2) Given the metal content reported here, it is important to be able to compare the specific activity of the enzyme reported here with earlier preparations. The authors do quote a Vmax of 16.52 µM/min/mg; however, these are incorrect units for Vmax, they should be µmol/min/mg. There is a further inconsistency between the text saying µM/min/mg and the Figure saying µM/min/µg.

(3) The consumption of formaldehyde to form methylene-THF is potentially interesting, but the authors say "HCHO levels decreased in the presence of THF", which could potentially be due to enzyme inhibition by THF. Is there evidence that this is a time-dependent and protein-dependent reaction? Also in Figure 1C, HCHO reduction (%) is not very helpful, because we don't know what concentration of formaldehyde is formed under these conditions; it would be better to quote in units of concentration, rather than %.

(4) Has this particular TMAO demethylase been reported before? It's not clear which Paracoccus strain the enzyme is from; the Experimental Section just says "Paracoccus sp.", which is not very precise. There has been published work on the Paracoccus PS1 enzyme; is that the strain used? Details about the strain are needed, and the accession for the protein sequence.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

Thach et al. report on the structure and function of trimethylamine N-oxide demethylase (TDM). They identify a novel complex assembly composed of multiple TDM monomers and obtain high-resolution structural information for the catalytic site, including an analysis of its metal composition, which leads them to propose a mechanism for the catalytic reaction.

In addition, the authors describe a novel substrate channel within the TDM complex that connects the N-terminal Zn²-dependent TMAO demethylation domain with the C-terminal tetrahydrofolate (THF)-binding domain. This continuous intramolecular tunnel appears highly optimized for shuttling formaldehyde (HCHO), based on its negative electrostatic properties and restricted width. The authors propose that this channel facilitates the safe transfer of HCHO, enabling its efficient conversion to methylenetetrahydrofolate (MTHF) at the C-terminal domain as a microbial detoxification strategy.

Strengths:

The authors provide convincing high-resolution cryo-EM structural evidence (up to 2 Å) revealing an intriguing complex composed of two full monomers and two half-domains. They further present evidence for the metal ion bound at the active site and articulate a plausible hypothesis for the catalytic cycle. Substantial effort is devoted to optimizing and characterizing enzyme activity, including detailed kinetic analyses across a range of pH values, temperatures, and substrate concentrations. Furthermore, the authors validate their structural insights through functional analysis of active-site point mutants.

In addition, the authors identify a continuous channel for formaldehyde (HCHO) passage within the structure and support this interpretation through molecular dynamics simulations. These analyses suggest an exciting mechanism of specific, dynamic, and gated channeling of HCHO. This finding is particularly appealing, as it implies the existence of a unique, completely enclosed conduit that may be of broad interest, including potential applications in bioengineering.

Weaknesses:

Although the idea of an enclosed channel for HCHO is compelling, the experimental evidence supporting enzymatic assistance in the reaction of HCHO with THF is less convincing. The linear regression analysis shown in Figure 1C demonstrates a THF concentration-dependent decrease in HCHO, but the concentrations used for THF greatly exceed its reported KD (enzyme concentration used in this assay is not reported). It has previously been shown that HCHO and THF can couple spontaneously in a non-enzymatic manner, raising the possibility that the observed effect does not require enzymatic channeling. An additional control that can rule out this possibility would help to strengthen the evidence. For example, mutating the THF binding site to prevent THF binding to the protein complex could clarify whether the observed decrease in HCHO depends on enzyme-mediated proximity effects. A mutation which would specifically disable channeling could be even more convincing (maybe at the narrowest bottleneck).

We agree with the reviewer that HCHO and THF can react spontaneously in a non-enzymatic manner, and our experiments were not intended to demonstrate enzymatic channeling. The linear regression analysis in Figure 1C was designed solely to confirm that HCHO reacts with THF under our assay conditions. Accordingly, THF was titrated over a broad concentration range starting from zero, and the observed THF concentration–dependent decrease in HCHO reflects this chemical reactivity.

We do not interpret these data as evidence that the enzyme catalyzes or is required for the HCHO–THF coupling reaction. Instead, the structural observation of an enclosed channel is presented as a separate finding. We have clarified this point in the revised text to avoid overinterpretation of the biochemical data (page 2, line 16).

Another concern is that the observed decrease in HCHO could alternatively arise from a reduced production of HCHO due to a negative allosteric effect of THF binding on the active site. From this perspective, the interpretation would be more convincing if a clear coupled effect could be demonstrated, specifically, that removal of the product (HCHO) from the reaction equilibrium leads to an increase in the catalytic efficiency of the demethylation reaction.

We agree that, in principle, a decrease in detectable HCHO could also arise from an indirect effect of THF binding on enzyme activity. However, in our study the experiment was not designed to assess catalytic coupling or allosteric regulation. The assay in question monitors HCHO levels under defined conditions and does not distinguish between changes in HCHO production and downstream consumption.

Additionally, we do not interpret the observed decrease in HCHO as evidence that THF binding enhances catalytic efficiency, or that removal of HCHO shifts the reaction equilibrium. Instead, the data are presented to establish that HCHO can react with THF under the assay conditions. Any potential allosteric effects of THF on the demethylation reaction, or kinetic coupling between HCHO removal and catalysis, are beyond the scope of the current study, and are not claimed.

While the enzyme kinetics appear to have been performed thoroughly, the description of the kinetic assays in the Methods section is very brief. Important details such as reaction buffer composition, cofactor identity and concentration (Zn²⁺), enzyme concentration, defined temperature, and precise pH are not clearly stated. Moreover, a detailed methodological description could not be found in the cited reference (6), if I am not mistaken.

Thank you for the suggestion. We have added reference [24] to the methodological description on page 8. The Methods section has been revised accordingly on page 8 under “TDM Activity Assay,” without altering the Zn²⁺ concentration.

The composition of the complex is intriguing but raises some questions. Based on SDS-PAGE analysis, the purified protein appears to be predominantly full-length TDM, and size-exclusion chromatography suggests an apparent molecular weight below 100 kDa. However, the cryo-EM structure reveals a substantially larger complex composed of two full-length monomers and two half-domains.

We appreciate the reviewer’s careful analysis of the apparent discrepancy between the biochemical characterization and the cryo-EM structure. This issue is addressed in Figure S1, which may have been overlooked.

As shown in Figure S1, the stability of TDM is highly dependent on protein and salt conditions. At 150 mM NaCl, SEC reveals a dominant peak eluting between 10.5 and 12 mL, corresponding to an estimated molecular weight of ~170–305 kDa (blue dot, Author response image 1). This fraction was explicitly selected for cryo-EM analysis and yields the larger complex observed in the reconstruction. At lower salt concentrations (50 mM) or higher (>150 mM NaCl), the protein either aggregates or elutes near the void volume (~8 mL).

SDS–PAGE analysis detects full-length TDM together with smaller fragments (~40–50 kDa and ~22–25 kDa). The apparent predominance of full-length protein on SDS–PAGE likely reflects its greater staining intensity per molecule and/or a higher population, rather than the absence of truncated species.

Author response image 1.

Given the lack of clear evidence for proteolytic fragments on the SDS-PAGE gel, it is unclear how the observed stoichiometry arises. This raises the possibility of higher-order assemblies or alternative oligomeric states. Did the authors attempt to pick or analyze larger particles during cryo-EM processing? Additional biophysical characterization of particle size distribution - for example, using interferometric scattering microscopy (iSCAT)-could help clarify the oligomeric state of the complex in solution.

Cryo-EM data were collected exclusively from the size-exclusion chromatography fraction eluting between 10.5 and 12 mL. This fraction was selected to isolate the dominant assembly in solution. Extensive 2D and 3D particle classification did not reveal distinct classes corresponding to smaller species or higher-order oligomeric assemblies. Instead, the vast majority of particles converged to a single, well-defined structure consistent with the 2 full-length + 2 half-domain stoichiometry.

A minor subpopulation (~2%) exhibited increased flexibility in the N-terminal region of the two full-length subunits, but these particles did not form a separate oligomeric class, indicating conformational heterogeneity rather than alternative assembly states (Author response image 2). Together, these data support the 2+2½ architecture as the predominant and stable complex under the conditions used for cryo-EM. Additional techniques, such as iSCAT, would provide complementary information, but are not required to support the conclusions drawn from the SEC and cryo-EM analyses presented here.

Author response image 2.

The authors mention strict symmetry in the complex, yet C2 symmetry was enforced during refinement. While this is reasonable as an initial approach, it would strengthen the structural interpretation to relax the symmetry to C1 using the C2-refined map as a reference. This could reveal subtle asymmetries or domain-specific differences without sacrificing the overall quality of the reconstruction.

We thank the reviewer for this thoughtful suggestion. In standard cryo-EM data processing, symmetry is typically not imposed initially to minimize potential model bias; accordingly, we first performed C1 refinement before applying C2 symmetry. The resulting C1 reconstructions revealed no detectable asymmetry or domain-specific differences relative to the C2 map. In addition, relaxing the symmetry consistently reduced overall resolution, indicating lower alignment accuracy and further supporting the presence of a predominantly symmetric assembly.

In this context, the proposed catalytic role of Zn²⁺ raises additional questions. Why is a 2:1 enzyme-to-metal stoichiometry observed, and how does this reconcile with previous reports? This point warrants discussion. Does this imply asymmetric catalysis within the complex? Would the stoichiometry change under Zn²⁺-saturating conditions, as no Zn²⁺ appears to be added to the buffers? It would be helpful to clarify whether Zn²⁺ occupancy is equivalent in both active sites when symmetry is not imposed, or whether partial occupancy is observed.

The observed ~2:1 enzyme-to-Zn²⁺ stoichiometry likely reflects the composition of the 2 full-length + 2 half-domain (2+2½) complex. In this assembly, only the core domains that are fully present in the complex contribute to metal binding. The truncated or half-domains lack the Zn²⁺ binding domain. As a result, only two metal-binding sites are occupied per assembled complex, consistent with the measured stoichiometry.

We note that Zn²⁺ was not deliberately added to the buffers, so occupancy may not reflect full saturation. Based on our cryo-EM and biochemical data, both metal-binding sites in the full-length subunits appear to be occupied to an equivalent extent, and no clear evidence of asymmetric catalysis is observed under these current experimental conditions. Full Zn²⁺ saturation could potentially increase occupancy, but was not explored in these experiments.

The divalent ion Zn²⁺ is suggested to activate water for the catalytic reaction. I am not sure if there is a need for a water molecule to explain this catalytic mechanism. Can you please elaborate on this more? As one aspect, it might be helpful to explain in more detail how Zn-OH and D220 are recovered in the last step before a new water molecule comes in.

Thank you for your suggestion. We revised our text in page 2 as bellow.

Based on our structural and biochemical data, we propose a structurally informed working model for TMAO turnover by TDM (Scheme 1). In this model, Zn²⁺ plays a non-redox role by polarizing the O–H bond of the bound hydroxyl, thereby lowering its pK_a. The D220 carboxylate functions as a general base, abstracting the proton to generate a hydroxide nucleophile. This hydroxide then attacks the electrophilic N-methyl carbon of TMAO, forming a tetrahedral carbinolamine (hemiaminal) intermediate. Subsequent heterolytic cleavage of the C–N bond leads to the release of HCHO. D220 then switches roles to act as a general acid, donating a proton to the departing nitrogen, which facilitates product release and regenerates the active site. This sequence allows a new water molecule to rebind Zn²⁺, enabling subsequent catalytic turnovers. This proposed pathway is consistent with prior mechanistic studies, in which water addition to the azomethine carbon of a cationic Schiff base generates a carbinolamine intermediate, followed by a rate-limiting breakdown to yield an amino alcohol and a carbonyl compound, in the published case, an aldehyde (Pihlaja et al., J. Chem. Soc. Perkin Trans. 2, 1983, 8, 1223–1226).

Overall, the authors were successful in advancing our structural and functional understanding of the TDM complex. They suggest an interesting oligomeric complex composition which should be investigated with additional biophysical techniques.

Additionally, they provide an intriguing hypothesis for a new type of substrate channeling. Additional kinetic experiments focusing on HCHO and THF turnover by enzymatic proximity effects would strengthen this potentially fundamental finding. If this channeling mechanism can be supported by stronger experimental evidence, it would substantially advance our understanding and knowledge of biologic conduits and enable future efforts in the design of artificial cascade catalysis systems with high conversion rate and efficiency, as well as detoxification pathways.

Reviewer #2 (Public review):

Summary:

The manuscript reports a cryo-EM structure of TMAO demethylase from Paracoccus sp. This is an important enzyme in the metabolism of trimethylamine oxide (TMAO) and trimethylamine (TMA) in human gut microbiota, so new information about this enzyme would certainly be of interest.

Strengths:

The cryo-EM structure for this enzyme is new and provides new insights into the function of the different protein domains, and a channel for formaldehyde between the two domains.

Weaknesses:

(1) The proposed catalytic mechanism in this manuscript does not make sense. Previous mechanistic studies on the Methylocella silvestris TMAO demethylase (FEBS Journal 2016, 283, 3979-3993, reference 7) reported that, as well as a Zn2+ cofactor, there was a dependence upon non-heme Fe²⁺, and proposed a catalytic mechanism involving deoxygenation to form TMA and an iron(IV)-oxo species, followed by oxidative demethylation to form DMA and formaldehyde.

In this work, the authors do not mention the previously proposed mechanism, but instead say that elemental analysis "excluded iron". This is alarming, since the previous work has a key role for non-heme iron in the mechanism. The elemental analysis here gives a Zn content of about 0.5 mol/mol protein (and no Fe), whereas the Methylocella TMAO demethylase was reported to contain 0.97 mol Zn/mol protein, and 0.35-0.38 mol Fe/mol protein. It does, therefore, appear that their enzyme is depleted in Zn, and the absence of Fe impacts the mechanism, as explained below.

The proposed catalytic mechanism in this manuscript, I am sorry to say, does not make sense to me, for several reasons:

(i) Demethylation to form formaldehyde is not a hydrolytic process; it is an oxidative process (normally accomplished by either cytochrome P450 or non-heme iron-dependent oxygenase). The authors propose that a zinc (II) hydroxide attacks the methyl group, which is unprecedented, and even if it were possible, would generate methanol, not formaldehyde.

(ii) The amine oxide is then proposed to deoxygenate, with hydroxide appearing on the Zn - unfortunately, amine oxide deoxygenation is a reductive process, for which a reducing agent is needed, and Zn2+ is not a redox-active metal ion;

(iii) The authors say "forming a tetrahedral intermediate, as described for metalloproteinase", but zinc metalloproteases attack an amide carbonyl to form an oxyanion intermediate, whereas in this mechanism, there is no carbonyl to attack, so this statement is just wrong.

So on several counts, the proposed mechanism cannot be correct. Some redox cofactor is needed in order to carry out amine oxide deoxygenation, and Zn²⁺cannot fulfil that role. Fe²⁺ could do, which is why the previously proposed mechanism involving an iron(IV)-oxo intermediate is feasible. But the authors claim that their enzyme has no Fe. If so, then there must be some other redox cofactor present. Therefore, the authors need to re-analyse their enzyme carefully and look either for Fe or for some other redox-active metal ion, and then provide convincing experimental evidence for a feasible catalytic mechanism. As it stands, the proposed catalytic mechanism is unacceptable.

We thank the reviewer for the detailed and thoughtful mechanistic critique. We fully agree that Zn²⁺ is not redox-active, and cannot directly mediate oxidative demethylation or amine oxide deoxygenation. We acknowledge that the oxidative step required for the conversion of TMAO to HCHO is not explicitly resolved in the present study. Accordingly, we have revised the manuscript to remove any implication of Zn²⁺-mediated redox chemistry, and have eliminated the previously imprecise analogy to zinc metalloproteases.

We recognize and now discuss prior biochemical work on TMAO demethylase from Methylocella silvestris (MsTDM), which proposed an iron-dependent oxidative mechanism (Zhu et al., FEBS 2016, 3979–3993). That study reported approximately one Zn²⁺ and one non-heme Fe²⁺ per active enzyme, implicated iron in catalysis through homology modeling and mutagenesis, and used crossover experiments suggesting a trimethylamine-like intermediate and oxygen transfer from TMAO, consistent with an Fe-dependent redox process. However, that system lacked experimental structural information, and did not define discrete metal-binding sites.

In contrast,

(1) Our high-resolution cryo-EM structures and metal analyses of TDM consistently reveal only a single, well-defined Zn²⁺-binding site, with no structural evidence for an additional iron-binding site as in the previous report (Zhu et al., FEBS 2016, 3979–3993).

(2) To investigate the potential involvement of iron, we expressed TDM in LB medium supplemented with Fe(NH₄)₂SO₄ and determined its cryo-EM structure. This structure is identical to the original one, and no EM density corresponding to a second iron ion was observed. Moreover, the previously proposed Fe²⁺-binding residues are spatially distant (Figure S6).

(3) ICP-MS analysis shows undetectable Iron, and only Zinc ion (Figure S5).

