eLife Assessment

This study reports a dynamic association/dissociation between malate dehydrogenase (MDH1) and citrate synthase (CIT1) in Saccharomyces cerevisiae under different metabolic conditions that control TCA pathway flux rate. The research question is timely, the use of the NanoBiT split-luciferase system to monitor protein-protein interactions is innovative, and the significance of the findings is valuable. However, the strength of evidence needed to support the conclusions was found to be incomplete based on a lack of critical control and mechanistic experiments.

Reviewer #1 (Public review):

Summary:

The study by the Obata group characterizes the dynamics of the canonical malate dehydrogenase-citrate synthase metabolon in yeast.

Strengths:

The study is well-written and appears to give clear demonstrations of this phenomenon.

Studies of the dynamics of metabolon formation are rare; if the authors can address the concern detailed below, then they have provided such for one of the canonical metabolons in nature.

Weaknesses:

There is a fundamental issue with the study, which is that the authors do not provide enough support or information concerning the split luciferase system that they use. Is the binding reversible or not? How the data is interpreted is massively influenced by this fact. What are the pros and cons of this method in comparison to, for example, FLIM-FRET? The authors state that the method is semi-quantitative - can they document this? All of the conclusions are based on the quality of this method. I know that it has been used by others, but at least some preliminary documentation to address these questions is required.

Reviewer #2 (Public review):

This study explores the dynamic association between malate dehydrogenase (MDH1) and citrate synthase (CIT1) in Saccharomyces cerevisiae, with the aim of linking this interaction to respiratory metabolism. Utilizing a NanoBiT split-luciferase system, the authors monitor protein-protein interactions in vivo under various metabolic conditions.

Major Concerns:

(1) NanoBiT Signal May Reflect Protein Abundance Rather Than Interaction Strength

In Figure 1C, the authors report increased MDH1-CIT1 interaction under respiratory (acetate) conditions and decreased interaction during fermentation (glucose), as indicated by NanoBiT luminescence. However, this signal appears to correlate strongly with the expression levels of MDH1 and CIT1, raising the possibility that the observed luminescence reflects protein abundance rather than specific interaction dynamics. To resolve this, NanoBiT signals should be normalized to the expression levels of both proteins to distinguish between abundance-driven and interaction-driven changes.

(2) Lack of Causal Evidence

The study presents a series of metabolic perturbation experiments (e.g., arsenite, AOA, antimycin A, malonate) and correlates changes in metabolite levels with NanoBiT signals. However, these data are correlative and do not establish a functional role for the MDH1-CIT1 interaction in metabolic regulation. To demonstrate causality, the authors should implement approaches to specifically disrupt the MDH1-CIT1 interaction. One strategy could involve using a 15-residue peptide (Pept1) derived from the Pro354-Pro366 region of CIT1, previously shown to mediate the interaction, or introducing the cit1Δ3 (Arg362Glu) mutation, which perturbs binding. Metabolic flux analysis using ^13C-labeled glucose and mitochondrial respiration assays (e.g., Seahorse) could then assess functional consequences.

(3) Absence of Protein Expression Controls Under Perturbation Conditions

In experiments involving acetate, arsenite, AOA, antimycin A, and malonate, the authors infer changes in MDH1-CIT1 association based solely on NanoBiT signals. However, no accompanying data are provided on MDH1 and CIT1 protein levels under these conditions. This omission weakens the conclusions, as altered expression rather than interaction strength could underlie the observed luminescence changes. Immunoblotting or quantitative proteomics should be used to confirm constant protein expression across conditions.

Conclusion:

Although the central question is compelling and the use of NanoBiT in live cells is a strength, the manuscript requires additional experimental rigor. Specifically, normalization of interaction signals, introduction of causative perturbations, and validation of protein expression are essential to substantiate the study's claims.

Reviewer #3 (Public review):

Summary:

Metabolons are multisubunit complexes that promote the physical association of sequential enzymes within a metabolic pathway. Such complexes are proposed to increase metabolic flux and efficiency by channeling reaction intermediates between enzymes. The TCA cycle enzymes malate dehydrogenase (MDH1) and citrate synthase (CIT1) have been linked to metabolon formation, yet the conditions under which these enzymes interact, and whether such interactions are dynamic in response to metabolic cues, remain unclear, particularly in the native cellular context. This study uses a nanoBIT protein-protein interaction assay to map the dynamic behavior of the MDH1-CIT1 interaction in response to multiple metabolic stimuli and challenges in yeast. Beyond mapping these interactions in real time, the authors also performed GC-MS metabolomics to map whole-cell metabolite alterations across experimental conditions. Finally, the authors use microscale thermophoresis to determine components that alter the MDH1-CIT1 interaction in vitro. Collectively, the authors synthesize their collected data into a model in which the MDH1-CIT1 metabolon dissociates in conditions of low respiratory flux, and is stimulated during conditions of high respiratory flux. While their data largely support these models, some key exceptions are found that suggest this model is likely oversimplified and will require further work to understand the complexities associated with MDH1-CIT1 interaction dynamics. Nonetheless, the authors put forth an interesting and timely toolkit to begin to understand the interaction kinetics and dynamics of key metabolic enzymes that should serve as a platform to begin disentangling these important yet understudied aspects of metabolic regulation.

Strengths:

(1) The authors address an important question: how do metabolon-associated protein-protein interactions change across altered metabolic conditions?

(2) The development and validation of the MDH1-CIT1 nanoBIT assay provides an important tool to allow the quantification of this protein-protein interaction in vivo. Importantly, the authors demonstrate that the assay allows kinetic and real time assessment of these protein interactions, which reveal interesting and dynamic behavior across conditions.

(3) The use of classic biochemical techniques to confirm that pH and various metabolites can alter the MDH1-CIT1 interaction in vitro is rigorous and supports the model put forth by the authors.

Weaknesses:

(1) Some of the data collected seem to be merely reported rather than synthesized and interpreted for the reader. This is particularly true for data that seem to reflect more complex trends, such as the GC-MS experiments that map metabolites across multiple experiments, or treatments that show somewhat counterintuitive results, such as the antimycin A treatment, which promotes rather than disrupts the MDH1-CIT1 interaction.

(2) Some of the assertions put forth in the manuscript are not substantiated by the data presented, and the authors are at times overly reliant on previous findings from the literature to support their claims. This is particularly notable for claims about "TCA cycle flux"; the authors do not perform flux analysis anywhere in their study and should be cautious when insinuating correlations between their observations and "flux".

(3) The manuscript presentation could be improved. For figures, at times, the axes do not have intuitive labels (example, Figure 1A), data points and details about the number of samples analyzed are missing (bar graphs and box plots), and molecular weight markers are not reported on western blots. The authors refer to the figures out of order in the text, which makes the manuscript challenging to navigate as a reader.

eLife Assessment

This useful study analyzed 335 Mycobacterium tuberculosis Complex genomes and found that MTBC has a closed pangenome with few accessory genes. The research provides solid evidence for gene presence-absence patterns which support the appending conclusions however, the main criticism regarding the dominance of genome reduction remains.

Reviewer #1 (Public review):

Summary:

In this paper, Behruznia and colleagues use long-read sequencing data for 339 strains of the Mycobacterium tuberculosis complex to study genome evolution in this clonal bacterial pathogen. They use both a "classical" pangenome approach that looks at the presence and absence of genes, and a pangenome graph based on whole genomes in order to investigate structural variants in non-coding regions. The comparison of the two approaches is informative and shows that much is missed when focusing only on genes. The two main biological results of the study are that 1) the MTBC has a small pangenome with few accessory genes, and that 2) pangenome evolution is driven by genome reduction. The second result is still questionable because it relies on a method that disregards paralogs.

Strengths:

The authors put together the so-far largest data set of long-read assemblies representing most lineages of the Mycobacterium tuberculosis context, and covering a large geographic area. They sequenced and assembled genomes for strains of M. pinnipedi, L9, and La2, for which no high-quality assemblies were available previously. State-of-the-art methods are used to analyze gene presence-absence polymorphisms (Panaroo) and to construct a pangenome graph (PanGraph). Additional analysis steps are performed to address known problems with misannotated or misassembled genes.

Weaknesses:

The main criticism regarding the dominance of genome reduction remains after two rounds of revisions. A method that systematically excludes paralogs is hardly suitable to draw conclusions about the relative importance of insertions/duplications and deletions in a clonal organism, where any insertion/duplication will result in a paralog. I understand that a re-analysis of the data might not be practical, and the authors have added a few sentences in the discussion that touch on this problem. However, the statements regarding the dominance of genome reduction remain too assertive given this basic flaw.

Here are the more detailed argument from the previous review:

In a fully clonal organism, any insertion/duplication will be an insertion/duplication of an existing sequence and thus produce a paralog. If I'm correctly understanding your methods section, paralogs are systematically excluded in the pangraph analysis. Genomic blocks are summarized at the sublineage level as follows (l.184 ): "The DNA sequences from genomic blocks present in at least one sub-lineage but completely absent in others were extracted to look for long-term evolution patterns in the pangenome." I presume this is done using blastn, as in other steps of the analysis.

So a sublineage-specific copy of IS6110 would be excluded here, because IS6110 is present somewhere in the genome in all sublineages. However, the appropriate category of comparison, at least for the discussion of genome reduction, is orthology rather than homology: is the same, orthologous copy of IS6110, at the same position in the genome, present or absent in other sublineages? The same considerations apply to potential sublineage-specific duplicates of PE, PPE, and Esx genes. These gene families play important roles in host-pathogen interactions, so I'd argue that the neglect of paralogs is not a finicky detail, but could be of broader biological relevance.

Author response:

The following is the authors’ response to the previous reviews

Public Reviews:

Reviewer #1 (Public review):

Summary:

In this paper, Behruznia and colleagues use long-read sequencing data for 339 strains of the Mycobacterium tuberculosis complex to study genome evolution in this clonal bacterial pathogen. They use both a "classical" pangenome approach that looks at the presence and absence of genes, and a pangenome graph based on whole genomes in order to investigate structural variants in non-coding regions. The comparison of the two approaches is informative and shows that much is missed when focussing only on genes. The two main biological results of the study are that 1) the MTBC has a small pangenome with few accessory genes, and that 2) pangenome evolution is driven by genome reduction. In the revised article, the description of the data set and the methods is much improved, and the comparison of the two pangenome approaches is more consistent. I still think, however, that the discussion of genome reduction suffers from a basic flaw, namely the failure to distinguish clearly between orthologs and homologs/paralogs.

Strengths:

The authors put together the so-far largest data set of long-read assemblies representing most lineages of the Mycobacterium tuberculosis context, and covering a large geographic area. They sequenced and assembled genomes for strains of M. pinnipedi, L9, and La2, for which no high-quality assemblies were available previously. State-of-the-art methods are used to analyze gene presence-absence polymorphisms (Panaroo) and to construct a pangenome graph (PanGraph). Additional analysis steps are performed to address known problems with misannotated or misassembled genes.

Weaknesses:

The revised manuscript has gained much clarity and consistency. One previous criticism, however, has in my opinion not been properly addressed. I think the problem boils down to not clearly distinguishing between orthologs and paralogs/homologs. As this problem affects a main conclusion - the prevalence of deletions over insertions in the MTBC - it should be addressed, if not through additional analyses, then at least in the discussion.

Insertions and deletions are now distinguished in the following way: "Accessory regions were further classified as a deletion if present in over 50% of the 192 sub-lineages or an insertion/duplication if present in less than 50% of sub-lineages." The outcome of this classification is suspicious: not a single accessory region was classified as an insertion/duplication. As a check of sanity, I'd expect at least some insertions of IS6110 to show up, which has produced lineage- or sublineage-specific insertions (Roychowdhury et al. 2015, Shitikov et al. 2019). Why, for example, wouldn't IS6110 insertions in the single L8 strain show up here?

In a fully clonal organism, any insertion/duplication will be an insertion/duplication of an existing sequence, and thus produce a paralog. If I'm correctly understanding your methods section, paralogs are systematically excluded in the pangraph analysis. Genomic blocks are summarized at the sublineage levels as follows (l.184 ): "The DNA sequences from genomic blocks present in at least one sub-lineage but completely absent in others were extracted to look for long-term evolution patterns in the pangenome." I presume this is done using blastn, as in other steps of the analysis.

So a sublineage-specific copy of IS6110 would be excluded here, because IS6110 is present somewhere in the genome in all sublineages. However, the appropriate category of comparison, at least for the discussion of genome reduction, is orthology rather than homology: is the same, orthologous copy of IS6110, at the same position in the genome, present or absent in other sublineages? The same considerations apply to potential sublineage-specific duplicates of PE, PPE, and Esx genes. These gene families play important roles in host-pathogen interactions, so I'd argue that the neglect of paralogs is not a finicky detail, but could be of broader biological relevance.

Reviewer #2 (Public review):

Summary:

The authors attempted to investigate the pangenome of MTBC by using a selection of state-of-the-art bioinformatic tools to analyse 324 complete and 11 new genomes representing all known lineages and sublineages. The aim of their work was to describe the total diversity of the MTBC and to investigate the driving evolutionary force. By using long read and hybrid approaches for genome assembly, an important attempt was made to understand why the MTBC pangenome size was reported to vary in size by previous reports. This study provides strong evidence that the MTBC pangenome is closed and that genome reduction is the main driver of this species evolution.

Strengths:

A stand-out feature of this work is the inclusion of non-coding regions as opposed to only coding regions which was a focus of previous papers and analyses which investigated the MTBC pangenome. A unique feature of this work is that it highlights sublineage-specific regions of difference (RDs) that was previously unknown. Another major strength is the utilisation of long-read whole genomes sequences, in combination with short-read sequences when available. It is known that using only short reads for genome assembly has several pitfalls. The parallel approach of utilizing both Panaroo and Pangraph for pangenomic reconstruction illuminated limitations of both tools while highlighting genomic features identified by both. This is important for any future work and perhaps alludes to the need for more MTBC-specific tools to be developed. Lastly, ample statistical support in the form of Heaps law and genome fluidity calculations for each pangenome to demonstrate that they are indeed closed.

Weaknesses:

There are no major weaknesses in the revised version of this manuscript.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

l. 27: "lineage-specific and -independent deletions": it is still not clear to me what a lineage-independent, or convergent, deletion is supposed to be. TBD1, for instance, is not lineage-specific, but it is also not convergent: it occurred once in the common ancestor of lineages 1, 2, and 3, while convergence implies multiple parallel occurrences.

We have changed this and in other places to more evolutionary terms, such as divergent (single event) and convergent (multiple events), or explain exactly what is meant where needed.

l. 118: "where relevant", what does that mean?

This was superfluous to the description and so is now removed.

l. 178ff.: It is not clear to me what issue is addressed by this correction of the pangenome graph. Also here there seems to be some confusion regarding orthologs and paralogs. A gene or IS copy can be present at one locus but absent at another, which is not a mistake of Pangraph that would require correction. It's rather the notion of "truly absent region" which is ambiguous.

We have changed the text to be more specific on the utility of this step. Since it is known that Panaroo mislabels some genes as being absent due to over splitting (see Ceres et al 2022 and our reclassification earlier in the paper), we wanted to see if the same occurred in Pangraph. We have modified the methods text to be more specific (line 181) and in the results included the percentage of total genes/regions affected by this correction.

In relation to copy number, Pangraph is not syntenic in its approach; if a region is present anywhere it is labelled as present in the genome. Pangraph will look for multiple copies of that region (e.g. an IS element) but indeed we did not look for specific syntenic changes across the genomes. This would be a great analysis and something we will consider in the future; we have indicated such in the discussion (line 454).

l. 305: "mislabelled as absent": see above, is this really 'mislabelled'?

See answer to question above

l. 372: "using the approach": something missing here.

This was superfluous to the description and so is now removed.

l. 381: the "additional analysis of paralogous blocks" (l. 381) seems to suffer from the same confusion of ortho- and paralogy described above: no new sub-lineage-specific accessory regions are found presumably because the analysis did consider any copy rather than orthologous copies.

Paralogous copies were looked for by Pangraph, and we did not find any sub-lineage where all members had additional copies compared to other sub-lineages. Indeed, single genomes could have these, and shorter timescales could see a lot of such insertions, but we looked at longer-scale (all genomes within a sub-lineage) patterns and did not find these. These limitations are already outlined in the discussion.

l. 415: see above. There is no diagnosis of a problem that would motivate a "correction". That's different from the correction of the Panaroo results, where fragmented annotations have been shown to be a problem.

Of interest, the refining of regions did re-label multiple regions as being core when Pangraph labelled it as absent from some genomes was at about the same rate as the correction to Pangraph (2% of genes/regions). This indicates there is a stringency issue with pangraph where blocks are mislabelled as absent. The underlying reason or this is not clear but the correction is evidently required in this version of Pangraph.

l. 430ff.: The issue of paralogy and that the "same" gene or region is defined in terms of homology rather than orthology should be addressed here. For me the given evidence does not support the claim that deletion is driving molecular evolution in the MTBC.

As outlined above, indeed paralogy may be driving some elements of the overall evolutionary patterns; our analysis just did not find this. Panaroo without merged paralogs did not find paralogous genes as a main differentiating factor for any sub-lineage. Pangraph also did not find multiple copies of blocks present in all genomes in a sub-lineage. As outlined above, indeed single genomes show such patterns but we did not include single genome analyses here, and outline that as a next steps in the discussion. We have also linked to a recent pangenome paper that showed duplication is present in the pangenome of Mtbc, although not related to any specific lineage (Discussion line 485).

l. 443 ff: "lineage-independent deletions (convergent evolution)": see above, I still think this terminology is unclear

This has now been made clearer to be specifically about convergent and divergent evolutionary patterns.

eLife Assessment

The authors investigate mechanisms of acquired resistance (AR) to KRAS-G12C inhibitors (sotorasib) in non-small cell lung cancer, proposing that resistance arises from signaling rewiring rather than additional mutations. While the study addresses a valuable clinical question, it is limited by several weaknesses in experimental rigor, data interpretation, and presentation, meaning the strength of evidence is incomplete

Reviewer #1 (Public review):

Summary:

In this study, the authors investigate mechanisms of acquired resistance (AR) to KRAS-G12C inhibitors (sotorasib) in NSCLC, proposing that resistance arises from signaling rewiring rather than additional mutations.

Strengths:

Using a panel of AR models - including cell lines, PDXs, CDXs, and PDXOs - they report activation of KRAS and PI3K/AKT/mTOR pathways, with elevated PI3K levels. Pharmacologic inhibition or CRISPR-Cas9 knockout of PI3K partially restores sotorasib sensitivity, and p-4EBP1 upregulation is implicated as an additional contributor, with dual mTORC1/2 inhibition more effective than mTORC1 inhibition alone.

Weaknesses:

While the study addresses an important clinical question, it is limited by several weaknesses in experimental rigor, data interpretation, and presentation. The mechanistic findings are not entirely novel, since the role of PI3K-AKT-mTOR signaling in therapeutic resistance is already well-established in the literature. Rather than uncovering new resistance mechanisms, the study largely confirms known pathways. Several key conclusions are not supported by the data, and critical alternative explanations - such as additional mutations or increased KRAS expression - are not thoroughly investigated or ruled out. Furthermore, while the authors use CRISPR-Cas9 to knock out PI3K and 4E-BP1 in H23-AR and H358-AR cells to restore sotorasib sensitivity, they do not perform reconstitution experiments to confirm that re-expressing PI3K or 4E-BP1 reverses the sensitization. This prevents full characterization of PI3K and p-4EBP1 upregulation as contributors to resistance. The manuscript also has several errors, poor figure quality, and a lack of proper quantification. Additional experimental validation, data improvement, and text revisions are required.

Reviewer #2 (Public review):

Summary:

In this manuscript, the authors focus on the identification of the mechanisms involved in the acquired resistance to Sotorasib in non-small lung KRASG12C mutant cells. To perform this study, the authors generate different clones of cell lines, cell-derived xenografts, patient-derived xenograft organoids, and patient-derived xenografts. In all these models, the authors generate resistant forms (i.e., resistant cell lines PDXs and organoids) and the genetic and molecular changes were characterised using whole-exome sequencing, proteomics, and phospho-proteomics. This analysis led to the identification of an important role of the PI3K/AKT/mTORC1/2 signalling network in the acquisition of resistance in several of the models tested. Molecular characterisation identified changes in the expression of some of the proteins in this network as key changes for the acquisition of resistance, and in particular, the authors show that changes in 4E-BP1 are common to some of the cells downstream of PI3K. Using pharmacological testing, they show that different drugs targeting PI3K, AKT, and MTORC1/2 sensitise some of the resistant models to Sotorasib. The analyses showed that the PI3K inhibitor copanlisib has an effect in NSCLC cells that, in some cases, seems to be synergistic with Sotorasib. Based on the work performed, the authors conclude that the PI3K/mTORC1/2 mediated 4E-BP1 phosphorylation is one of the mechanisms associated with the acquisition of resistance to Sotorasib and that targeting this signalling module could result in effective treatments for NSCLC patients.

The work as presented in the current manuscript is very interesting, provides cell models that benefit the community, and can be used to expand our knowledge of the mechanism of resistance to KRAS targeting therapies. Overall, the techniques and methodology seem to be performed in agreement with standard practice, and the results support most of the conclusions made by the authors. However, there are some points that, if addressed, would increase the value and relevance of the findings and further extend the impact of this work. Some of the recommendations for changes relate to the way things are explained and presented, which need some work. Other changes might require the performance of additional experiments or reanalysis of the existing data.

Strengths:

(1) One of the stronger contributions of this article is the different models used to study the acquisition of resistance to Sotorasib. The resistant cell lines, PDXs and PDXOs, and the fact that the authors have different clones for each, made this collection especially relevant, as they seem to show different mechanisms that the cells used to become resistant to Sotorasib. Although logically, the authors focus on one of these mechanisms, the differential responses of the different clones and models to the treatments used in this work show that some of the clones used additional mechanisms of resistance that can be explored in other studies. Importantly, as they use in vitro and in vivo models, the results also consider the tumour microenvironment and other factors in the response to the treatments.

(2) Another strength is the molecular characterisation of the different Sotorasib-resistant tumour cells by WES, which shows that these cells do not seem to acquire secondary mutations.

(3) The use of MS-based proteomics also identifies proteome signatures that are associated with the acquisition of resistance, including PI3K/mTORC1/2. The combination of proteomics and phospho-proteomics results should allow the identification of several mechanisms that are deregulated in Sotorasib-resistant cells.

(4) The results show a strong response of the NSCLC cells and PDXs to copanlisib, a drug for which there is limited information in this cancer type.

(5) The way they develop the PDX-resistant and the PDXO seems to be appropriate.

Weaknesses:

In general, the data is of good quality, but due to the sheer amount of data included and the way it is presented and discussed, several of the claims or conclusions are not clear.

(1) The abstract is rather long and gives details that are not usually included in one. This makes it very complicated to identify the most relevant findings of the work. The use of acronyms PDX, PDXO, and CDX without defining them makes it complicated for the non-specialist to know what the models are. Rewriting and reorganisation of the abstract would benefit the manuscript.

(2) Expression, presentation, and grammar should be reviewed in all sections of the manuscript.

(3) In the different parts of the result section where the models shown in Figure 2 are described the authors indicate "Whole-exome sequencing (WES) confirmed that XXX model retained the KRASG12C mutation with no additional KRAS mutations detected" however, it is not indicated where this data is shown and in not all the cases there is explanation to other possible modifications that might relate to mechanisms of resistance. This information should be included in the manuscript, and the WES made publicly available.

(4) The way the proteomics analysis of the TC303 and TC314 parental and resistant PDX is described in the text is confusing. The addition of an experimental layout figure would facilitate the understanding. As it is written, it is not obvious that the parental PDX were also analysed. For instance, the authors say, "The global and phosphoproteomic analyses identified over 8,000 and 4,000 gene protein products (GPPs), respectively". Is this comparing only resistant cells, or from the comparison of the parental and resistant pairs? And where are these numbers presented in the figures? Also, there is information that seems more adequate for the materials and methods sections, i.e., "Samples were analyzed using label-free nanoscale liquid chromatography coupled with tandem mass spectrometry (nanoLC-MS/MS) on a Thermo Fusion Mass Spectrometer. The resulting data were processed and quantified using the Proteome Discoverer 2.5 interface with the Mascot search engine, referencing the NCBI RefSeq protein database (Saltzman, Ruprecht). Two-component analysis is better named principal component analysis."

(5) While the presentation of the proteomics data could be done in different ways, the way the data is presented in Figure 3 does not allow the reader to get an idea of many of the findings from this experiment. Although it is indicated that a table with the data will be made available, this should be central to the way the data is presented and explained. A table (ie, Excel doc) where the raw data and all the analysis are presented should be included and referenced. Additionally, heat maps for the whole proteomes identified should be included. In the text, it is said, "Global proteomic heatmap analysis revealed unique protein profiles in TC303AR and TC314AR PDXs compared to their sensitive counterparts (Figure 3C)." However, this figure only shows the histogram of the differentially regulated cells. Inclusion of the histogram showing all the cells is necessary, and it might be informative to include the histogram comparing the two isogenic pairs, which could identify common mechanisms and differences between both sets. In Figure 3C, the protein names should be readable, or a reference to tables where the proteins are listed should be included.