(4) Our enzyme kinetics analysis with the TDM without Iron is comparable to that of from MsTDM (Figure 1A). The differences in Km and Vmax we propose is due to the difference in the overall sequence of the enzymes. Please also see comment at the end on a new published paper on MsTDM.

While we cannot comment on the MsTDM results, our ‘experimental’ results do not support the presence of an iron-binding site. Our data indicate that this chemistry is unlikely to be mediated by a canonical non-heme iron center as proposed for MsTDM. We therefore revised our model as a structural framework that rationalizes substrate binding, metal coordination, and product stabilization, while clearly delineating the limits of mechanistic inference supported by the current data.

The scheme 1 and proposal mechanism section were revised in page 4. Figure S6 was added.

(2) Given the metal content reported here, it is important to be able to compare the specific activity of the enzyme reported here with earlier preparations. The authors do quote a Vmax of 16.52 µM/min/mg; however, these are incorrect units for Vmax, they should be µmol/min/mg. There is a further inconsistency between the text saying µM/min/mg and the Figure saying µM/min/µg.

Thank you for the correction. We converted the V_max unit to nmol/min/mg. and revised the text in page 2. We also compared with the value of the previous report in the TDM enzyme by revising the text on page 2. See also the note on a newly published manuscript and its comparison.

(3) The consumption of formaldehyde to form methylene-THF is potentially interesting, but the authors say "HCHO levels decreased in the presence of THF", which could potentially be due to enzyme inhibition by THF. Is there evidence that this is a time-dependent and protein-dependent reaction? Also in Figure 1C, HCHO reduction (%) is not very helpful, because we don't know what concentration of formaldehyde is formed under these conditions; it would be better to quote in units of concentration, rather than %.

We appreciate this important point. We have revised Figure 1C to present HCHO levels in absolute concentration units. While the current data demonstrate reduced detectable HCHO in the presence of THF, we agree that distinguishing between HCHO consumption and potential THF-mediated enzyme inhibition would require dedicated time-course and protein-dependence experiments. We have therefore revised the description to avoid overinterpretation and limit our conclusions to the observed changes in HCHO concentration in page 2, line 18-19.

(4) Has this particular TMAO demethylase been reported before? It's not clear which Paracoccus strain the enzyme is from; the Experimental Section just says "Paracoccus sp.", which is not very precise. There has been published work on the Paracoccus PS1 enzyme; is that the strain used? Details about the strain are needed, and the accession for the protein sequence.

Thank you for this comment. We now indicate that the enzyme is derived from Paracoccus sp. DMF and provide the accession number for the protein sequence (WP_263566861) in the Experimental Section (page 8, line 4).

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

(1) The ITC experiment requires a ligand-into-buffer titration as an additional control. Also, maybe I misunderstood the molar ratio or the concentrations you used, but if you indeed added a total of 4.75 μL of 20 μM THF into 250 μL of 5 μM TDM, it is not clear to me how this leads to a final molar ratio of 3.

We thank the reviewer for this suggestion. A ligand-into-buffer control ITC experiment was performed and is now included in Figure S8C, which shows no realizable signal.

Regarding the molar ratio, it is our mistake. The experiment used 2.45 μL injections of 80 μM THF into 250 μL of 5 μM TDM. This corresponds to a final ligand concentration of ~12.8 μM, giving a ligand-to-protein molar ratio of ~2.6. We revised our text in page 9, ITC section.

(2) Characterization/quality check of all mutant enzymes should be performed by NanoDSF, CD spectroscopy or similar techniques to confirm that proteins are properly folded and fit for kinetic testing.

We appreciate the reviewer’s suggestion. All mutant proteins, including D220A, D367A, and F327A, were purified with yields similar to the wild-type enzyme. Additionally, cryo-EM maps of the mutants show well-defined density and overall structural integrity consistent with the wild-type. These findings indicate that the introduced mutations do not significantly affect protein folding, supporting their use for kinetic analysis. While NanoDSF might reveal differences in thermal stability due to mutations, it does not provide structural information. Our conclusions are not based on minor differences in thermostability. Our cryo-EM structures of the mutants offer much more reliable structural data than CD spectroscopy.

(3) Best practice would suggest overlapping pH ranges between different buffer systems in the pH-dependence experiments to rule out buffer-specific effects independent of pH.

We thank the reviewer for this helpful suggestion. We agree that overlapping pH ranges between different buffer systems can be valuable for excluding buffer-specific effects. In this study, the pH-dependence experiments were intended to provide a qualitative assessment of pH sensitivity rather than a detailed analysis of buffer-independent pKa values. While we cannot fully exclude minor buffer-specific contributions, the overall trends observed were reproducible and sufficient to support the conclusions drawn. We have added a clarifying statement to the revised manuscript to reflect this consideration, page 2, line 12.

(4) Structural comparison revealed high similarity to a THF-binding protein, with superposition onto a T protein.": It would be nice to show this as an additional figure, as resolution and occupancy for THF are low.

We thank the reviewer for this suggestion. To address this point, we have revised Figure S6 by adding an additional panel (C, now is Figure S7C) showing the structural superposition of TDM with the THF-binding T protein. This comparison is included to better illustrate the structural similarity, despite the limited resolution and partial occupancy of THF density in our map.

(5) Editing could have been done more thoroughly. Some spelling mistakes, e.g. "RESEULTS", "redius", "complec"; kinetic rate constants should be written in italic (not uniform between text and figures); Prism version is missing; Vmax of 16.52 µM/min/mg - doublecheck units; Figure S1B: The "arrow on the right" might have gone missing.

We corrected the spelling in page 2 ~ line 10, page 5 ~ line 34, page 6 ~ line40. Prism version was added. The arrow was added into figure S1B. The Vmax unit is corrected to nmol/min/mg.

Reviewer #2 (Recommendations for the authors):

(1) The authors must re-examine the metal content of their purified enzyme, looking in particular for Fe or another redox-active metal ion, which could be involved in a reasonable catalytic mechanism.

We thank the reviewer for this suggestion and have carefully re-examined the metal content of TDM. Elemental analyses by EDX and ICP-MS consistently detected Zn²⁺ in purified TDM (Zn:protein ≈ 1:2), whereas Fe was below the detection limit across multiple independent preparations (Fig. S5A,B). To assess whether iron could be incorporated or play a functional role, we expressed TDM in E. coli grown in LB medium supplemented with Fe(NH₄SO₄)₂ and performed activity assays in the presence of exogenous Fe²⁺. Neither condition resulted in enhanced enzymatic activity.

Consistent with these biochemical data, all cryo-EM structures reveal a single, well-defined metal-binding site coordinated by three conserved cysteine residues and occupied by Zn²⁺, with no evidence for an additional iron species or other redox-active metal site.

(2) The specific activity of the enzyme should be quoted in the same units as other literature papers, so that the enzyme activity can be compared. It could be, for example, that the content of Fe (or other redox-active metal) is low, and that could then give rise to a low specific activity.

Thank you for the suggestion, we quoted the enzyme units as similar with previous report. and revised the text in in page 2.

Since the submission of our paper a new report on MsTDM has been published (Cappa et al., Protein Science 33(11), e70364). It further supports our findings. First, the reported kinetic parameters using ITC (Vmax = 0.309 μmol/s, approximately 240 nmol/min/mg; Km = 0.866 mM) are comparable to our observed (156 nmol/min/mg and 1.33 mM, respectively) in the absence of exogenous iron. Second, the optimal pH for enzymatic activity similar to that observed in our paraTDM. Third, the reported two-state unfolding behavior is consistent with our cryo-EM structural observations, in which the more dynamic subunits appear to destabilize prior to unfolding of the core domains. Based on these findings, we now propose that Zn²⁺ appears to function primarily as an organizational cofactor at the core catalytic domain (revised Scheme 1).

eLife Assessment

This study provides a useful contribution to understanding how wearable augmentation devices interact with human proprioception, using a longitudinal design over a single session. Results demonstrate that the perceptual representation of the biological finger and augmentation device changes across different phases of device exposure and use. The evidence supporting a representational change over time is solid, although it is still not clear whether these changes reflect three distinct phases of sensorimotor plasticity, as argued, versus 'washout' or adaptation effects. This work will be of interest to researchers studying body representation, sensorimotor learning, and human-technology interaction.

Reviewer #1 (Public review):

This study by Radziun and colleagues investigates the effects of using a hand-augmentation device on mental body representations. The authors use a proprioceptive localisation task to measure metric representations of finger length before and after participants wear the device, and then before and after they learn to use the device, which extends the lengths of the fingers by 10 cm. The authors find changes between different time points, which they interpret as evidence for three distinct forms of plasticity: one related to simply wearing the device, one related to learning to use it, and an aftereffect after taking the device off. A control experiment with a similar device, which does not lengthen the fingers, showed the first and third of these forms of plasticity, but not the second.

This study takes an interesting approach to a timely and theoretically significant issue. The study appears to be appropriately designed and conducted. There are, however, some points which require clarification.

(1) The nature of the localization task is unclear. On its face, the task appears to involve localization of each landmark within the 2-dimensional surface of the touchscreen. However, the regression analysis presupposes that localization is made in a 1-dimensional space. Figure S2 shows that three lines are presented on the screen above the index, middle, and ring fingers, which I imagine the participant is meant to use as a guide. But it is at least conceivable that the perceived location or orientation of the finger might not correspond exactly to these lines. While the method can deal gracefully with proximal-distal translations of the fingers (i.e., with the intercept parameter of the regression), it isn't clear how the participant is supposed to respond if their proprioceptive perception of finger location is translated left-right or rotated relative to the lines on the screen. I also worry that presenting a long, thin line to represent each finger on the screen may not be a neutral method and may prime participants to represent the finger as long and thin.

(2) The task used here fits within a wider family of tasks in the literature using localization judgments of multiple landmarks to map body representations. I feel that some discussion of this broader set of tasks and their use to measure body representation and plasticity is notably absent from the paper. It is also striking to me that some of the present authors have themselves recently criticized the use of landmark localization methods as a measure of represented body size and shape (Peviani et al, 2024, Current Biology). It is therefore surprising to see them use this task here as a measure of represented finger length without commenting on this issue.

(3) 18 participants strikes me as a relatively small sample size for this type of study. It weakens the manuscript that the authors do not provide any justification, or even comment on, the sample size. This is especially true as participants are excluded from the entire sample, and from specific analyses, on rather post-hoc grounds.

(4) I have some concerns about the interpretation of contraction in stage 2. The authors claim that wearing the finger extended produces "a contraction",i.e., an "under-representation" (page 12). But in both experiments, regression slopes in stage 2 were not significantly different from 1 (i.e., 0.98 [SE: 0.07] in Exp 1a and 1.04 [SE: 0.09] in Experiment 1b). So how can that be interpreted as "under-representation"?

(5) I also have concerns about the interpretation of the stretch that is claimed to occur following training. In Exp 1a, regression slopes in stage 3 are on average 1.15. That is LESS than in the pretest at stage 1 (mean: 1.16). The idea of stretch only comes about because of the lower slopes in stage 2, which the authors have interpreted as reflecting contraction. So what the authors call stretch and a 2nd form of plasticity could just be the contraction from stage 2 wearing off or dissipating, since perceived finger length in stage 3 just appears to return to the baseline level seen in stage 1. While the authors describe their results in terms of three distinct forms of plasticity, these are not in fact statistically independent. The dip in regression slopes in stage 2 is interpreted as evidence for two distinct plasticity effects, which I do not find convincing.

(6) The distinction between plasticity at stage 3 (which appears specific to augmentation) and plasticity at stage 4 (which does not appear specific, as it also occurs in Experiment 1b) feels strained. This feels like a very subtle distinction, and the theoretical significance of it is not convincingly developed.

(7) The reporting of statistics is not always consistent. For example, 95%CIs are presented for regression slopes in stages 1, 3, and 4, but not for stage 2. Statistics are performed on regression slopes, except for one t-test on page 7 comparing lengths in cm. Estimates of effect size would be nice additions to statistical tests.

(8) Minor point: On page 4, the authors write, "These included sorting colored blocks, stacking a Jenga tower, and sorting pegs into holes; the latter task required fine-grained manipulation and was used as our outcome measure of motor learning." This suggests that peg sorting was the outcome measure, but in Figure 1D, Jenga is presented as the outcome measure.

Reviewer #2 (Public review):

Summary:

This study aimed to explore dynamic changes in the somatosensory representation of both the body and artificial body parts. The study investigated how proprioceptive localisation along the finger changes when participants wear, actively use, and then remove a hand augmentation device - a rigid finger-extension. By mapping perceived target locations along the biological finger and the extension across multiple stages, the authors aim to characterise how the somatosensory system updates our spatial body representation during and after interaction with body augmentation technology.

Strengths:

The manuscript addresses an interesting question of how augmentation devices alter proprioceptive localisation abilities. Conceptually, the work moves beyond classic tool-use paradigms by focusing on a device that is used with the hand to extend the fingers' abilities (versus a tool that is simply used by the hand), and by attempting to map perceived spatial structure across both biological and artificial segments within the same framework.

A major strength is the multi-stage design, which samples localisation abilities at baseline, the beginning of device wear, post-training, and immediately post-removal. This provides a richer characterisation of short-term adaptation compared to a simple pre/post comparison. The dense sampling across stages and target locations generates a rich behavioural dataset that will be valuable to readers interested in somatosensory body representation. The within-subject, counterbalanced control session further strengthens interpretability, providing a useful comparison for interpreting stage-dependent effects, and to probe how functional training shapes changes in the perceptual representations. Finally, the augmentation device itself appears carefully engineered, with thoughtful design decisions regarding wearability, including comfort and customised fit. The manuscript is also communicated clearly, with transparent reporting of analyses and succinct figures that make the pattern changes across stages straightforward to evaluate.

Weaknesses:

There is conceptual ambiguity in how the regression outcomes are interpreted in relation to perceived length and spatial integration. The manuscript treats regression slope as a proxy for "length perception" and discards the intercept as "spatial bias," but in this localisation task translation (intercept) and scaling (slope) are coupled: changes in anchoring at the proximal baseline (intercept) or distal endpoint can generate slope differences without uniform rescaling across the mapped surface. Relatedly, the analyses do not establish whether the reported effects are global across targets or disproportionately driven by the most distal locations. This limits the strength of inferences about "partitioning" or "reallocation" of representational space across biological and artificial segments. Some interpretive statements also appear stronger than the evidence supports (e.g., describing the stage 2 bio-extension map as "geometrically accurate", despite Bayes factors that provide only anecdotal support for no difference from true length). Extensive repeated judgements to a fixed set of locations may additionally stabilise response strategies or anchoring even without feedback, complicating the separation of body-representation change from task-specific calibration.

The manuscript would also benefit from clearer conceptual framing of what the device is and what its training probes are. The device is described variably as an "artificial finger" versus a rigid "finger extension," with different implications for perception and function. In addition, the training tasks appear to emphasise manipulation and dexterity more than scenarios requiring an extended reachable workspace (indeed, participants appear to have performed at least as well, if not better, in the control training), which brings into question whether participants explored the device's intended functionality and possible proprioceptive consequences. The control experiment is thoughtfully designed to test whether functional training contributes to the stage 3 changes, but because localisation is not performed while wearing the short device, the design does not resolve whether the stage 2 change and the post-removal aftereffect are specific to the augmentative extension versus more general consequences of wearing a device on the finger (and the following possible distorted distal cues).

Finally, the immediate post-removal aftereffects are intriguing, but the mechanistic interpretation remains underspecified. As presented within the internal model framework, the magnitude and consistency of the aftereffect following brief exposure are difficult to reconcile with the stability expected from a lifetime biological finger model, and because the aftereffect is assessed only immediately after removal, its time course and functional significance remain unclear.

Reviewer #3 (Public review):

Summary:

The study aims to investigate sensorimotor plasticity mechanisms by exposing a cohort of 20 subjects to manipulation activities while using wearable finger extensions. With a series of experiments involving localization and motor tasks, the authors provide evidence that the finger extensions are integrated into the body representation of the subjects.

Strengths:

The study deserves attention, and the psychophysical protocols are carefully designed, and the statistical analyses are solid.

Weaknesses:

However, the current version of the manuscript, in my opinion, makes an exaggerated use of the term plasticity, and this should be amended. This is because the authors support the plasticity claims with psychophysical experiments, without providing evidence of neural-plasticity mechanisms (e.g., neuroimaging methods are not used).

The authors are recommended to revise the wording of the manuscript and possibly perform additional experiments with brain imaging methods (e.g., EEG or fMRI).

eLife Assessment

This paper investigates the Achilles' heel of an aggressive pediatric bone cancer known as Ewing sarcoma. The authors aimed to better understand how its previously undruggable drivers mediate oncogenic mechanisms using several omics approaches. Transcriptomic changes aligned with their findings provide convincing evidence for the role of a short alpha helix in the DNA binding domain of FLI1 in modulating binding to GGAA microsatellites and promoting enhancer activity. The study provides valuable new insights into the underlying oncogenic mechanisms in Ewing sarcoma.

Reviewer #1 (Public review):

Summary:

Ewing sarcoma is an aggressive pediatric cancer driven by the EWS-FLI oncogene. Ewing sarcoma cells are addicted to this chimeric transcription factor, which represents a strong therapeutic vulnerability. Unfortunately, targeting EWS-FLI has proven to be very difficult and better understanding how this chimeric transcription factor works is critical to achieving this goal. Towards this perspective, the group had previously identified a DBD-𝛼4 helix (DBD) in FLI that appears to be necessary to mediate EWS-FLI transcriptomic activity. Here, the authors used multi-omic approaches, including CUT&tag, RNAseq, and MicroC to investigate the impact of this DBD domain. Importantly, these experiments were performed in the A673 Ewing sarcoma model where endogenous EWS-FLI was silenced, and EWS-FLI-DBD proficient or deficient isoforms were re-expressed (isogenic context). They found that the DBD domain is key to mediate EWS-FLI cis activity (at msat) and to generate the formation of specific TADs. Furthermore, cells expressing DBD deficient EWS-FLI display very poor colony forming capacity, highlighting that targeting this domain may lead to therapeutic perspectives.

Strengths:

The group has strong expertise in Ewing sarcoma genetics and epigenetics and also in using and analyzing this model (Theisen et al., 2019; Boone et al., 2021; Showpnil et al., 2022).

They aim at better understanding how EWS-FLI mediated its oncogenic activity, which is critical to eventually identifying novel therapies against this aggressive cancer.

They use the most recent state-of-the-art omics methods to investigate transcriptome, epigenetics, and genome conformation methods. In particular, Micro-C enables achieving up to 1kb resolved 3D chromatin structures, making it possible to investigate a large number of TADs and sub-TADs structures where EWS-FLI1 mediates its oncogenic activity.

They performed all their experiments in an Ewing sarcoma genetic background (A673 cells) which circumvents bias from previously reported approaches when working in non-orthologous cell models using similar approaches.

Weaknesses:

The main weakness stems from the poor reproducibility of the Micro-C data. Indeed, the distances and clustering observed between replicates appear to be similar to, or even greater than, those observed between biological conditions. For instance, in Figure 1B, we do not observe any clear clustering among DBD1, DBD2, DBD+1, and DBD+2. Although no further experiments were performed, the authors tempered their claims by rephrasing aspects related to this issue and the reviewer also acknowledged that the transcriptomic data are convincing and support their findings.

Regarding DBD stability and the cycloheximide experiments requested to rule out any half-life bias of DBD (as higher stability of the re-expressed DBD+ could also partially explain the results independently of a 3D conformational change), the reviewer acknowledged that the WB, RNA-seq data and agar assays presented by the authors appear reproducible across experiments.

Author response:

The following is the authors’ response to the previous reviews

Public Review:

Reviewer #1 (Public review):

Ewing sarcoma is an aggressive pediatric cancer driven by the EWS-FLI oncogene. Ewing sarcoma cells are addicted to this chimeric transcription factor, which represents a strong therapeutic vulnerability. Unfortunately, targeting EWS-FLI has proven to be very difficult and better understanding how this chimeric transcription factor works is critical to achieving this goal. Towards this perspective, the group had previously identified a DBD-𝛼4 helix (DBD) in FLI that appears to be necessary to mediate EWS-FLI transcriptomic activity. Here, the authors used multi-omic approaches, including CUT&tag, RNAseq, and MicroC to investigate the impact of this DBD domain. Importantly, these experiments were performed in the A673 Ewing sarcoma model where endogenous EWS-FLI was silenced, and EWS-FLI-DBD proficient or deficient isoforms were re-expressed (isogenic context). They found that the DBD domain is key to mediate EWS-FLI cis activity (at msat) and to generate the formation of specific TADs. Furthermore, cells expressing DBD deficient EWS-FLI display very poor colony forming capacity, highlighting that targeting this domain may lead to therapeutic perspectives.

This new version of the study comprises as requested new data from an additional cell line. The new data has strengthened the manuscript. Nevertheless, some of the arguments of the authors pertaining to the limitations of immunoblots to assess stability of the DBD constructs or the poor reproducibility of the Micro C data remain problematic. While the effort to repeat MicroC in a different cell line is appreciated, the data are as heterogeneous as those in A673 and no real conclusion can be drawn. The authors should tone down their conclusions. If DBD has a strong effect on chromatin organization, it should be reproducible and detectable. The transcriptomic and cut and tag data are more consistent and provide robust evidence for their findings at these levels. 

We agree that the Micro-C data have more apparent heterogeneity within and across cell lines as compared to other analyses such as our included CUT&Tag and RNA-seq. We addressed the possible limitations of the technique as well as inherent biology that might be driving these findings in our previous responses. Despite the poor clustering on the PCA plots, our analysis on differential interacting regions, TADs and loops remain consistent across both cell lines. We are confident that these findings reflect the context of transcriptional regulation by the constructs, therefore the role of the alpha-helix in modulating chromatin organization. To address the concerns raised by the editors and reviewers for the strength of the conclusions we drew from the Micro-C findings we have made changes to the language used to describe them throughout the manuscript. Find these changes outlined below.

• On lines 70-71, "is required to restructure" was changed to "is implicated in restructuring of"

• On line 91, "is required for" was changed to "participates in"

• On line 98, "is required for" changed to "is potentially required for"

• On line 360-361, "is required for restructuring" changed to "participates in restructuring"

Concerning the issue of stability of the DBD and DBD+ constructs, a simple protein half-life assay (e.g. cycloheximide chase assay) could rule out any bias here and satisfactorily address the issue.

While we generally agree that a cycloheximide assay is a relatively simple approach to look at protein half-life, as we discussed last me the assays included in this paper are performed at equilibrium and rely on the concentration of protein at the me of the assay. This is particularly true for assays involving crosslinking, like Micro-C. As discussed in our prior response, western blots are semi quantitative at best, even when normalized to a housekeeping protein. In analyzing the relative protein concentration of DBD vs. DBD+ with relative protein intensities first normalized to tubulin and using the wildtype EWSR1::FLI1 rescue as a reference point, we find that there is no statistical difference in the samples used for micro-C here (Author responseimage 1A) or across all of the samples that we have used for publication (Author response image 1B). This does show that DBD generally has more variable expression levels relative to wildtype EWSR1::FLI1, and this is consistent with our experience in the lab.

Nonetheless, we did attempt to perform the requested cycloheximide chase experiment to determine protein stability. Unfortunately, despite an extensive number of troubleshooting attempts, we have not been able to get good expression of DBD for these experiments. The first author who performed this work has left the lab and we have moved to a new lab space since the benchwork was performed. We continue to try to troubleshoot to get this experimental system for DBD and DBD+ to work again. When we tried to look at stability of DBD+ following cycloheximide treatment, there did appear to be some difference in protein stability (Author response image 2). However, these conditions are not the same conditions as those we published, they do not meet our quality control standards for publication, and we are concerned about being close to the limit of detection for DBD throughout the later timepoints. Additional studies will be needed with more comparable expression levels between DBD and DBD+ to satisfactorily address the reviewer concerns.

Author response image 1.

Expression Levels of DBD and DBD+ Across Experiments. Expression levels of DBD and DBD+ protein based on western blot band intensity normalized by tubulin band intensity. Expression levels are relative to wildtype EWSR1::FLI1 rescue levels and are calculated for (A) A673 samples used for micro-C and (B) all published studies of DBD and DBD+. P-values were calculated with an unpaired t-test.

Author response image 2.

CHX chase assay to determine the stability of DBD and DBD+. (A) Knock-down of endogenous EWSR1::FLI1 detected with FLI1 ab and rescue with DBD and DBD+ detected with FLAG ab. (B) CHX chase assay to determine the stability of DBD and DBD+ in A-673 cells with quantification of the protein levels (n=3). Error bars represent standard deviation. The half-lives (t1/2) of DBD and DBD+ were listed in the table.

Suggestions:

The Reviewing Editor and a referee have considered the revised version and the responses of the referees. While the additional data included in the new version has consolidated many conclusions of the study, the MicroC data in the new cell line are also heterogeneous and as the authors argue, this may be an inherent limitation of the technique. In this situation, the best would be for the authors to avoid drawing robust conclusions from this data and to acknowledge its current limitations.

As discussed above, we have changed the language regarding our conclusions from micro-C data to soften the conclusions we draw per the Editor’s suggestion.

The referee and Reviewing Editor also felt that the arguments of the authors concerning a lack of firm conclusions on the stability of EWS-FLI1 under +/-DBD conditions could be better addressed. We would urge the authors to perform a cycloheximide chase type assay to assess protein half-life. These types of experiments are relatively simple to perform and should address this issue in a satisfactory manner.

As discussed above, we do not feel that differences in protein stability would affect the results here because the assays performed required similar levels of protein at equilibrium. Our additional analyses in this response shows that there are not significant differences between DBD and DBD+ levels in samples that pass quality control and are used in published studies. However, we attempted to address the reviewer and editor comments with a cycloheximide chase assay and were unable to get samples that would have passed our internal quality control standards. These data may suggest differences in protein stability, but it is unclear that these conditions accurately reflect the conditions of the published experiments, or that this would matter with equal protein levels at equilibrium.

eLife Assessment

This fundamental manuscript describes how the posterolateral cortical amygdala (plCoA) generates appetitive or aversive behaviors in response to odors. By combining optogenetic stimulation, single-cell RNA sequencing, and spatial analysis, the authors identify a topographically organized circuit within plCoA that governs these behaviors. The manuscript shows convincingly that multiple features (spatial, genetic, and projection) contribute to overall population encoding of valence. Overall, the authors conduct many challenging experiments, each of which contains the relevant controls, and the results are interpreted within the framework of their experiments.

Reviewer #1 (Public review):

Summary:

This study by Howe and colleagues investigates the role of the posterolateral cortical amygdala (plCoA) in mediating innate responses to odors, specifically attraction and aversion. By combining optogenetic stimulation, single-cell RNA sequencing, and spatial analysis, the authors identify a topographically organized circuit within plCoA that governs these behaviors. They show that specific glutamatergic neurons in the anterior and posterior regions of plCoA are responsible for driving attraction and avoidance, respectively, and that these neurons project to distinct downstream regions, including the medial amygdala and nucleus accumbens, to control these responses.

Strengths:

The major strength of the study is the thoroughness of the experimental approach, which combines advanced techniques in neural manipulation and mapping with high-resolution molecular profiling. The identification of a topographically organized circuit in plCoA and the connection between molecularly defined populations and distinct behaviors is a notable contribution to understanding the neural basis of innate motivational responses. Additionally, the use of fucntional manipulations adds depth to the findings, offering valuable insights into the functionality of specific neuronal populations.

Weaknesses:

Previously described weaknesses in the study's methods and interpretation were fully addressed during revision. Locomotor behavior of the mice during head-fixed imaging experiments was added and analysis of the correlation of locomotion with neural activity was also added.

This work provides significant insights into the neural circuits underlying innate behaviors and opens new avenues for further research. The findings are particularly relevant for understanding the neural basis of motivational behaviors in response to sensory stimuli, and the methods used could be valuable for researchers studying similar circuits in other brain regions. If the authors address the methodological issues raised, this work could have a substantial impact on the field, contributing to both basic neuroscience and translational research on the neural control of behavior.

Reviewer #2 (Public review):

Summary:

The manuscript by the Root laboratory and colleagues describes how the posterolateral cortical amygdala (plCoA) generates valenced behaviors. Using a suite of methods, the authors demonstrate that valence encoding is mediated by several factors, including spatial localization of neurons within the plCoA, glutamatergic markers, and projection. The manuscript shows convincingly that multiple features (spatial, genetic, and projection) contribute to overall population encoding of valence. Overall, the authors conduct many challenging experiments, each of which contains the relevant controls, and the results are interpreted within the framework of their experiments.

Strengths:

- The manuscript is well constructed, containing lots of data sets and clearly presented, in spite of the abundance of experimental results.

- The authors should be commended for their rigorous anatomical characterizations and post-hoc analysis. In the field of circuit neuroscience, this is rarely done so carefully, and when it is, often new insights are gleaned as is the case in the current manuscript.

- The combination of molecular markers, behavioral readouts and projection mapping together substantially strengthens the results.

- The focus on this relatively understudied brain region in the context is valence is well appreciated, exciting and novel.

Weaknesses:

The weaknesses noted in the primary review have all been addressed adequately.

Reviewer #3 (Public review):

Summary:

Combining electrophysiological recording, circuit tracing, single cell RNAseq, and optogenetic and chemogenetic manipulation, Howe and colleagues have identified a graded division between anterior and posterior plCoA and determined the molecular characteristics that distinguish the neurons in this part of the amygdala. They demonstrate that the expression of slc17a6 is mostly restricted to the anterior plCoA whereas slc17a7 is more broadly expressed. Through both anterograde and retrograde tracing experiments, they demonstrate that the anterior plCoA neurons preferentially projected to the MEA whereas those in the posterior plCoA preferentially innervated the nucleus accumbens. Interestingly, optogenetic activation of the aplCoA drives avoidance in a spatial preference assay whereas activating the pplCoA leads to preference. The data support a model that spatially segregated and molecularly defined populations of neurons and their projection targets carry valence specific information for the odors. Moreover, the intermingling of neurons in the plCoA is consistent with prior observations. The presence of a gradient rather than a distinct separation of the cells fits the model being proposed. The discoveries represent a conceptual advance in understanding plCoA function and innate valence coding in the olfactory system.

Strengths:

The strongest evidence supporting the model comes from single-cell RNASeq, genetically facilitated anterograde and retrograde circuit tracing, and optogenetic stimulation. The evidence clear demonstrates two molecularly defined cell populations with differential projection targets. Stimulating the two populations produced opposite behavioral responses.

Weaknesses:

The weaknesses noted in primary review have all been addressed adequately.

Author response:

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

Public Reviews:

Reviewer #1 (Public review):

Summary:

This study by Howe and colleagues investigates the role of the posterolateral cortical amygdala (plCoA) in mediating innate responses to odors, specifically attraction and aversion. By combining optogenetic stimulation, single-cell RNA sequencing, and spatial analysis, the authors identify a topographically organized circuit within plCoA that governs these behaviors. They show that specific glutamatergic neurons in the anterior and posterior regions of plCoA are responsible for driving attraction and avoidance, respectively, and that these neurons project to distinct downstream regions, including the medial amygdala and nucleus accumbens, to control these responses.

Strengths:

The major strength of the study is the thoroughness of the experimental approach, which combines advanced techniques in neural manipulation and mapping with high-resolution molecular profiling. The identification of a topographically organized circuit in plCoA and the connection between molecularly defined populations and distinct behaviors is a notable contribution to understanding the neural basis of innate motivational responses. Additionally, the use of functional manipulations adds depth to the findings, offering valuable insights into the functionality of specific neuronal populations.

Weaknesses:

There are some weaknesses in the study's methods and interpretation. The lack of clarity regarding the behavior of the mice during head-fixed imaging experiments raises the possibility that restricted behavior could explain the absence of valence encoding at the population level.

We agree with idea that head-fixation may alter the state of the animal and the neural encoding of odor. To address this, we have provided further analysis of walking behavior during the imaging sessions, which is provided in Figure S2. Overall, we could not identify any clear patterns in locomotor behavior that are odor-specific. Moreover, when neural activity was sorted depending on the behavioral state (walking, pausing or fleeing) we didn’t observe any apparent patterns in odor-evoked neural activity. This is now discussed in the Results and Limitations sections of the manuscript.

Furthermore, while the authors employ chemogenetic inhibition of specific pathways, the rationale for this choice over optogenetic inhibition is not fully addressed, and this could potentially affect the interpretation of the results.

The rationale was logistical. First, inhibition of over a timescale of minutes is problematic with heat generation during prolonged optical stimulation. Second, our behavioral apparatus has a narrow height between the ceiling and floor, making tethering difficult. This is now explained the results section. The trade-off of using chemogenetics is that we are silencing neurons and not specific projections. However, because we find that NAc- and MeA- projecting neurons have little shared collateralization, we believe the conclusion of divergent pathways still stands. This is now discussed in the Limitations section.

Additionally, the choice of the mplCoA for manipulation, rather than the more directly implicated anterior and posterior subregions, is not well-explained, which could undermine the conclusions drawn about the topographic organization of plCoA.

We targeted the middle region of plCoA because it contains a mixture of cell types found in both the anterior and posterior plCoA, allowing us to test the hypothesis that cell types, not intra plCoA location, elicit different responses. Had we targeted the anterior or posterior regions, we would expect to simply recapitulate the result from activation of random cells in each region. As a result, we think stimulation in the middle plCoA is a better test for the contribution of cell types. We have now clarified this in the text.

Despite these concerns, the work provides significant insights into the neural circuits underlying innate behaviors and opens new avenues for further research. The findings are particularly relevant for understanding the neural basis of motivational behaviors in response to sensory stimuli, and the methods used could be valuable for researchers studying similar circuits in other brain regions. If the authors address the methodological issues raised, this work could have a substantial impact on the field, contributing to both basic neuroscience and translational research on the neural control of behavior.