(6) In Figure 3, the pathway enrichment tool and GO used should be mentioned in the text. The tables with all significant tables should also be provided. The proteomics data seems to convincingly identify mTOR as one of the pathways deregulated in resistant cells, but there is little explanation of what is considered a significant FDR value and if there are other pathways or networks that are also modified, which might not be common to both isogenic models. In MS-based Phosphoproteome could help with the identification of differentially regulated pathways, but it is not really presented in the current manuscript. Most of the analysis of phospho-proteomics comes from the RPPA analysis, which is targeted proteomics. With the way the data is presented, the authors show evidence for a role of mTOR in the acquisition of resistance, but unfortunately, they do not discuss or allow the reader to explore if other pathways might also contribute to this change.

(7) Where is the proteomics data going to be deposited, and will it be made public to comply with FAIR principles?

(8) The authors claim that the resistance shown for H23AR and H353AR cells is due to reactivation of KRAS signalling. This is done by looking to phosphorylation of ERK as a surrogate, as they claim, "KRAS inhibition is commonly assessed by evaluating the inhibition of ERK phosphorylation (p-ERK)". While this might be true in many cases, the data presented does not demonstrate that the increase in p-ERK is due to reactivation of KRAS. To make this claim, the authors should measure activation of KRAS (and possibly H- and NRAS) using GST-pull down or an image-based method.

(9) The experiments in Figure 4 are very confusing, and some controls are missing. There is no blot where they show the effect of Sotorasib treatment in H23 and H358 parental cells. Is the increase shown in resistant cells shown in parental or is it exclusive for resistant cells only (and therefore acquired)? Experiment 4B should include this control. What is clear is that there is an increase in the expression of AKT and PI3K.

(10) The main point here is whether this is acquired resistance or the sensitivity to the drug is already there, and there was no need to do an omics experiment to find this. In some cases, it seems that the single treatment with PI3K inhibitors is as effective as Sotorasib treatment, promoting the death of the parental cells. This is in line with previous data in H23 and H353 that show sensitivity to PI3K inhibition ( i.e., H358 10.1016/j.jtcvs.2005.06.051 ; 10.1016/j.jtcvs.2005.06.051H23 10.20892/j.issn.2095-3941.2018.0361). The data is clear, especially for copanlisib, but would it be the case that this treatment could be used for the treatment of NSCLC alone or directly in combination with Sotorasib and prevent resistance? The results shown in Figure 4C strongly support that a single treatment might be effective in cases that do not respond to Sotorasib. The data in figure 4D-F (please correct typo "inhibition" in labels) seem to support that PI3K treatment of parental cells is as effective as in the resistant cells.

(11) The experiments presented in Figure 7 show synergy between Sotorasib and copanlisib treatment in some of the resistant cells. But in Figure 7G, the single treatment of H23AR is as effective as the combination. Did the authors check the effect of this drug on the parental cells? As they do not include this control, it is not possible to know if this is acquired sensitivity to PI3K inhibition or if the parental cells were already sensitive (as indicated by the Figure 4 results).

Author response:

Reviewer #1 (Public review):

Summary:

In this study, the authors investigate mechanisms of acquired resistance (AR) to KRAS-G12C inhibitors (sotorasib) in NSCLC, proposing that resistance arises from signaling rewiring rather than additional mutations.

Strengths:

Using a panel of AR models - including cell lines, PDXs, CDXs, and PDXOs - they report activation of KRAS and PI3K/AKT/mTOR pathways, with elevated PI3K levels. Pharmacologic inhibition or CRISPR-Cas9 knockout of PI3K partially restores sotorasib sensitivity, and p-4EBP1 upregulation is implicated as an additional contributor, with dual mTORC1/2 inhibition more effective than mTORC1 inhibition alone.

Weaknesses:

While the study addresses an important clinical question, it is limited by several weaknesses in experimental rigor, data interpretation, and presentation. The mechanistic findings are not entirely novel, since the role of PI3K-AKT-mTOR signaling in therapeutic resistance is already well-established in the literature. Rather than uncovering new resistance mechanisms, the study largely confirms known pathways. Several key conclusions are not supported by the data, and critical alternative explanations - such as additional mutations or increased KRAS expression - are not thoroughly investigated or ruled out. Furthermore, while the authors use CRISPR-Cas9 to knock out PI3K and 4E-BP1 in H23-AR and H358-AR cells to restore sotorasib sensitivity, they do not perform reconstitution experiments to confirm that re-expressing PI3K or 4E-BP1 reverses the sensitization. This prevents full characterization of PI3K and p-4EBP1 upregulation as contributors to resistance. The manuscript also has several errors, poor figure quality, and a lack of proper quantification. Additional experimental validation, data improvement, and text revisions are required.

Acquired resistance to KRAS^G12C inhibitors such as sotorasib or adagrasib remains a significant clinical challenge. Therefore, the identification of mechanisms of acquired resistance, along with the development of alternative therapeutic strategies, including combination therapies with KRAS inhibitors, represents an urgent unmet clinical need. The emergence of secondary KRAS mutations or new mutations in other oncogenic drivers has been observed as a primary cause of acquired resistance in a fraction of patients. No identifiable mutations were detected in more than half of the tumors from patients who developed acquired resistance after treatment with sotorasib or adagrasib.

Using a discovery-based approach that integrated global proteomic and phosphoproteomic analyses in the TC303AR and TC314AR PDX models, we identified distinct protein signatures associated with KRAS reactivation, upregulation of mTORC1 signaling, and activation of the PI3K/AKT/mTOR pathway. These findings prompted further investigation into these mechanisms of resistance and evaluation of novel therapeutic combinations to overcome resistance. Notably, the combination of sotorasib with copanlisib (a PI3K inhibitor), or the combination of sotorasib with AZD8055 or sapanisertib (mTORC1/2 dual inhibitors) demonstrated strong potential for future clinical use. These regimens effectively restored sotorasib sensitivity in both in vitro and in vivo models and produced robust, synergistic antitumor effects across various acquired resistance models.

CRISPR-Cas9-mediated PI3K and 4E-BP1 knockout clones were generated in more than one resistant cell line that expressed a robust level of the knockout target, and multiple independent clones in each cell line were evaluated with and without gene disruption. Given the thorough nature of this analysis, additional reconstitution experiments were deemed unnecessary, as they would not yield further insight.

Whole exome sequencing was performed on resistant cells or PDX models to confirm retention of the KRAS^G12C mutation and to identify secondary KRAS mutations, none of which were found. We acknowledge that additional resistance mechanisms may be involved. These will be the focus of future investigations.

The revised manuscript will feature improved figure quality, complete and clarified figure legends, and corrected textual errors to enhance overall clarity and presentation.  

Reviewer #2 (Public review):

Summary:

In this manuscript, the authors focus on the identification of the mechanisms involved in the acquired resistance to Sotorasib in non-small lung KRASG12C mutant cells. To perform this study, the authors generate different clones of cell lines, cell-derived xenografts, patient-derived xenograft organoids, and patient-derived xenografts. In all these models, the authors generate resistant forms (i.e., resistant cell lines PDXs and organoids) and the genetic and molecular changes were characterised using whole-exome sequencing, proteomics, and phospho-proteomics. This analysis led to the identification of an important role of the PI3K/AKT/mTORC1/2 signalling network in the acquisition of resistance in several of the models tested. Molecular characterisation identified changes in the expression of some of the proteins in this network as key changes for the acquisition of resistance, and in particular, the authors show that changes in 4E-BP1 are common to some of the cells downstream of PI3K. Using pharmacological testing, they show that different drugs targeting PI3K, AKT, and MTORC1/2 sensitise some of the resistant models to Sotorasib. The analyses showed that the PI3K inhibitor copanlisib has an effect in NSCLC cells that, in some cases, seems to be synergistic with Sotorasib. Based on the work performed, the authors conclude that the PI3K/mTORC1/2 mediated 4E-BP1 phosphorylation is one of the mechanisms associated with the acquisition of resistance to Sotorasib and that targeting this signalling module could result in effective treatments for NSCLC patients.

The work as presented in the current manuscript is very interesting, provides cell models that benefit the community, and can be used to expand our knowledge of the mechanism of resistance to KRAS targeting therapies. Overall, the techniques and methodology seem to be performed in agreement with standard practice, and the results support most of the conclusions made by the authors. However, there are some points that, if addressed, would increase the value and relevance of the findings and further extend the impact of this work. Some of the recommendations for changes relate to the way things are explained and presented, which need some work. Other changes might require the performance of additional experiments or reanalysis of the existing data.

Strengths:

(1) One of the stronger contributions of this article is the different models used to study the acquisition of resistance to Sotorasib. The resistant cell lines, PDXs and PDXOs, and the fact that the authors have different clones for each, made this collection especially relevant, as they seem to show different mechanisms that the cells used to become resistant to Sotorasib. Although logically, the authors focus on one of these mechanisms, the differential responses of the different clones and models to the treatments used in this work show that some of the clones used additional mechanisms of resistance that can be explored in other studies. Importantly, as they use in vitro and in vivo models, the results also consider the tumour microenvironment and other factors in the response to the treatments.

(2) Another strength is the molecular characterisation of the different Sotorasib-resistant tumour cells by WES, which shows that these cells do not seem to acquire secondary mutations.

(3) The use of MS-based proteomics also identifies proteome signatures that are associated with the acquisition of resistance, including PI3K/mTORC1/2. The combination of proteomics and phospho-proteomics results should allow the identification of several mechanisms that are deregulated in Sotorasib-resistant cells.

(4) The results show a strong response of the NSCLC cells and PDXs to copanlisib, a drug for which there is limited information in this cancer type.

(5) The way they develop the PDX-resistant and the PDXO seems to be appropriate.

Weaknesses:

In general, the data is of good quality, but due to the sheer amount of data included and the way it is presented and discussed, several of the claims or conclusions are not clear.

(1) The abstract is rather long and gives details that are not usually included in one. This makes it very complicated to identify the most relevant findings of the work. The use of acronyms PDX, PDXO, and CDX without defining them makes it complicated for the non-specialist to know what the models are. Rewriting and reorganisation of the abstract would benefit the manuscript.

We will revise the abstract to ensure that the key findings and overall message are clearly communicated and easily understood by readers.

2) Expression, presentation, and grammar should be reviewed in all sections of the manuscript.

Will be done accordingly in the revised version

(3) In the different parts of the result section where the models shown in Figure 2 are described the authors indicate "Whole-exome sequencing (WES) confirmed that XXX model retained the KRASG12C mutation with no additional KRAS mutations detected" however, it is not indicated where this data is shown and in not all the cases there is explanation to other possible modifications that might relate to mechanisms of resistance. This information should be included in the manuscript, and the WES made publicly available.

WES was done for KRAS to identify secondary mutations in the KRAS as well as to verify the retention of the KRAS^G12C mutation in these AR models. WES data will be provided as supplements

(4) The way the proteomics analysis of the TC303 and TC314 parental and resistant PDX is described in the text is confusing. The addition of an experimental layout figure would facilitate the understanding. As it is written, it is not obvious that the parental PDX were also analysed. For instance, the authors say, "The global and phosphoproteomic analyses identified over 8,000 and 4,000 gene protein products (GPPs), respectively". Is this comparing only resistant cells, or from the comparison of the parental and resistant pairs? And where are these numbers presented in the figures? Also, there is information that seems more adequate for the materials and methods sections, i.e., "Samples were analyzed using label-free nanoscale liquid chromatography coupled with tandem mass spectrometry (nanoLC-MS/MS) on a Thermo Fusion Mass Spectrometer. The resulting data were processed and quantified using the Proteome Discoverer 2.5 interface with the Mascot search engine, referencing the NCBI RefSeq protein database (Saltzman, Ruprecht). Two-component analysis is better named principal component analysis."

The texts will be revised accordingly

(5) While the presentation of the proteomics data could be done in different ways, the way the data is presented in Figure 3 does not allow the reader to get an idea of many of the findings from this experiment. Although it is indicated that a table with the data will be made available, this should be central to the way the data is presented and explained. A table (ie, Excel doc) where the raw data and all the analysis are presented should be included and referenced. Additionally, heat maps for the whole proteomes identified should be included. In the text, it is said, "Global proteomic heatmap analysis revealed unique protein profiles in TC303AR and TC314AR PDXs compared to their sensitive counterparts (Figure 3C)." However, this figure only shows the histogram of the differentially regulated cells. Inclusion of the histogram showing all the cells is necessary, and it might be informative to include the histogram comparing the two isogenic pairs, which could identify common mechanisms and differences between both sets. In Figure 3C, the protein names should be readable, or a reference to tables where the proteins are listed should be included.

The raw data associated with the proteomics and global proteomics will be added as supplements.

(6) In Figure 3, the pathway enrichment tool and GO used should be mentioned in the text. The tables with all significant tables should also be provided. The proteomics data seems to convincingly identify mTOR as one of the pathways deregulated in resistant cells, but there is little explanation of what is considered a significant FDR value and if there are other pathways or networks that are also modified, which might not be common to both isogenic models. In MS-based Phosphoproteome could help with the identification of differentially regulated pathways, but it is not really presented in the current manuscript. Most of the analysis of phospho-proteomics comes from the RPPA analysis, which is targeted proteomics. With the way the data is presented, the authors show evidence for a role of mTOR in the acquisition of resistance, but unfortunately, they do not discuss or allow the reader to explore if other pathways might also contribute to this change.

The authors agree that other pathways may be involved, and this will be the subject of future studies. The raw data will be added as supplements.

(7) Where is the proteomics data going to be deposited, and will it be made public to comply with FAIR principles?

will be uploaded according to the journal guidelines

(8) The authors claim that the resistance shown for H23AR and H353AR cells is due to reactivation of KRAS signalling. This is done by looking to phosphorylation of ERK as a surrogate, as they claim, "KRAS inhibition is commonly assessed by evaluating the inhibition of ERK phosphorylation (p-ERK)". While this might be true in many cases, the data presented does not demonstrate that the increase in p-ERK is due to reactivation of KRAS. To make this claim, the authors should measure activation of KRAS (and possibly H- and NRAS) using GST-pull down or an image-based method.

We agree that KRAS activation can be assessed through various methods. In this manuscript, which primarily focuses on mechanisms of resistance, pathway analysis revealed upregulation of KRAS signaling. This finding correlated with the incomplete inhibition of p-ERK by sotorasib in resistant cells. Notably, p-ERK status is widely recognized and routinely used as a surrogate marker for KRAS pathway activation.

(9) The experiments in Figure 4 are very confusing, and some controls are missing. There is no blot where they show the effect of Sotorasib treatment in H23 and H358 parental cells. Is the increase shown in resistant cells shown in parental or is it exclusive for resistant cells only (and therefore acquired)? Experiment 4B should include this control. What is clear is that there is an increase in the expression of AKT and PI3K.

H23 and H358 cells are highly sensitive to sotorasib, as demonstrated by the cell viability assays presented in Figure 2. As shown in Figure 3—figure supplement 3, sotorasib treatment led to complete inhibition of p-ERK in these parental cell lines. In contrast, p-ERK inhibition was incomplete in the resistant H23AR and H358AR cells. Moreover, these AR cells were continuously cultured under sotorasib pressure to maintain resistance.

(10) The main point here is whether this is acquired resistance or the sensitivity to the drug is already there, and there was no need to do an omics experiment to find this. In some cases, it seems that the single treatment with PI3K inhibitors is as effective as Sotorasib treatment, promoting the death of the parental cells. This is in line with previous data in H23 and H353 that show sensitivity to PI3K inhibition ( i.e., H358 10.1016/j.jtcvs.2005.06.051 ; 10.1016/j.jtcvs.2005.06.051H23 10.20892/j.issn.2095-3941.2018.0361). The data is clear, especially for copanlisib, but would it be the case that this treatment could be used for the treatment of NSCLC alone or directly in combination with Sotorasib and prevent resistance? The results shown in Figure 4C strongly support that a single treatment might be effective in cases that do not respond to Sotorasib. The data in figure 4D-F (please correct typo "inhibition" in labels) seem to support that PI3K treatment of parental cells is as effective as in the resistant cells.

We agree. Based on our in vitro (Figure 4) and in vivo (Figure 7) data, copanlisib was able to overcome sotorasib resistance, demonstrating either synergistic or additive effects depending on the specific model. These findings support the potential of combining PI3K inhibition with KRAS^G12C inhibition as a promising strategy to address acquired resistance.

(11) The experiments presented in Figure 7 show synergy between Sotorasib and copanlisib treatment in some of the resistant cells. But in Figure 7G, the single treatment of H23AR is as effective as the combination. Did the authors check the effect of this drug on the parental cells? As they do not include this control, it is not possible to know if this is acquired sensitivity to PI3K inhibition or if the parental cells were already sensitive (as indicated by the Figure 4 results).

Both H23 and H23AR cells showed high sensitivity to copanlisib, as shown in Figure 4. Combination index analysis for the copanlisib + sotorasib treatment (Figure 7A) revealed synergistic effects on cell viability at specific concentrations. However, in the in vivo experiment (Figure 7G), we did not observe a clear synergistic effect of the combination treatment against H23AR xenografts. This may be attributed to the dose of copanlisib used, which was potentially sufficient on its own to produce a strong antitumor response, thereby masking any additional benefit from the combination.

eLife Assessment

This important work substantially advances our understanding of how accessory olfactory bulb neurons respond to social odor cues across the estrous cycle, showing that responses vary with the strain and sex of the odor source but display no consistent differences between estrous and non-estrous states. It employs a unique electrophysiology preparation that activates the vomeronasal organ pump via electric stimulation, enabling precise recordings of accessory olfactory bulb cell responses to different chemosignals in anesthetized mice. Overall, the study presents convincing findings on the stability and variability of accessory olfactory bulb response patterns, indicating that while accessory olfactory bulb detects social signals, it does not appear to interpret them based on reproductive state. This work will be of interest to those studying olfaction, social behavior, reproductive cycles, and systems neuroscience more broadly.

Reviewer #1 (Public review):

Summary:

In this detailed study, Cohen and Ben-Shaul characterized the AOB cell responses to various conspecific urine samples in female mice across the estrous cycle. The authors found that AOB cell responses vary with strains and sexes of the samples. Between estrous and non-estrous females, no clear or consistent difference in responses was found. The cell response patterns, as measured by the distance between pairs of stimuli, are largely stable. When some changes do occur, they are not consistent across strains or male status. The authors concluded that AOB detects the signals without interpreting them. Overall, this study will provide useful information for scientists in the field of olfaction.

Strengths:

The study uses electrophysiological recording to characterize the responses of AOB cells to various urines in female mice. AOB recording is not trivial as it requires activation of VNO pump. The team uses a unique preparation to activate the VNO pump with electric stimulation, allowing them to record AOB cell responses to urines in anesthetized animals. The study comprehensively described the AOB cell responses to social stimuli and how the responses vary (or not) with features of the urine source and the reproductive state of the recording females. The dataset could be a valuable resource for scientists in the field of olfaction.

Weaknesses:

The study will be significantly strengthened by understanding the "distance" of chemical composition in different urine. This could be an important future direction.

Author response:

The following is the authors’ response to the original reviews

Public Reviews:

Reviewer #1 (Public review):

Summary:

In this detailed study, Cohen and Ben-Shaul characterized the AOB cell responses to various conspecific urine samples in female mice across the estrous cycle. The authors found that AOB cell responses vary with the strains and sexes of the samples. Between estrous and non-estrous females, no clear or consistent difference in responses was found. The cell response patterns, as measured by the distance between pairs of stimuli, are largely stable. When some changes do occur, they are not consistent across strains or male status. The authors concluded that AOB detects the signals without interpreting them. Overall, this study will provide useful information for scientists in the field of olfaction.

Strengths:

The study uses electrophysiological recording to characterize the responses of AOB cells to various urines in female mice. AOB recording is not trivial as it requires activation of VNO pump. The team uses a unique preparation to activate the VNO pump with electric stimulation, allowing them to record AOB cell responses to urines in anesthetized animals. The study comprehensively described the AOB cell responses to social stimuli and how the responses vary (or not) with features of the urine source and the reproductive state of the recording females. The dataset could be a valuable resource for scientists in the field of olfaction.

Weaknesses:

(1) The figures could be better labeled.

We revised all figures (except the model figure, Fig. 8), and among other improvements (many of which were suggested by the reviewers in other comments), added more labelling and annotation within the figures.

(2) For Figure 2E, please plot the error bar. Are there any statistics performed to compare the mean responses?

We added error bars (standard errors of the mean). We had not originally performed statistical comparisons between the stimuli, but now we have. The analysis of responses strength now appears in a new table (Table 1)

(3) For Figure 2D, it will be more informative to plot the percentage of responsive units.

Done.

(4) Could the similarity in response be explained by the similarity in urine composition? The study will be significantly strengthened by understanding the "distance" of chemical composition in different urine.

We agree. As we wrote in the Discussion: “Ultimately, lacking knowledge of the chemical space associated with each of the stimuli, this and all the other ideas developed here remain speculative.” We note however, that chemical distance (which in itself is hard to define) will provide only part of the picture. The other part is the “projection” of chemical space on the receptor array. This is an idea that we develop in the Discussion and in Figure 8. Specifically, that it is the combination of stimulus composition, and receptor tuning properties that will determine stimulus distances in neuronal space.

That said, a better understanding of the chemical distance is an important aspect that we are working to include in our future studies. For this dataset unfortunately, we have no such data.

(5) If it is not possible for the authors to obtain these data first-hand, published data on MUPs and chemicals found in these urines may provide some clues.

This comment is directly related to the previous one. Measurements about some classes of molecules may be found for some of the stimuli that we used here, but not for all. We are not aware of any single dataset that contains this information for any type of molecule across the entire stimulus set that we have used and pooling results from different studies has limited validity because of the biological and technical variability across studies. In order to reliably interpret our current recordings, it would be necessary to measure the urinary content of the very same samples that were used for stimulation. Unfortunately, we are not able to conduct this analysis at this stage.

(6) It is not very clear to me whether the female overrepresentation is because there are truly more AOB cells that respond to females than males or because there are only two female samples but 9 male samples.

The definitive answer to this comment is given in our response to the next one.

Nevertheless, we agree that this is an important point. It is true that the number of neurons fulfilling each of the patterns depends on the number of individual stimuli that define it (and on the frequency of neurons that respond to those stimuli). However, our measure of “over representation” was designed to overcome this bias, by using bootstrapping to reveal if the observed number of patterns is larger than expected by chance.  The higher frequency of responses to female, as compared to male stimuli, is observed in other studies by others and by us, also when the number of male and female stimuli is matched (e.g., Bansal et al BMC Biol 2021, Ben-Shaul et al, PNAS 2010, Hendrickson et al, JNS, 2008). However, here, by overrepresentation, we do not refer to the higher frequency of female responding neurons, but rather that given the number of responding neurons, the female pattern is more common than expected by chance.

(7) If the authors only select two male samples, let's say ICR Naïve and ICR DOM, combine them with responses to two female samples, and do the same analysis as in Figure 3, will the female response still be overrepresented?

Following this suggestion, we have performed this analysis, and we were glad to see that the result is the one we had anticipated. Below, we provide an image of the results, following the same approach that we applied before, and showed in Figure 3C. Here, we defined a female pattern (using the two female samples) and compared it to a male pattern (using the ICR naïve and ICR DOM as suggested). It is as if we had only four stimuli in our set. As in the article, we calculated the expected distribution with 100,000 shuffles. We denoted this pattern as F/M ICR. The results are shown below.

Under the present conditions, the distribution of the number of female selective patterns is larger (i.e., shifted to the right, compare to the female category in Figure 3C. This is expected, since now the criterion is more permissive. Specifically, now to qualify as a “female pattern”, the two responses to female urine must be stronger only than the responses to the two male stimuli included in this analysis (and to all other responses). Notably, although the null distribution shifted to the right, the actual number of neurons fulfilling this pattern is also larger, so that the actual number remains significantly larger than expected by chance. This is also true for the reverse category (as is the case in the ~female category Figure 3C).  Thus, we conclude that overrepresentation of the female pattern is not a trivial consequence of the number of male and female stimuli.

Author response image 1.

(8) In Figure 4B and 4C, the pairwise distance during non-estrus is generally higher than that during estrus, although they are highly correlated. Does it mean that the cells respond to different urines more distinctively during diestrus than in estrus?

This is an important observation (!) and we had originally overlooked it.  It is true that higher distance (as they are in estrus) imply more distinct population level responses and hence better discrimination among stimuli. However, this is inconsistent with all our other analyses that do not point to enhanced selectivity or discrimination in either state. If anything, we find somewhat higher sparseness in estrus.  Yet, there may be technical explanations for the differences.

For Euclidean distances, the explanation may be trivial. The distance depends on the number of dimensions (i.e., units), and since our sample contains more neurons recorded during non-estrus, the larger distance is expected.

In fact, there is a similar dependence on sample size for the correlation distance. Smaller samples are associated with higher (spurious) correlations, and hence larger samples are be associated with larger distances. To demonstrate this, we conducted a simple simulation, where we calculated the absolute correlation coefficients of random samples from standard normal distributions (using the MATLAB function randn), changing the size of the population. For each sample size, we conducted 1000 tests. We considered sample sizes from 10 to 100000, including 200 and 300 (which are similar to our sample sizes). The results are shown in the figure below. Note that the absolute value of the correlation coefficient decreases with sample size, while the p-value for the observed correlation is stable at ~0.5.