Reviewer #2 (Public review):

Summary:

The manuscript by the Root laboratory and colleagues describes how the posterolateral cortical amygdala (plCoA) generates valenced behaviors. Using a suite of methods, the authors demonstrate that valence encoding is mediated by several factors, including spatial localization of neurons within the plCoA, glutamatergic markers, and projection. The manuscript shows convincingly that multiple features (spatial, genetic, and projection) contribute to overall population encoding of valence. Overall, the authors conduct many challenging experiments, each of which contains the relevant controls, and the results are interpreted within the framework of their experiments.

Strengths:

- For a first submission the manuscript is well constructed, containing lots of data sets and clearly presented, in spite of the abundance of experimental results.

- The authors should be commended for their rigorous anatomical characterizations and posthoc analysis. In the field of circuit neuroscience, this is rarely done so carefully, and when it is, often new insights are gleaned as is the case in the current manuscript.

- The combination of molecular markers, behavioral readouts and projection mapping together substantially strengthen the results.

- The focus on this relatively understudied brain region in the context is valence is well appreciated, exciting and novel.

Weaknesses:

- Interpretation of calcium imaging data is very limited and requires additional analysis and behavioral responses specific to odors should be considered. If there are neural responses behavioral epochs and responses to those neuronal responses should be displayed and analyzed.

We have now considered this, see response above.

- The effect of odor habituation is not considered.

We considered this, but we did not find any apparent differences in valence encoding as measured by the proportion of neurons with significant valence scores across trials (see Figure 1J).

- Optogenetic data in the two subregions relies on very careful viral spread and fiber placement. The current anatomy results provided should be clear about the spread of virus in A-P, and D-V axis, providing coordinates for this, to ensure readers the specificity of each sub-zone is real.

We were careful to exclude animals for improper targeting. The spread of virus is detailed in Figures S3, S8 & S9.

- The choice of behavioral assays across the two regions doesn't seem balanced and would benefit from more congruency.

The choice of the 4-quadrant assay was used because this study builds off of our prior experiments that demonstrate a role for the plCoA in innate behavior. It is noteworthy that the responses to odor seen in this assay are generally in agreement with other olfactory behavioral assays, so one wouldn’t predict a different result. Moreover, the approach and avoidance responses measured in this assay are precisely the behaviors we wish to understand. We did examine other non-olfactory behavioral readouts (Figures S3, S8), and didn’t observe any effect of manipulation of these pathways.

- Rationale for some of the choices of photo-stimulation experiment parameters isn't well defined.

The parameters for photo-stimulation were based on those used in our past work (Root et al., 2014). We used a gradient of frequency from 1-10 Hz based on the idea that odor likely exists in a gradient and this was meant to mimic a potential gradient, though we don’t know if it exists. The range in stimulation frequencies appears to align with the actual rate of firing of plCoA neurons (Iurilli et al., 2017).

Reviewer #3 (Public review):

Summary:

Combining electrophysiological recording, circuit tracing, single cell RNAseq, and optogenetic and chemogenetic manipulation, Howe and colleagues have identified a graded division between anterior and posterior plCoA and determined the molecular characteristics that distinguish the neurons in this part of the amygdala. They demonstrate that the expression of slc17a6 is mostly restricted to the anterior plCoA whereas slc17a7 is more broadly expressed. Through both anterograde and retrograde tracing experiments, they demonstrate that the anterior plCoA neurons preferentially projected to the MEA whereas those in the posterior plCoA preferentially innervated the nucleus accumbens. Interestingly, optogenetic activation of the aplCoA drives avoidance in a spatial preference assay whereas activating the pplCoA leads to preference. The data support a model that spatially segregated and molecularly defined populations of neurons and their projection targets carry valence specific information for the odors. The discoveries represent a conceptual advance in understanding plCoA function and innate valence coding in the olfactory system.

Strengths:

The strongest evidence supporting the model comes from single cell RNASeq, genetically facilitated anterograde and retrograde circuit tracing, and optogenetic stimulation. The evidence clear demonstrates two molecularly defined cell populations with differential projection targets. Stimulating the two populations produced opposite behavioral responses.

Weaknesses:

There are a couple of inconsistencies that may be addressed by additional experiments and careful interpretation of the data.

Stimulating aplCoA or slc17a6 neurons results in spatial avoidance, and stimulating pplCoA or slc17a7 neurons drives approach behaviors. On the other hand, the authors and others in the field also show that there is no apparent spatial bias in odor-driven responses associated with odor valence. This discrepancy may be addressed better. A possibility is that odor-evoked responses are recorded from populations outside of those defined by slc17a6/a7. This may be addressed by marking activated cells and identifying their molecular markers. A second possibility is that optogenetic stimulation activates a broad set of neurons that and does not recapitulate the sparseness of odor responses. It is not known whether sparsely activation by optogenetic stimulation can still drive approach of avoidance behaviors.

We agree that marking specific genetic or projection defined neurons could help to clarify if there are some neurons have more selective valence responses. However, we are not able to perform these experiments at the moment. We have included new data demonstrating that sparser optogenetic activation evokes behaviors similar in magnitude as the broader activation (see Figure S4).

The authors show that inhibiting slc17a7 neurons blocks approaching behaviors toward 2-PE. Consistent with this result, inhibiting NAc projection neurons also inhibits approach responses. However, inhibiting aplCOA or slc17a6 neurons does not reduce aversive response to TMT, but blocking MEA projection neurons does. The latter two pieces of evidence are not consistent with each other. One possibility is that the MEA projecting neurons may not be expressing slc17a6. It is not clear that the retrogradely labeling experiments what percentage of MEA- and NACprojecting neurons express slc17a6 and slc17a7. It is possible that neurons expressing neither VGluT1 nor VGluT2 could drive aversive or appetitive responses. This possibility may also explain that silencing slc17a6 neurons does not block avoidance.

We have now performed RNAscope staining on retrograde tracing to better define this relationship. Although the VGluT1 and VGluT2 neurons have biased projections to the MeA and NAc, respectively, there is some nuance detailed in Figure S10. Generally, MeA projecting neurons are predominately VGluT2+, whereas NAc projecting have about 20% that express both. Some (less than 35%) retrogradely labeled neurons were not detected as VGluT1 or VGluT2 positive, suggesting that other populations could also contribute. We agree that the discrepancy between MeA-projection and VGluT2 silencing is likely due to incomplete targeting of the MeA-projecting population with the VGluT2-cre line. This is included in the Discussion section.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

Main:

(1) For the head-fixed imaging experiments, what is the behavior of the mice during odor exposure? Could the weak reliability of individual neurons be due to a lack of approach or avoidance behavior? Could restricted behavior also explain the lack of valence encoding at the population level?

We agree that this is a limitation of head-fixed recordings. In the revised manuscript we did attempt to characterize their behavioral response, and look for correlations in odor representation. Although we did find different patterns of odor-evoked walking behavior, these patterns were not reliable or specific to particular odors (Figure S2). For example, one might expect aversive odors to pause walking or elicit a fast fleeing-like response, but we did not observe any apparent differences for locomotion between odors as all odors evoked a mixture of responses (Figure S2A-D, text lines 208-232). We then examined responses to odor depending on the behavioral state (walking, pausing or fleeing) and didn’t observe any apparent patterns in odor responses (Figure S2E,F). Lastly, we acknowledge in the text that the lack of valence encoding may be an artifact of head-fixation (see lines 849-857).

(2) For the optogenetic manipulations of Vglut1 and Vglut2 neurons, why was the injection and fiber targeted to the medial portion of the plCoA, if the hypothesis was that these glutamatergic neuron populations in different regions (anterior or posterior) are responsible for approach and avoidance? 

We targeted the middle region of plCoA because it contains a mixture of cell types found in both the anterior and posterior plCoA, allowing us to test the hypothesis that cell types, not intraplCoA location, elicit different responses. Had we targeted the anterior or posterior regions, we would expect to simply recapitulate the result from activation of random cells in each region. As a result, we think stimulation in the middle plCoA is a better test for the contribution of cell types. We have clarified this in the text (Lines 417-419).

Could this explain the lack of necessity with the DREADD experiments? 

For the loss of function experiments, a larger volume of virus was injected to cover a larger area and we did confirm targeting of the appropriate areas. Though, it is always possible that the lack of necessity is due to incomplete silencing.

Further, why was an optogenetic inhibition approach not utilized? 

Although optogenetic inhibition could have plausibly been used instead, we chose chemogenetic inhibition for two reasons: First, for minutes-long periods of inhibition, optical illumination poses the risk of introducing heat related effects (Owen et al., 2019). In fact, we first tried optical inhibition but controls were exhibited unusually large variance. Second, it is more feasible in our assay as it has a narrow height between the floor and lid that complicates tethering to an optic fiber. Past experiments overcame this with a motorized fiber retraction system (Root et al., 2014), but this is highly variable with user-dependent effects, so we found chemogenetics to be a more practical strategy. We have added a sentence to explain the rationale (see lines 561-563).

(3) The specific subregion of the nucleus accumbens that was targeted should be named, as distinct parts of the nucleus accumbens can have very different functions. 

We attempted to define specific subregions of the nucleus accumbens and found that plCoA projection is not specific to the shell or core, anterior or posterior, rather it broadly innervates the entire structure. We have added a note about this in manuscript (see lines 470-471). Given that we did not find notable subregion-specific outputs within the NAc, targeting was directed to the middle region of NAc, with coordinates stated in the methods. 

(4) Why was an intersectional DREADD approach used to inhibit the projection pathways, as opposed to optogenetic inhibition? The DREADD approach could potentially affect all projection targets, and the authors might want to address how this could influence the interpretation of the results.

This is partly addressed above in point 2. As for interpretation, we acknowledge that the intersectional approach silences the neurons projecting to a given target and not the specific projection and we have been careful with the wording. Although this may complicate the conclusion, we did map the collaterals for NAc and MeA projecting neurons and find that neurons do not appreciably project to both targets and have minimal projections to other targets. We have now taken care to state that we silence the neurons projecting to a structure, not silencing the projection, and we acknowledge this caveat. However, since the MeA- and NAcprojecting neurons appear to be distinct from each other (largely not collateralizing to each other), the conclusion that these divergent pathways are required still stands. We have added discussion of this in the Limitations section (see lines 859-863).

Minor:

(1) Line 402 needs a reference.

We have added the missing reference (now line 441).

(2) The Supplemental Figure labeling in the main text should be checked carefully.

Thank you for pointing this out. We have fixed the prior errors.

(3) Panel letter D is missing from Figure 2.

This has been fixed.

Reviewer #2 (Recommendations for the authors):

Major Concerns, additional experiments:

- In the calcium imaging experiments mice were presented with the same odor many times. Overall responses to odor presentations were quite variable and appear to habituate dramatically (Figure S1F). The general conclusion from these experiments are a lack of consistent valence-specific responses of individual neurons, but I wonder if this conclusion is slightly premature. A few potential explanatory factors that may need additional attention are: -First, despite recording video of the mouse's face during experiments, no behavioral response to any odor is described. Is it possible these odors when presented in head-fixed conditions do not have the same valence?

Yes, we agree that this is a possibility. We have added a discussion in the Limitations section (see lines 849-857). We have also added additional behavioral analysis discussed below.

On trials with neural responses are there behavioral responses that could be quantified? 

We have now added data in which we attempt to characterize their behavioral response, to look for correlations in odor representation (see lines 208-228). Although we did observe different patterns of odor-evoked walking behavior, these patterns were not reliable or specific to particular odors (Figure S2). One might expect aversive odors to pause walking or elicit a fast fleeing-like response, but we did not observe any apparent differences for locomotion between odors (Figure S2A-D). Next, we examined responses to odor depending on the behavioral state (walking, pausing or fleeing) and didn’t observe any meaningful differences in odor responses (Figure S2E,F). Lastly, we acknowledge that the odor representation may be different in freely moving animals that exhibit dynamic responses to odor (see lines 859-857).

- Habituation seems to play a prominent role in the neural signals, is there a larger contribution of valence if you look only at the first delivery (or some subset of the 20 presentations) of an odor type for a given trial? 

Indeed, we considered this, but we did not find any apparent differences in valence encoding as measured by the proportion of neurons with significant valence scores across trials (see Figure 1J).

- Is it reasonable to exclude valence encoding as a possibility when largely neurons were unresponsive to the positive valence odors (2PE and peanut) chosen when looking at the average cluster response (Figure 1F)? 

It is true that we see fewer neurons responding to the appetitive odors (Figure 1H) and smaller average responses within the cluster, but some neurons do respond robustly. If these were valence responses, we would predict that neural responses should be similarly selective, but we do not observe any such selectivity. The sparseness of responses to appetitive odors does cause the average cluster analysis (Figure 1F) to show muted responses to these odors, consistent with the decreased responsivity to appetitive odors. Moreover, single neuron response analysis reveals that a given neuron is not more likely to respond to appetitive or aversive odors with any selectivity greater than chance. For these reasons, we think it is reasonable to conclude an absence of valence responses, which is consistent with the conclusion from another report (Iurilli et al., 2017).

- While the preference and aversion assay with 4 corners is an interesting set-up and provides a lot of data for this particular manuscript. It would be helpful to test additional behaviors to determine whether these circuits are more conserved. As it stands the current manuscript relies on very broad claims using a single behavioral readout. Some attempts to use head-fixed approaches with more defined odor delivery timelines and/or additional valenced behavioral readouts is warranted.

We appreciate the suggestion, but are not able to perform these experiments at the moment. The choice of the 4-quadrant assay was used because it built off of our prior experiments that demonstrate a role for the plCoA in innate behavior. It is noteworthy that the responses to odor seen in this assay are generally in agreement with other olfactory behavioral assays, so one wouldn’t predict a different result. The approach and avoidance responses measured in this assay are precisely the behaviors we wish to understand. Moreover, we did examine other nonolfactory behavioral readouts (Figures S3, S8), and didn’t observe any effect of manipulation of these pathways. Lastly, we have tried to define parameters for head-fixed behavior that would permit correlation of neural responses with behavior, including longer stimulations and closed loop locomotion control of odor concentration, but were unsuccessful at establishing parameters that generated reliable behavioral responses. We acknowledge that one limitation of the study is the limited behavioral tests with two odors and whether the circuits are more broadly necessary for other odors. 

Minor comments:

• Please define PID in the Results when it is first introduced.

Done (see line 154)

• Line 412 Figure S5C-N should be Figure S6C-N.

Fixed. Now Figure S8C-N due to additional figures (see line 451).

• Throughout the Discussion it would be helpful if the authors referred to specific Figure panels that support their statements (e.g. lines 654-656 "[...] which is supported by other findings presented here showing that both VGluT2+ and VGluT1+ neurons project to MeA, while the projection to NAc is almost entirely composed of VGluT1+ neurons".

Thank you for the suggestion. We have added figure references in the discussion.

• Line 778 "producing" should be "produce".

Corrected (see line 840)

• The figures are very busy, especially all the manipulations. The authors are commended for including each data point, but they might consider a more subtle design (translucent lines only for each animal, and one mean dot for the SEM), just to reduce the overall clutter of an already overwhelming figure set. But this is ultimately left to the authors to resolve and style to their liking. 

Thank you for the suggestion. We have tried some different styles but like the original best.

Reviewer #3 (Recommendations for the authors):

If within reach, I suggest that the author determine the percentage of retrogradely labeled neurons to NAc or MEA that expresses GluT1 and GluT2. 

We have done this for the middle region plCoA that has the greatest mixture of cell types (See Figure S10, lines 504-517). We find that the MeA projecting neurons are mostly VGluT2+ with a minority that express both VGluT1 and VGlut2. NAc-projecting neurons are primarily VGluT1+ with about 20% expressing VGlut2 as well.

It would also be nice to sparse label of aplCoA and pplCoA using ChR2 to see if sparse activation drives approach or avoidance. 

We agree that it would be useful to vary the sparseness of the ChR2 expression, to see if produces similar results. We examined this using sparsely labeled odor ensembles, as previously done (Root et al., 2014). Briefly, we used the Arc-CreER mouse to label TMT responsive neurons with a cre-dependent ChR2 AAV vector targeted to the anterior or posterior regions, while previously we had broadly targeted the entirety of plCoA. We had established that this labeling method captures about half of the active cells detected by Arc expression, which is on the order of hundreds of neurons rather than thousands by broad cre-independent expression. Remarkably, we get effects similar in magnitude that are not significantly different from that with broader activation of the anterior or posterior domains (see new Figure S4, lines 267-288). It still remains possible that there is a threshold number of neurons that are necessary to elicit behavior, but that is beyond the scope of the current study. However, these data indicate that the effect of activating anterior and posterior domains is not an artifact of broad stimulation.

eLife Assessment

This is an important study with direct implications for the rational selection of antimalarial drug combinations. The authors present data demonstrating antagonism between 4-aminoquinoline antimalarials and peroxide drugs under physiologically relevant conditions, including robust effects at the trophozoite stage and for chloroquine at the ring stage. While the conclusions are based on in vitro assays and further work will be needed to fully resolve the underlying mechanism, the findings are convincing and provide a strong rationale for evaluating drug combinations in relevant preclinical models prior to clinical testing.

Reviewer #1 (Public review):

Summary:

This study set out to investigate potential pharmacological drug-drug interactions between the two most common antimalarial classes, the artemisinins and quinolines. There is strong rationale for this aim, because drugs from these classes are already widely-used in Artemisinin Combination Therapies (ACTs) in the clinic, and drug combinations are an important consideration in the development of new medicines. Furthermore, whilst there is ample literature proposing many diverse mechanisms of action and resistance for the artemisinins and quinolines, it is generally accepted that the mechanisms for both classes involve heme metabolism in the parasite, and that artemisinin activity is dependent on activation by reduced heme. The study was designed to measure drug-drug interactions associated with a short pulse exposure (4 h) that is reminiscent of the short duration of artemisinin exposure obtained after in vivo dosing. Clear antagonism was observed between dihydroartemisinin (DHA) and chloroquine, which became even more extensive in chloroquine-resistant parasites. Antagonism was also observed in this assay for the more clinically-relevant ACT partner drugs piperaquine and amodiaquine, but not for other ACT partners mefloquine and lumefantrine, which don't share the 4-aminoquinoline structure or mode of action. Interestingly, chloroquine induced an artemisinin resistance phenotype in the standard in vitro Ring-stage Survival Assay, whereas this effect was not as extensive for piperaquine.

The authors also utilised a heme-reactive probe to demonstrate that the 4-aminoquinolines can inhibit heme-mediated activation of the probe within parasites, which suggests that the mechanism of antagonism involves the inactivation of heme, rendering it unable to activate the artemisinins. Measurement of protein ubiquitination showed reduced DHA-induced protein damage in the presence of chloroquine, which is also consistent with decreased heme-mediated activation, and/or with decreased DHA activity more generally.

Overall, the study clearly demonstrates a mechanistic antagonism between DHA and 4-aminoquinoline antimalarials in vitro. It is interesting that this combination is successfully used to treat millions of malaria cases every year, which may raise questions about the clinical relevance of this finding. However, the conclusions in this paper are supported by multiple lines of evidence and the data is clearly and transparently presented, leaving no doubt that DHA activity is compromised by the presence of chloroquine in vitro. It is perhaps fortunate the that the clinical dosing regimens of 4-aminoquinoline-based ACTs have been sufficient to maintain clinical efficacy despite the non-optimal combination. Nevertheless, optimisation of antimalarial combinations and dosing regimens is becoming more important in the current era of increasing resistance to artemisinins and 4-aminoquinolines. Therefore, these findings should be considered when proposing new treatment regimens (including Triple-ACTs) and the assays described in this study should be performed on new drug combinations that are proposed for new or existing antimalarial medicines.

Strengths:

This manuscript is clearly written and the data presented is clear and complete. The key conclusions are supported by multiple lines of evidence, and most findings are replicated with multiple drugs within a class, and across multiple parasite strains, thus providing more confidence in the generalisability of these findings across the 4-aminoquinoline and peroxide drug classes.

A key strength of this study was the focus on short pulse exposures to DHA (4 h in trophs and 3 h in rings), which is relevant to the in vivo exposure of artemisinins. Artemisinin resistance has had a significant impact on treatment outcomes in South-East Asia, and is now emerging in Africa, but is not detected using a 'standard' 48 or 72 h in vitro growth inhibition assay. It is only in the RSA (a short pulse of 3-6 h treatment of early ring stage parasites) that the resistance phenotype can be detected in vitro. Therefore, assays based on this short pulse exposure provide the most relevant approach to determine whether drug-drug interactions are likely to have a clinically-relevant impact on DHA activity. These assays clearly showed antagonism between DHA and 4-aminoquinolines (chloroquine, piperaquine, amodiaquine and ferroquine) in trophozoite stages. Interestingly, whilst chloroquine clearly induced an artemisinin-resistant phenotype in the RSA, piperaquine only had a minor impact on the early ring stage activity of DHA, which may be fortunate considering that piperaquine is a currently recommended DHA partner drug in ACTs, whereas chloroquine is not.

The evaluation of additional drug combinations at the end of this paper is a valuable addition, which increases the potential impact of this work. The finding of antagonism between piperaquine and OZ439 in trophozoites is consistent with the general interactions observed between peroxides and 4-aminoquinolines, and it may be interesting to see whether piperaquine impacts the ring-stage activity of OZ439.

The evaluation of reactive heme in parasites using a fluorescent sensor, combined with the measurement of K48-linked ubiquitin, further support the findings of this study, providing independent read-outs for the chloroquine-induced antagonism.<br /> The in-depth discussion of the interpretation and implications of the results are an additional strength of this manuscript. Whilst the discussion section is rather lengthy, there are important caveats to the interpretation of some of these results, and clear relevance to the future management of malaria that require these detailed explanations.

Overall, this is a high quality manuscript describing an important study that has implications for the selection of antimalarial combinations for new and existing malaria medicines.

Weaknesses:

This study is an in vitro study of parasite cultures, and therefore caution should be taken when applying these findings to decisions about clinical combinations. The drug concentrations and exposure durations in these assays are intended to represent clinically relevant exposures, although it is recognised that the in vitro system is somewhat simplified and there may be additional factors that influence in vivo activity. This limitation is reasonably well acknowledged in the manuscript.

It is also important to recognise that the majority of the key findings regarding antagonism are based on trophozoite-stage parasites, and one must show caution when generalising these findings to other stages or scenarios. For example, piperaquine showed clear antagonism in trophozoite stages, but minimal impact in ring stages under these assay conditions.

A key limitation is the interpretation of the mechanistic studies that implicate heme-mediated artemisinin activation as the mechanism underpinning antagonism by chloroquine. This study did not directly measure the activation of artemisinins. The data obtained from the activation of the fluorescent probe are generally supportive of chloroquine suppressing the heme-mediated activation of artemisinins, and I think this is the most likely explanation, but there are significant caveats to consider. Primarily, the inconsistency between the fluorescence profile in the chemical reactions and the cell-based assay raise questions about the accuracy of this readout. In the chemical reaction, mefloquine and chloroquine showed identical inhibition of fluorescence, whereas piperaquine had minimal impact. On the contrary, in the cell, chloroquine and piperaquine had similar impacts on fluorescence, but mefloquine had minimal impact. This inconsistency indicates that the cellular fluorescence based on this sensor does not give a simple direct readout of the reactivity of ferrous heme, and therefore, these results should be interpreted with caution. Indeed, the correlation between fluorescence and antagonism for the tested drugs is a correlation, not causation. There could be several reasons for the disconnect between the chemical and biological results, either via additional mechanisms that quench fluorescence, or the presence of biomolecules that alter the oxidation state or coordination chemistry of heme or other potential catalysts of this sensor. It is possible that another factor that influences the H-FluNox fluorescence in cells also influences the DHA activity in cells, leading to the correlation with activity. It should be noted that H-FluNox is not a chemical analogue of artemisinins. It's activation relies on Fenton-like chemistry, but with a N-O rather that O-O bond, and it possesses very different steric and electronic substituents around the reactive centre, which are known to alter reactivity to different iron sources. Despite these limitations, the authors have provided reasonable justification for the use of this probe to directly visualise heme reactivity in cells, and the results are still informative.

Another interesting finding that was not elaborated by the authors is the impact of chloroquine in the DHA dose-response curves from the ring stage assays. Detection of artemisinin resistance in the RSA generally focuses on the % survival at high DHA concentrations (700 nM) as there is minimal shift in the IC50 (see Fig 2), however, chloroquine clearly induces a shift in the IC50 (~5-fold), where the whole curve is shifted to the right, whereas the increase in % survival is relatively small. This different profile suggests that the mechanism of chloroquine-induced antagonism may be different to the mechanism of artemisinin resistance. Current evidence regarding the mechanism of artemisinin resistance generally points towards decreased heme-mediated drug activation due to a decrease in hemoglobin uptake, which should be analogous to the decrease in heme-mediated drug activation caused by chloroquine. However, these different dose response curves suggest different mechanisms are primarily responsible. Additional mechanisms have been proposed for artemisinin resistance, involving redox or heat stress responses, proteostatic responses, mitochondrial function, dormancy and PI3K signalling among others. Whilst the H-FluNox probe generally supports the idea that chloroquine suppresses heme-mediated DHA activation, it remains plausible that chloroquine could induce these, or other, cellular responses that suppress DHA activity.

Impact:

This study has important implications for the selection of drugs to form combinations for the treatment of malaria. The overall findings of antagonism between peroxide antimalarials and 4-aminoquinolines in the trophozoite stage are robust, and the this carries across to the ring stage for chloroquine.

The manuscript also provides a plausible mechanism to explain the antagonism, although future work will be required to further explore the details of this mechanism and to rule out alternative factors that may contribute.

Overall, this is an important contribution to the field and provides a clear justification for the evaluation of potential drug combinations in relevant in vitro assays before clinical testing.

Reviewer #2 (Public review):

Summary:

This manuscript by Rosenthal and Goldberg investigates interactions between artemisinins and its quinoline partner drugs currently used for treating uncomplicated Plasmodium falciparum malaria. The authors show that chloroquine (CQ), piperaquine, and amodiaquine antagonize dihydroartemisinin (DHA) activity, and in CQ-resistant parasites, the interaction is described as "superantagonism," linked to the pfcrt genotype. Mechanistically, application of the heme-reactive probe H-FluNox indicates that quinolines render cytosolic heme chemically inert, thereby reducing peroxide activation. The work is further extended to triple ACTs and ozonide-quinoline combinations, with implications for artemisinin-based combination therapy (ACT) design, including triple ACTs.

Strengths:

The manuscript is clearly written, methodologically careful, and addresses a clinically relevant question. The pulsing assay format more accurately models in vivo artemisinin exposure than conventional 72-hour assays, and the use of H-FluNox and Ac-H-FluNox probes provides mechanistic depth by distinguishing chemically active versus inert heme. These elements represent important refinements beyond prior studies, adding nuance to our understanding of artemisinin-quinoline interactions.

Weaknesses:

Several points warrant consideration. The novelty of the work is somewhat incremental, as antagonism between artemisinins and quinolines is well established. Multiple prior studies using standard fixed-ratio isobologram assays have shown that DHA exhibits indifferent or antagonistic interactions with chloroquine, piperaquine, and amodiaquine (e.g., Davis et al., 2006; Fivelman et al., 2007; Muangnoicharoen et al., 2009), with recent work highlighting the role of parasite genetic background, including pfcrt and pfmdr1, in modulating these interactions (Eastman et al., 2016). High-throughput drug screens likewise identify quinoline-artemisinin combinations as mostly antagonistic. The present manuscript adds refinement by applying pulsed-exposure assays and heme probes rather than establishing antagonism de novo.

The dataset focuses on several parasite lines assayed in vitro, so claims about broad clinical implications should be tempered, and the discussion could more clearly address how in vitro antagonism may or may not translate to clinical outcomes. The conclusion that artemisinins are predominantly activated in the cytoplasm is intriguing but relies heavily on Ac-H-FluNox data, which may have limitations in accessing the digestive vacuole and should be acknowledged explicitly. The term "superantagonism" is striking but may appear rhetorical; clarifying its reproducibility across replicates and providing a mechanistic definition would strengthen the framing. Finally, some discussion points, such as questioning the clinical utility of DHA-PPQ, should be moderated to better align conclusions with the presented data while acknowledging the complexity of in vivo pharmacology and clinical outcomes.

Despite these mild reservations, the data are interesting and of high quality and provide important new information for the field.

Editor's Review of the Revision: The authors have provided a well-reasoned rebuttal to the comments of the three reviewers. Most of the changes were incorporated in their revised Discussion. Their data with the active heme probe H-FluNox are novel and the authors reveal interesting interactions between peroxide and 4-aminoquinoline-based antimalarials that open new avenues of research especially when considering antimalarial combinations that combine these chemical scaffolds. This study will be of broad interest to investigators studying and developing antimalarial drugs and combinations and the impact of Plasmodium falciparum resistance mechanisms. A minor recommendation would be that the authors state H-FluNox when referring to their small molecule probe in the abstract, so that it is captured in PubMed searches.

Reviewer #3 (Public review):

Summary:

The authors present an in vitro evaluation of drug-drug interactions between artemisinins and quinoline antimalarials, as an important aspect for screening the current artemisinin-based combination therapies for Plasmodium falciparum. Using a revised pulsing assay, they report antagonism between dihydroartemisinin (DHA) and several quinolines, including chloroquine, piperaquine (PPQ), and amodiaquine. This antagonism is increased in CQ-resistant strains in isobologram analyses. Moreover, CQ co-treatment was found to induce artemisinin resistance even in parasites lacking K13 mutations during the ring-stage survival assay. This implies that drug-drug interactions, not just genetic mutations, can influence resistance phenotypes. By using a chemical probe for reactive heme, the authors demonstrate that quinolines inhibit artemisinin activation by rendering cytosolic heme chemically inert, thereby impairing the cytotoxic effects of DHA. The study also observed negative interactions in triple-drug regimens (e.g., DHA-PPQ-Mefloquine) and in combinations involving OZ439, a next-generation peroxide antimalarial. Taken together, these findings raise significant concerns regarding the compatibility of artemisinin and quinoline combinations, which may promote resistance or reduce efficacy.

With the additive profile as the comparison and a lack of synergistic effect in any of the comparisons, it is hard to contextualize the observed antagonism. Including a known synergistic pair (e.g., artemisinin + lumefantrine) would have provided a useful benchmark to assess the relative impact of the drug interactions described.

Strengths:

This study demonstrates the following strengths:

• The use of a pulsed in vitro assay that is more physiologically relevant over the traditional 48h or 72h assays

• Small molecule probes, H-FluNox, and Ac-H-FluNox to detect reactive cytosolic heme, demonstrating that quinolines render heme inert and thereby block DHA activation.

• Evaluates not only traditional combinations but also triple-drug combinations and next-generation artemisinins like OZ439. This broad scope increases the study's relevance to current treatment strategies and future drug development.

• By using the K13 wild-type parasites, the study suggests that resistance phenotypes can emerge from drug-drug interactions alone, without requiring genetic resistance markers.

Weaknesses:

• The study would benefit from a future characterization of the molecular basis for the observed heme inactivation by quinolines to support this hypothesis - while the probe experiments are valuable, they do not fully elucidate how quinolines specifically alter heme chemistry at the molecular level.

• Suggestion of alternative combinations that show synergy could have improved the significance of the work. The invitro study did not include pharmacokinetic/pharmacodynamic modeling, hence it leaves questions about how the observed antagonism would manifest under real-world dosing conditions, necessitating furture work based on these findings.

Author response:

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

eLife Assessment

We appreciate the positive assessment. We recognize that since all of the work in this manuscript was done in vitro, there are reasonable concerns about the translatability of these data to clinical settings. These results should not directly inform malaria policy, but we hope that these data bring new considerations to the approach for choosing strategic antimalarial combinations. We have modified the manuscript to clarify this distinction.

Public Reviews

Reviewer #1 (Public Review):

We thank the reviewer for their thoughtful summary of this manuscript. It is important to note that DHA-PPQ did show antagonism in RSAs. In this modified RSA, 200 nM PPQ alone inhibited growth of PPQ-sensitive parasites approximately 20%. If DHA and PPQ were additive, then we would expect that addition of 200 nM PPQ would shift the DHA dose response curve to the left and result in a lower DHA IC50. Please refer to Figure 4a and b as examples of additive relationships in dose-response assays. We observed no significant shift in IC50 values between DHA alone and DHA + PPQ. This suggests antagonism, albeit not to the extent seen with CQ. We have modified the manuscript to emphasize this point. As the reviewer pointed out, it is fortunate that despite being antagonistic, clinically used artemisinin-4-aminoquinoline combinations are effective, provided that parasites are sensitive to the 4-aminoquinoline. It is possible that superantagonism is required to observe a noticeable effect on treatment efficacy (Sutherland et al. 2003 and Kofoed et al. 2003), but that classical antagonism may still have silent consequences. For example, if PPQ blocks some DHA activation, this might result in DHA-PPQ acting more like a pseudo-monotherapy. However, as the reviewer pointed out, while our data suggest that DHA-PPQ and AS-ADQ are “non-optimal” combinations, the clinical consequences of these interactions are unclear. We have modified the manuscript to emphasize the later point.