While this is not a rigorous analysis of this issue, and while it does not exactly reflect the scenario in our data, where correlations are generally positive, it shows that the observed correlation (and hence correlation distance) is also affected by sample size.

For these reasons, we focus on comparison of these distances, rather than the absolute values of the correlation distances.

Author response image 2.

Following this comment, we now write in the manuscript:

“We first note that distances are generally larger during non-estrus, suggesting enhanced discrimination during this stage. However, further analyses of sparseness and selectivity do not support this idea (see below). Furthermore, we note that both Euclidean and correlation distances generally depend on sample size. In both cases, distances are expected to increase as a function of sample size, which in our dataset, is larger for the non-estrus (n = 305) as compared to the estrus (n = 241) neurons. Because of this factor, we focus here on the similarity of the relative within-state distances across the states (and not on their absolute magnitudes). Specifically, we find a positive and significant correlation among pairwise population distances under the two states. Thus, at the population level, representational space remains broadly stable across the estrus cycle. Nevertheless, several points in Fig. 4D, E clearly diverge from a linear relationship, implying that representational space differs under the two states. We next examine such state-dependent changes in more detail.”

(9) The correlation analysis is not entirely intuitive when just looking at the figures. Some sample heatmaps showing the response differences between estrous states will be helpful.

If we understand correctly, the idea is to show the correlation matrices from which the values in 4B and 4C are taken. The relevant images are now included in figure 4B, C and are references within the main text.

Reviewer #2 (Public review):

Summary:

Many aspects of the study are carefully done, and in the grand scheme this is a solid contribution. I have no "big-picture" concerns about the approach or methodology. However, in numerous places the manuscript is unnecessarily vague, ambiguous, or confusing. Tightening up the presentation will magnify their impact.

We have reviewed the text and made substantial editing changes. Along with other specific comments by made both reviewers, we hope that these changes improve the presentation.

Strengths:

(1) The study includes urine donors from males of three strains each with three social states, as well as females in two states. This diversity significantly enhances their ability to interpret their results.

(2) Several distinct analyses are used to explore the question of whether AOB MCs are biased towards specific states or different between estrus and non-estrus females. The results of these different analyses are self-reinforcing about the main conclusions of the study.

(3) The presentation maintains a neutral perspective throughout while touching on topics of widespread interest.

Weaknesses:

(1) Introduction:

The discussion of the role of the VNS and preferences for different male stimuli should perhaps include Wysocki and Lepri 1991

We assume that the reviewer is referring to “Consequences of removing the vomeronasal organ” by Wysocki CJ, Lepri JJ, a review article in J Steroid Biochem from 1991. We were not familiar with this specific article and have now read it. The article discusses various male behaviors, and some effects on female behavior and physiology (e.g., puberty acceleration, maternal behaviors, ovulation) but we could not find any mention of the preference of female mice in this article. We also expanded our search to all pubmed articles authored by Wysocki and Lepri and then all articles by Wysocki (with the keyword Vomeronasal). Despite our best intentions to give due credit, we found nothing that seems directly related to this statement. Please correct us if we had missed anything.

(2) Results:

a) Given the 20s gap between them, the distinction between sample application and sympathetic nerve trunk stimulation needs to be made crystal clear; in many places, "stimulus application" is used in places where this reviewer suspects they actually mean sympathetic nerve trunk stimulation.

We realize that this is confusing, and we also agree that at least in one place, we have not been sufficiently clear about the distinction. To clarify, we distinguish between stimulus application (physical application of stimulus to the nostril), and stimulation (which refers to SNT stimulation, which typically induces VNO suction). The general term stimulus presentation refers to the entire process. As explained in the text, in our analysis, we consider the entire window starting at application and ending 40s after stimulation. This is because we sometimes observe immediate responses following application. One such responses is seen in Figure 2D, and this is directly related to a detailed comment made below (on Figure 1D, part c). Indeed, for this figure time 0 indicates stimulus application. This was indicated previously, but we have now rearranged order of the panels to make the distinction between this response and other clearer. We have also revised the figure caption and the text to clarify this issue.

b) There appears to be a mismatch between the discussion of Figure 3 and its contents. Specifically, there is an example of an "adjusted" pattern in 3A, not 3B.

True. we have revised the text to correctly refer to the figure. Thanks.

c) The discussion of patterns neglects to mention whether it's possible for a neuron to belong to more than one pattern. For example, it would seem possible for a neuron to simultaneously fit the "ICR pattern" and the "dominant adjusted pattern" if, e.g., all ICR responses are stronger than all others, but if simultaneously within each strain the dominant male causes the largest response.

This is true. In the legend to Figure 3B, we actually wrote: “A neuron may fulfill more than one pattern and thus may appear in more than one row.”, but we now also write in the main text:

“We note that criteria for adjusted patterns are less stringent than for the standard patterns defined above. Furthermore, some patterns are not mutually exclusive, and thus, a neuron may fulfil more than a single pattern.”

(3) Discussion:

a) The discussion of chemical specificity in urine focuses on volatiles and MUPs (citation #47), but many important molecules for the VNS are small, nonvolatile ligands. For such molecules, the corresponding study is Fu et al 2015.

Agreed. We now cite this work and several others that were not included before in the context of chemical and electrophysiological analyses.

b) "Following our line of reasoning, this scarcity may represent an optimal allocation of resources to separate dominant from naïve males": 1 unit out of 215 is roughly consistent with a single receptor. Surely little would be lost if there could be more computational capacity devoted to this important axis than that? It seems more likely that dominance is computed from multiple neuronal types with mixed encoding.

We fully agree, and we are not claiming that dominance, nor any other feature, is derived using dedicated feature selective neurons. Our discussion of resource allocation is inevitably speculative. Our main point in this context is that a lack of overrepresentation does not imply that a feature is not important. As a note, we do not think that there is good reason to suppose that AOB neurons reflect the activity of single receptors.

To present this potential confusion, we now added the following sentences in the Discussion subsection titled “Response patterns of AOB-MCs”:

“We stress that we do not suggest that features such as physiological state are encoded by the activity of single neurons. In fact, we believe that most ethologically relevant features are encoded by the activity of multiple neurons. Nevertheless, such population level representations ultimately depend on the response properties of individual neurons, and we thus ask: what can we learn from our analysis of response pattern frequency?”

(4) Methods:

a) Male status, "were unambiguous in most cases": is it possible to put numerical estimates on this? 55% and 99% are both "most," yet they differ substantially in interpretive uncertainty.

Upon reexamination, we realized that this sentence is incorrect. Ambiguous cases were not considered as dominant for urine collection. We only classified mice as dominant if they “won” the tube test and exhibited dominant behavior in the subsequent observation period in the cage. The phrasing has now been corrected in the manuscript (Methods section).

b) Surgical procedures and electrode positioning: important details of probes are missing (electrode recording area, spacing, etc).

This information has been added to the Methods subsection “Surgical procedures and electrode positioning”

c) Stimulus presentation procedure: Are stimuli manually pipetted or delivered by apparatus with precise timing?

They are delivered manually. This has now been clarified in the text.

d) Data analysis, "we applied more permissive criteria involving response magnitude": it's not clear whether this is what's spelled out in the next paragraph, or whether that's left unspecified. In either case, the next paragraph appears to be about establishing a noise floor on pattern membership, not a "permissive criterion."

True, the next paragraph is not the explanation for the more permissive criteria. The more permissive criteria involving response magnitude are actually those described in Figure 3A and 3B. The sentence that was quoted above merely states that before applying those criteria, we had also searched for patterns defined by binary designation of neurons as responsive, or not responsive, to each of the stimuli (this is directly related to the next comment below). Using those binary definitions, we obtained a very small number of neurons for each pattern and thus decided to apply the approach actually used and described in the manuscript.

To clarify this confusion, we thoroughly derived the description of this paragraph, and the beginning of the next one in the Methods section.

e) Data analysis, method for assessing significance: there's a lot to like about the use of pooling to estimate the baseline and the use of an ANOVA-like test to assess unit responsiveness.

But:

i) for a specific stimulus, at 4 trials (the minimum specified in "Stimulus presentation procedure") kruskalwallis is questionable. They state that most trials use 5, however, and that should be okay.

The exact values are now given in the text. The mean number of repeated presentations per stimulus: 5.1± 0.9, mean ± sd. In 72% of the cases, stimuli were given 5 or more times. Otherwise, they were presented 4 times. In the context of the statistical test, we note that we are not comparing 5 (or 4) values with another set of 5 (or 4 values), but with a much larger sample (~44-55 baseline trials – given 11 trials and 4-5 repeats of each). Under this scenario, we think that the statistical approach is sound. However, the more important consideration, in our opinion, is given below.

ii) the methods statement suggests they are running kruskalwallis individually for each neuron/stimulus, rather than once per neuron across all stimuli. With 11 stimuli, there is a substantial chance of a false-positive if they used p < 0.05 to assess significance. (The actual threshold was unstated.) Were there any multiple comparison corrections performed? Or did they run kruskalwallis on the neuron, and then if significant assess individual stimuli? (Which is a form of multiple-comparisons correction.)

First, we indeed failed to mention that our criterion was 0.05. This has been corrected, by adding the information to the results and the Methods sections. No, we did not apply any multiple comparison measures. We consider each neuron-stimulus pair as an independent entity, and we are aware that this leads to a higher false positive rate. On the other hand, applying multiple comparisons would be problematic, as the same number of stimuli used in different studies varies. Application of multiple comparison corrections would thus lead to different response criteria across different studies, which would be very problematic. This raises the almost philosophical question regarding the use of multiple comparisons (as well as one and two tailed tests), but practically, most, if not all of our conclusions involve comparisons across conditions. For this purpose, we think that our procedure is valid. More generally, while selection of responses according to significance has some obvious advantages, the decision to use any particular criterion is entirely arbitrary. Therefore, we do not attach any special meaning to the significance threshold used here. Rather, we think of it as a simple criterion that allows us to exclude weakly responding or non-responsive neurons, and to compare frequencies of neurons that fulfill this criterion, under different conditions and contexts.

Recommendations for the authors:

Reviewer #2 (Recommendations for the authors):

Results:

"are represented more than represented by chance" seems to have a misplaced word

True. Thanks. Corrected.

Figure 1D:

a) Indicate the meaning of the number that appears in the top left for each unit (10, 5, 40, 5, 5) (I'm guessing it's the vertical scale for the PSTH, but best to spell it out explicitly.)

This information has been added.

b) "The red vertical line indicates stimulus application": is it the application of the chemical stimulus or SNT shock?

Please see our answer to c

c) "For unit 2, time 0 indicate stimulus application, as in this case, responses began after stimulus application, prior to stimulation." First, the meaning of time 0 for the other units is not clearly specified (we infer that unit 2 is an exception, but we don't know what most of them mean). Second, it seems as if the response (?) to ICR naive begins even before stimulus application.

This issue was also mentioned above as the 2nd weakness raised by this reviewer. To explain the meaning of the red lines, and resolve this confusion, we revised the figure caption text to indicate that for all units (except the former unit 2) time 0 indicates SNT stimulation. We also changed the order of the unit examples, placing the former unit 2 in the rightmost position. It is true that for this unit, there is a firing rate change prior to stimulus application, which actually appears as rate attenuation following stimulus application. In this specific case, we consider this activity as “noise”, and note that this neuron-stimulus combination would not be classified as a response (since there is no consistent change across stimulus presentation).

As a note, while reviewing this figure, we noted an error. We have previously written that the ITI was 10 s, whereas it was actually 18 s long. This has been corrected in the Figure and in the text.

Figure 2B:

"The mean error due to the reduced 2-D representation is 0.29 (arbitrary units)." This is unclear. MDS is often described in terms of % of variance explained, is that what this means? If so, the units are not arbitrary; otherwise, it's unclear whether specifying a value with arbitrary units adds any value.

This is a very good point, and we thank the reviewer for identifying this mistake. The units are not arbitrary! They are units of correlation distance. We now added a scale bar (a square) to panel 2B to indicate what a distance of 0.1. Following this comment, we also calculated the mean error in the original data, and noted the ratio between the mean absolute error (due to considering only two dimensions) and the mean original distances. We also now report the value of the first two eigenvalues. Specifically, we now write:

“Note that like all dimensionally reduced representations, the representation in Fig. 2B is an approximation. Here, the first two eigenvalues of account for 44.6% of the variance of the original distances (30.4% and 14.2%, respectively for the first and second dimension). Another way to evaluate the representation is via the mean error due to the reduced 2-D representation. Here, it is 0.29, whereas the mean of the original distances is 0.73.”

Figure 3A:

a) There is a truncated label (or something) above the panel letter.

Thanks. Corrected. This was part of the “Figure” label

b) The graphic for the "adjusted pattern" also fits the criterion of the "pattern": for example, in the top row the activity for ICR is still higher than for any other stimulus, thus fulfilling the criterion of a "pattern" and not just an "adjusted pattern."

That was not our intention. An adjusted pattern does not necessarily fulfill the (non-adjusted) “pattern” (while the opposite is true). We have now revised the rightmost panel in figure 3A, adding both “&s” to indicate that all three conditions must be fulfilled, and in attempt for a more intuitive representation, applied a different background denoting stimuli with irrelevant responses. We also changed the terms in the legend within the panel, making them more accurate: (Thus, “strong activity” was changed to “stronger responses”). In addition, we revised the text and figure legends in attempt to better clarify these definitions.

Figure 3B:

I'm assuming that the columns of the heatmap correspond to different urine stimuli, and that the color is normalized firing rate. But readers should not have to guess.

True, and agreed. We added legends to clarify this.

Figure 4B:

The caption should mention that the pairwise measures are between the stimulus columns of panel A.

We revised the caption to indicate this. Note that we also added two additional panels to this figure.

Figure 5A&B:

Instead of a multiple-comparisons correction, it seems likely to be better to use a 2-way ANOVA. At a minimum, the nature of the multiple-comparisons correction needs to be specified (many are conservative, but they differ in the extent of how conservative they are).

We now write in the text that we used a Bonferroni correction (this information previously appeared only in the caption). We also found an error in the caption. We previously wrote that we used a binomial exact test for both panels A and B. However, only the data in panel A was calculated with a binomial exact test. The data in panel B was calculated with a one-way ANOVA.

We now also applied a 2-way ANOVA to response magnitudes (i.e., panel B). We find a main effect of stimulus, but not of state, and no effect of interaction between the two. This is consistent with our previous analyses. This analysis is now included in the text. We thank the reviewer for this suggestion.

Editor's note:

Should you choose to revise your manuscript, if you have not already done so, please include full statistical reporting including exact p-values wherever possible alongside the summary statistics (test statistic and df) and, where appropriate, 95% confidence intervals. These should be reported for all key questions and not only when the p-value is less than 0.05 in the main manuscript.

eLife Assessment

The study presents a valuable resource of proline hydroxylation proteins for molecular biology studies in oxygen-sensing and cell signaling with the characterization of Repo-man proline hydroxylation site. The evidence supporting the claim of the authors is solid, although further clarification of the overall efficiency of the HILIC analysis, the specificity/sensitivity of immonium ion analysis, as well as quantification of proline hydroxylation identifications will be helpful. The work will be of interest to researchers studying post-translational modification, oxygen sensing, and cell signaling.

Reviewer #1 (Public review):

Summary:

The manuscript by Hao Jiang et al described a systematic approach to identify proline hydroxylation proteins. The authors implemented a proteomic strategy with HILIC-chromatographic separation and reported an identification of 4993 sites from HEK293 cells (4 replicates) and 3247 sites from RCC4 sites (3 replicates) with 1412 sites overlapping between the two cell lines. From the analysis, the authors identified 225 sites and 184 sites respectively from 293 and RCC4 cells with HyPro diagnostic ion. The identifications were validated by analyzing a few synthetic peptides, with a specific focus on Repo-man (CDCA2) through comparing MS/MS spectra, retention time, and diagnostic ions. With SILAC analysis and recombinant enzyme assay, the study showed that Repo-man HyPro604 is a target of the PHD1 enzyme.

Strengths:

The study involved extensive LC-MS analysis and was carefully implemented. The identification of over 4000 confident proline hydroxylation sites would be a valuable resource for the community. The characterization of Repo-man proline hydroxylation is a novel finding.

Weaknesses:

However, as a study mainly focused on methodology, the findings from the experimental data did not convincingly demonstrate the sensitivity and specificity of the workflow for site-specific identification of proline hydroxylation in global studies.

Major concerns:

(1) The study applied HILIC-based chromatographic separation with a goal of enriching and separating hydroxyproline-containing peptides. However, as the authors mentioned, such an approach is not specific to proline hydroxylation. In addition, many other chromatography techniques can achieve deep proteome fractionation such as high pH reverse phase fractionation, strong-cation exchange etc. There was no data in this study to demonstrate that the strategy offered improved coverage of proline hydroxylation proteins, as the identifications of the HyPro sites could be achieved through deep fractionation and a highly sensitive LCMS setup. The data of Figure 2A and S1A were somewhat confusing without a clear explanation of the heat map representations.

(2) The study reported that the HyPro immonium ion is a diagnostic ion for HyPro identification. However, the data showed that only around 5% of the identifications had such a diagnostic ion. In comparison, acetyllysine immonium ion was previously reported to be a useful marker for acetyllysine peptides (PMID: 18338905), and the strategy offered a sensitivity of 70% with a specificity of 98%. In this study, the sensitivity of HyPro immonium ion was quite low. The authors also clearly demonstrated that the presence of immonium ion varied significantly due to MS settings, peptide sequence, and abundance. With further complications from L/I immonium ions, it became very challenging to implement this strategy in a global LC-MS analysis to either validate or invalidate HyPro identifications.

(3) The study aimed to apply the HILIC-based proteomics workflow to identify HyPro proteins regulated by the PHD enzyme. However, the quantification strategy was not rigorous. The study just considered the HyPro proteins not identified by FG-4592 treatment as potential PHD targeted proteins. There are a few issues. First, such an analysis was not quantitative without reproducibility or statistical analysis. Second, it did not take into consideration that data-dependent LC-MS analysis was not comprehensive and some peptide ions may not be identified due to background interferences. Lastly, FG-4592 treatment for 24 hrs could lead to wide changes in gene expressions and protein abundances. Therefore, it is not informative to draw conclusions based on the data for bioinformatic analysis.

(4) The authors performed an in vitro PHD1 enzyme assay to validate that Repo-man can be hydroxylated by PHD1. However, Figure 9 did not show quantitatively PHD1-induced increase in Repo-man HyPro abundance and it is difficult to assess its reaction efficiency to compare with HIF1a HyPro.

Reviewer #2 (Public review):

Summary:

In this manuscript, Jiang et al. developed a robust workflow for identifying proline hydroxylation sites in proteins. They identified proline hydroxylation sites in HEK293 and RCC4 cells, respectively. The authors found that the more hydrophilic HILIC fractions were enriched in peptides containing hydroxylated proline residues. These peptides showed differences in charge and mass distribution compared to unmodified or oxidized peptides. The intensity of the diagnostic hydroxyproline iminium ion depended on parameters including MS collision energy, parent peptide concentration, and the sequence of amino acids adjacent to the modified proline residue. Additionally, they demonstrate that a combination of retention time in LC and optimized MS parameter settings reliably identifies proline hydroxylation sites in peptides, even when multiple proline residues are present

Strengths:

Overall, the manuscript presents an advanced, standardized protocol for identifying proline hydroxylation. The experiments were well designed, and the developed protocol is straightforward, which may help resolve confusion in the field.

Weaknesses:

(1) The authors should provide a summary of the standard protocol for identifying proline hydroxylation sites in proteins that can easily be followed by others.

(2) Cockman et al. proposed that HIF-α is the only physiologically relevant target for PHDs. Their approach is considered the gold standard for identifying PHD targets. Therefore, the authors should discuss the major progress they made in this manuscript that challenges Cockman's conclusion.

Reviewer #3 (Public review):

Summary:

The authors present a new method for detecting and identifying proline hydroxylation sites within the proteome. This tool utilizes traditional LC-MS technology with optimized parameters, combined with HILIC-based separation techniques. The authors show that they pick up known hydroxy-proline sites and also validate a new site discovered through their pipeline.

Strengths:

The manuscript utilizes state-of-the-art mass spectrometric techniques with optimized collision parameters to ensure proper detection of the immonium ions, which is an advance compared to other similar approaches before. The use of synthetic control peptides on the HILIC separation step clearly demonstrates the ability of the method to reliably distinguish hydroxy-proline from oxidized methionine - containing peptides. Using this method, they identify a site on CDCA2, which they go on to validate in vitro and also study its role in regulation of mitotic progression in an associated manuscript.

Weaknesses:

Despite the authors' claim about the specificity of this method in picking up the intended peptides, there is a good amount of potential false positives that also happen to get picked (owing to the limitations of MS-based readout), and the authors' criteria for downstream filtering of such peptides require further clarification. In the same vein, greater and more diverse cell-based validation approach will be helpful to substantiate the claims regarding enrichment of peptides in the described pathway analyses.

Author response:

Reviewer #1 (Recommendations for the authors):

We appreciate the reviewer recognising that our study has been carefully performed and provides a valuable resource for the community. The characterization of Repo-man proline hydroxylation is also recognised as a novel finding.

With respect to Concerns raised by reviewer 1:

(1) The study applied HILIC-based chromatographic separation with a goal of enriching and separating hydroxyproline-containing peptides. However, as the authors mentioned, such an approach is not specific to proline hydroxylation. In addition, many other chromatography techniques can achieve deep proteome fractionation such as high pH reverse phase fractionation, strong-cation exchange etc. There was no data in this study to demonstrate that the strategy offered improved coverage of proline hydroxylation proteins, as the identifications of the HyPro sites could be achieved through deep fractionation and a highly sensitive LCMS setup. The data of Figure 2A and S1A were somewhat confusing without a clear explanation of the heat map representations.

We do not agree that the apparent concern raised here, i.e., that the method we present is not 100% specific for enriching only hydroxylated peptides, is a serious issue. We show specifically that our method indeed enriches samples for hydroxylated peptides, thereby increasing the chances of identifying proline hydroxylated peptides in a cell extract. We never claimed that it was mono-specific for enrichment of hydroxylated peptides. Further, we note that almost no chromatographic method we know of, including those commonly used to enrich for different types of post translationally-modified peptides (including phospho-peptides) is completely mono-specific for a single type of modified peptide. The reviewer comments that it could have been possible to use alternative methods to identify proline-hydroxylated peptides. This may be true, but we know of no published examples, or previous studies, where this has been demonstrated experimentally on a scale comparable to that we show here. Of course there is always more than one way to approach technical challenges and it may be that future methods will be demonstrated that achieve equivalent, or even superior, results with respect to the detection of proline hydroxylated peptides. To the best of our knowledge, however, our current study provides a robust methodology that goes well beyond any previously published analysis of proline hydroxylation.

(2) The study reported that the HyPro immonium ion is a diagnostic ion for HyPro identification. However, the data showed that only around 5% of the identifications had such a diagnostic ion. In comparison, acetyllysine immonium ion was previously reported to be a useful marker for acetyllysine peptides (PMID: 18338905), and the strategy offered a sensitivity of 70% with a specificity of 98%. In this study, the sensitivity of HyPro immonium ion was quite low. The authors also clearly demonstrated that the presence of immonium ion varied significantly due to MS settings, peptide sequence, and abundance. With further complications from L/I immonium ions, it became very challenging to implement this strategy in a global LC-MS analysis to either validate or invalidate HyPro identifications.

We feel that the reviewer’s initial comment is potentially misleading - it implies that we were proposing here that the 'HyPro immonium ion is a diagnostic ion for HyPro identification’. In contrast, this concept was already widely held in the field before we started this project. Indeed, the fact that the diagnostic HyPro immonium ion is often difficult to detect, has been used as one of the arguments by other researchers to support the view that HIF-α is the only physiologically relevant target for PHD enzymes, a controversy referenced explicitly by Reviewer 2 below. What we actually show here are novel data that help to explain why the diagnostic HyPro immonium ion is often difficult to detect, when standard approaches and technical parameters for MS analysis are used. We beleive that this observation, along with other data we present, is a useful contribution to the field that can help to resolve the previous controversies concerning the true prevalence and biological roles of PHD-catalysed proline hydroxylation on protein targets.

(3) The study aimed to apply the HILIC-based proteomics workflow to identify HyPro proteins regulated by the PHD enzyme. However, the quantification strategy was not rigorous. The study just considered the HyPro proteins not identified by FG-4592 treatment as potential PHD targeted proteins. There are a few issues. First, such an analysis was not quantitative without reproducibility or statistical analysis. Second, it did not take into consideration that data-dependent LC-MS analysis was not comprehensive and some peptide ions may not be identified due to background interferences. Lastly, FG-4592 treatment for 24 hrs could lead to wide changes in gene expressions and protein abundances. Therefore, it is not informative to draw conclusions based on the data for bioinformatic analysis.

We agree that this study is not quantifying or addressing the stoichiometry of proline hydroxylation across the very large number of new PHD target sites we identify. That was not claimed and was not the objective of our study. Nonetheless, we feel the comments of the referee do not adequately take into account the SILAC data we included (cf Figure 8) or the full range of experimental data presented in this study. We would further refer the reviewer also to the data presented in the companion paper by Druker et al., which we cross-referenced extensively in our study and have also made available previously on biorxiv.