While the Ac-H-FluNox and ubiquitin data point to a likely mechanism for DHA-quinoline antagonism, we agree that there are other possible mechanisms to explain this interaction.  We have addressed this limitation in the discussion section. Though we tried to measure DHA activation in parasites directly, these attempts were unsuccessful. We acknowledge that the chemistry of DHA and Ac-H-FluNox activation is not identical and that caution should be taken when interpreting these data. Nevertheless, we believe that Ac-H-FluNox is the best currently available tool to measure “active heme” in live parasites and is the best available proxy to assess DHA activation in live parasites. These points are now addressed in the discussion section. Both in vitro and in parasite studies point to a roll for CQ in modulating heme, though an exact mechanism will require further examination. Similar to the reviewer, we were perplexed by the differences observed between in vitro and in parasite assays with PPQ and MFQ. We proposed possible hypotheses to explain these discrepancies in the discussion section. Interestingly, our data corelate well with hemozoin inhibition assays in which all three antimalarials inhibit hemozoin formation in solution, but only CQ and PPQ inhibit hemozoin formation in parasites. In both assays, in-parasite experiments are likely to be more informative for mechanistic assessment.

It remains unclear why K13 genotype influences RSA values, but not early ring DHA IC50 values. In K13^WT parasites, both RSA values and DHA IC50 values were increased 3-5 fold upon addition of CQ. This suggests that CQ-mediated resistance is more robust than that conferred by K13 genotype. However, this does not necessarily suggest a different resistance mechanism. We acknowledge that in addition to modulating heme, it is possible that CQ may enhance DHA survival by promoting parasite stress responses. Future studies will be needed to test this alternative hypothesis. This limitation has been acknowledged in the manuscript. We have also addressed the reviewer’s point that other factors, including poor pharmacokinetic exposure, contributed to OZ439-PPQ treatment failure.

Reviewer #2 (Public Review):

We appreciate the positive feedback. We agree that there have been previous studies, many of which we cited, assessing interactions of these antimalarials. We also acknowledge that previous work, including our own, has shown that parasite genetics can alter drug-drug interactions. We have included the author’s recommended citations to the list of references that we cited. Importantly, our work was unique not only for utilizing a pulsing format, but also for revealing a superantagonistic phenotype, assessing interactions in an RSA format, and investigating a mechanism to explain these interactions. We agree with the reviewer that implications from this in vitro work should be cautious, but hope that this work contributes another dimension to critical thinking about drug-drug interactions for future combination therapies. We have modified the manuscript to temper any unintended recommendations or implications.

The reviewer notes that we conclude “artemisinins are predominantly activated in the cytoplasm”. We recognize that the site of artemisinin activation is contentious. We were very clear to state that our data combined with others suggest that artemisinins can be activated in the parasite cytoplasm. We did not state that this is the primary site of activation. We were clear to point out that technical limitations may prevent Ac-H-FluNox signal in the digestive vacuole, but determined that low pH alone could not explain the absence of a digestive vacuole signal.

With regard to the “reproducibility” and “mechanistic definition” of superantagonism, we observed what we defined as a one-sided superantagonistic relationship for three different parasites (Dd2, Dd2 PfCRT^Dd2, and Dd2 K13^R539T) for a total of nine independent replicates. In the text, we define that these isoboles are unique in that they had mean ΣFIC50 values > 2.4 and peak ΣFIC50 values >4 with points extending upward instead of curving back to the axis. As further evidence of the reproducibility of this relationship, we show that CQ has a significant rescuing effect on parasite survival to DHA as assessed by RSAs and IC50 values in early rings.

Reviewer #3 (Public Review):

We thank the reviewer for their positive feedback. We acknowledge that no combinations tested in this manuscript were synergistic. However, two combinations, DHA-MFQ and DHA-LM, were additive, which provides context for contextualizing antagonistic relationships. We have previously reported synergistic and additive isobolograms for peroxide-proteasome inhibitor combinations using this same pulsing format (Rosenthal and Ng 2021). These published results are now cited in the manuscript.

We believe that these findings are specific to 4-aminoquinoline-peroxide combinations, and that these findings cannot be generalized to antimalarials with different mechanisms of action. Note that the aryl amino alcohols, MFQ and LM, were additive with DHA. Since the mechanism of action of MFQ and LM are poorly understood, it is difficult to speculate on a mechanism underlying these interactions.

We agree with the reviewer that while the heme probe may provide some mechanistic insight to explain DHA-quinoline interactions, there is much more to learn about CQ-heme chemistry, particularly within parasites.

The focus of this manuscript was to add a new dimension to considerations about pairings for combination therapies. It is outside the scope of this manuscript to suggest alternative combinations. However, we agree that synergistic combinations would likely be more strategic clinically.

An in vitro setup allows us to eliminate many confounding variables in order to directly assess the impact of partner drugs on DHA activity. However, we agree that in vivo conditions are incredibly more complex, and explicitly state this.

We agree that in the future, modeling studies could provide insight into how antagonism may contribute to real-world efficacy. This is outside the scope of our studies.

Recommendations for the Authors:

Reviewer #1 (Recommendations for the Authors):

The key weaknesses identified in this manuscript are described in the 'weaknesses' section of the public review. The major one is the inconsistency around the H-FluNox response in the chemical vs biological experiments. I can't think of a simple experiment to resolve this issue, but it is good that this data is openly provided in the manuscript. I believe there could be more discussion to clarify this limitation with the current study, and the conclusions, and particularly the title, should be softened regarding the mechanism of antagonism being based on heme reactivity.

We have softened the title and conclusions to take into account the limitations of our studies.

(1) Please double-check the definitions for isobologram interpretation. In most antimicrobial interaction studies, I see the threshold for antagonism at sumFIC50 of 1.5, or even 2. 1.25 is often interpreted as additive in many studies.

We acknowledge that different studies use various cutoff values. Our interpretations for additive versus antagonistic versus superantagonistic were based not only on mean ΣFIC50 values, but also isobologram shape. For example, the flat isoboles for MFQ-DHA were clearly distinct from the curved isoboles of PPQ-DHA. It is unclear what cutoff value(s) would be most clinically relevant.

(2) For the MFQ-PPQ interaction study, please make it clear that these drugs have very long half-lives (weeks), so the 4 h pulse assay isn't really relevant to their overall activity. It probably shows a slower onset of action, but there is plenty of drug remaining for many days in the clinical scenario, so perhaps the data from the traditional 48h assay is more relevant. The same consideration applies to OZ439, which may impact the interpretation of that data.

We have now included the half-lives of these compounds in the discussion section. Our intent was to use a pulsing format to make these isobolograms comparable with the other assays. It is important to note that pulses can reveal stronger phenotypes that might be missed with traditional methods. Thus, while 48 h assays may better mimic in vivo conditions, they could also mask important phenotypes.

Reviewer #3 (Recommendations for the Authors):

I have included most of my concerns in the public review. Below are some additional specific points for consideration:

(1) It is expected to include a synergistic combination as a control (e.g., artemisinin + lumefantrine) to contextualize the degree of antagonism observed. The experimental design should show some synergistic profiles in comparison. Adding a few experiments by including a synergistic control is needed.

Both MFQ-DHA and LM-DHA combinations were additive, which provides context for antagonistic combinations. This is now stated in the results section pertaining to Figure 1. We have also included a reference to our previous publication in which we demonstrated that proteasome inhibitor-peroxide combinations are synergistic to additive using this same pulsing format.

(2) Consider in vivo validation or pharmacokinetic/pharmacodynamic modeling to strengthen the translational relevance of the findings when it comes to doses and the IC50 correlations.

We agree that this would be useful to do in future, but it is outside the scope of the current study.

(3) It would be beneficial to include a discussion section on how the findings are generalizable to different Plasmodium falciparum genotypes (3D7, Dd2, MRA-1284) and their relevance.

Findings were consistent across three parasite backgrounds depending on PfCRT genotype. This point has been included in the discussion section. The background of these parasites is also provided in Table 1.

(4) Potential evaluation criteria to understand where certain combinations should be reconsidered can be included as a suggestion for the wider audience.

Our in vitro studies suggest that pulsing isobolograms would be a useful assay to include when evaluating combination therapies. While we believe that synergistic combinations would be more strategic than antagonistic combinations, we cannot provide evaluation criteria or make recommendations for reconsidering currently used combinations.

(5) Further elaborate on the mechanistic basis of heme inactivation by quinolines. If data are available, please include more data on the specificity of the process.

Despite our best efforts, we were unable to evaluate quinoline-heme interactions in parasites. Even in vitro, this interaction has remined elusive for decades. We agree that this would be an important future step towards supporting a specific mechanism for quinoline-DHA antagonism.

eLife Assessment

In this study, the authors identify EOLA1 as a novel mitochondrial protein required for mitochondrial translation and normal cardiac function. The characterization of the molecular role of EOLA1 is still incomplete, and additional controls will be necessary. Nevertheless, the identification of a novel factor critical for mitochondrial gene expression and oxidative phosphorylation will be useful for cell biologists working on mitochondrial dysfunction.

Reviewer #1 (Public review):

Summary:

Mitochondria encode a small set of proteins that are made inside the organelle by specialized ribosomes. When this mitochondrial translation system fails, oxidative phosphorylation is impaired, an outcome that is particularly harmful to energy-demanding tissues such as the heart. In this manuscript, the authors use a targeted CRISPR/Cas9 screen in cultured cells grown on galactose (a condition that forces reliance on oxidative phosphorylation) to identify genes required for mitochondrial activity. They highlight EOLA1, previously studied mainly in inflammatory contexts, as a top candidate.

Strengths:

The authors present data suggesting that EOLA1 is imported into mitochondria via an N-terminal targeting sequence and resides in the mitochondrial matrix. Loss of EOLA1 reduces oxygen consumption and is associated with altered mitochondrial ultrastructure. Mechanistically, affinity purification suggests interaction with mitochondrial elongation factors TUFM (mtEF-Tu), and RNA immunoprecipitation experiments enrich 12S mt-rRNA, consistent with a relationship to the small ribosomal subunit. Multiple assays, including sucrose-gradient profiling, reduced abundance of selected mtDNA-encoded proteins, and a click-chemistry labeling approach, support the conclusion that mitochondrial protein synthesis is decreased in EOLA1-deficient cells. Finally, whole-body Eola1 knockout mice show echocardiographic findings consistent with dilated cardiomyopathy and reduced levels of representative mitochondrially encoded proteins in cardiac tissue.

How to interpret the work:

The data support a role for EOLA1 in maintaining mitochondrial gene expression and oxidative phosphorylation capacity, and they plausibly implicate mitochondrial translation.

Weaknesses:

The main caveat is that the study does not yet establish how EOLA1 acts, whether it directly modulates translation elongation through TUFM, whether it is primarily required for mitoribosome biogenesis/rRNA stability, or whether it influences translation indirectly through mitochondrial stress pathways. The in vivo phenotype is intriguing, but without tissue-specific deletion/rescue and deeper cardiac pathology/mitochondrial functional measurements, it remains uncertain how directly the heart phenotype reflects a cardiomyocyte-autonomous defect in mitochondrial translation.

Reviewer #2 (Public review):

Summary:

In this study, the authors identify a previously uncharacterised regulator of mitochondrial function using a genetic screen and propose a role for this protein in supporting mitochondrial protein production. They provide evidence that the protein localises to mitochondria, interacts with components of the mitochondrial translation machinery, and is required for normal heart function in an animal model.

Strengths:

A major strength of the work is the use of multiple independent approaches to assess mitochondrial activity and protein production, which together provide support for the central conclusions. The in vivo data linking loss of this factor to impaired heart function are particularly compelling and elevate the relevance of the study beyond a purely cell-based context.

Weaknesses:

Given prior reports placing this protein outside mitochondria, its mitochondrial localisation would benefit from more rigorous and quantitative validation, and the proposed mechanism of the interaction with the mitochondrial translation machinery remains only partially explored. In addition, the physiological analysis is largely limited to the heart, leaving open questions about how broadly this pathway operates across tissues.

Major comments:

(1) Evidence for mitochondrial localization of EOLA1<br /> EOLA1 has previously been reported as a nuclear and cytosolic protein and is not annotated in MitoCarta 3.0, making rigorous validation of its mitochondrial localization particularly important. Although the authors provide several lines of evidence, interpretation is complicated by the use of different cell lines across localization, interaction, and functional experiments. Greater consistency in the cellular models used would strengthen the conclusions. The immunofluorescence analysis of tagged EOLA1 would also benefit from quantification across more cells and the inclusion of an additional mitochondrial marker (e.g., an outer membrane marker such as TOM20), as HSP60 staining can vary with mitochondrial state.

(2) Normalization of OCR measurements<br /> Clarification of how Seahorse oxygen consumption rate measurements were normalized (e.g., cell number or protein content) would aid interpretation, particularly given potential effects of Eola1 loss on cell growth.

(3) Linking interaction data to functional phenotypes<br /> Loss-of-function analyses are performed in mouse cell lines, whereas localization and interactome studies are conducted in human HEK293T cells. The absence of a human EOLA1 knockout model makes it difficult to directly connect the interaction data to the observed functional phenotypes. Additional validation or discussion of species conservation would improve clarity.

(4) Mechanistic interpretation of the EOLA1-TUFM-12S rRNA interaction<br /> The identification of TUFM and 12S mt-rRNA as EOLA1 interactors is an interesting finding; however, the basis for prioritizing TUFM among the many mitochondrial proteins identified in the interactome is not fully explained. Providing enrichment statistics and functional categorization of mitochondrial interactors would increase transparency. In addition, the proposed role of the ASCH domain in RNA binding would be strengthened by structure-informed or mutational analysis of the conserved RNA-binding motif.

(5) Interpretation of mitochondrial translation and protein abundance data<br /> Several assays supporting impaired mitochondrial translation would benefit from additional controls and quantification. The de novo mitochondrial translation assay (Fig. 3h) is not quantified, making it difficult to assess the magnitude and reproducibility of the effect. In addition, western blots showing reduced levels of mitochondrially encoded OXPHOS subunits (Figure 3g) lack a mitochondrial loading control (e.g., TOM20 or VDAC). Since loss of EOLA1 may affect mitochondrial mass, normalization to a mitochondrial marker is necessary. Relatedly, it would be informative to assess whether steady-state levels of mitoribosomal proteins (e.g., MRPS15, MRPL37) and nuclear-encoded OXPHOS subunits are altered upon Eola1 loss, both in knockout cell lines and in the knockout mouse.

(6) Physiological scope of the in vivo analysis<br /> The cardiac phenotype observed in the whole-body Eola1 knockout mouse is compelling, but the focus on a single tissue limits interpretation of EOLA1's broader physiological role. Examination of additional high-energy-demand tissues would help clarify whether the observed effects are heart-specific or more general. In addition, the presence of residual EOLA1 protein bands in western blots (Figure 4a) and remaining Eola1 transcripts in qRT-PCR analyses (Extended Figure 4e) from knockout tissues should be addressed. The authors should clarify whether these signals reflect incomplete knockout, alternative isoforms, antibody cross-reactivity, or technical background.

(7) Relationship to previously reported MT2A interaction<br /> Given prior reports of EOLA1 interaction with MT2A, a brief comment on whether MT2A was detected in the authors' co-immunoprecipitation experiments and how this relates to the proposed mitochondrial role would be useful.

Reviewer #3 (Public review):

The authors identified EOLA1 in a CRISPR/Cas9 screen for essential mitochondrial genes in a mouse B16-F10 cell line; however, no information on the library used for this screen or the list of all identified essential genes is provided. What was the p-value for EOLA1 in Figure 1b?

The authors show that EOLA1 is indeed a mitochondrial protein (using both mouse and human cell lines). It is valuable that the authors use different cell lines to investigate the function of this protein; however, this also presents a challenge, as four different cell lines (two mouse and two human) are used across individual experiments, with no consistency between them. Knock-out (KO) experiments were performed in mouse cell lines only, and human cell lines were used in overexpression experiments, in which EOLA1 was tagged with FLAG-HA. It would be beneficial if a knock-out were also generated in a human cell line to confirm the effect on the expression of mitochondria-encoded proteins, along with a rescue experiment in which the EOLA1 protein is reintroduced into KO cells.

Functional analysis of EOLA1: The authors performed affinity immunoprecipitation of FLAG-HA-tagged EOLA1 from stably overexpressing cells, and identified 202 co-immunoprecipitating proteins, of which 71 were known mitochondrial proteins; however, no list of these proteins is provided. Why did the authors choose TUFM? Were any mitochondrial ribosomal proteins co-immunoprecipitated, if EOLA1 is suggested to regulate translation? Were levels of TUFM affected in EOLA1-KO cells?

The authors continued to analyze mitochondrial ribosomes using sucrose gradient fractionation and in-vitro mitochondrial translation. However, there are several technical problems with the presented data: It has been established that mitochondrial ribosomes do not form polysomes in mammalian cells but rather perform translation as monosomes. The authors indirectly confirm this: almost no 12S or 16S rRNA (Fig. 3f) or MRP proteins (Extended data 3c) are present in "polysome" fractions. Although indeed 12S and 16S rRNAs are decreased in monosome fractions, the levels of mRNAs are not different between KO and WT cells, and neither is the migration of mitochondrial ribosomal proteins. As there is no loading control provided for the sucrose gradients blots (such as SDHA, VDAC), it is not possible to assess the overall levels of mitochondrial ribosomes. The gel presented for mitochondrial translation is of poor quality, as it is impossible to identify any of the expected 13 polypeptides. Although the intensity of the signal is weaker for KO, so is the intensity in the portion of Coomassie stained gel. A better-quality gel and quantification need to be provided to support the claims.

What is the difference between endogenous and exogenous RIP-qPCR? EOLA1 pulled down 12S rRNA without cross-linking (Figure 3d) or with UV-crosslinking (Figure 3e), however, both 12S and 16S rRNAs were enriched in UV-crosslinked cells (Figure 3c) and by UV-RIP-seq (Extended data 3b; although no control is provided here). Is no discussion offered for this observation? Is it possible that EOLA1 plays a role in the maturation of the mito-ribosome, rather than translation? Does EOLA1 co-migrate with the mito-ribosome on sucrose gradients?

Altogether, there is insufficient evidence to support the conclusion that EOLA1 plays a role in mitochondrial translation.

To investigate EOLA1 biological function, the authors created a whole-body EOLA1-/- mouse that exhibited no overall developmental abnormalities; however presented with an abnormal cardiac function. This is an ideal model to confirm prior observations in cellular models; however, apart from one western-blot for three mitochondrial encoded subunits, no other experiments were provided (such as measurements of the levels of 12S, or 16S rRNA, TUFM levels, ribosomes profile, mitochondrial translation, OXPHOS assembly, respirometry).

In Figure 2 g-i: TEM images are presented, but the method is not described, nor is any information on the cells used provided, nor is it clear how the circularity was determined. KO cells certainly look abnormal; however, are the authors sure that the indicated structures are mitochondria? They rather resemble autophagosomes/lysosomes with lamellar inclusions.

Reviewer #3 (Public review):

This work by Du et al. addresses a critical problem in cryo-electron microscopy. To date, there are few ways of generating phase contrast during cryo-EM imaging while remaining in focus. Cryo-EM practitioners today must generate contrast by collecting out-of-focus exposures, a process that introduces aberrations in the resulting image data. Recent work has shown that standing wave lasers are capable of using the ponderomotive effect to shift the phase of electrons in transmission electron microscopy to generate in-focus phase contrast imaging for cryo-EM. A limitation of this 'laser phase plate' is the high laser power required, which can damage optical mirrors and necessitate high laser safety. Thus, alternative approaches are needed for phase contrast imaging in cryo-EM.

In this manuscript, Du et al. exploit their expertise in ultrafast electron microscopy to explore the ability to shift the phase of electrons using pulsed electrons and lasers. The motivation for exploring pulsed laser phase plates stems from the fact that femtosecond pulses from 9W lasers can generate extremely high power (as much as the standing-wave laser phase plate, > 1 gigawatt) at the back focal plane. If successful, this type of instrument will likely be much more affordable and easier to deploy worldwide.

The work outlined here shows a proof of principle, highlighting that an ultrafast scanning electron microscopy beam at 30 kV can have the electron packets phase shift by 430 radians (24637 degrees), which is much greater than the required 1.5 radians (90 degrees) needed for phase contrast imaging. The data presented do not use any biological samples; instead, they measure the spread of the electron beam on a test sample to assess the ability to target pulsed lasers onto electron packets and the amount of electron spread (which relates to the phase shift). They were also able to take their system a step further to measure how changes to the system in terms of laser power affect performance, and show that the system can be stable for 10+ hours.

The only weaknesses relate to the broad readability of the text. Improved textual clarity will help ensure a wider readership.

Overall, this work is an important step toward developing lower-cost alternatives to the standing-wave laser phase plate.

Reviewer #2 (Public review):

Summary:

In this manuscript, the authors present the development and characterization of a pulsed ponderomotive phase plate for transmission electron microscopy (TEM). The primary goal is to overcome the long-standing challenge of generating stable, tunable phase contrast for weakly scattering biological specimens - a capability that has remained elusive despite decades of development. While the commercially available Volta Phase Plate offers phase enhancement, it suffers from a lack of control and stability. More recent efforts have focused on continuous-wave (CW) laser phase plates; however, these systems face significant practical hurdles, including extreme optical power requirements, thermal instability of mirrors, and the necessity for high-finesse optical cavities that act as diffraction gratings for the electron beam. The authors aim to demonstrate that a pulsed, free-space laser interaction can circumvent these limitations, offering a more robust path toward practically usable phase plates

Strengths:

The most significant strength of this work is the elegant use of a free-space pulsed interaction, which fundamentally simplifies the hardware requirements compared to cavity-based designs. By utilizing a high-intensity pulsed laser focus rather than a standing wave inside a resonator, the authors eliminate the need for complex locking feedback loops and avoid the thermal mirror deformation that currently limits CW systems.

Furthermore, this approach provides a critical theoretical advantage regarding image quality. Current CW cavity-based designs must grapple with the Kapitza-Dirac effect, where the standing wave creates a diffraction grating that generates unwanted "ghost images," delocalizing the signal. Recent proposals have had to resort to complex crossed-beam geometries to mitigate these artifacts. In contrast, the traveling-wave nature of the pulsed interaction described here inherently avoids the creation of a standing wave grating, thereby eliminating ghost images entirely without requiring elaborate compensation strategies.

The authors successfully demonstrate a proof-of-concept implementation, reporting a pronounced peak phase shift of approximately 430 radians and a stable angular deflection of the electron beam. The stability data, covering a 10-hour period, suggests that this approach is robust enough for data collection sessions typical in structural biology.

Weaknesses:

However, the strength of the evidence is modestly tempered by limitations in data presentation and analysis. The agreement between the experimental data and the theoretical simulation in Figure 2b is imperfect; the simulation underestimates the depth of the central signal trough. While the authors acknowledge this "muted" prediction, the discrepancy suggests that the theoretical model or the estimation of experimental parameters (such as electron beam size or laser intensity) requires refinement to fully describe the interaction.

While the authors claim stability over many hours, the data in Figure 3c reveal a significant drift in the baseline reference signal. Although attributed to a weakening electron beam, this drift complicates the reader's ability to assess the true stability of the laser-induced phase shift. A drift-corrected analysis would have provided more compelling evidence of the "stable angular kick" described.

Despite these specific weaknesses in data presentation, the work represents a fundamental step forward. The authors have effectively demonstrated that the trade-off between beam current and spatiotemporal resolution (driven by space-charge effects) can be managed to achieve significant phase modulation. By moving the field away from the tight constraints of optical cavities and toward free-space pulsed interactions, this work establishes a potentially more viable route for integrating laser phase plates into routine biological imaging workflows. This study will be of high value to biophysicists and microscopists seeking to push the boundaries of contrast in cryo-EM

Reviewer #1 (Public review):

Summary:

Du, Daniel X. et al studied the interaction of the ultrashort electron and laser pulses inside a scanning electron microscopy (SEM), aiming to build a foundation for pulsed laser phase plate electron microscopy, in which the contrast of cryo samples can be significantly increased. The author modified a commercial SEM to accommodate optics to introduce a laser beam inside the instrument to overlap with the electron beam and performed multiple experiments aimed to characterize the electron-light interaction, particularly reaching an extremely high phase shift of >400 rad. Moreover, the authors built a theoretical model for this interaction and estimated the laser beam parameters needed to reach 90 degrees phase shift in transmission electron microscopy (TEM).

Strengths:

The conclusion on the interaction of the electron pulses and laser pulses is well described and supported by the experiment.

The presented instrument can serve as a great tool for studying fundamental interactions of electrons with extremely intense light pulses.

Weaknesses:

The authors motivate the project by using the pulsed electron beam with a phase shift for improving the contrast in cryo-EM, and while they indicate the low current in UEM, they do not discuss the limitations of the laser beam properties.

Such, even for 1 ps electron pulses with the repetition rate of 100 GHz (duty cycle of 10%), they will need to use 100 GHz laser pulses with pulse energies of at least ~1 uJ a second (the lowest pulse energy reported in the simulations in Figure 4), which would mean that ~10 kW of optical power needs to enter the electron microscope and be dumped somewhere after leaving the instrument. This significantly complicates the system and, in my view, makes it harder to use a pulsed laser phase plate in cryo-EM due to either low acquisition rate at lower repetition rates or extreme difficulties to operate multi kW ultrafast laser system.

I would also expect the unscattered electron beam diameter to be <1 micron, which would significantly change the plot in 4b for the 300 keV electron beam.

Adding experimental parameters for a typical cryo-EM experiment with the pulsed phase plate, including the repetition rate, electron pulse duration, number of electrons per pulse, electron beam size, and the parameters of the laser beam (wavelength, laser pulse duration, pulse energy), will help readers better understand technical requirements for the proposed cryo-EM experiments.

eLife Assessment

This important study introduces a pulsed laser phase plate that generates stable phase contrast in electron microscopy, offering a practical alternative to continuous-wave designs that suffer from optical instabilities and diffraction artifacts. The experimental results demonstrate a controllable and stable electron phase shift, and the evidence supporting the feasibility of this approach for phase-contrast electron microscopy is convincing. Clarifying the agreement between experiment and theory and further elaborating on possible applications would strengthen the manuscript.

eLife Assessment

This important study reports an endometrial organoid culture system mimicking the window of implantation. The evidence supporting the conclusion drawn is convincing. The data will be of interest to embryologists and investigators working on reproductive biology and medicine.

Reviewer #2 (Public review):

Zhang et al. have developed an advanced three-dimensional culture system of human endometrial cells, termed a receptive endometrial assembloid, that models the uterine lining during the crucial window of implantation (WOI). During this mid-secretory phase of the menstrual cycle, the endometrium becomes receptive to an embryo, undergoing distinctive changes. In this work, endometrial cells (epithelial glands, stromal cells, and immune cells from patient samples) were grown into spheroid assembloids and treated with a sequence of hormones to mimic the natural cycle. Notably, the authors added pregnancy-related factors (such as hCG and placental lactogen) on top of estrogen and progesterone, pushing the tissue construct into a highly differentiated, receptive state. The resulting WOI assembloid closely resembles a natural receptive endometrium in both structure and function. The cultures form characteristic surface structures like pinopodes and exhibit abundant motile cilia on the epithelial cells, both known hallmarks of the mid-secretory phase. The assembloids also show signs of stromal cell decidualization and an epithelial mesenchymal transition, like process at the implantation interface, reflecting how real endometrial cells prepare for possible embryo invasion.

Although the WOI assembloid represents an important step forward, it still has limitations: the supportive stromal and immune cell populations decrease over time in culture, so only early-passage assembloids retain full complexity. Additionally, the differences between the WOI assembloid and a conventional secretory-phase organoid are more quantitative than absolute; both respond to hormones and develop secretory features, but the WOI assembloid achieves a higher degree of differentiation due to the addition of "pregnancy" signals. Overall, while it's a reinforced model (not an exact replica of the natural endometrium), it provides a valuable in vitro system for implantation studies and testing potential interventions, with opportunities to improve its long-term stability and biological fidelity in the future.

[Editors' note: the authors have responded to the previous round of recommendations.]

Author response:

The following is the authors’ response to the previous reviews

Public Reviews:

Reviewer #1 (Public review):

Summary:

This study generated 3D cell constructs from endometrial cell mixtures that were seeded in the Matrigel scaffold. The cell assemblies were treated with hormones to induce a "window of implantation" (WOI) state. Although many bioinformatic analyses point in this direction, there are major concerns that must be addressed.

Strengths:

The addition of 3 hormones to enhance the WOI state (although not clearly supported in comparison to the secretory state).

Comments on revisions:

The authors did their best to revise their study according to the Reviewers' comments. However, the study remains unconvincing, incomplete and at the same time still too dense and not focused enough.

Reviewer #2 (Public review):

Zhang et al. have developed an advanced three-dimensional culture system of human endometrial cells, termed a receptive endometrial assembloid, that models the uterine lining during the crucial window of implantation (WOI). During this mid-secretory phase of the menstrual cycle, the endometrium becomes receptive to an embryo, undergoing distinctive changes. In this work, endometrial cells (epithelial glands, stromal cells, and immune cells from patient samples) were grown into spheroid assembloids and treated with a sequence of hormones to mimic the natural cycle. Notably, the authors added pregnancy-related factors (such as hCG and placental lactogen) on top of estrogen and progesterone, pushing the tissue construct into a highly differentiated, receptive state. The resulting WOI assembloid closely resembles a natural receptive endometrium in both structure and function. The cultures form characteristic surface structures like pinopodes and exhibit abundant motile cilia on the epithelial cells, both known hallmarks of the mid-secretory phase. The assembloids also show signs of stromal cell decidualization and an epithelial mesenchymal transition, like process at the implantation interface, reflecting how real endometrial cells prepare for possible embryo invasion.

Although the WOI assembloid represents an important step forward, it still has limitations: the supportive stromal and immune cell populations decrease over time in culture, so only earlypassage assembloids retain full complexity. Additionally, the differences between the WOI assembloid and a conventional secretory-phase organoid are more quantitative than absolute; both respond to hormones and develop secretory features, but the WOI assembloid achieves a higher degree of differentiation due to the addition of "pregnancy" signals. Overall, while it's a reinforced model (not an exact replica of the natural endometrium), it provides a valuable in vitro system for implantation studies and testing potential interventions, with opportunities to improve its long-term stability and biological fidelity in the future.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

This study generated 3D cell constructs (i.e., assembloids) that were treated with hormones to induce a 'window of implantation' (WOI) state. While the authors have made large efforts to address the reviewers' feedback, the study's findings remain unconvincing and incomplete.

(1) The authors have appropriately revised the terminology from 'organoids' to 'assembloids' in several parts of the manuscript. However, this revision remains incomplete, as the main title, figure legends, and figure titles still contain the incorrect term. A thorough review of the entire manuscript is recommended to ensure consistent and accurate use of terminology.

Thank you for your meticulous review. We have now conducted a full check and confirmed that terminology is used consistently and accurately throughout the text.

(1) Previous comments raised concerns about the feasibility of robustly passaging assembloid structures - comprising epithelial, stromal and immune cells - under epithelial growth conditions. The authors responded by stating that they optimized the expansion medium with a stromal cell-promoting factor. Additionally, rather than conducting scRNA-seq on both early and late passages (P6-P10) as suggested, they performed immunofluorescence staining, which confirmed the persistence of stromal cells at passage 6. However, the presence of immune cells was not addressed. Confirmation of their presence is essential for all further claims. Moreover, a more zoomed-out view of the immunostaining would help clarify the overall cellular composition across the entire well and facilitate comparison with corresponding brightfield images.

Whole-mount immunofluorescence of the 6th - generation assembloids revealed that CD45⁺ immune cells surrounded FOXA2⁺ glands, with a more zoomed-out view provided.

Author response image 1.

Whole-mount immunofluorescence showed that CD45⁺ cells (immune cells) were arranged around the glandular spheres that were FOXA2⁺. Scale bar =50 μm (left) and 30 μm (right).

In their response, the authors mention using the first three passages to ensure optimal cell diversity and viability. However, the manuscript states that 'assembloids derived from the first generation are used for experiments' (line 106). This discrepancy must be clarified.

Thank you for your suggestion. We have revised the relevant content to “The assembloids derived from the first three generation are used for experiments” (Line 90-91).

(2) The authors have made a commendable effort to bring more focus to the manuscript, which has improved readability.

We thank you for your insightful suggestions, which have greatly improved the quality of our manuscript.

(3) The "embryo implantation" part remains very unconvincing. How did authors define "the blastoids could grow within the endometrial assembloids and interact with them"? What did they mean with "grow"? Did blastoids further differentiate? Normally, blastoids cannot further "grow". "Survival rates of blastoids" is not equal to "growth". It is not clear how the survival rate was quantified. Besides, regarding the "interaction rates", how did authors define and quantify it? Actually, blastoids are able to attach to Matrigel efficiently (even without any endometrial cells), so authors cannot simply define the "interaction" as the co-localization of blastoids and assembloids via brightfield images. In addition, for the assembloids as the 3D structures grow in the Matrigel, the epithelial parts are normally apical-in, while the blastoids attach to the apical (lumen) side of the epithelial cells, so physiologically, blastoids should interact with the apical part of the epithelial cells instead of the outside of the assembloids.

(1) What did they mean with "grow"? Did blastoids further differentiate?

On the one hand, volume and morphology undergo continuous dynamic changes; on the other hand, only the inner cell mass and trophectoderm exist at the blastocyst stage, with the ICM further differentiating into OCT4⁺ epiblast and GATA6⁺ hypoblast.

(2) Survival rates of blastoids" is not equal to "growth". It is not clear how the survival rate was quantified.

The definition of "survival rate" is as follows: morphologically, the blastocoel remains noncollapsed and the cell boundaries are distinct (with no obvious cell detachment); molecularly, the markers of epiblast, hypoblast and trophectoderm are expressed. The survival rate is calculated as the ratio of viable embryoids to the total number of embryoids.