(4) The authors performed an in vitro PHD1 enzyme assay to validate that Repo-man can be hydroxylated by PHD1. However, Figure 9 did not show quantitatively PHD1-induced increase in Repo-man HyPro abundance and it is difficult to assess its reaction efficiency to compare with HIF1a HyPro.

Here again we refer to the recent controversy referenced explicitly by Reviewer 2 below, concerning the view expressed by some researchers that only HIF-α is a physiological substrate for PHD enzymes in cells. We were challenged to show that any of the novel protein targets of PHDs we identified were indeed hydroxylated by PHD enzymes in vitro and that is what we demonstrated in Figure 9. This was not an experiment performed to quantify stoichiometry and indeed, it is not possible to draw any firm conclusions about efficiency or stiochiometry in vitro when using catalytic PHD subunits alone, given that we do not yet know whether PHDs may show different properties in cells, dependent on interactions with other factors and/or modifications.

Reviewer #2 (Recommendations for the authors):

We appreciate the reviewer’s comments that our manuscript presents an advanced, standardized protocol for identifying proline hydroxylation, with well designed experiments, which may help resolve confusion in the field.

With respect to Concerns raised by reviewer 2:

(1) The authors should provide a summary of the standard protocol for identifying proline hydroxylation sites in proteins that can easily be followed by others.

We agree and plan to provide a clearly described, step by step guide to assist other researchers who wish to employ our methods for proline hydroxylation analysis in their own studies.

(2) Cockman et al. proposed that HIF-α is the only physiologically relevant target for PHDs. Their approach is considered the gold standard for identifying PHD targets. Therefore, the authors should discuss the major progress they made in this manuscript that challenges Cockman's conclusion.

We agree that our study provides valuable information germane to the recent controversy in the field and the views published by Cockman et al., to the effect that HIF-α is the only physiologically relevant target for PHDs. We will carefully review our statements when preparing a suitably revised version of record with the aim of providing a balanced and objective discussion of this issue.

Reviewer #3 (Recommendations for the authors):

We appreciate the reviewer’s comments that our study employs state-of-the-art mass spectrometric techniques with optimized collision parameters to ensure proper detection of the immonium ions, along with their recognition that our study is, 'an advance compared to other similar approaches before.’ We also appreciate their reference to our companion study by Druker et al, in which we characterise the mechanism and biological role in regulation of mitotic progression of the hydroxylation of P604 in the target protein RepoMan (CDCA2), that is identified in this study.

With respect to the Concern raised by reviewer 3:

Despite the authors' claim about the specificity of this method in picking up the intended peptides, there is a good amount of potential false positives that also happen to get picked (owing to the limitations of MS-based readout), and the authors' criteria for downstream filtering of such peptides require further clarification. In the same vein, greater and more diverse cell-based validation approach will be helpful to substantiate the claims regarding enrichment of peptides in the described pathway analyses..

We agree that this study, which has a focus on methodology and technical approaches for detecting sites of PHD- catalysed proline hydroxylation, cannot exhaustively validate the biological significance of all of the putative sites and targets identified. As the reviewer notes, we have performed a detailed functional characterisation of one such novel PHD-catalyed proline hydroxylation site, i.e. P604 in the protein RepoMan (CDCA2). This functional analysis is presented in the companion paper by Druker et al., which has also been reviewed by eLife and placed on biorxiv (doi: https://doi.org/10.1101/2025.05.06.652400). We hope that publication of our identification of many new putative PHD target sites will encourage other researchers to pursue characterisation of their functional reoles in different biological mechanisms and have tried here to provide some degree of guidance to focus attention on the identification of those sites for which we currently have highest confidence.

eLife Assessment

This valuable study advances our understanding of how bactofilin cytoskeletal proteins associate with cell membranes by identifying and characterizing a conserved membrane-targeting sequence. The evidence is solid, with a well-integrated combination of mutagenesis, biophysical analysis, molecular simulations, and bioinformatics supporting the mechanistic model. The work will be of particular interest to microbiologists and structural biologists studying bacterial cytoskeletons and membrane-protein interactions.

Reviewer #2 (Public review):

Summary:

The authors of this study investigated the membrane-binding properties of bactofilin A from Caulobacter crescentus, a classic model organism for bacterial cell biology. BacA was the progenitor of a family of cytoskeletal proteins that have been identified as ubiquitous structural components in bacteria, performing a range of cell biological functions. Association with the cell membrane is a frequent property of the bactofilins studied and is thought to be important for functionality. However, almost all bactofilins lack a transmembrane domain. While membrane association has been attributed to the unstructured N-terminus, experimental evidence had yet to be provided. As a result, the mode of membrane association and the underlying molecular mechanics remained elusive.

Liu at al. analyze the membrane binding properties of BacA in detail and scrutinize molecular interactions using in-vivo, in-vitro and in-silico techniques. They show that few N-terminal amino acids are important for membrane association or proper localization and suggest that membrane association promotes polymerization. Bioinformatic analyses revealed conserved lineage-specific N-terminal motifs indicating a conserved role in protein localization. Using HDX analysis they also identify a potential interaction site with PbpC, a morphogenic cell wall synthase implicated in Caulobacter stalk synthesis. Complementary, they pinpoint the bactofilin-interacting region within the PbpC C-terminus, known to interact with bactofilin. They further show that BacA localization is independent of PbpC.

Although the phenotypic effects of an abolished BacA-PbpC interaction are mild, these data significantly advance our understanding of bactofilin membrane binding, polymerization, and function at the molecular level. The major strength of the comprehensive study is the combination of complementary in vivo, in vitro and bioinformatic/simulation approaches, the results of which are consistent.

Author response:

The following is the authors’ response to the original reviews

Public Reviews:

Reviewer #1 (Public review):

Summary:

The investigators undertook detailed characterization of a previously proposed membrane targeting sequence (MTS), a short N-terminal peptide, of the bactofilin BacA in Caulobacter crescentus. Using light microscopy, single molecule tracking, liposome binding assays, and molecular dynamics simulations, they provide data to suggest that this sequence indeed does function in membrane targeting and further conclude that membrane targeting is required for polymerization. While the membrane association data are reasonably convincing, there are no direct assays to assess polymerization and some assays used lack proper controls as detailed below. Since the MTS isn't required for bactofilin polymerization in other bacterial homologues, showing that membrane binding facilitates polymerization would be a significant advance for the field.

We agree that additional experiments were required to consolidate our results and conclusions. Please see below for a description of the new data included in the revised version of the manuscript.

Major concerns

(1) This work claims that the N-termina MTS domain of BacA is required for polymerization, but they do not provide sufficient evidence that the ∆2-8 mutant or any of the other MTS variants actually do not polymerize (or form higher order structures). Bactofilins are known to form filaments, bundles of filaments, and lattice sheets in vitro and bundles of filaments have been observed in cells. Whether puncta or diffuse labeling represents different polymerized states or filaments vs. monomers has not been established. Microscopy shows mis-localization away from the stalk, but resolution is limited. Further experiments using higher resolution microscopy and TEM of purified protein would prove that the MTS is required for polymerization.

We do not propose that the MTS is directly involved in the polymerization process and state this more clearly now in the Results and Discussion sections of the revised manuscript. To address this point, we performed transmission electron microscopy studies comparing the polymerization behavior of wild-type and mutant BacA variants. The results clearly show that the MTS-free BacA variant (∆2-8) forms polymers that are indistinguishable from those formed by the wild-type protein, when purified from an E. coli overproduction strain (new Figure 1–figure supplement 1). This finding is consistent with structural work showing that bactofilin polymerization is exclusively mediated by the conserved bactofilin domain (Deng et al, Nat Microbiol, 2019). However, at native expression levels, BacA only accumulates to ~200 molecules per cell (Kühn et al, EMBO J, 2006). Under these conditions, the MTS-mediated increase in the local concentration of BacA at the membrane surface and, potentially, steric constraints imposed by membrane curvature, may facilitate the polymerization process. This hypothesis has now been stated more clearly in the Results and Discussion sections.

For polymer-forming proteins, defined localized signals are typically interpreted as slow-moving or stationary polymeric complexes. A diffuse localization, by contrast, suggests that a protein exists in a monomeric or, at most, (small) oligomeric state in which it diffuses rapidly within the cell and is thus no longer detected as distinct foci by widefield microscopy. Our single-molecule data show that BacA variants that are no longer able to interact with the membrane (as verified by cell fractionation studies and in vitro liposome binding assays) have a high diffusion rate, similar to that measured for the non-polymerizing and non-membrane-bound F130R variant. These results demonstrate that a defect in membrane binding strongly reduces the ability of BacA to form polymeric assemblies. To support this hypothesis, we have now repeated all single-particle tracking experiments and included mVenus as a freely diffusible reference protein. Our data confirm that the mobilities of the ∆2-8 and F130R variants are similar and approach those of free mVenus, supporting the idea that the deficiency to interact with the membrane prevents the formation of extended polymeric structures (which should show much lower mobilities). To underscore the relevance of membrane binding for BacA assembly, we have now included a new experiment, in which we used the PbpC membrane anchor (PbpC_1-132-mcherry) to restore the recruitment of the ∆2-8 variant to the membrane (Figure 9 and Figure 9–figure supplement 1). The results obtained show that the ∆2-8 variant transitions from a diffuse localization to polar foci upon overproduction of PbpC_1-132-mcherry. The polymerization-impaired F130R variant, by contrast, remains evenly distributed throughout the cytoplasm under all conditions. These findings further support the idea that polymerization and membrane-association are mutually interdependent processes.

(2) Liposome binding data would be strengthened with TEM images to show BacA binding to liposomes. From this experiment, gross polymerization structures of MTS variants could also be characterized.

We do not have the possibility to perform cryo-electron microscopy studies of liposomes bound to BacA. However, the results of the cell fractionation and liposome sedimentation assays clearly support a critical role of the MTS in membrane binding.

(3) The use of the BacA F130R mutant throughout the study to probe the effect of polymerization on membrane binding is concerning as there is no evidence showing that this variant cannot polymerize. Looking through the papers the authors referenced, there was no evidence of an identical mutation in BacA that was shown to be depolymerized or any discussion in this study of how the F130R mutation might to analogous to polymerization-deficient variants in other bactofilins mentioned in these references.

Residue F130 in the C-terminal polymerization interface of BacA is conserved among bactofilin homologs, although its absolute position in the protein sequence may vary, depending on the length of the N-terminal unstructured tail. The papers cited in our manuscript show that an exchange of this conserved phenylalanine residue abolishes polymer formation. Nevertheless, we agree that it is important to verify the polymerization defect of the F130R variant in the system under study. We have now included size-exclusion chromatography data showing that BacA-F130R forms a low-molecular-weight complex, whereas the wild-type protein largely elutes in the exclusion volume, indicating the formation of large, polymeric species (new Figure 1–figure supplement 1). In addition, we performed transmission electron microscopy analyses of BacA-F130R, which verified the absence of larger oligomers (new Figure 1–figure supplement 2).

(4) Microscopy shows that a BacA variant lacking the native MTS regains the ability to form puncta, albeit mis-localized, in the cell when fused to a heterologous MTS from MreB. While this swap suggests a link between puncta formation and membrane binding the relationship between puncta and polymerization has not been established (see comment 1).

We show that a BacA variant lacking the MTS (∆2-8) regains the ability to form membrane-associated foci when fused to the MTS of MreB. By contrast, a similar variant that additionally carries the F130R exchange (preventing its polymerization) shows a diffuse cytoplasmic localization. In addition, we show that the F130R exchange leads to a loss of membrane binding and to a considerable increase in the mobility of the variants carrying the MTS of E. coli MreB. As described above, we now provide additional data demonstrating that elevated levels of the PbpC membrane anchor can reinstate polar localization for the ∆2-8 variant, whereas it fails to do so for the polymerization-deficient F130R variant (Figure 9 and Figure 9–figure supplement 1). Together, these results support the hypothesis that membrane association and polymerization act synergistically to establish localized bactofilin assemblies at the stalked cell pole.

(5) The authors provide no primary data for single molecule tracking. There is no tracking mapped onto microscopy images to show membrane localization or lack of localization in MTS deletion/ variants. A known soluble protein (e.g. unfused mVenus) and a known membrane bound protein would serve as valuable controls to interpret the data presented. It also is unclear why the authors chose to report molecular dynamics as mean squared displacement rather than mean squared displacement per unit time, and the number of localizations is not indicated. Extrapolating from the graph in figure 4 D for example, it looks like WT BacA-mVenus would have a mobility of 0.5 (0.02/0.04) micrometers squared per second which is approaching diffusive behavior. Further justification/details of their analysis method is needed. It's also not clear how one should interpret the finding that several of the double point mutants show higher displacement than deleting the entire MTS. These experiments as they stand don't account for any other cause of molecular behavior change and assume that a decrease in movement is synonymous with membrane binding.

We now provide additional information on the single-particle analysis. A new supplemental figure now shows a mapping of single-particle tracks onto the cells in which they were recorded for all proteins analyzed (Figure 2–figure supplement 1). Due to the small size of C. crescentus, it is difficult to clearly differentiate between membrane-associated and cytoplasmic protein species. However, overall, slow-diffusing particles tend to be localized to the cell periphery, supporting the idea that membrane-associated particles form larger assemblies (apart from diffusing more slowly due to their membrane association). In addition, we have included a movie that shows the single-particle diffusion dynamics of all proteins in representative cells (Figure 2-video 1). Finally, we have included a table that gives an overview of the number of cells and tracks analyzed for all proteins investigated (Supplementary file 1). Figure 2A and 4D show the mean squared displacement as a function of time, which makes it possible to assess whether the particles observed move by normal, Brownian diffusion (which is the case here). We repeated the entire single-particle tracking analysis to verify the data obtained previously and obtained very similar results. Among the different mutant proteins, only the K4E-K7E variant consistently shows a higher mobility than the MTS-free ∆2-8 variant, with MSD values similar to that of free mVenus. The underlying reason remains unclear. However, we believe that an in-depth analysis of this phenomenon is beyond the scope of this paper. We re-confirmed the integrity of the construct encoding the K4E/K7E variant by DNA sequencing and once again verified the size and stability of the fusion protein by Western blot analysis, excluding artifacts due to errors during cloning and strain construction.

We agree that the single-molecule tracking data alone are certainly not sufficient to draw firm conclusions on the relationship between membrane binding and protein mobility. However, they are consistent with the results of our other in vivo and in vitro analyses, which together indicate a clear correlation between the mobility of BacA and its ability to interact with the membrane and polymerize (processes that promote each other synergistically).

(6) The experiments that map the interaction surface between the N-terminal unstructured region of PbpC and a specific part of the BacA bactofilin domain seem distinct from the main focus of the paper and the data somewhat preliminary. While the PbpC side has been probed by orthogonal approaches (mutation with localization in cells and affinity in vitro), the BacA region side has only been suggested by the deuterium exchange experiment and needs some kind of validation.

The results of the HDX analysis per se are not preliminary and clearly show a change in the solvent accessibility of backbone amides in the C-terminal region in the bactofilin domain in the presence of the PbpC_1-13 peptide. However, we agree that additional experiments would be required to verify the binding site suggested by these data. We agree that further research is required to precisely map and verify the PbpC binding site. However, as this is not the main focus of the paper, we would like to proceed without conducting further experiments in this area.

We now provide additional data showing that elevated levels of the PbpC membrane anchor are able to recruit the MTS-free BacA variant (∆2-8) to the cytoplasmic membrane and stimulate its assembly at the stalked pole (Figure 9). These results now integrate Figure 8 more effectively into the overall theme of the paper.

Reviewer #2 (Public review):

Summary:

The authors of this study investigated the membrane-binding properties of bactofilin A from Caulobacter crescentus, a classic model organism for bacterial cell biology. BacA was the progenitor of a family of cytoskeletal proteins that have been identified as ubiquitous structural components in bacteria, performing a range of cell biological functions. Association with the cell membrane is a common property of the bactofilins studied and is thought to be important for functionality. However, almost all bactofilins lack a transmembrane domain. While membrane association has been attributed to the unstructured N-terminus, experimental evidence had yet to be provided. As a result, the mode of membrane association and the underlying molecular mechanics remained elusive.

Liu at al. analyze the membrane binding properties of BacA in detail and scrutinize molecular interactions using in-vivo, in-vitro and in-silico techniques. They show that few N-terminal amino acids are important for membrane association or proper localization and suggest that membrane association promotes polymerization. Bioinformatic analyses revealed conserved lineage-specific N-terminal motifs indicating a conserved role in protein localization. Using HDX analysis they also identify a potential interaction site with PbpC, a morphogenic cell wall synthase implicated in Caulobacter stalk synthesis. Complementary, they pinpoint the bactofilin-interacting region within the PbpC C-terminus, known to interact with bactofilin. They further show that BacA localization is independent of PbpC.

Strengths:

These data significantly advance the understanding of the membrane binding determinants of bactofilins and thus their function at the molecular level. The major strength of the comprehensive study is the combination of complementary in vivo, in vitro and bioinformatic/simulation approaches, the results of which are consistent.

Thank you for this positive feedback.

Weaknesses:

The results are limited to protein localization and interaction, as there is no data on phenotypic effects. Therefore, the cell biological significance remains somewhat underrepresented.

We agree that it is interesting to investigate the phenotypic effects caused by the reduced membrane binding activity of BacA variants with defects in the MTS. We have now included phenotypic analyses that shed light on the role of region C1 in the localization of PbpC and its function in stalk elongation under phosphate-limiting conditions (see below).

Recommendations for the authors:

Reviewer #2 (Recommendations for the authors):

To address the missing estimation of biological relevance, some additional experiments may be carried out.

For example, given that BacA localizes PbpC by direct interaction, one might expect an effect on stalk formation if BacA is unable to bind the membrane or to polymerize. The same applies to PbpC variants lacking the C1 region. As the mutant strains are available, these data are not difficult to obtain but would help to compare the effect of the deletions with previous data (e.g. Kühn et al.) even if the differences are small.

We have now analyzed the effect of the removal of region C1 on the ability of mVenus-PbpC to promote stalk elongation in C. crescentus under phosphate starvation. Interestingly, our results show that the lack of the BacA-interaction motif impairs the recruitment of the fusion protein to the stalked pole, but it does not interfere with its stimulatory effect on stalk biogenesis. Thus, the polar localization of PbpC does not appear to be critical for its function in localized peptidoglycan synthesis at the stalk base. These results are now shown in Figure 8–Figure supplement 4. The results obtained may be explained by residual transient interactions of mVenus-PbpC with proteins other than BacA at the stalked pole. Notably, PbpC has also been implicated in the attachment of the stalk-specific protein StpX to components of the outer membrane at the stalk base. The polar localization of PbpC may therefore be primarily required to ensure proper StpX localization, consistent with previous work by Hughes et al. (Mol Microbiol, 2013) showing that StpX is partially mislocalized in a strain producing an N-terminally truncated PbpC variant that no longer localizes to the stalk base.

We have also attempted to investigate the ability of the Δ2-8 and F130R variants of BacA-mVenus to promote stalk elongation under phosphate starvation. However, the levels of the WT, Δ2-8 and F130R proteins and their stabilities were dramatically different after prolonged incubation of the cells in phosphate-limited medium, so that it was not possible to draw any firm conclusions from the results obtained (not shown).

In addition, the M23-like endopeptidase LdpA is proposed to be a client protein of BacA (in C. crescentus, Billini et al. 2018, and H. neptunium or R. rubrum, Pöhl et al. 2024). In H. neptunium, it is suggested that the interaction is mediated by a cytoplasmic peptide of LmdC reminiscent of PbpC. This should at least be commented on. It would be interesting to see, if LpdA in C. crescentus is also delocalized and if so, this could identify another client protein of BacA.

We agree that it would be interesting to study the role of BacA in LdpA function. However, we have not yet succeeded in generating a stable fluorescent protein fusion to LdpA, which currently makes it impossible to study the interplay between these two proteins in vivo. The focus of the present paper is on the mode of interaction between bactofilins and the cytoplasmic membrane and on the mutual interdependence of membrane binding and bactofilin polymerization. Given that PbpC is so far the only verified interaction partner of BacA in C. crescentus, we would like to limit our analysis to this client protein.

Further comments:

L105: analyze --> analyzed

Done.

L169: Is there any reason why the MTS of E. coli MreB was doubled?

Previous work has shown that two tandem copies of the N-terminal amphiphilic helix of E. coli MreB were required to partially target a heterologous fusion partner protein (GFP) to the cytoplasmic membrane of E. coli cells (Salje et al, 2011).

Fig. S3:

a) Please decide which tag was used (mNG or mVenus) and adapt the figure or legend accordingly.<br /> b) In the legend for panel (C), please describe how the relative amounts were calculated, as the fractions arithmetically cannot add to > 100%. I guess each band was densiometrically rated and independently normalized to the whole-cell signal?

The fluorescent tag used was mNeonGreen, as indicated in the figure. We have now corrected the legend accordingly. Thank you for making us aware of the wrong labeling of the y-axis. We have now corrected the figure and describe the method used to calculate the plotted values in the legend.

Legend of Fig 1b: It is not clear to me, to which part of panel B the somewhat cryptic LY... strain names belong. I suggest putting them either next to the images, to delete them, or at least to unify the layout (compare, e.g. to Fig S7). (I would delete the LY numbers and stay with the genes/mutations throughout. This is just a suggestion).

These names indicate the strains analyzed in panel B, and we have now clarified this in the legend. It is more straightforward to label the images according to the mutations carried by the different strains. Nevertheless, we would like to keep the strain names in the legend, so that the material used for the analysis can be clearly identified.

Fig. 2a: As some of the colors are difficult to distinguish, I suggest sorting the names in the legend within the graph according to the slope of the curves (e.g. K4E K7E (?) on top and WT being at the bottom).

Thank you for this suggestion. We have now rearranged the labels as proposed.

In the legend (L924), correct typo "panel C" to "panel B".

Done.

Fig. 3: In the legend, I suggest deleting the abbreviations "S" and "P" as they do not show up in the image. In line 929, I suggest adding: average "relative" amount... or even more precisely: "average relative signal intensities obtained..."

We have removed the abbreviations and now state that the bars indicate the “average relative signal intensities” obtained for the different fractions.

Fig 4d: same suggestion as for Fig. 2a.

Done.

Fig 8: In the legend (L978), delete 1x "the"

Done.

L258 and Fig. S5: The expression "To account for biases in the coverage of bacterial species" seems somewhat unclear. I suggest rephrasing and adding information from the M+M section here (e.g. from L593, if this is meant).

We now state that this step in the analysis pipeline was performed “To avoid biases arising from the over-representation of certain bacterial species in UniProt”.

I appreciate the outline of the workflow in panel (a) of Fig. S5. It would be even more useful when some more details about the applied criteria for filtering would be provided (e.g. concerning what is meant with "detailed taxonomic information" or "filter out closely related sequences". Does the latter mean that only one bactofilin sequence per species was used? (As quite many bacteria have more than one but similar bactofilins.)

We removed sequences from species with unclear phylogeny (e.g. candidate species whose precise taxonomic position has not yet been determined). For many pathogenic species, numerous strains have been sequenced. To account for this bias, only one sequence from clusters of highly similar bactofilin sequences (>90% identity) was retained per species. This information has now been included in the diagram. It is true that many bacteria have more than one bactofilin homolog. However, the sequences of these proteins are typically quite different. For instance, the BacA and BacB from C. crescentus only share 52% identity. Therefore, our analysis does not systematically eliminate bactofilin paralogs that coexist in the same species.

L281: Although likely, I am not sure if membrane binding has ever been shown for a bactofilin from these phyla. (See also L 380.) Is there an example? Otherwise, membrane binding may not be a property of these bactofilins.

To our knowledge, the ability of bactofilins from these clades to interact with membranes has not been investigated to date. We agree that the absence of an MTS-like motif may indicate that they lack membrane binding activity, and we have now stated this possibility in the Results and Discussion.

L285: See comment above concerning the M23-like peptidase LpdA. Although not yet directly shown for C. crescentus, it seems likely that BacACc does also localize this peptidase in addition to PbpC. I suggest rephrasing, e.g. "known" --> "shown"

We now use the word “reported”.

L295 and Fig S8: PbpC is ubiquitous. Which criteria/filters have been applied to select the shown sequences?

C. crescentus PbpC is different from E. coli Pbp1C. It is characterized by distinctive, conserved N- and C-terminal tails and only found in C. crescentus and close relatives. The C. crescentus homolog of E. coli PbpC is called PbpZ (Yakhnina et al, J Bacteriol, 2013; Strobel et al, J Bacterol, 2014), whereas C. crescentus PbpC is related to E. coli PBP1A. We have now added this information to the text to avoid confusion.

L311: may replace "assembly" by "polymerization"

Done.

L320: bactofilin --> bactofilin domain?

Yes, this was supposed to read “bactofilin domain”. Thank you for spotting this issue.

L324: The HDX analysis of BacA suggests that the exchange is slowed down in the presence of the PbpC peptide, which is indicative of a physical interaction between these two molecules. To corroborate the claim that BacA polymerization is critical for interaction with the peptide (resp. PbpC), this experiment should be carried out with the polymerization defective BacA version F130R.

(Or tone this statement down, e.g. show --> suggest.)

“suggest”

L386: undergoes --> undergo

Done.