(3) Besides, regarding the "interaction rates", how did authors define and quantify it? Actually, blastoids are able to attach to Matrigel efficiently (even without any endometrial cells), so authors cannot simply define the "interaction" as the co-localization of blastoids and assembloids via brightfield images.

The criteria for determining interaction include not only attachment between the blastoids and assembloids observed via brightfield images, but also their sustained tight adhesion against external mechanical perturbations (e.g., medium replacement, immunostaining procedures).

(4) In addition, for the assembloids as the 3D structures grow in the Matrigel, the epithelial parts are normally apical-in, while the blastoids attach to the apical (lumen) side of the epithelial cells, so physiologically, blastoids should interact with the apical part of the epithelial cells instead of the outside of the assembloids.

You are absolutely correct. In vivo, the embryo indeed makes initial contact with the apical side of the epithelial cells. The introduction of the blastoid co-culture model herein is intended to demonstrate that this receptive endometrial assembloids can better support blastoid growth and development.

(4) Previous comments highlighted the absence of distinct shifts in gene expression profiles between SEC assembloids and WOI assembloids, which contrasts with findings from primary endometrial tissue reported by Wang et al. (2020). While the authors have expanded their analysis using the Mfuzz algorithm and identified changes in mitochondria- and cilia-associated genes, the manuscript still lacks evidence of significant transcriptional changes in key WOI marker genes, as described in Wang et al. This discrepancy must be addressed and discussed in greater depth to clarify the biological relevance of their model.

The endometrium in vivo involves complex crosstalk among multiple cell types and is tightly regulated by the hypothalamic-pituitary-ovarian (HPO) axis, thus exhibiting distinct shifts in gene expression during the peri-implantation period.

In our in vitro model, alterations in mitochondria- and cilia-related genes were observed, which to a certain extent demonstrates that these window of implantation (WOI) assembloids possess receptive-phase characteristics and can be employed to investigate WOI-associated scientific questions or conduct in vitro drug screening.

However, substantial efforts are still required to optimize the current model for fully recapitulating the dynamic changes in endometrial gene expression across different phases in vivo, and this aspect is further addressed in the Limitations section of our discussion (Line 342-353).

“However, our WOI endometrial assembloids also exhibit some limitations. It is undeniable that the assembloids cannot perfectly replicate the in vivo endometrium, which comprises functional and basal layers with a greater abundance of cell subtypes, under superior regulation by hypothalamic-pituitary-ovarian (HPO) axis. Specifically, stromal and immune cells are challenging to stably passage, and their proportion is lower than in the in vivo endometrium. While the in vivo peri-implantation period exhibits intricate gene expression dynamics driven by systemic regulation, our models only partially recapitulate these changes, primarily in mitochondria- and cilia-associated genes. Nevertheless, to some extent, these WOI assembloids possess receptivity characteristics and can be utilized for investigating receptivity-related scientific questions or conducting in vitro drug screening. Further refinements are required to fully simulate the dynamic endometrial gene expression patterns across all menstrual cycle stages. We are looking forward to integrating stem cell induction, 3D printing, and microfluidic systems to modify the culture environment.”

(5) In the authors' response document, they present data integrating their results with those of Garcia Alonso et al. (2021). However, these integrated analyses are not included in the revised manuscript (which should be, if answering a major concern).

Thanks for your valuable suggestions. We have now integrated the findings of Garcia Alonso et al. (2021) into the revised manuscript (Line 132) and Figure S2E–F.

(8) Fig 2D: The authors have clarified that CD45+ staining is used. However, they have not yet adapted the typo in the figure legend of the right picture.

Thanks for your thorough review. The left panel of Figure 2D is stained with CD45 to label immune cells, while the right panel is stained with CD44. These details have been clearly indicated in both the manuscript and the figure legend.  

(9) All quantification analyses (as described in the authors' response document) should be clearly described in the Materials & Methods section.  

Thanks for your valuable suggestions. All quantification analyses have now been added to the Supporting Materials and Methods section (Line 94-104, Line 110-111, Line 241244).

(10) The authors have provided clarification regarding their method for quantifying immunofluorescence staining (e.g., OLFM4 expression in Fig. 3C) in their response document. However, these methodological details are not included in the revised manuscript. It is important that such information is incorporated into the manuscript itself to ensure transparency and reproducibility for others.

Thanks for your valuable suggestions. All quantification analyses have now been added to the Supporting Materials and Methods section (Line 94-104).

(13) It is needed to include the author's response to the comment about literature showing the opposite of increased number of cilia during the WOI into the discussion part of the paper.

We appreciate your suggestions. The relevant content has now been added to the Discussion section (Lines 319–323).

(14) In the authors' response, they explain the difference between pinopodes and microvilli. They should include this explanation briefly in the manuscript. Moreover, Fig. 3F lacks a picture of cilia structure in CTRL condition. In addition, the structures that are indicated as cilia with an orange arrow seem to not be attached to the endometrial cells (anymore). It would be useful to show another more representative picture for the cilia.

(1) Thank you for your valuable suggestions. The distinction between pinopodes and microvilli has now been added to the Supporting Materials and Methods section (Line 230-236).

(2) You are probably referring to Figure 2F—we did not observe ciliary structures in the CTRL group.

(3) The cilia structure was visualized via transmission electron microscopy (TEM), which requires ultrathin sectioning. Thus, the cilia shown in the image correspond to a single cross-section of the captured assembloids. Owing to technical limitations, three-dimensional visualization of cilia on the cells cannot be achieved.

(17) The results on co-culturing blastoids with the WOI assembloids is not convincing. The blastoids are exposed to the basolateral side of the endometrial epithelial cells, while in vivo, blastocysts interact with the apical side of the endometrial epithelial cells first (apposition and attachment), followed by invasion into the endometrium. This means that the interaction shown here is not physiological. Therefore, it is not justified to say that this platform holds promise to investigate maternal-fetal interactions.

We agree with your perspective that discrepancies exist between this model and the physiological processes in vivo. However, such differences do not negate the scientific value of the model.

The core merit of this study lies in the successful establishment of co-culture systems for blastoids and WOI assembloids. Notably, genuine cross-talk occurs between the two components, thereby providing a practical and operational tool for subsequent research.

Although the current contact orientation differs from that observed in vivo, future optimization of the cell culture protocol (via modulation of cell polarity) will enable the model to better recapitulate physiological conditions. Therefore, the innovation and operability of this model within specific research contexts still render it a robust platform for investigating maternal-fetal interactions.

Overall, it is highly recommended that the authors carefully review the manuscript for grammatical errors, inconsistencies and issues with scientific phrasing. The language throughout the text requires substantial editing to improve clarity, readability and precision. 

We appreciate your suggestions. A full manuscript check was performed to rectify grammatical errors, inconsistencies, and inappropriate scientific phrasing, with further language refinement by a native English-speaking specialist.

Fig 1A: This overview is unclear. How many days do the assembloids grow before being stimulated with hormones? Are CTRL assembloids only kept in culture until day 2 and SEC and WOI assembloids until day 8? This is also not clear form the Materials and Methods section. Should be clarified.

Thanks for your valuable suggestions. We have now updated the overview (Figure 1A) and Materials and Methods section (Line 370-371, Line 379-381).

“Hormonal treatment was initiated following the assembly of the endometrial assembloids (about 7-day growth period).”

“The CTRL group was cultured in ExM without hormone supplementation and subjected to parallel culture for 8 days along with the two aforementioned groups.”

Fig 1B: From these brightfield images, it appears that the size of the assembloids remains relatively consistent from Day 0 to Day 3 and up to Day 11 (especially in CTRL). However, in Fig S1A, the assembloids on Day 11 appear significantly larger compared to those on Day 2 (or Day 4). Authors should clarify this discrepancy (since both of the figures are shown as "brightfield of endometrial assembloids").

You are probably referring to the observation that the assembloids at Day 11 in Fig. S1A are smaller in size than those at Day 2 (or Day 4) in Fig. 1B. This discrepancy arises because the time points in Fig. 1B are calculated starting from the initiation of hormone treatment for the SEC and WOI groups, rather than from the beginning of the overall culture as in Fig. S1A. In addition, assembloids exhibit size variability during the same culture period due to individual heterogeneity.

To eliminate ambiguity, we have now labeled “Hormone Day 0, Day 2, Day 8” in Fig. 1B and revised the corresponding figure legend to read: “Endometrial assembloids from the CTRL, SEC, and WOI groups, which were subjected to hormone treatment on Days 0, 2, and 8, exhibited comparable growth patterns throughout the culture period.”

Fig 2G: authors still used the description "organoids" here instead of "assembloids".

We appreciate your careful review. Corrections have been made accordingly.

Fig. 3C: For the OLFM4 staining quantification, in the Y-axis authors wrote "proportion of OLFM4 (+) cells (OLFM4 (+)/total", but in the rebuttal letter they mention "its fluorescence intensity (quantified as mean grey value) was significantly stronger in both the SEC and WOI groups compared to the CTRL group". This is confounding and should be clarified.

We apologize for incorrectly writing "fluorescence intensity" in the rebuttal letter; the correct term should be the "proportion of OLFM4 (+) cells (OLFM4 (+)/total)" as shown in Fig. 3C.

Fig 5D: Acetyl-α-tubulin is the marker of ciliated cells and should be expressed in the cilia instead of the whole cells. It is very strange to quantify as "mean fluorescence intensity (acetyl-αtubulin/DAPI)" to assess the cilia. Please clarify.

Thank you for your insightful comment. To clarify, the ratio "mean fluorescence intensity (acetyl-α-tubulin/DAPI)" was calculated within individual acetyl-α-tubulin⁺ ciliated cells. Acetyl-αtubulin fluorescence was normalized to the DAPI signal of the same cell nucleus, not the wholecell population. This corrected for variations in cell number and staining efficiency to ensure data accuracy.

Fig 5F: it is very bizarre that unciliated epithelium was transformed from ciliated epithelium, and CTRL was transformed from SEC and WOI. Should be clarified and discussed.

Pseudotime analysis sorts discrete cells along a "pseudotime axis" based on similarities and differences in cellular gene expression, thereby simulating cell state transitions.

Ciliated epithelium → unciliated epithelium: During the menstrual cycle, ciliated and unciliated epithelia undergo mutual transformation from the secretory phase (or mid-secretory phase) to the menstrual phase, and then to the proliferative phase. Here, we demonstrate the transition of ciliated cells to unciliated cells from the SEC and WOI stages to the CTRL stage.

Notably, the two cell types coexist, and what is presented here merely reflects a transformation trend. Relative content has been incorporated into the Discussion section (Line 319-321).

“Throughout the menstrual cycle, ciliated and unciliated epithelia undergo mutual transformation from the secretory phase (or mid-secretory phase) to the menstrual phase, and then to the proliferative phase.”

Fig 5H: To show "enhanced invasion ability", authors must provide some quantification and statistic analysis. It is very hard to see the difference between the CTRL and SEC regarding ROR2Wnt5A.

We appreciate your suggestion. Quantification and statistic analysis have been added to Figure 5H.

Fig 6A: please elaborate the "mIVC1" and "mIVC2" in the figure legends.

Additions have been made to the figure legends accordingly, as follows: "mIVC1: modified In Vitro Culture Medium 1; mIVC2: modified In Vitro Culture Medium 2."

Fig S1D: Is the PAS staining also done in CTRL assembloids? In addition, it is stated that the assembloids secrete glycogen because of a positive PAS staining, while it could also be neutral mucins, glycoproteins, etc, which are all detected by PAS staining. So, the authors should be more careful in stating that it is glycogen, or a PAS staining with diastase digestion should be done.

The PAS staining results for the CTRL group are presented in Fig. S1I. In addition, results of PAS staining with diastase digestion are included in Figure S1.

Line 120: references?

The reference has been added accordingly.

Line 178: The term 'Endometrial Receptivity Test (ERT)' is used. Do the authors mean Endometrial Receptivity Analysis (ERA) test? ERA is the commonly used abbreviation for this test. Moreover, the authors describe ERA as 'a kind of gene analysis-based test.' This should be rephrased more scientifically correct.

Thank you for your valuable suggestion. We have revised the term to ERA, and modified the phrase "a kind of gene analysis-based test" to "gene expression profiling-based diagnostic assay" (Lines 160–163).

“We performed Endometrial Receptivity Analysis (ERA), a gene expression profiling-based diagnostic assay that integrates high-throughput sequencing and machine learning to quantify the expression of endometrial receptivity-associated genes.”

Line 83: assemblies à assembloids

We appreciate your suggestion. The text has been updated to “the endometrial assembloids progressed from epithelial organoids, to assemblies of epithelial and stromal cells and then to stem cell-laden 3D artificial endometrium”.

The Materials and Methods section currently lacks the needed details. Authors should substantially expand this section to clearly describe all experimental and analytical procedures, including, aùmong others, immunofluorescence staining, quantification methods, bioinformatics analyses and statistical approaches. Providing comprehensive methodological information is essential.

A detailed description of these methods is provided in the Supporting Materials and Methods section.

Reviewer #2 (Recommendations for the authors): 

The revised manuscript is much improved in clarity, focus, and experimental support. The authors have thoughtfully addressed the major concerns from the previous review. In particular, the logic and flow of the paper are clearer, it now guides the reader through the rationale (constructing a WOI model), the comparative analysis against in vivo tissue and simpler organoids, and the key features that distinguish the WOI assembloid. The added functional validation (especially the blastoid co-culture experiment) significantly strengthens the work by showing a tangible outcome of "receptivity" beyond molecular profiling. The distinction between the standard secretory-phase organoid and the WOI assembloid is now more convincing, as the authors highlight several specific differences in morphology (more cilia, pinopodes), metabolism, and implantation success that favor the WOI model. The manuscript also reads cleaner with the bioinformatic sections condensed to the most important findings (excess detail was trimmed or moved to supplements) and the rationale for gene/pathway selection explicitly stated.

The manuscript has been significantly strengthened through the addition of functional assays (like the blastoid co-culture), clearer transcriptomic and proteomic data, and detailed analyses of hormone treatments, cilia biology, and stromal and immune cell behavior in early passages. These updates confirm that the WOI assembloid supports embryo attachment and outperforms standard secretory organoids, while integrating external references and clarifications on terminology. Minor suggestions remain, such as clarifying statistical significance and adding functional interpretations for certain observations, but overall, the manuscript is now more robust and biologically convincing.

Remaining points for clarification: There are a few minor points that still merit attention:

- Use of the Endometrial Receptivity Test (ERT): As previously mentioned, if the authors have ERT data for the SEC organoid group, including that information would further support the claim that the WOI assembloid is uniquely receptive. If not, it would be helpful to add a statement clarifying that the ERT was employed specifically as a confirmatory test for the WOI assembloids, rather than as a comparative measure across all groups.

Thank you for your valuable suggestion. We have now supplemented the description in the Supporting Materials and Methods section (Lines 160–162) as follows: “ERA was employed specifically as a confirmatory test for the WOI assembloids, rather than as a comparative measure across all groups.”

- Because the assembloids are created from primary tissue samples, it would be helpful to briefly comment on how consistent the findings were across different patient-derived samples. For example, did all biological replicates show similar expression of receptivity markers and comparable capacity to support blastoid attachment? Although this seems implied, including a sentence in the Methods or Results sections that specifies the number of donor lines tested would help readers assess the model's variability and reproducibility.

We appreciated your advice. The relevant statement has been added to the Supporting Materials and Methods section. (Line 312-313).

“All biological replicates (fourteen individuals) of endometrial assembloids show similar expression of receptivity markers and comparable capacity to support blastoid attachment.”

- The authors mention promising future directions, such as integrating 3D printing and microfluidics to further enhance the model, which is an excellent forward-looking statement. It would also be valuable to suggest the inclusion of additional cell types, like more robust immune cell populations or endothelial components, as future improvements to create an even more comprehensive model of the endometrial lining.

Thank you for your valuable suggestion. 3D printing and microfluidics serve as approaches for introducing multiple cell types. We have supplemented the following statement in the manuscript: “We are looking forward to integrating stem cell induction, 3D printing, and microfluidic systems to modify the culture environment.” (Line 352-353).

We are grateful for your valuable feedback and constructive criticism, which have helped us improve the quality of our work in terms of content and presentation. We have diligently revised the manuscript and made necessary changes. Here, we have attached the revised manuscript, figures, and all supplementary materials for your re-evaluation. Thank you again for your continued support and look forward to your favorable decision.

eLife Assessment

The authors developed a fundamental computational method, which is intended to automatically process bioluminescence imaging-derived tumour images across anatomical regions and over time. This allows quantitative analysis of such data, and the authors applied it to describe the spatiotemporal distribution of tumour cells in response to CD19-targeted CAR-T cells that contained either CD28 or 4-1BB costimulatory domains. Some operational limitations were identified, which relate to the pipeline's reliance on predefined regions of interest instead of aligning signal sites with anatomical information, scaling, and limitations in taking animal pose into account. Overall, the authors provide compelling evidence for the functionality of their computational approach towards automated analysis of bioluminescence imaging data, while applying it to a current topic of wide interest in cell therapy research.