L391-400: This idea is tempting but the suggested mechanism then would be restricted to bactofilins of C. crescentus and close relatives. The bactofilin of Rhodomicrobium, for example, was shown to localize dynamically and not to stick to a positively curved membrane.

In the vast majority of species investigated so far, bactofilins were found to associate with specifically curved membrane regions and to contribute to the establishment of membrane curvature. Unfortu­nately, the sequences of the three co-polymerizing bactofilin paralogs of R. vannielii DSM 166 studied by Richter et al (2023) have not been reported and the genome sequence of this strain is not publicly available. However, in related species with three bactofilin paralogs, only one paralog shows an MTS-like N-terminal peptide and another paralog typically contains an unusual cadherin-like domain of unknown function, as also reported for R. vannielii DSM 166. Therefore, the mechanism controlling the localization dynamics of bactofilins may be complex in the Rhodomicrobium lineage. Nevertheless, at native expression levels, the major bactofilin (BacA) of R. vannielii DSM 166 was shown to localize predominantly to the hyphal tips and the (incipient) bud necks, suggesting that regions of distinct membrane curvature could also play a role in its recruitment. We do not claim that all bactofilins recognize positive membrane curvature, which is clearly not the case. It rather appears as though the curvature preference of bactofilins varies depending on their specific function.

L405-406: I agree that localization of BacA has been shown to be independent of PbpC. However, this does not generally preclude an effect on BacA localization by other "client" or interacting proteins. (See also comment above about the putative BacA interactor LpdA). I suggest either to corroborate or to change this statement from "client binding" to "PbpC binding".

Thank you for pointing out the imprecision of this statement. We now conclude that “PbpC binding” is not critical for BacA assembly and positioning.

Suppl. Fig. S11: In the legend, please correct the copy-paste mismatch (...VirB...).

Done.

L482: delete 1x "at"

Done.

L484: may be better "soluble and insoluble fractions"?

We now describe the two fractions as “soluble and membrane-containing insoluble fractions” to make clear to all readers that membrane vesicles are found in the pellet after ultracentrifugation.

L489-490: check spelling immunoglobulin – immuneglobulin

Done.

L500 and 504: º_C --> ºC

Done.

Suppl. file X (HDX data): please check the table headline, table should be included in Suppl. file 1

We have now included a headline in this file (now Supplementary file 3).

eLife Assessment

This manuscript offers valuable structural and mechanistic insights into the structure and assembly of the Type II internal ribosome entry site (IRES) from encephalomyocarditis virus (EMCV) and the translation initiation complex, revealing a direct interaction between the IRES and the 40S ribosomal subunit. While a solid cryo-EM method was used, enhancing the overall resolution or adding complementary biochemical data would further improve the clarity and impact of this study. This manuscript will attract researchers in cap-independent translation, host-pathogen interactions, and virology.

Reviewer #1 (Public review):

Summary:

The authors have studied how a virus (EMCV) uses its RNA (Type 2 IRES) to hijack the host's protein-making machinery. They use cryo-EM to extract structural information about the recruitment of viral Type 2 IRES to ribosomal pre-IC. The authors propose a novel interaction mechanism in which the EMCV Type 2 IRES mimics 28S rRNA and interacts with ribosomal proteins and initiator tRNA (tRNAi).

Strengths:

(1) Getting structural insights about the Type 2 IRES-based initiation is novel.

(2) The study allows a good comparison of other IRES-based initiation systems.

(3) The manuscript is well-written and clearly explains the background, methods, and results.

Weaknesses:

(1) The main weakness of the work is the low resolution of the structure. This limits the possibility of data interpretation at the molecular level.

However, despite the moderate resolution of the cryo-EM reconstructions, the model fits well into the density. The analysis of the EMCV IRES-48S PIC structure is thorough and includes meaningful comparisons to previously published structures (e.g., PDB IDs - 7QP6 and 7QP7). These comparisons showed that Map B1 represents a closed conformation, in contrast to Map A in the open state (Figure 2). Additionally, the proposed 28S rRNA mimicry strategy supported by structural superposition with the 80S ribosome and sequence similarity between the I domain of the IRES and the h38 region of 28S rRNA (Fig. 4) is well-justified.

(2) The lack of experimental validation of the functional importance of regions like the GNRA and RAAA loops is another limitation of this study.

(3) Minor modifications related to data processing and biochemical studies will further validate and strengthen the findings.

a) In the cryo-EM data section, the authors should include an image showing rejected particles during 2D classification. This would help readers understand why, despite having over 22k micrographs with sufficient particle distribution and good contrast, only a smaller number of particles were used in the final reconstruction. Additionally, employing map-sharpening tools such as Ewald sphere correction, Bayesian polishing, or reference-based motion correction might further improve the quality of the maps. Targeting high-resolution structures would be particularly informative.

b) The strategic modelling of different IRES domains into the density, particularly the domain into the region above the 40S head, is appreciable. However, providing the full RNA tertiary structure (RNAfold) of the EMCV IRES (nucleotides 280-905) would better explain the logic behind the model building and its molecular interpretation.

c) Although the authors compare their findings with other types of IRESs (Types 1, 3, and 4), there is no experimental validation of the functional importance of regions like the GNRA and RAAA loops. Including luciferase-based assays or mutational studies of these regions for validation of structural interpretations is strongly recommended.

Reviewer #2 (Public review):

Summary:

The field of protein translation has long sought the structure of a Type 2 Internal Ribosome Entry Site (IRES). In this work, Das and Hussain pair cryo-EM with algorithmic RNA structure prediction to present a structure of the Type 2 IRES found in Encephalomyocarditis virus (EMCV). Using medium to low resolution cryo-EM maps, they resolve the overall shape of a critical domain of this Type 2 IRES. They use algorithmic RNA prediction to model this domain onto their maps and attempt to explain previous results using this model.

Strengths:

(1) This study reveals a previously unknown/unseen binding modality used by IRESes: a direct interaction of the IRES with the initiator tRNA.

(2) Use of an IRES-associated factor to assemble and pull down an IRES bound to the small subunit of the ribosome from cellular extracts is innovative.

(3) Algorithmic modeling of RNA structure to complement medium to low resolution cryo-EM maps, as employed here, can be implemented for other RNA structures.

Weaknesses:

(1) Maps at the resolution presented prevent unambiguous modelling of the EMCV-IRES. This, combined with the lack of any biochemical data, calls into question any inferences made at the level of individual nucleotides, such as the GNRA loop and CAAA loop (Figure 4).

(2) The EMCV IRES contains an upstream AUG at position 826, where the PIC can assemble (Pestova et al 1996; PMID 8943341). It is unclear if this start codon was mutated in this study. If it were not mutated, placement of AUG-834 over AUG-826 in the P-site is unexplained.

(3) The claims the authors make about (i) the general overall shape and binding site of the IRES, (ii) its gross interaction with the two ribosomal proteins, (iii) the P-in state of the 48S, (iv) the rearrangement of the ternary complex are all warranted. Their claims about individual nucleotides or smaller stretches of the IRES-without any supporting biochemical data-is not warranted by the data.

Reviewer #3 (Public review):

Summary:

Type II IRES, such as those from encephalomyocarditis virus (EMCV) and foot-and-mouth disease virus (FMDV), mediate cap-independent translation initiation by using the full complement of eukaryotic initiation factors (eIFs), except the cap-binding protein eIF4E. The molecular details of how IRES type II interacts with the ribosome and initiation factors to promote recruitment have remained unclear. Das and Hussain used cryo-electron microscopy to determine the structure of a translation initiation complex assembled on the EMCV IRES. The structure reveals a direct interaction between the IRES and the 40S ribosomal subunit, offering mechanistic insight into how type II IRES elements recruit the ribosome.

Strengths:

The structure reveals a direct interaction between the IRES and the 40S ribosomal subunit, offering mechanistic insight into how type II IRES elements recruit the ribosome.

Weaknesses:

While this reviewer acknowledges the technical challenges inherent in determining the structure of such a highly flexible complex, the overall resolution remains insufficient to fully support the authors' conclusions, particularly given that cryo-EM is the sole experimental approach presented in the manuscript.

The study is biologically significant; however, the authors should improve the resolution or include complementary biochemical validation.

Author response:

Reviewer #1 (Public review):

Summary:

The authors have studied how a virus (EMCV) uses its RNA (Type 2 IRES) to hijack the host's protein-making machinery. They use cryo-EM to extract structural information about the recruitment of viral Type 2 IRES to ribosomal pre-IC. The authors propose a novel interaction mechanism in which the EMCV Type 2 IRES mimics 28S rRNA and interacts with ribosomal proteins and initiator tRNA (tRNAi).

Strengths:

(1) Getting structural insights about the Type 2 IRES-based initiation is novel.

(2) The study allows a good comparison of other IRES-based initiation systems.

(3) The manuscript is well-written and clearly explains the background, methods, and results.

We thank Reviewer 1 for appreciating our efforts and finding structural insights about the type 2 IRES-based initiation presented in this study as novel.

Weaknesses:

(1) The main weakness of the work is the low resolution of the structure. This limits the possibility of data interpretation at the molecular level.

However, despite the moderate resolution of the cryo-EM reconstructions, the model fits well into the density. The analysis of the EMCV IRES-48S PIC structure is thorough and includes meaningful comparisons to previously published structures (e.g., PDB IDs - 7QP6 and 7QP7). These comparisons showed that Map B1 represents a closed conformation, in contrast to Map A in the open state (Figure 2). Additionally, the proposed 28S rRNA mimicry strategy supported by structural superposition with the 80S ribosome and sequence similarity between the I domain of the IRES and the h38 region of 28S rRNA (Fig. 4) is welljustified.

We agree that the low resolution of the map has compromised the data interpretation at the molecular level, and we thank the reviewer for appreciating our findings at this resolution. Due to the compromise in resolution, we have reported findings related to stretches or regions such as loops and stems, rather than individual nucleotides and interactions.  

(2) The lack of experimental validation of the functional importance of regions like the GNRA and RAAA loops is another limitation of this study.

We agree with the lack of any additional experiments other than Cryo-EM for probing the importance of regions such as GNRA and RAAA loops in this study. However, we have cited earlier reports that demonstrate the importance of these regions for overall IRES activity. The essentiality of RAAA loop for type 2 IRES was demonstrated in earlier report López de Quinto and Martínez-Salas, 1997 (Cited in manuscript). Further, the conservation of this loop across the type 2 IRES family adds to the importance of this loop (Manuscript Figure 6B). This loop and its flanking G-C stem are similar to h38 of 28S rRNA, and it appears that RAAA loop adopts a mimicry mechanism to interact with the 40S ribosomal protein- uS19, thus highlighting its importance for interaction with 40S. Experiments destabilising the G-C stem also compromise IRES activity, as shown in the case of FMDV IRES (Fernández et al 2011). Previous studies related to the mutation of the GNRA or GCGA loop in EMCV IRES have shown a deficiency in IRES activity (Roberts and Belsham, 1997; Robertson et al 1999), suggesting the importance of these regions in the viral IRES biology, and these reports are cited in the manuscript. Not only EMCV IRES, but mutation in the GUAA (representative of GNRA) loop of FMDV IRES also showed significant reduction in IRES activity (López de Quinto and Martínez-Salas, 1997). In our study, we observe that GCGA loop interacts with tRNA_i in EMCV IRES-48S PIC, thus implicating the importance of this loop. Moreover, incubation of FMDV IRES with 40S ribosomes has shown a decrease in SHAPE reactivity in domain 3 apex (position 170- 200 nucleotides) (Lozano et al 2018), which corresponds to EMCV IRES domain I apex. Further, we will attempt to address the concern of lack of experimental validation of GNRA and RAAA loops by performing biochemical assays.

(3) Minor modifications related to data processing and biochemical studies will further validate and strengthen the findings.

a) In the cryo-EM data section, the authors should include an image showing rejected particles during 2D classification. This would help readers understand why, despite having over 22k micrographs with sufficient particle distribution and good contrast, only a smaller number of particles were used in the final reconstruction. Additionally, employing mapsharpening tools such as Ewald sphere correction, Bayesian polishing, or reference-based motion correction might further improve the quality of the maps. Targeting high-resolution structures would be particularly informative.

We thank the reviewer for the suggestions, and we would employ suggested processes that may help improve the quality of the maps further. We will include image for rejected 2D classes in the revised manuscript. We agree with the Reviewer’s query related to the substantial number of micrographs and smaller number of particles for the final reconstruction. The total number of micrographs is the summation of multiple datasets, prepared and collected at various times. Among these, around 8000 micrographs have extremely poor particle number and distribution. As a result, the number of particles per micrograph is heterogeneous in the compiled dataset. We obtained only 237054 ‘good particles’ after multiple rounds of 2D & 3D classifications, and the final reconstruction has 28439 particles (~12%). This class was obtained after masked classification for IRES and ternary complex density. Hence, only the particles that show the best density for both IRES and ternary complex are used for reconstructing this map. Another set of particles that have only a portion of IRES and tRNA but NO density for eIF2 forms another map (26792 particles, 11.3%). Thus, we obtained a total of 55231 particles (23.3%) with IRES density.  

b) The strategic modelling of different IRES domains into the density, particularly the domain into the region above the 40S head, is appreciable. However, providing the full RNA tertiary structure (RNAfold) of the EMCV IRES (nucleotides 280-905) would better explain the logic behind the model building and its molecular interpretation.

We thank the reviewer for appreciating the modelling of the domain I apex in the cryo-EM density. We tried to predict the full tertiary structure of the IRES, however, inclusion of the full-length sequence from 280-905 gave models of extremely low confidence, and few domains do not abide by the secondary structure of EMCV IRES as reported in Duke et al 1992. Hence, we used individual domains of EMCV IRES and predicted the tertiary structure independent of other IRES domains. Furthermore, 3D models of FMDV IRES domains 2, 3, and 4 (corresponding to EMCV IRES domains- H, I, and J-K) were predicted from SHAPE reactivity values and RNAComposer server (Figure 3 in Lozano et al 2018). The predicted architecture of domain 3 apex (FMDV IRES) coincides with our I domain apex model (EMCV IRES).

c)  Although the authors compare their findings with other types of IRESs (Types 1, 3, and 4), there is no experimental validation of the functional importance of regions like the GNRA and RAAA loops. Including luciferase-based assays or mutational studies of these regions for validation of structural interpretations is strongly recommended.

We have discussed the possibility of how the other IRESs, such as type 1 and type 5 (Aichi virus), might use similar strategies as EMCV IRES to assemble the 48S PIC, given the similarity in the motif sequence and position across the viral IRESs. Like EMCV IRES, the type 1 IRES (e.g. Poliovirus, Coxsackie virus) also harbours the GNRA loop, preceded by a C-rich loop at its longest domain, known for long-range RNA-RNA interactions. The segment harbouring GNRA loop is highly conserved across the type 1 family of IRESs (Kim et al 2015).The Aichi viral IRES (type 5) harbours a GNRA loop in its longest domain, which is domain J. Deletion of the GNRA loop has compromised the IRES activity; however, substitution mutations in this region either elevated the IRES activity or it remained unaltered (Yu et al 2011). We have hypothesized that these IRESs (type 1 and type 5) might use the GNRA motifs in their longest domain (domain IV in type 1, and domain J in type 5) similar to that of EMCV IRES, where GNRA is present in the longest domain (I) and preceded by a C-rich loop. Thus, GNRA can potentially mediate long-range interactions with tRNA_i as all these IRESs require eIF2-ternary complex for the formation of 48S PIC. Parallelly, like EMCV IRES, type 1 and type 5 IRESs also have similar placement of GNRA motif-containing domain before the eIF4G-binding domain (domain J-K in EMCV IRES, domain V in poliovirus, domain K in Aichi virus). Hence, we suggest the possibility of a similar strategy by these IRESs to interact with tRNA_i during the formation of 48S PIC.  

Reviewer #2 (Public review):

Summary:

The field of protein translation has long sought the structure of a Type 2 Internal Ribosome Entry Site (IRES). In this work, Das and Hussain pair cryo-EM with algorithmic RNA structure prediction to present a structure of the Type 2 IRES found in Encephalomyocarditis virus (EMCV). Using medium to low resolution cryo-EM maps, they resolve the overall shape of a critical domain of this Type 2 IRES. They use algorithmic RNA prediction to model this domain onto their maps and attempt to explain previous results using this model.

Strengths:

(1) This study reveals a previously unknown/unseen binding modality used by IRESes: a direct interaction of the IRES with the initiator tRNA.

(2) Use of an IRES-associated factor to assemble and pull down an IRES bound to the small subunit of the ribosome from cellular extracts is innovative.

(3) Algorithmic modeling of RNA structure to complement medium to low resolution cryoEM maps, as employed here, can be implemented for other RNA structures.

We thank Reviewer 2 for positive and encouraging comments on our work, appreciating our ‘innovative’ approach of using IRES-associated factor to assemble and pull down IRES-bound ribosomal complex.  

Weaknesses:

(1) Maps at the resolution presented prevent unambiguous modelling of the EMCV-IRES. This, combined with the lack of any biochemical data, calls into question any inferences made at the level of individual nucleotides, such as the GNRA loop and CAAA loop (Figure 4).

We understand the concerns raised by the reviewer related to the resolution of the EMCV IRES-48S PIC map. However, we would like to mention that we refrained from commenting on individual nucleotides or molecular interactions in the manuscript. Instead, we discuss about loops, RNA stretches or motifs that could be inferred with more confidence as shown in Manuscript Figure 4. The EMCV IRES can directly interact with the 40S ribosome using its domain H and I (Chamond et al 2014), however, the details this interaction was unknown. We observe that the CAAA loop of domain I apex interacts with 40S ribosome based on the placement of portion of domain I in the cryo-EM map. This is also reflected in the earlier reported SHAPE data (Supplementary figures 2, and 8 in Chamond et al 2014), where a decrease in reactivity is evident in the presence of 40S ribosome. In addition, incubation of EMCV IRES with rabbit reticulocyte lysate (RRL) offered protection to domain I apex regions, which included the CAAA loop (Figure 4b in Maloney and Joseph, 2024).

Furthermore, this decrease in SHAPE reactivity pattern is also evident for FMDV IRES domain 3 apex (like domain I in EMCV IRES) in the presence of 40S ribosome (Lozano et al 2018).

Thus, these studies are consistent with the placement of IRES model in the cryo-EM map.

We aim to improve the resolution of the maps for better clarity and add biochemical experiments to justify the possible interactions.

(2) The EMCV IRES contains an upstream AUG at position 826, where the PIC can assemble (Pestova et al 1996; PMID 8943341). It is unclear if this start codon was mutated in this study. If it were not mutated, placement of AUG-834 over AUG-826 in the P-site is unexplained.

We thank the reviewer for bringing up this point, as we missed mentioning this in the manuscript. The EMCV IRES does not require scanning and directly positions the AUG-834 at the P site (Pestova et al 1996). In Pestova et al 1996, the intensity of the toeprint at AUG-834 is much more intense than that of AUG-826. Further, AUG-834 lies in the Kozak context, whereas AUG-826 has a poor Kozak context. Furthermore, the synthesis of the polypeptide requires placement of AUG-834 at the P site. In our cryo-EM map, we observed that the tRNA_i is in a P_IN state, which indicates the recognition of the start codon, and we reasoned that it is more likely that AUG-834 is placed at the P site than AUG-826. We will mention this in the revised manuscript, as we had NOT mutated AUG-826.

(3) The claims the authors make about (i) the general overall shape and binding site of the IRES, (ii) its gross interaction with the two ribosomal proteins, (iii) the P-in state of the 48S, (iv) the rearrangement of the ternary complex are all warranted. Their claims about individual nucleotides or smaller stretches of the IRES-without any supporting biochemical data-is not warranted by the data.

We thank the reviewer for warranting major claims, and we wish to make further improvements to support our assessment of small stretches and individual nucleotides.

Reviewer #3 (Public review):

Summary:

Type II IRES, such as those from encephalomyocarditis virus (EMCV) and foot-and-mouth disease virus (FMDV), mediate cap-independent translation initiation by using the full complement of eukaryotic initiation factors (eIFs), except the cap-binding protein eIF4E. The molecular details of how IRES type II interacts with the ribosome and initiation factors to promote recruitment have remained unclear. Das and Hussain used cryo-electron microscopy to determine the structure of a translation initiation complex assembled on the EMCV IRES. The structure reveals a direct interaction between the IRES and the 40S ribosomal subunit, offering mechanistic insight into how type II IRES elements recruit the ribosome.

Strengths:

The structure reveals a direct interaction between the IRES and the 40S ribosomal subunit, offering mechanistic insight into how type II IRES elements recruit the ribosome.

Weaknesses:

While this reviewer acknowledges the technical challenges inherent in determining the structure of such a highly flexible complex, the overall resolution remains insufficient to fully support the authors' conclusions, particularly given that cryo-EM is the sole experimental approach presented in the manuscript.

The study is biologically significant; however, the authors should improve the resolution or include complementary biochemical validation.

We thank Reviewer 3 for acknowledging the technical challenges in this study and finding our study biologically significant. We understand the concerns related to low resolution and the requirement of complementary biochemical validation for our reported observations and interpretations in the manuscript. We are attempting to improve the resolution and complement the interpretations with biochemical experiments.

eLife Assessment

This valuable investigation provides new and solid evidence for a specific cognitive deficit in cerebellar degeneration patients. The authors use three tasks that modulate complexity and violations of cognitive expectations. They show specific slowing of reaction times in the presence of violations but not with task complexity. While some alternative interpretations of the results are possible and are discussed, the work provides a new, invaluable data point in describing the cognitive contribution of cerebellar processing.

Reviewer #1 (Public review):

Summary:

The authors test the hypothesis that the contribution of the cerebellum to cognitive tasks is similar to motor tasks, and is related to the processing of prediction errors (here: violation of expectations, VE). In three experiments, they find that cerebellar patients show differences compared to controls in measures of VE, but not task complexity. The findings show that cerebellar disease results in deficits in VE processing in cognitive tasks, and makes a valuable contribution of the field. The authors were able to test a large number of patients with cerebellar disease which is known to primarily affect the cerebellum (i.e. SCA6).

Strengths:

A strength of the study is that it is hypothesis-driven and that the three experiments are very well thought out. Furthermore, a comparatively large group of patients with spinocerebellar ataxia type 6 (SCA6) was tested, a disease which affects primarily the cerebellum.

Weaknesses:

- Acquisition of brain MRI scans would have been useful to perform lesion-behaviour-mapping. But this does not limit the significance of the behavioural findings.<br /> - Exp. 1 and 2: The lack of difference in accuracy was that an unexpected finding? How meaningful are the used paradigms when accuracy was the same in cerebellar patients and controls?<br /> - Exp. 1 and 2: Cerebellar patients have motor dysfunction which impacts reaction time. Can the authors exclude that this contributed at least in part to their findings? Any correlations to SARA score (upper limb function) or oculomotor dysfunction (e.g. presence of nystagmus)?<br /> - Data on the attention probes which have been done would be of interest. Were there any differences in attention between patients and controls, any correlations with the findings?

Comments on revisions:

I am not sure if I can follow the interpretation of the authors that the cerebellum contributes to prediction errors, but not predictions; These two are tightly connected? It may rather be that in patients with slowly progressive chronic disease there is a lot of compensation? It is not so rare that in cognitive tasks cerebellar patients do not perform differently from controls, even though one would expect a difference (e.g. based on fMRI data in controls)? Another factor which likely adds is age, Patients and controls are often middle-aged and elderly, adding to variability, decreasing the chance to see group differences?

Author response:

The following is the authors’ response to the original reviews

Joint Public Review:

Summary:

In this study, Daniel et al. used three cognitive tasks to investigate behavioral signatures of cerebellar degeneration. In the first two tasks, the authors found that if an equation was incorrect, reaction times slowed significantly more for cerebellar patients than for healthy controls. In comparison, the slowing in the reaction times when the task required more operations was comparable to normal controls. In the third task, the authors show increased errors in cerebellar patients when they had to judge whether a letter string corresponded to an artificial grammar.

Strengths:

Overall, the work is methodologically sound and the manuscript well written. The data do show some evidence for specific cognitive deficits in cerebellar degeneration patients.

Thank you for the thoughtful summary and constructive feedback. We are pleased that the methodological rigor and clarity of the manuscript were appreciated, and that the data were recognized as providing meaningful evidence regarding cognitive deficits in cerebellar degeneration.

Weaknesses:

The current version has some weaknesses in the visual presentation of results. Overall, the study lacks a more precise discussion on how the patterns of deficits relate to the hypothesized cerebellar function. The reviewers and the editor agreed that the data are interesting and point to a specific cognitive deficit in cerebellar patients. However, in the discussion, we were somewhat confused about the interpretation of the result: If the cerebellum (as proposed in the introduction) is involved in forming expectations in a cognitive task, should they not show problems both in the expected (1+3 =4) and unexpected (1+3=2) conditions? Without having formed the correct expectation, how can you correctly say "yes" in the expected condition? No increase in error rate is observed - just slowing in the unexpected condition. But this increase in error rate was not observed. If the patients make up for the lack of prediction by using some other strategy, why are they only slowing in the unexpected case? If the cerebellum is NOT involved in making the prediction, but only involved in detecting the mismatch between predicted and real outcome, why would the patients not show specifically more errors in the unexpected condition?

Thank you for asking these important questions and initiating an interesting discussion. While decision errors and processing efficiency are not fully orthogonal and are likely related, they are not necessarily the same internal construct. The data from Experiments 1 and 2 suggest impaired processing efficiency rather than increased decision error. Reaction time slowing without increased error rates suggests that the CA group can form expectations but respond more slowly, possibly due to reduced processing efficiency. Thus, this analysis of our data suggests that the cerebellum is not essential for forming expectations, but it plays a critical role in processing their violations.

Relatedly, a few important questions remain open in the literature concerning the cerebellum’s role in expectation-related processes. The first is whether the cerebellum contributes to the formation of expectations or the processing of their violations. In Experiments 1 and 2, the CA group did not show impairments in the complexity manipulation. Solving these problems requires the formation of expectations during the reasoning process. Given the intact performance of the CA group, these results suggest that they are not impaired in forming expectations. However, in both Experiments 1 and 2, patients exhibited selective impairments in solving incorrect problems compared to correct problems. Since expectation formation is required in both conditions, but only incorrect problems involve a VE, we hypothesize that the cerebellum is involved in VE processes. We suggest that the CA group can form expectations in familiar tasks, but are impaired in processing unexpected compared to expected outcomes. This supports the notion that the cerebellum contributes to VE, rather than to forming expectations.

In Experiment 3, during training, the participant is learning a novel rule (grammar), forming new expectations on how strings of letters should be. Afterwards, during testing, the participant is requested to identify if a novel string is following the rule or not. We examined sensitivity to distinguish between grammatical and non‐grammatical strings of letters, thus taking into account a baseline ability to identify expected strings. Additionally, both in the low‐similarity and highsimilarity conditions, there are expectations regarding whether the strings are following the rule or not. However, in the high‐similarity condition, there is more uncertainty regarding which strings are following the grammatical rule, as demonstrated in a lower sensitivity (d prime). Given the group differences only in the low similarity condition, these results suggest the CA group is impaired only when the rules are more certain. Given these results, we suggest that forming cognitive expectations is not necessarily dependent on the cerebellum. Rather, we propose that the cerebellum is critical for processing rule-based VE (detection or processing of detected errors) under conditions of more certainty. One remaining question for future studies is whether the cerebellum contributes to detection of a mismatch between the expectation and sensory evidence, or the processing of a detected VE. 

We suggest that these key questions are relevant to both motor and non-motor domains and were not fully addressed even in the previous, well-studied motor domain. Importantly, while previous experimental manipulations17,19,40,94–96 have provided important insights regarding the cerebellar role in these processes, some may have confounded these internal constructs due to task design limitations (e.g., lack of baseline conditions). Notably, some of these previous studies did not include control conditions, such as correct trials, where there was no VE. In addition, other studies did not include a control measure (e.g., complexity effect), which limits their ability to infer the specific cerebellar role in expectation manipulation. 

Thus, the current experimental design used in three different experiments provides a valuable novel experimental perspective, allowing us to distinguish between some, but not all, of the processes involved in the formation of expectations and their violations. For instance, to our knowledge, this is the first study to demonstrate a selective impairment in rule-based VE processing in cerebellar patients across both numerical reasoning and artificial grammar tasks. If feasible, we propose that future studies should disentangle different forms of VE by operationalizing them in experimental tasks in an orthogonal manner. This will allow us to achieve a more detailed and well-defined cerebellar motor and non-motor mechanistic account.

Recommendations for the authors:

Editors comments:

The Figures are somewhat sub-standard and should be improved before the paper is made the VOR. Ensure consistent ordering of the group factor (CA, NT) and experimental factor across Figure 3,4, and 6 (panels A). Having the patient group as columns in Figure 4a and in rows in Figure 6a is very confusing.

We have standardized the layout across Figures 2, 4, and 6 so that the group factor (CA, NT) and experimental conditions are consistently ordered. In all panels, the group factor now appears as a column.

Subpanels should be numbered A,B,C... not A, B1, B2.

Subpanel labels have been updated to follow the standard A, B, C format across all figures.

Fonts should have a 100% aspect ratio - they should not be stretched (Figure 6B).

We have corrected the font aspect ratios in all figures (e.g., Figure 6B) to ensure proper proportions and readability. 

Colors should be more suitable to print - use a CYMK color scheme (i.e. avoid neon colors such as the neon green for the CA).

The color scheme across all figures has been revised to be print-friendly using CMYKcompatible, colorblind-accessible palettes. Neon green for the CA group was replaced with a more muted, distinguishable color.

Abstract: "The CA group exhibited a disproportionate cost when comparing expected problems compared to unexpected problems" - I recommend switching unexpected and expected, as the disproportional cost in on the former.

We have changed the wording of the sentence accordingly. 

Upon re-reading the details for the AGL task were not clear to us. Please do not rely on the reference (78) for the details - your paper should contain enough information to have the reader understand the experimental details. For you to appreciate the depth of our not-understanding, here a simple question: The test strings either followed the grammar in Fig 5 or they did not. If they did not, how exactly was similarity to the grammar measured? If they did, what was the difference between the “Grammatical-high” and “Grammatical-low” trials? If the string was grammatical, there should not be a notion of similarity, no? Or where these trials arbitrary split in half? 

We have clarified that 50% of the test strings followed the grammar of the training strings. We also elaborated on the calculation of chunk strength as a measure of similarity between the training and testing strings, similar to the previous papers. The differences between low and high similarity are explained in the paper. Specifically, for each test string, we calculated chunk strength by summing the frequencies of all relevant substrings (e.g., bigrams and trigrams) that appeared in the training set. The test strings whose chunk‐strength values fell above the median for grammatical items were classified as “high similarity,” while those falling below the median were classified as “low similarity.” Also, grammatical strings can be of both low and high similarity; this is precisely the beautiful aspect of this experimental manipulation, showing the importance of uncertainty. We have utilized a 2 × 2 fully orthogonal design (grammaticality × similarity).

Experimental details of the task should be added to the Method section. In the results you should only mention the experimental details that are necessary for understanding the experiments, but details such as the number of trials, etc, can be moved to the methods. 

We have now moved the experimental task details to the Method sections.

Reviewer #1 (Recommendations for the author):

Studies have been done online and not in the lab. Could that have affected the results?

We addressed this in the Methods section, referring to established protocols for online neuropsychological testing[9–12]. Our results align with similar in-lab findings in both the subtraction and AGL tasks, supporting the online approach's robustness. 

Figure 2, B1; Figure 4, B1; Figure 6B: How many patients performed worse than the (worst-performing) controls? There appears to be quite some overlap between patients and controls. In the patients who performed worse, was there any difference from the other patients (e.g. disease severity as assessed by SARA score, repeat length, data of attention probes)?

We appreciate the reviewer’s thoughtful comment. We considered conducting individual-level comparisons to identify patients who performed worse than the lowest-performing controls. However, defining "worse" based on the performance of the lowest control is only one possible criterion. Other definitions—such as a specific number (1/2/3?) of standard deviations below the control mean—are also commonly used in literature, and each may yield different conclusions. This variability highlights the lack of a standardized threshold for what constitutes “worse” or "impaired" performance at the individual level. Given this ambiguity, and in line with prior studies that focus on average group differences rather than “impairment” prevalence, we chose not to include these individual-level comparisons. We believe this approach better aligns with the goals and design of the current study. That said, we agree that examining individual variability is important and may be more appropriate in future studies with larger samples so that percentage is a more robust measure. However, given the rarity of the disease, this would also be a challenge for future studies.  

SARA ataxia scale does not include oculomotor function. In SCA6 oculomotor deficits are frequent, eg, downbeat nystagmus. Please include information on oculomotor dysfunction.

We thank the reviewer for this important observation. While it is true that the SARA scale does not explicitly assess oculomotor function, our experimental design – in all three experiments – has control conditions that help account for general processing differences, including those that could arise from oculomotor deficits. These conditions, such as the correct trials and the complexity effects, allow us to isolate effects specifically related to the violation of expectation while minimizing the influence of broader performance factors, such as eye movement abnormalities. We also note that, while some patients can experience oculomotor symptoms such as downbeat nystagmus, none of our tasks required precise visual tracking or gaze shifts. In our experimental tasks, stimuli were centrally presented, and no visual tracking or saccadic responses were required. Moreover, the response time windows and stimulus durations (>2–5 s) were sufficient to mitigate the effects of delayed visual processing due to oculomotor impairment.

Why was MoCA used and not the CCAS-Schmahmann scale to assess cognitive function?

We selected the MoCA due to its broad clinical utility, time efficiency, and ability to detect mild cognitive impairment specifically in CA[101,102].  

Were there any signs of depression in the patient group that could have affected the results?

None of the patients had a clinical diagnosis of depression or were undergoing psychiatric treatment.  

Additionally, the interaction between group and expectancy was insignificant when RT was the depended vaibale .." = variable

This has been corrected to "variable" in the revised manuscript.

Reviewer #2 (Recommendations for the authors):

The terms 'unexpected' and 'expected' conditions are confusing. [...] Terming this 'violation of expectation' seems unnecessarily complicated to me. 

We thank the reviewer for raising this important concern. We recognize that the terms "expected" and "unexpected" can be ambiguous without clarification, and that "violation of expectation" (VE) may initially appear unnecessary. Our choice to use VE terminology is grounded in an established theoretical framework that distinguishes between mere stimulus correctness and prediction mechanisms. Specifically, VE captures the internal processing of mismatches between anticipated and observed outcomes, which we believe is central to the cerebellar function under investigation. While simpler, technical alternatives (e.g., "correct" vs. "incorrect") could describe the stimuli, we find that VE more accurately reflects the mental constructs under study and is consistent with previous literature in both motor and cognitive domains. 

Both tasks provide an error (or violation of expectation) that is non-informative and therefore unlikely to be used to update a forward model. The authors draw on motor literature to formulate a cognitive task where the presence of an error would engage the cerebellum and lead to longer reaction times in cerebellar patients. But in the motor domain, mismatch of sensory feedback and expectations would lead to an updating of the internal forward model. It seems unlikely to me in the arithmetic and alphabetic addition tasks that patients would update their internal model of addition according to an error presented at the end of each trial. If the error processed in these tasks will not lead to the updating of the internal forward model, can the authors discuss to what extent the cerebellum will be engaged similarly in these tasks, and what exactly connects cerebellar processing in these motor and cognitive tasks.

We thank the reviewer for this thoughtful and important comment. We fully agree that the current tasks do not directly probe learning-related updating of internal models. As stated in the paper, the goal of the present study was not to support or refute a specific claim regarding the cerebellum’s role in learning processes. Rather, our focus was on examining cerebellar involvement in the processing of VE. While we were inspired by models from the motor domain, our design was not intended to induce learning or adaptation per se, but to isolate the processing of unexpected outcomes. We agree that the tasks in their current form are unlikely to engage forward model updating in the same way as in sensorimotor adaptation paradigms. That said, we believe the current findings can serve as a basis for future research exploring the relationship between cerebellar prediction error processing and learning over time. As we also noted in the paper, this is a direction we propose, and actively pursuing, in ongoing research work.

The colour scheme is difficult for anyone with colour blindness or red-green visual impairment. Please adjust.

All figures have been revised to use CMYK-compatible, colorblind-safe palettes, and neon colors have been removed.

The introduction is a bit difficult to understand, because the authors draw on a number of different theories about cerebellar functioning, without clearly delineating how these relate to each other. For example: a) In the paragraph beginning with 'notably': If the cerebellum is required for sequential operations, why does it show the impairment with the rotation of the letters?

We understand the concern that if the cerebellum is involved in sequential operations, its involvement in mental letter rotation, which can be assumed as “continuous transformation,” may appear contradictory. We note that the boundary between continuous and stepwise, procedural operations is not always clear-cut and may vary depending on the participant's strategy or previous knowledge, which is not fully known to the researchers. Furthermore, to our knowledge, prior work on mental rotation has not directly investigated the impact of VE during this task. However, these are two debatable considerations. 

More importantly, a careful reading of our paper suggests that our experiments were designed to examine VE within tasks that involve sequential processing. Notably, we are not claiming that the cerebellum is involved in sequential or procedural processing per se. Rather, our findings point to a more specific role for the cerebellum in processing VE that arises during the construction of multistep procedural tasks. In fact, the results indicate that while the cerebellum may not be directly involved in the procedural process itself, it is critical when expectations are violated within such a context. This distinction is made possible in our study by the inclusion of a control condition (the complexity effect), which allows for a unique dissociation in our experimental design—one that, to our knowledge, has not been sufficiently addressed in previous studies.

Additionally, in the case of arithmetic problem solving—such as the tasks used in prior studies cited in our manuscript21—there is substantial evidence that these problems are typically solved through stepwise, procedural operations. Arithmetic reasoning, used in Experiments 1 and 2, has been robustly associated with procedural, multi-step strategies, which may be more clearly aligned with traditional views of cerebellar involvement in sequential operations. Thus, we propose that the role of the cerebellum in continuous transformations should be further examined. 

We suggest a more parsimonious theory that the cerebellum contributes to VE,  a field that was highly examined before. Yet, to reconcile ours and previous findings, we propose that the cerebellum’s contribution may not be limited to either continuous or stepwise operations per se, but rather to a domain-general process: the processing of VE. This theoretical framework can explain performance patterns across both mental rotation tasks and stepwise, procedural arithmetic.   

The authors mention generation prediction as a function of the cerebellum, processing of prediction errors (or violations of expectations), sequentially, and continuous transformations - but it is unclear whether the authors are trying to dissociate these from each other or whether ALL of these functions have informed task design.

We propose that the cerebellum’s contribution may not be limited to either continuous transformations or stepwise, procedural operations per se, but rather to a domain-general process: the processing of VE. We would like to clarify that we do not claim the cerebellum contributes to continuous transformations only, as suggested in some earlier work[21]. Rather, it could be that the cerebellum may contribute to continuous transformations, but we propose that it also supports multi-step, procedural processes. Given that framework, in the current study, across three separate experiments, we demonstrated that the cerebellum can also contribute to procedural, multi-step reasoning tasks.  

Minor Comments

Typo under paragraph beginning with 'notably' - cerebellum role should be cerebellar role.

Corrected as suggested.

When mentioning sequences as a recruiting feature for the cerebellum in the introduction, Van Overwalle's extensive work in the social domain should be referenced for completeness.

Thank you for the suggestion. We have now cited Van Overwalle’s work on cerebellar involvement in sequence processing within the social domain in the revised Introduction.

eLife Assessment

This study provides fundamental insights into eukaryotic phosphate homeostasis by demonstrating how yeast vacuoles dynamically regulate cytosolic phosphate levels. The conclusions are convincing, supported by an elegant combination of in vitro assays and in vivo measurements. This study will be of interest to cell biologists, particularly for those who are working in the field of phosphate metabolism.

Reviewer #1 (Public review):

The manuscript by Bru et al. focuses on the role of vacuoles as a phosphate buffering system for yeast cells. The authors describe here the crosstalk between the vacuole and the cytosol using a combination of in vitro analyses of vacuoles and in vivo assays. They show that the luminal polyphosphatases of the vacuole can hydrolyze polyphosphates to generate inorganic phosphate, yet they are inhibited by high concentrations. This balances the synthesis of polyphosphates against the inorganic phosphate pool. Their data further show that the Pho91 transporter provides a valve for the cytosol as it gets activated by a decline in inositol pyrophosphate levels. The authors thus demonstrate how the vacuole functions as a phosphate buffering system to maintain a constant cytosolic inorganic phosphate pool.

This is a very consistent and well-written manuscript with a number of convincing experiments, where the authors use isolated vacuoles and cellular read-out systems to demonstrate the interplay of polyphosphate synthesis, hydrolysis, and release. The beauty of this system the authors present is the clear correlation between product inhibition and the role of Pho91 as a valve to release Pi to the cytosol to replenish the cytosolic pool. I find the paper overall an excellent fit and only have a few issues, including :

(1) Figure 3: The authors use in their assays 1 mM ZnCl2 or 1mM MgCl2. Is this concentration in the range of the vacuolar luminal ion concentration? Did they also test the effect of Ca2+, as this ion is also highly concentrated in the lumen?

(2) Regarding the concentration of 30 mM K-PI, did the authors also use higher and lower concentrations? I agree that there is inhibition by 30 mM, but they cannot derive conclusions on the luminal concentration if they use just one in their assay. A titration is necessary here.

(3) What are the consequences on vacuole morphology if the cells lack Pho91?

(4) Discussion: The authors do not refer to the effect of calcium, even though I would expect that the levels of the counterion should affect the phosphate metabolism. I would appreciate it if they would extend their discussion accordingly.

(5) I would appreciate a brief discussion on how phosphate sensing and control are done in human cells. Do they use a similar lysosomal buffer system?

Reviewer #2 (Public review):

Summary:

This manuscript presents a well-conceived and concise study that significantly advances our understanding of polyphosphate (polyP) metabolism and its role in cytosolic phosphate (Pi) homeostasis in a model unicellular eukaryote. The authors provide evidence that yeast vacuoles function as dynamic regulatory buffers for Pi homeostasis, integrating polyP synthesis, storage, and hydrolysis in response to cellular metabolic demands. The work is methodologically sound and offers valuable insights into the conserved mechanisms of phosphate regulation across eukaryotes.

Strengths:

The results demonstrate that the vacuolar transporter chaperone (VTC) complex, in conjunction with luminal polyphosphatases (Ppn1/Ppn2) and the Pi exporter Pho91, establishes a finely tuned feedback system that balances cytosolic Pi levels. Under Pi-replete conditions, inositol pyrophosphates (InsPPs) promote polyP synthesis and storage while inhibiting polyP hydrolysis, leading to vacuolar Pi accumulation.

Conversely, Pi scarcity triggers InsPP depletion, activating Pho91-mediated Pi export and polyP mobilization to sustain cytosolic phosphate levels. This regulatory circuit ensures metabolic flexibility, particularly during critical processes such as glycolysis, nucleotide synthesis, and cell cycle progression, where phosphate demand fluctuates dramatically.

From my viewpoint, one of the most important findings is the demonstration that vacuoles act as a rapidly accessible Pi reservoir, capable of switching between storage (as polyP) and release (as free Pi) in response to metabolic cues. The energetic cost of polyP synthesis-driven by ATP and the vacuolar proton gradient-highlights the evolutionary importance of this buffering system. The study also draws parallels between yeast vacuoles and acidocalcisomes in other eukaryotes, such as Trypanosoma and Chlamydomonas, suggesting a conserved role for these organelles in phosphate homeostasis.

Weaknesses:

While the manuscript is highly insightful, referring to yeast vacuoles as "acidocalcisome-like" may warrant further discussion. Canonical acidocalcisomes are structurally and chemically distinct (e.g., electron-dense, in most cases spherical, and not routinely subjected to morphological changes, and enriched with specific ions), whereas yeast vacuoles have well-established roles beyond phosphate storage. A comment on this terminology could strengthen the comparative analysis and avoid potential confusion in the field.

Reviewer #3 (Public review):

Bru et al. investigated how inorganic phosphate (Pi) is buffered in cells using S. cerevisiae as a model. Pi is stored in cells in the form of polyphosphates in acidocalcisomes. In S. cerevisiae, the vacuole, which is the yeast lysosome, also fulfills the function of Pi storage organelle. Therefore, yeast is an ideal system to study Pi storage and mobilization.

They can recapitulate in their previously established system, using isolated yeast vacuoles, findings from their own and other groups. They integrate the available data and propose a working model of feedback loops to control the level of Pi on the cellular level.

This is a solid study, in which the biological significance of their findings is not entirely clear. The data analysis and statistical significance need to be improved and included, respectively. The manuscript would have benefited from rigorously testing the model, which would also have increased the impact of the study.

Author response:

Reviewer #1 (Public review): 

The manuscript by Bru et al. focuses on the role of vacuoles as a phosphate buffering system for yeast cells. The authors describe here the crosstalk between the vacuole and the cytosol using a combination of in vitro analyses of vacuoles and in vivo assays. They show that the luminal polyphosphatases of the vacuole can hydrolyse polyphosphates to generate inorganic phosphate, yet they are inhibited by high concentrations. This balances the synthesis of polyphosphates against the inorganic phosphate pool. Their data further show that the Pho91 transporter provides a valve for the cytosol as it gets activated by a decline in inositol pyrophosphate levels. The authors thus demonstrate how the vacuole functions as a phosphate buffering system to maintain a constant cytosolic inorganic phosphate pool. 

This is a very consistent and well-written manuscript with a number of convincing experiments, where the authors use isolated vacuoles and cellular read-out systems to demonstrate the interplay of polyphosphate synthesis, hydrolysis, and release. The beauty of this system the authors present is the clear correlation between product inhibition and the role of Pho91 as a valve to release Pi to the cytosol to replenish the cytosolic pool. I find the paper overall an excellent fit and only have a few issues, including: 

(1) Figure 3: The authors use in their assays 1 mM ZnCl2 or 1mM MgCl2. Is this concentration in the range of the vacuolar luminal ion concentration? Did they also test the effect of Ca2+, as this ion is also highly concentrated in the lumen? 

The concentrations inside vacuoles can reach those values. However, given that polyP is a potent chelator of divalent metal ions, what would matter are the concentrations of free Zn²⁺ or Mg²⁺ inside the organelle. These are not known. This is not critical since we use those two conditions only as a convenient tool to differentiate Ppn1 and Ppn2 activity in vitro. In our initial characterisation of Ppn2 (10.1242/jcs.201061), we had also tested Mn, Co, Ca, Ni, Cu. Only Zn and Co supported activity. Ca did not. Andreeva et al. (10.1016/j.biochi.2019.06.001) reached similar conclusions and extended our results.

(2) Regarding the concentration of 30 mM K-PI, did the authors also use higher and lower concentrations? I agree that there is inhibition by 30 mM, but they cannot derive conclusions on the luminal concentration if they use just one in their assay. A titration is necessary here. 

The concentration of 30 mM was not arbitrarily chosen. It is the luminal P_i concentration that the vacuoles could reach through when they entered a plateau of luminal Pi. We consider this as an upper limit because polyP kept increasing which luminal P_i did not. Thus, there is in principle no physiological motivation for trying higher values. But we will probably add a titration to the revised version.

(3) What are the consequences on vacuole morphology if the cells lack Pho91? 

We had not observed significant abnormalities during a screen of the genome-wide deletion collection of yeast (10.1371/journal.pone.0054160)

(4) Discussion: The authors do not refer to the effect of calcium, even though I would expect that the levels of the counterion should affect the phosphate metabolism. I would appreciate it if they would extend their discussion accordingly. 

We will pick this up in the discussion. However, the situation is much more complex because major pools of counterions (up to hundreds of mM) are constituted by vacuolar lysine, arginine, polyamines, Mg, Zn etc. Their interplay with polyP is probably complex and worth to be treated in a dedicated project.

(5) I would appreciate a brief discussion on how phosphate sensing and control are done in human cells. Do they use a similar lysosomal buffer system? 

Mammalian cells have their Pi exporter XPR1 mainly on a lysosome-like compartment (10.1016/j.celrep.2024.114316). Whether and how it functions there for Pi export from the cytosol is not entirely clear. We will address this situation in the revision.

Reviewer #2 (Public review): 

Summary: 

This manuscript presents a well-conceived and concise study that significantly advances our understanding of polyphosphate (polyP) metabolism and its role in cytosolic phosphate (Pi) homeostasis in a model unicellular eukaryote. The authors provide evidence that yeast vacuoles function as dynamic regulatory buffers for Pi homeostasis, integrating polyP synthesis, storage, and hydrolysis in response to cellular metabolic demands. The work is methodologically sound and offers valuable insights into the conserved mechanisms of phosphate regulation across eukaryotes. 

Strengths: 

The results demonstrate that the vacuolar transporter chaperone (VTC) complex, in conjunction with luminal polyphosphatases (Ppn1/Ppn2) and the Pi exporter Pho91, establishes a finely tuned feedback system that balances cytosolic Pi levels. Under Pi-replete conditions, inositol pyrophosphates (InsPPs) promote polyP synthesis and storage while inhibiting polyP hydrolysis, leading to vacuolar Pi accumulation. 

Conversely, Pi scarcity triggers InsPP depletion, activating Pho91-mediated Pi export and polyP mobilization to sustain cytosolic phosphate levels. This regulatory circuit ensures metabolic flexibility, particularly during critical processes such as glycolysis, nucleotide synthesis, and cell cycle progression, where phosphate demand fluctuates dramatically. 

From my viewpoint, one of the most important findings is the demonstration that vacuoles act as a rapidly accessible Pi reservoir, capable of switching between storage (as polyP) and release (as free Pi) in response to metabolic cues. The energetic cost of polyP synthesis-driven by ATP and the vacuolar proton gradient-highlights the evolutionary importance of this buffering system. The study also draws parallels between yeast vacuoles and acidocalcisomes in other eukaryotes, such as Trypanosoma and Chlamydomonas, suggesting a conserved role for these organelles in phosphate homeostasis. 

Weaknesses: 

While the manuscript is highly insightful, referring to yeast vacuoles as "acidocalcisome-like" may warrant further discussion. Canonical acidocalcisomes are structurally and chemically distinct (e.g., electron-dense, in most cases spherical, and not routinely subjected to morphological changes, and enriched with specific ions), whereas yeast vacuoles have well-established roles beyond phosphate storage. A comment on this terminology could strengthen the comparative analysis and avoid potential confusion in the field. 

Yeast vacuoles show all major chemical features of acidocalcisomes. They are acidified, contain high concentrations of Ca, polyP (which make them electron-dense, too), other divalent ions, such as Mg, Zn, Mn etc, and high concentrations of basic amino acids. Thus, they clearly have an acidocalcisome-like character. In addition, they have hydrolytic, lysosome-like functions and, depending on the strain background, they can be larger than acidocalcisomes described e.g. in protists. We will elaborate this point, which is obvious to us but probably not to most readers, in the revised version.

Reviewer #3 (Public review): 

Bru et al. investigated how inorganic phosphate (Pi) is buffered in cells using S. cerevisiae as a model. Pi is stored in cells in the form of polyphosphates in acidocalcisomes. In S. cerevisiae, the vacuole, which is the yeast lysosome, also fulfills the function of Pi storage organelle. Therefore, yeast is an ideal system to study Pi storage and mobilization. 

They can recapitulate in their previously established system, using isolated yeast vacuoles, findings from their own and other groups. They integrate the available data and propose a working model of feedback loops to control the level of Pi on the cellular level. 

This is a solid study, in which the biological significance of their findings is not entirely clear. The data analysis and statistical significance need to be improved and included, respectively. The manuscript would have benefited from rigorously testing the model, which would also have increased the impact of the study.

It is not clear to us what the reviewer would see as a more rigorous test of the model.

eLife Assessment

This important study suggests that adolescent mice exhibit less accuracy than adult mice in a sound discrimination task when the sound frequencies are very similar. The evidence supporting this observation is solid and suggests that it arises from cognitive control differences between adolescent and adult mice. The adolescent period is largely understudied, despite its contribution to shaping the adult brain, which makes this study interesting for a broad range of neuroscientists.

Reviewer #1 (Public review):

Summary:

Praegel et al. explore the differences in learning an auditory discrimination task between adolescent and adult mice. Using freely-moving (Educage) and head-fixed paradigms, they compare behavioral performance and neuronal responses over the course of learning. The mice were initially trained for seven days on an easy pure frequency tone Go/No-go task (frequency difference of one octave), followed by seven days of a harder version (frequency difference of 0.25 octave). While adolescents and adults showed similar performance on the easy task, adults performed significantly better on the harder task. Quantifying the lick bias of both groups, the authors then argue that the difference in performance is not due to a difference in perception, but rather to a difference in cognitive control. The authors then used neuropixel recordings across 4 auditory cortical regions to quantify the neuronal activity related to the behavior. At the single cell level, the data shows earlier stimulus-related discrimination for adults compared to adolescents in both the easy and hard tasks. At the neuronal population level, adults displayed a higher decoding accuracy and lower onset latency in the hard task as compared to adolescents. Such differences were not only due to learning, but also to age as concluded from recordings in novice mice. After learning, neuronal tuning properties had changed in adults but not in adolescent. Overall, the differences between adolescent and adult neuronal data correlates with the behavior results in showing that learning a difficult task is more challenging for younger mice.

Strengths:

The behavioral task is well designed, with the comparison of easy and difficult tasks allowing for a refined conclusion regarding learning across age. The experiments with optogenetics and novice mice are completing the research question in a convincing way.

The analysis, including the systematic comparison of task performance across the two age groups, is most interesting and reveals differences in learning (or learning strategies?) that are compelling.

Neuronal recording during both behavioral training and passive sound exposure is particularly powerful, and allows interesting conclusions.

Weaknesses:

The weaknesses listed by this reviewer were addressed by adequate revisions.

Reviewer #2 (Public review):

Summary:

The authors aimed to find out how and how well adult and adolescent mice discriminate tones of different frequencies and whether there are differences in processing at the level of the auditory cortex that might explain differences in behavior between the two groups. Adolescent mice were found to be worse at sound frequency discrimination than adult mice. The performance difference between the groups was most pronounced when the sounds are close in frequency and thus difficult to distinguish and could, at least in part, be attributed to the younger mice' inability to withhold licking in no-go trials. By recording the activity of individual neurons in the auditory cortex when mice performed the task or were passively listening as well as in untrained mice the authors identified differences in the way that the adult and adolescent brains encode sounds and the animals' choice that could potentially contribute to the differences in behavior.

Strengths:

The study combines behavioural testing in freely-moving and head-fixed mice, optogenetic manipulation and high density electrophysiological recordings in behaving mice to address important open questions about age differences in sound-guided behavior and sound representation in the auditory cortex.

Weaknesses:

The weaknesses listed by this reviewer were addressed by adequate revisions.

Author response:

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

Reviewer #1 (Public review):

Summary:

Praegel et al. explore the differences in learning an auditory discrimination task between adolescent and adult mice. Using freely-moving (Educage) and head-fixed paradigms, they compare behavioral performance and neuronal responses over the course of learning. The mice were initially trained for seven days on an easy pure frequency tone Go/No-go task (frequency difference of one octave), followed by seven days of a harder version (frequency difference of 0.25 octave). While adolescents and adults showed similar performance on the easy task, adults performed significantly better on the harder task. Quantifying the lick bias of both groups, the authors then argue that the difference in performance is not due to a difference in perception, but rather to a difference in cognitive control. The authors then used neuropixel recordings across 4 auditory cortical regions to quantify the neuronal activity related to the behavior. At the single cell level, the data shows earlier stimulus-related discrimination for adults compared to adolescents in both the easy and hard tasks. At the neuronal population level, adults displayed a higher decoding accuracy and lower onset latency in the hard task as compared to adolescents. Such differences were not only due to learning, but also to age as concluded from recordings in novice mice. After learning, neuronal tuning properties had changed in adults but not in adolescent. Overall, the differences between adolescent and adult neuronal data correlates with the behavior results in showing that learning a difficult task is more challenging for younger mice.

Strengths:

The behavioral task is well designed, with the comparison of easy and difficult tasks allowing for a refined conclusion regarding learning across age. The experiments with optogenetics and novice mice are completing the research question in a convincing way.

The analysis, including the systematic comparison of task performance across the two age groups, is most interesting, and reveals differences in learning (or learning strategies?) that are compelling.

Neuronal recording during both behavioral training and passive sound exposure is particularly powerful, and allows interesting conclusions.

Weaknesses:

The presentation of the paper must be strengthened. Inconsistencies, missing information or confusing descriptions should be fixed.

We have carefully re-read the manuscript and reviewed it for inconsistencies. We made several corrections in the figures. For example, we removed redundant lines from violin plots and statistics, applied consistent labels, matched y- and x-limits of graphics, and adjusted labels. We also clarified descriptions of some experiment by adding explanations to the text.

The recording electrodes cover regions in the primary and secondary cortices. It is well known that these two regions process sounds quite differently (for example, one has tonotopy, the other not), and separating recordings from both regions is important to conclude anything about sound representations. The authors show that the conclusions are the same across regions for Figure 4, but is it also the case for the subsequent analysis? Comparing to the original manuscript, the authors have now done the analysis for AuDp and AUDv separately, and say that the differences are similar in both regions. The data however shows that this is not the case (Fig S7). And even if it were the case, how would it compatible with the published literature?

To address this and previous concerns about regional differences, the manuscript now includes 4 figures (4-1, 4-3, 6-2, 7-1) and 5 supplemental tables (3,4, 5, 6, 8) that explicitly compare results across brain regions.

Following the reviewer’s request for subsequent analysis, we now added a new supplemental figure (Fig. S6-2) and two new supplementary tables (Tables S5, S6). We show that similar to expert mice (supplementary Table 3, and supplementary Table 4), the firing properties of adolescent and adult novice mice differ across auditory subregions (supplementary Table 5). We also show that the different auditory subregions have different firing properties (supplementary Table 6). With respect to task engagement, we show that (similar to Fig. S4-2) the neuronal discriminability in different auditory subregions is similar in both novice and expert mice (Fig. S6-2).

Following the comment on Fig. S7-1, we made three changes to the revised manuscript. First, we now highlight that the differences firing properties between adolescent and adult neurons in AUDp and AUDv were distinct, but not significantly different within age-group comparisons. Second, we clearly state that the learning related changes in the measured parameters are different between AUDp and AUDv. Note, however, the greater changes in adult neurons after learning remains consistent between AUDp and AUDv. Third, we softened our original claim but still highlighted the stronger learning-induced plasticity in adults.

Regarding the concern that different regions should show different patterns due to their known differences (e.g. tonotopy). Of course we agree that different areas differ functionally (as shown in our own previous work and here as well). However, it is still plausible, and biologically reasonable, that developmental changes may proceed in a similar direction across different areas, even if their baseline coding properties differ.

Reviewer #2 (Public review):

Summary:

The authors aimed to find out how and how well adult and adolescent mice discriminate tones of different frequencies and whether there are differences in processing at the level of the auditory cortex that might explain differences in behavior between the two groups. Adolescent mice were found to be worse at sound frequency discrimination than adult mice. The performance difference between the groups was most pronounced when the sounds are close in frequency and thus difficult to distinguish and could, at least in part, be attributed to the younger mice' inability to withhold licking in no-go trials. By recording the activity of individual neurons in the auditory cortex when mice performed the task or were passively listening as well as in untrained mice the authors identified differences in the way that the adult and adolescent brains encode sounds and the animals' choice that could potentially contribute to the differences in behavior.

Strengths:

The study combines behavioural testing in freely-moving and head-fixed mice, optogenetic manipulation and high density electrophysiological recordings in behaving mice to address important open questions about age differences in sound-guided behavior and sound representation in the auditory cortex.

Weaknesses:

For some of the analyses that the authors conducted it is unclear what the rationale behind them is and, consequently, what conclusion we can draw from them.

We have carefully re-read the manuscript and reviewed it for analyses that lacked a clear rationale or conclusion. To address this, we have made several changes to clarify the reasoning and strengthen the interpretation of the results.

Reviewer #1 (Recommendations for the authors):

It would have helped if the authors had highlighted the changes they made to the manuscript compared to the original version - especially since many replies to the reviewers' comments were as vague as "...we fixed some of the wording so it adheres to the data shown", or "we refined our interpretation", without further details.

The revised version has improved substantially, and the main claims have been discussed in a more objective way. Important new analyses have been added to allow for a refined interpretation of the results. However, the presentation of the data could still be strengthened significantly (in response to comment A from last review).

We apologize for the lack of detail in some of our previous responses. Our intention was to keep the replies concise, assuming that the side-by-side version with tracked changes would make the edits sufficiently clear. However, we understand the need for greater transparency. Thus, below we provide the following five lists describing the major changes: (1) List of specific reviewer recommendations, (2) list of corrections in figures, (3) list of clarity issues, (4) list of fixed mistakes, (5) list of new figures. We hope this breakdown makes the revisions clearer and more accessible.

List of specific reviewer recommendations:

l.108 mentions a significant change in the vertical line of Fig 1F - Could this significance be indicated and quantified in the figure?

We quantified and indicated the significance of the vertical line in Fig. 1f and Fig. 1i.

Fig.1G - the thick and thin lines should be defined, as well as the grey and white dots (same values for adolescents, not for adults).

(a) We removed the thin inner lines from the violin plot. We define the bar (thick line) of the violin plot in an additional sentence in the methods section under data analysis (LL820-823). b) We adjusted the marker outlines in the adult data (Fig. 1G).

the figure axis legends should be consistent (trails in Fig D vs # trails in Fig 1F)

We adjusted the axis legend to # trials in Fig. 1D.

l.110: is d' always calculated based on the 100 last trials of a session, or is it just for Figure 1F? -etc...

d’ is always calculated based on the last 100 trials. To clarify this, we added a description in the methods section (L830).

List of corrections in the figures:

(1) We removed the internal lines from violin plots in throughout Fig. 1-7.

(2) We removed the underline of the statistics throughout Fig. 1-7.

(3) We consistently applied ‘adolescent’ and ‘adult’ figure labels and titles with lowercase letters throughout Fig. 1-7.

(4) We applied consistent labelling of ‘time (ms)’ throughout Fig. 1-7.

(5) We matched the size of dashed lines throughout Fig. 1-6.

(6) We adjusted the x-label of Fig. 1d, Fig. S-1-1 a, Fig. 3c, Fig. 3h-i, Fig, 4d to ‘# trials’.

(7) We removed the x-label of ‘Experimental Group’ from Fig. 1 to enhance consistency with other figures.

(8) We removed misaligned dots from the violin plots in Fig. 1g, Fig. 2f, Fig. 3f,g.

(9) We corrected the plot in Fig. S1-1b.

(10) We adjusted the y-limits of Fig. S1-1c to be consistent with Fig. S1-1d,e.

(11) We adjusted the x-labels and y-labels of Fig, 2, Fig. S3-1, Fig, S3-2 and Fig. 3b to ‘freq. (kHz)’.

(12) We added the age of adolescent and adult mice to the schematic timeline in Fig. 2a.

(13) We added a label of the reinforcement delay to the schematic trial structure in Fig. 3b.

(14) We added within-group statistics to Fig. 3e and the figure legend.

(15) We adjusted the x-label of Fig. 3d to ‘# sessions’.

(16) We adjusted the x-label of Fig. 3d and Fig. S3-1b to ‘# licks’.

(17) We changed the y-label in Fig. S3-1a, and Fig. S3-2d, e to ‘lick ratio’ to avoid confusion with the lick rate (Hz) that was calculated in Fig. 4 and Fig. 6.

(18) We replaced the titles ‘CAMKII’ with ‘dTomato’ in Fig. S3-2 to correctly highlight that both the experimental and control injection were CAMKII injections.

(19) We adjusted the x-labels and y-labels of Fig, 2, Fig. S3-1, Fig, S3-2 and Fig. 3b to ‘freq. (kHz)’.

(20) We adjusted the y-label of Fig. S4-1c to ‘# neurons’.

(21) We matched the x-ticks in Fig. 4e,f.

(22) We matched the x-ticks in Fig. 6d-g.

(23) We changed the x-label in Fig. 4g, S4-2 and S6-2 to ‘duration (ms)’ to match the figure label with the manuscript.

(24) We consistently label ‘Hit’, ‘Miss’, ‘FA’ and ‘CR’ with capital letters in Fig. 4d-e.

(25) We replaced the double figure label ‘C.’ in Fig. S4-2 with ‘D.’.

(26) We adjusted the dot-size in Fig. 5 to be equal for all graphs.

(27) We added ticks to the experimental timeline in Fig. 6a.

(28) We corrected the y-label in Fig.7c. Now it correctly reflects 5 attenuations from 72-32 dB SPL.

(29) We matched the y-label of Fig. 7e-h and Fig. S7-1.

List of clarity issues:

(1) We replaced the term ‘lower response bias’ with ‘higher lick bias’ (L24) to accurately describe the more negative (lower) criterion-bias, which highlights a higher tendency to lick.

(2) We replaced the term ‘response bias’ with ‘lick bias’ to consistently describe the calculated criterion-bias (L24, L149, L164, L455, L456, L468).

(3) We clarify that the age-related differences were ‘more pronounced’ instead of simply ‘higher’ to accurately reflect not simply the increase in adolescent lick-bias, but also the decrease in adult lick-bias (L31).

(4) We clarified that adolescent sound representations are not merely ’distinct’, but ‘not fully mature’ in L83.

(5) We clarified in L180 that the impulsive responses we observed in adolescent mice could be related to being ‘less impacted by punishments’.

(6) We clarified the differences in firing properties of auditory sub-regions analyzed in Supplementary Table 3 (L287-295).

(7) We explained and clarified the reference to Fig. 3j (LL252-253).

(8) We added statistics to Fig.S4-2 to support our claim that there are no differences in the onset-latency, duration of discriminability and maximal discriminability between different sub-regions within age-groups (LL 314-315).

(9) We expanded our explanation of the results in Table 3 (LL370-379).

(10) We separated the reference to Fig. 6b and Fig. 6c to clarify their meaning (LL358-361).

(11) We clarified the differences in basic firing properties during the FRA protocol in Fig. 7 (LL409-418).

(12) We expanded our explanation of the differences of the learning related firing properties in AUDp and AUDv of Fig. S7-1 (LL426-433).

(13) We changed the term ‘plasticity profiles’ to ‘learning related plasticity’ to further clarify our limitation that L5/6 and L2/3 may exhibit distinct learning related changes (L496).

(14) We changed the term ‘sluggish’ (L481) to ‘delayed’ to more precisely explain differences between adolescent and adult tuning properties.

(15) We clarified that the running d’ was calculated in bins of 25 trials, instead of ‘the last 25 trials’ (LL845-846).

List of fixed mistakes:

(1) We corrected and matched the age to more accurately reflect the age mice were recorded (P37-42 and P77-82).

(2) We corrected the attenuation range from 72-42 to 72-32 dB SPL to correctly reflect the 5 attenuations used in the protocol.

(3) We corrected the number of channels shown in the voltage trace from 10 to 11 (Fig. S4-1a)

(4) We corrected the number of neurons recorded in novice adolescent mice in the legend of Fig. 6 from 140 to 130 (Fig. 6b).

(5) We removed redundant, or double brackets, commas, dots, and semi-colons in the figure legends.

(6) We corrected the LME statistics Table 2.

List of new figures and tables:

(1) We added a new supplementary figure to accompany Figure 6. Specifically, Fig. S6-2, shows the interaction of the three measured discriminability properties (onset delay, duration of discriminability, and maximal discriminability) in novice compared to expert mice in the easy and hard task (Go compared to No Go). The figure compares the different auditory sub-regions (similar to Fig. S4-2). We show that the discriminability properties within different groups is not significantly different among the four different sub-regions.

(2) Supplementary Table 5: We compared the firing properties in different auditory subregions in novice mice, and found (similar to expert mice) that the firing properties differ between adult and adolescent mice across the four different sub-regions.

(3) Supplementary Table 6: We compared the firing properties between different subregions, separately for adolescent and adult novice mice. Similar to expert mice, we found that different auditory subregions differ in their auditory firing properties.

Reviewer #2 (Recommendations for the authors):

The authors largely addressed my suggestions.

Comparing hit vs correct rejection trials in the population decoding analysis (L313-314): The authors acknowledge that comparing these two trial types conflates choice and stimulus decoding but I am not convinced that the changes to the manuscript text make this clear enough to the reader.

Thank you for pointing this out. We have made additional revisions to clarify this, and other issues more explicitly, as follows:

(1) We have expanded the explanation of how our population decoding analysis conflates stimulus and choice, and we acknowledge the limitations of this approach in the Abstract (L28), the Results section (L324-326, LL367-370) and the Discussion (LL516-519).

(2) We replaced the analysis of impulsivity on the head-fixed task. Instead of analyzing all it is, we focus only on ITIs following FA trials (Fig. S3-1c,d). This is more consistent with the analysis in the Educage (Fig. S2-1), where we show that adolescents exhibit increased impulsivity after FA trials. We found a similar result for ITIs following FA trials in the head-fixed task.

(3) To provide complementary insight, we now further justify our use of the Fisher separation metric alongside decoding accuracy in Figure 5, with a clearer rationale provided in LL343-345

(4) We also clarified our reasoning for focusing on 62 dB SPL in the FRA-based analysis in LL400-403.

Dalam konteks ini, 'tos' merujuk pada ungkapan penghormatan yang diucapkan saat mengangkat gelas, biasanya selama acara sosial atau pertemuan keluarga. toast /toʊst/

kata benda Sebuah ungkapan penghormatan yang diucapkan ketika mengangkat gelas. Contoh: We all raised our glasses for a toast to the bride and groom. Kami semua mengangkat gelas untuk sebuah tos bagi pengantin baru.

Dalam konteks ini, 'bodoh' digunakan untuk menggambarkan seseorang yang kurang cerdas atau tidak memahami sesuatu, sehingga dianggap tidak layak untuk menjadi murid dalam sekte abadi. dumb /dʌm/

Adjektiva Kondisi atau sifat orang yang tidak memiliki kecerdasan atau tidak mampu berbicara. Contoh: He was too dumb to understand the instructions. Dia terlalu bodoh untuk memahami instruksi tersebut.

Dalam konteks ini, 'nilai' merujuk pada kualitas atau prestasi seseorang yang dapat dianggap sebagai kontribusi positif, meskipun yang bersangkutan tidak memiliki banyak prestasi yang mencolok.

merit /ˈmɛrɪt/

noun Kualitas atau prestasi yang patut dihargai atau dipertimbangkan. Contoh: His merit in the team was recognized by everyone. Nilai dia di tim diakui oleh semua orang.

Dalam konteks kalimat, 'cluelessly' menggambarkan keadaan di mana Tie Zhu tidak memiliki pemahaman atau informasi tentang situasi yang terjadi, dan 'bodoh-bodoh saja' menunjukkan ketidakpahaman atau kepolosan dalam pertanyaan yang ia ajukan. cluelessly /ˈkluː.les.li/

adverb tanpa pemahaman; dengan cara yang menunjukkan ketidakpahaman Contoh: He nodded cluelessly after hearing the instructions. Ia mengangguk bodoh-bodoh saja setelah mendengar instruksi.

Dalam konteks ini, 'memaksa' digunakan untuk menunjukkan bahwa orang tua Tie Zhu merasa didorong pergi oleh pria tua tersebut, menandakan rasa tidak hormat dan tekanan yang dirasakan. forced /fɔːrst/

adjektiva dilakukan dengan paksaan atau tekanan; tidak sukarela. Contoh: He was forced to leave the party. Dia dipaksa untuk meninggalkan pesta. kata kerja (past tense) memaksa seseorang untuk melakukan sesuatu, seringkali dengan cara yang tidak diinginkan. Contoh: They forced him to sign the contract. Mereka memaksanya untuk menandatangani kontrak.

Dalam konteks ini, 'bound' digunakan untuk menunjukkan bahwa ada sesuatu yang ditakdirkan atau pasti terjadi, dalam hal ini, bahwa Tie Zhu akan lebih baik daripada ayahnya. bound /baʊnd/

kata sifat ditakdirkan untuk menjadi Contoh: He is bound to succeed. Dia ditakdirkan untuk berhasil. kata kerja mengikat atau membatasi Contoh: The dog is bound in the yard. Anjing itu terikat di halaman.

Dalam konteks ini, 'kebanyakan' merujuk pada sebagian besar anggota keluarga Wang yang diundang ke pesta. most /moʊst/

adjektiva Sebagian terbesar dari sesuatu. Contoh: Most people enjoy the festival. Kebanyakan orang menikmati festival tersebut. kata ganti Bagian terbesar dari sesuatu. Contoh: Most of them left early. Kebanyakan dari mereka pergi lebih awal.

Dalam konteks ini, 'keberatan' merujuk pada tidak adanya penolakan atau bantahan dari Tie Zhu terhadap pernyataan bahwa dia dibesarkan di bawah pengawasan guru, sehingga menunjukkan penerimaan dan kebenaran pernyataan tersebut. objections /əbˈdʒɛkʃənz/

Kata benda Sesuatu yang menentang atau tidak setuju dengan suatu pernyataan atau situasi. Contoh: She raised her objections to the proposed plan. Dia mengajukan keberatannya terhadap rencana yang diusulkan.

Dalam konteks ini, 'kesedihan' menggambarkan perasaan berat yang dialami oleh ayah Tie Zhu akibat perlakuan buruk dari kerabat di masa lalu dan bagaimana kedatangan mereka sekarang membawa kebahagiaan menggantikan rasa sakit tersebut. sorrow /ˈsɔːroʊ/

kata benda perasaan sedih atau duka akibat kehilangan, kekecewaan, atau penderitaan. Contoh: He felt deep sorrow after the loss of his friend. Dia merasakan kesedihan yang mendalam setelah kehilangan temannya.

ignorant /ɪɡˈnɔːrənt/ tidak tahu\ Dalam konteks ini, 'tidak tahu' menunjukkan bahwa Tie Zhu tidak memiliki pengetahuan atau pemahaman tentang para immortal, meskipun ia menyadari pentingnya situasi tersebut berdasarkan ekspresi orang tuanya.

1. Adjektiva: Tidak memiliki pengetahuan atau informasi tentang sesuatu
2. He was ignorant of the rules of the game.
3. Ia tidak tahu aturan permainan.

sternly /ˈstɜːrnli/ dengan tegas\ Dalam konteks ini, 'dengan tegas' digunakan untuk menunjukkan cara yang serius dan kuat dalam menyampaikan pesan penting kepada Tie Zhu tentang tanggung jawabnya terhadap keluarganya.

1. Adverb: Dengan cara yang serius dan penuh ketegasan.
2. He spoke sternly to the children.
3. Dia berbicara dengan tegas kepada anak-anak.

heavily patted /ˈhɛvɪli ˈpætɪd/ menepuk dengan kuat\ Dalam konteks ini, 'menepuk dengan kuat' menunjukkan tindakan yang penuh emosi dan kekuatan, sebagai ungkapan dukungan atau keprihatinan yang mendalam.

1. verb: Menepuk dengan kuat, seringkali untuk menunjukkan dukungan atau penghiburan.
2. He heavily patted his friend on the back to comfort him.
3. Dia menepuk dengan kuat temannya di punggung untuk menghiburnya.

akan melakukan

Dalam konteks ini, 'about' digunakan untuk menunjukkan bahwa seseorang akan melakukan tindakan tertentu yang berkaitan dengan orang lain, dalam kasus ini adalah membungkuk sebagai tanda penghormatan. about /əˈbaʊt/

Preposisi Menyatakan topik atau subjek yang sedang dibicarakan. Contoh: They talked about the weather. Mereka berbicara tentang cuaca. Preposisi Digunakan untuk menunjukkan arah atau bagian dari suatu tindakan atau peristiwa. Contoh: She was about to leave. Dia akan pergi.

hesitantly /ˈhɛzɪtəntli/ ragu-ragu\ Dalam konteks ini, 'hesitantly' berarti Tie Zhu berbicara dengan ketidakpastian atau keraguan karena dia tidak memahami situasi yang sedang berlangsung.

1. Adverbia: Dengan ragu-ragu; dengan ketidakpastian atau keraguan.
2. She answered the question hesitantly.
3. Dia menjawab pertanyaan itu dengan ragu-ragu.

settled /ˈsɛtld/ diselesaikan\ Penggunaan 'diselesaikan' dalam konteks ini menunjukkan bahwa permasalahan atau situasi telah disepakati atau dijadikan final, memungkinkan untuk melanjutkan ke langkah berikutnya tanpa adanya perselisihan lebih lanjut.

1. adjektiva: telah ditentukan atau disepakati, dalam hal ini merujuk pada penyelesaian masalah.
2. The issue has been settled amicably.
3. Masalah telah diselesaikan dengan baik.

delighted /dɪˈlaɪtɪd/ senang\ Dalam konteks kalimat, 'senang' digunakan untuk menggambarkan perasaan bahagia dan puas yang dirasakan oleh ibu Tie Zhu. Hal ini menunjukkan rasa suka cita yang mendalam terhadap sesuatu yang terjadi.

1. adjective: sangat senang; merasa bahagia karena sesuatu yang menyenangkan
2. She was delighted by the surprise party.
3. Dia sangat senang dengan pesta kejutan itu.

spot /spɒt/ tempat\ Dalam konteks ini, 'tempat' merujuk pada posisi atau kesempatan yang dianggap sangat berharga untuk diisi oleh calon murid di secta immortal.

1. noun: Tempat atau posisi tertentu.
2. She found a nice spot to have a picnic.
3. Dia menemukan tempat yang bagus untuk piknik.
4. verb: Melihat atau menemukan sesuatu.
5. I spotted a deer in the woods.
6. Saya melihat seekor rusa di hutan.

pondered /ˈpɒndərd/ merenungkan\ Dalam konteks ini, 'merenungkan' menggambarkan tindakan berpikir secara mendalam dan serius, mencerminkan pergumulan batin sebelum memberikan pernyataan penting.

1. verb: berpikir dengan mendalam atau merenungkan sesuatu
2. He pondered the meaning of life.
3. Dia merenungkan makna hidup.

berlalu begitu saja

passed = berlalu.

passed by = berlalu begitu saja / terlewat tanpa sadar.

10 years have passed. → 10 tahun telah berlalu. (netral, pernyataan fakta)

10 years have passed by. → 10 tahun sudah lewat begitu saja. (lebih terasa “mengalir / cepat hilang tanpa terasa”)

🔹 pass (tanpa “by”) = melewati / lulus / menyerahkan → banyak arti tergantung konteks.

Contoh:

I passed the exam. → aku lulus ujian.

Pass me the salt, please. → tolong operin garam.

The car passed the bus. → mobil itu menyalip/melewati bus.

🔹 pass by (phrasal verb) = lewat, berlalu (tanpa interaksi, cuma numpang lewat).

Contoh:

The car passed by quickly. → mobil itu melintas/berlalu dengan cepat.

Ten years passed by so fast. → sepuluh tahun berlalu begitu cepat.

🔑 Bedanya gampang: pass = bisa banyak makna (lulus, ngoper, nyalip, melewati).

pass by = khusus: lewat / berlalu / melintas.

De toepassing van intelligente systemen in het hoger onderwijs is een actief onderzoeksgebied, gedreven door de overvloed aan beschikbare data en de urgentie om effectieve, datagestuurde strategieën te definiëren om uitval van studenten tegen te gaan en hun studieprestaties te verbeteren. Dit werk past machine learning-technieken toe om voorspellingsmodellen te ontwikkelen die kunnen bijdragen aan de vroege detectie van studenten die het risico lopen hun studie af te breken of hun diploma niet op tijd af te ronden. Het evalueert ook het beste moment om de voorspelling uit te voeren in het inschrijvingsjaar van de student. De modellen zijn gebaseerd op data van bachelorstudenten van een Polytechnische Universiteit in Portugal, ingeschreven tussen 2009 en 2017, en bevatten academische, sociaal-demografische en macro-economische informatie in drie verschillende fasen van het eerste studiejaar van de studenten. Vijf machine learning-algoritmen worden gebruikt om voorspellingsmodellen in elke fase te trainen, en de meest relevante kenmerken voor de best presterende modellen worden geïdentificeerd. De resultaten tonen aan dat de beste modellen gebruikmaken van Random Forest, waarbij strategieën worden toegepast om om te gaan met de onevenwichtige aard van de data, of waarbij dergelijke strategieën op dataniveau worden toegepast. De beste resultaten worden behaald aan het einde van het eerste semester, wanneer er al enige informatie beschikbaar is over de studieprestaties na inschrijving. De algehele resultaten zijn redelijk vergelijkbaar met die van vergelijkbare studies die zich richten op de vroege voorspelling van uitval of studieprestaties van studenten.

eLife Assessment

This study presents a valuable finding on the representational structure of task encoding in the prefrontal cortex. The evidence supporting the claims of the authors is solid, representing an impressive data collection effort and best-practice fMRI analyses. However, at least including visual regions as a control and controlling for behavioral differences in the task in representation analyses would have strengthened the study. The work will be of interest to cognitive neuroscientists interested in the neural basis of cognitive control.

Reviewer #1 (Public review):

Summary:

Bhandari and colleagues present tour-de-force analyses that compare the representational geometry in the lateral prefrontal cortex and primary auditory cortex between two complex cognitive control tasks, with one having a "flat" structure where subjects are asked to form rote memory of all the stimulus-action mappings in the task and one having a "hierarchical" task structure that allows clustering of task conditions and that renders certain stimulus dimensions irrelevant for choices. They discovered that the lPFC geometry is high-dimensional in nature in that it allows above-chance separation between different dichotomies of task conditions. The separability is significantly higher for task-relevant features than task-irrelevant ones. They also found task features that are represented in an "abstract" format (e.g., audio features), i.e., the neural representation generalizes across specific task conditions that share this variable. The neural patterns in lPFC are highly relevant for behaviors as they are correlated with subjects' reaction times and choices.

Strengths:

Typically, geometry in coding patterns is reflected in single-unit firings; this manuscript demonstrates that such geometry can be recovered using fMRI BOLD signals, which is both surprising and important. The tasks are well designed and powerful in revealing the differences in neural geometry, and analyses are all done in a rigorous way. I am thus very enthusiastic about this paper and identify no major issues.

I am curious about the consequence of dimensionality collapse in lPFC. The authors propose a very interesting idea that separability is critical for cognitive control; indeed, separability is high for task-relevant information. What happens when task-relevant separation is low or task-irrelevant separation is high, and will this lead to behavioral errors? Maybe a difference score between the separability of task-relevant and task-irrelevant features is a signature of the strength of cognitive control?

Weaknesses:

The authors show a difference between flat and hierarchical tasks, but the two tasks are different in accuracy, with the flat task having more errors. Will this difference in task difficulty/errors contribute to the task differences in results reported?

Reviewer #2 (Public review):

Summary:

The authors study the influence of tasks on the representational geometry of the lPFC and auditory cortex (AC). In particular, they use two context-dependent tasks: a task with a hierarchical structure and a task with a flat structure, in which each context/stimulus maps to a specific response. Their primary finding is that the representational geometry in the lPFC, in contrast to AC, aligns with the optimal organization of the task. They conclude that the geometry of representations adapts, or is tailored, to the task in the lPFC, therefore supporting control processes.

Strengths:

(1) Dataset:<br /> The dataset is impressive and well-sampled. Having data from both tasks collected in the same subjects is a great property. If it is publicly available, it will be a significant contribution to the community.

(2) Choice of methods:<br /> The choice of analyses are largely well-suited towards the questions at hand - cross-condition generalization, RSA + regression, in combination with ANOVAs, are well-suited to characterizing task representations.

(3) I found some of their results, in particular, those presented in Figures 4 and 5, to be particularly compelling.

(4) The correlation analysis with behavior is also a nice result.

Weaknesses:

(1) Choice of ROIs:<br /> A strength of fMRI is its spatial coverage of the whole brain. In this study, however, the authors focus on only two ROIs: the lPFC and auditory cortex. Though I understand the justification for choosing lPFC from decades of research, the choice of AC as a control feels somewhat arbitrary - AC is known to have worse SNR in fMRI data, and limiting a 'control' to a single region seems arbitrary. For example, why not also include visual regions, given that the task also involves two visual features?

(2) Construction of ROIs:<br /> The choice and construction of the ROIs feel a bit arbitrary, as the lPFC region was constructed out of 10 parcels from Schaefer, while the AC was constructed from a different methodology (neurosynth). Did both parcels have the same number of voxels/vertices? It would be helpful to include a visualization of these masks as a figure.

(3) Task dimensionality:<br /> In some ways, the main findings - that representation dimensionality is tailored to the task - seem to obviously follow from the choice of two tasks, particularly from a normative modeling perspective. For example, the flat task is effectively a memorization task, and is incompressible in the sense that there are no heuristics to solve it. In contrast, the hierarchical task can have several strategies, an uncompressed (memorized) strategy, and a compressed strategy. This is analogous to other studies evaluating representations during 'rich' vs. 'lazy'/kernel learning in ANNs. However, it seems unlikely (if not impossible) to form a 'rich' representation in the flat task. Posed another way, the flat task will always necessarily have a higher dimensionality than the hierarchical task. Thus, is their hypothesis - that representational geometry is tailored to the task - actually falsifiable? I understand the authors posit alternative hypotheses, e.g., "a fully compressed global axis with no separation among individual stimulus inputs could support responding [in the flat task]" (p. 36). But is this a realistic outcome, for example, in the space of all possible computational models performing this task? I understand that directly addressing this comment is challenging (without additional data collection or modeling work), but perhaps some additional discussion around this would be helpful.

(4) Related to the above:<br /> The authors have a section on p. 27: "Local structure of lPFC representational geometry of the flat task shows high separability with no evidence for abstraction" - I understand a generalization analysis can be done in the feature space, but in practice, the fact that the flat task doubles as a memorization task implies that there are no useful abstractions, so it seems to trivially follow that there would be no abstract representations. In fact, the use of task abstractions in the stimulus space would be detrimental to task performance here. I could understand the use of this analysis as a control, but the phrasing of this section seems to indicate that this is a surprising result.

(5) Statistical inferences:<br /> Throughout the manuscript, the authors appear to conflate failure to reject the null with acceptance of the null. For example, p. 24: "However, unlike left lPFC, paired t-tests showed no reliable difference in the separability of the task-relevant features vs the orthogonal, task-irrelevant features... Therefore, the overall separability of pAC representations is not shaped by either task-relevance of task structure."

Reviewer #3 (Public review):

Summary:

In this paper, Bhandari, Keglovits, et al. explore the representational structure of task encoding in the lateral prefrontal cortex. Through an impressive fMRI data-collection effort, they compare and contrast neural representations across tasks with different high-level stimulus-response structures. They find that the lateral prefrontal cortex shows enhanced encoding of task-relevant information, but that most of these representations do not generalize across conditions (i.e., have low abstraction). This appears to be driven in part by the representation of task conditions being clustered by the higher-order task properties ('global' representations), with poor generalization across these clusters ('local' representations). Overall, this paper provides an interesting account of how task representations are encoded in the PFC.

Strengths:

(1) Impressive dataset, which may provide further opportunities for investigating prefrontal representations.

(2) Clever task design, allowing the authors to confound several features within a complex paradigm.

(3) Best-practice analysis for decoding, similarity analyses, and assessments of representational geometry.

(4) Extensive analyses to quantify the structure of PFC task representations.

Weaknesses:

(1) The paper would benefit from improved presentational clarity: more scaffolding of design and analysis decisions, clearer grounding to understand the high-level interpretations of the analyses (e.g., context, cluster, abstraction), and better visualizations of the key findings.

(2) The paper would benefit from stronger theoretical motivation for the experimental design, as well as a refined discussion on the implications of these findings for theories of cognitive control.

Author response:

We thank the reviewers and editors for their careful and constructive assessment of our manuscript. We have provided a provisional response to the eLife assessment and the reviewer’s public comments below, addressing their main concerns and outlining our planned revisions that we believe will substantially strengthen our paper.  

eLife Assessment

This study presents a valuable finding on the representational structure of task encoding in the prefrontal cortex. The evidence supporting the claims of the authors is solid, representing an impressive data collection effort and best-practice fMRI analyses. However, at least including visual regions as a control and controlling for behavioral differences in the task in representation analyses would have strengthened the study. The work will be of interest to cognitive neuroscientists interested in the neural basis of cognitive control.

We plan to address both specific methodological weaknesses mentioned in the assessment in our forthcoming revision. First, the revision will include analyses of an early visual cortex ROI as an additional control region, allowing us to test whether the primary auditory cortex findings generalize to the sensory cortex across input modalities. Preliminary results indicate that the early visual cortex ROI exhibits a similar pattern of results, with evidence for coding both task-relevant and task-irrelevant visual dimensions across both tasks, as well as the context dimension specifically in the hierarchy task. Second, we will include behavioral performance as a covariate for the relevant statistical comparison across tasks to mitigate concerns over performance-related confounds. In addition, we will include a set of control analyses that demonstrate that equating the amount of data for pattern analyses across the two tasks by subsampling from the hierarchy task, while reducing our overall power, does not appreciably alter our results. We note that our analyses of representational geometries relied only on neural data from correct trials and, in the first-level modelling of the fMRI data, already controlled for differences in trial-by-trial response times. Therefore, our analyses of decoding and representation similarity are not directly affected by differences in performance across the two tasks. Finally, we have provided clarifications regarding Reviewer 2’s questions about the size and construction of the regions of interest employed in the study, as well as about the language employed to discuss null results.  

Public Reviews:

Reviewer #1 (Public review):

Summary:

Bhandari and colleagues present tour-de-force analyses that compare the representational geometry in the lateral prefrontal cortex and primary auditory cortex between two complex cognitive control tasks, with one having a "flat" structure where subjects are asked to form rote memory of all the stimulus-action mappings in the task and one having a "hierarchical" task structure that allows clustering of task conditions and that renders certain stimulus dimensions irrelevant for choices. They discovered that the lPFC geometry is high-dimensional in nature in that it allows above-chance separation between different dichotomies of task conditions. The separability is significantly higher for task-relevant features than task-irrelevant ones. They also found task features that are represented in an "abstract" format (e.g., audio features), i.e., the neural representation generalizes across specific task conditions that share this variable. The neural patterns in lPFC are highly relevant for behaviors as they are correlated with subjects' reaction times and choices.

Strengths:

Typically, geometry in coding patterns is reflected in single-unit firings; this manuscript demonstrates that such geometry can be recovered using fMRI BOLD signals, which is both surprising and important. The tasks are well designed and powerful in revealing the differences in neural geometry, and analyses are all done in a rigorous way. I am thus very enthusiastic about this paper and identify no major issues.

I am curious about the consequence of dimensionality collapse in lPFC. The authors propose a very interesting idea that separability is critical for cognitive control; indeed, separability is high for task-relevant information. What happens when task-relevant separation is low or task-irrelevant separation is high, and will this lead to behavioral errors? Maybe a difference score between the separability of task-relevant and taskirrelevant features is a signature of the strength of cognitive control?

We appreciate the reviewers’ positive evaluation of our paper.

Weaknesses:

The authors show a difference between flat and hierarchical tasks, but the two tasks are different in accuracy, with the flat task having more errors. Will this difference in task difficulty/errors contribute to the task differences in results reported?

To address the Reviewer’s concern about the difference in behavioural performance between the two tasks influencing our results, we will take several approaches. First, we will include behavioral performance as a covariate for the relevant statistical comparison across tasks. This should ensure that any differences we observe across tasks are over and above those that can be explained by the difference in behavioral performance. Second, we will include a set of decoding analyses that control for differences in performance across the tasks. We note that all our analyses of representational geometries relied on neural data from correct trials only. In addition, the first-level modelling of the fMRI data already controlled for trial-by-trial variability in response times. Therefore, our decoding and representation similarity analyses should not directly be affected by differences in performance across the two tasks. However, one possible issue with this approach is that the larger number of errors in the flat task means that less data was available for estimating multivoxel patterns in the flat task compared to the hierarchy task, resulting in differential power to detect decoding effects across the two tasks. We note that the on average, this difference was not substantial: on average, 21.7 runs were available per participant for the flat task, while 23.8 runs per participant were available for the hierarchy task. Moreover, rerunning our analyses with the number of runs equated for each participant does not meaningfully alter the pattern of results. These additional analyses will be included in the supplement in the forthcoming revised manuscript.  

Reviewer #2 (Public review):

Summary:

The authors study the influence of tasks on the representational geometry of the lPFC and auditory cortex (AC). In particular, they use two context-dependent tasks: a task with a hierarchical structure and a task with a flat structure, in which each context/stimulus maps to a specific response. Their primary finding is that the representational geometry in the lPFC, in contrast to AC, aligns with the optimal organization of the task. They conclude that the geometry of representations adapts, or is tailored, to the task in the lPFC, therefore supporting control processes.

Strengths:

(1) Dataset:

The dataset is impressive and well-sampled. Having data from both tasks collected in the same subjects is a great property. If it is publicly available, it will be a significant contribution to the community.

(2) Choice of methods:

The choice of analyses are largely well-suited towards the questions at hand - crosscondition generalization, RSA + regression, in combination with ANOVAs, are well-suited to characterizing task representations.

(3) I found some of their results, in particular, those presented in Figures 4 and 5, to be particularly compelling.

(4) The correlation analysis with behavior is also a nice result.

We thank the reviewer for noting the strengths of the paper. We respond to the weaknesses noted below. 

Weaknesses:

(1) Choice of ROIs:

A strength of fMRI is its spatial coverage of the whole brain. In this study, however, the authors focus on only two ROIs: the lPFC and auditory cortex. Though I understand the justification for choosing lPFC from decades of research, the choice of AC as a control feels somewhat arbitrary - AC is known to have worse SNR in fMRI data, and limiting a 'control' to a single region seems arbitrary. For example, why not also include visual regions, given that the task also involves two visual features?

We agree with the reviewer that the whole-brain fMRI data certainly provide ample opportunities to explore the nature of these representations across the brain. Our focus in this paper is squarely on the principles of coding and flexibility in the lPFC. We believe that a whole-brain exploration addresses a separate question that would be out of the scope of this study. To clarify, we are not arguing that the lPFC is the only region in the brain that employs the coding principles that our study brings to light. Our contention is only that lPFC employs these principles, and it differs at least from the primary sensory cortex. The questions of whether these principles generalize beyond lPFC (quite likely) and, if so, how broadly, are distinct from the ones addressed in the manuscript. We intend to follow up with another manuscript that addresses these questions.

Nevertheless, given the focus of this paper, we agree that a second control region, which allows one to test if the primary auditory cortex findings generalize to the sensory cortex more broadly, would strengthen our claims. We will include an early visual cortex ROI in our forthcoming revision. Preliminary results indicate that the early visual cortex ROI shows a similar set of findings – with evidence for coding of task-relevant and taskirrelevant visual dimensions across both tasks, but also specifically the context dimension in the hierarchy task. These results will be detailed in the forthcoming revision

(2) Construction of ROIs:

The choice and construction of the ROIs feel a bit arbitrary, as the lPFC region was constructed out of 10 parcels from Schaefer, while the AC was constructed from a different methodology (neurosynth). Did both parcels have the same number of voxels/vertices? It would be helpful to include a visualization of these masks as a figure.

We defined the lPFC ROIs by selecting Schaefer parcels in the frontal lobe that were previously mapped onto the Control A resting state network identified by Yeo et al. (2011). This network aligns with the multiple-demand network, which has also been identified in the macaque, where it includes the lPFC regions that abut the principal sulcus. Prior results from these regions in the monkey brain provide the scientific premise for our hypotheses. The two lPFC ROIs in each hemisphere were constructed out of 5 Schaefer parcels in each hemisphere. These parcels cluster into the same functional network and tend to behave similarly in univariate analyses. Given that our hypotheses do not distinguish between the different parcels, we elected to improve power by merging them into left and right dlPFC ROIs. 

On the other hand, the same approach could not be used to identify the primary auditory cortex. As Yeo et al. noted in their paper, the 17 resting state networks they identify did not adequately parcellate somatomotor and auditory cortices into distinct networks, likely due to their proximity (see Fig 14 and related text in Yeo et al. (2011)). We therefore relied on a different approach to define the primary auditory cortex, using an association test in Neurosynth to obtain a map of regions associated with the term “primary auditory”. In the revised manuscript, we will also include a primary auditory cortex ROI, defined again using a term-based association test in Neurosynth.

Our lPFC ROIs and pAC ROIs are of similar size. In the left hemisphere, the lPFC ROI (constructed from merging Schaefer parcels 128-thru-132) has, on average, 624.55 voxels. The left pAC ROI (defined with Neurosynth) has, on average, 628 voxels. In the right hemisphere, the lPFC ROI (constructed from merging Schaefer parcels 330-thru334), has 470.8 voxels on average. The right pAC ROI has, on average, 568 voxels. A table reporting the size of our parcels and ROIs was included in the supplement. In our forthcoming revision, we will additionally include a supplementary figure visualizing the ROI masks. 

(3) Task dimensionality:

In some ways, the main findings - that representation dimensionality is tailored to the task - seem to obviously follow from the choice of two tasks, particularly from a normative modeling perspective. For example, the flat task is effectively a memorization task, and is incompressible in the sense that there are no heuristics to solve it. In contrast, the hierarchical task can have several strategies, an uncompressed (memorized) strategy, and a compressed strategy. This is analogous to other studies evaluating representations during 'rich' vs. 'lazy'/kernel learning in ANNs. However, it seems unlikely (if not impossible) to form a 'rich' representation in the flat task. Posed another way, the flat task will always necessarily have a higher dimensionality than the hierarchical task. Thus, is their hypothesis - that representational geometry is tailored to the task - actually falsifiable? I understand the authors posit alternative hypotheses, e.g., "a fully compressed global axis with no separation among individual stimulus inputs could support responding [in the flat task]" (p. 36). But is this a realistic outcome, for example, in the space of all possible computational models performing this task? I understand that directly addressing this comment is challenging (without additional data collection or modeling work), but perhaps some additional discussion around this would be helpful.

We thank the reviewer for this comment, which gives us a chance to clarify our argument.

As noted by the reviewer, whether a network takes advantage of the compressibility of a task depends on its learning regime (i.e. rich vs lazy). One way to frame our question regarding the lPFC’s coding strategy, then, is to ask whether it operates in a rich or a lazy learning regime (which would predict, respectively, task-tailored vs task-agnostic representations). The reviewer’s concern is that the two task structures we employed are differentially compressible, and therefore, it is inevitable that we observe tailored representations and therefore, our hypotheses are not falsifiable.

First, it is important to clarify the theoretical premise behind our design and how it relates logically to our hypotheses. Under a lazy learning regime, a network would encode highdimensional representations of both tasks, regardless of their compressibility. On the other hand, under a rich learning regime, representational dimensionality will likely be shaped by the tasks’ structure. If the two tasks differ in their compressibility, only in the rich learning regime would the network learn representations of different dimensionality. Therefore, observing representations with dimensionality tailored to the task structure rules out the possibility that the lPFC is operating in a lazy regime. Therefore, the hypotheses are certainly testable.

The second point of clarification is that, contrary to the reviewer’s assertion, the flat task is, in fact, compressible – the task can be solved with a categorical representation of the response categories, with no sensitivity to the different specific stimuli within each category. Indeed, it is possible to train a simple, three-layer feedforward artificial neural network to perform the flat task perfectly with only 2 units in the hidden layer, demonstrating this compressibility. While we agree with the reviewer that in the space of all possible architectures one might consider the two tasks may differ in compressibility, particularly at the local levels, as we noted above, this does not imply that our hypotheses are not testable.

Finally, as a third point of clarification, our focus in this paper is on understanding the nature of coding in the lPFC in particular. Arguments based on a normative modelling perspective properly apply to the representations learned by an agent (such as an ANN or a human) as a whole. In a minimal feedforward ANN with a single hidden layer trained in a regime which encourages compression (i.e. a rich learning regime), it would indeed be the case that the representational dimensionality in that hidden layer would be higher for less compressible tasks. However, when applied to humans, such an argument applies to the brain as a whole rather than to an individual region of the brain like the lPFC. As such, it is less straightforward to predict how a single region might represent a task without additional information about the region’s inputs, outputs and broader position in a network. Even for a highly compressible task, a particular brain region may nevertheless be sensitive to all task dimensions. Conversely, even when a task is not compressible, a particular population within the brain may be invariant to some task features. For example, the primary auditory cortex is expected to be invariant to visual task dimensions.

Therefore, how a task is represented in the lPFC in particular (as opposed to the whole brain) depends on its computational function and coding principles, which remain debated. For instance, as some accounts (such as the guided activation theory) posit, if the primary function of the lPFC is to encode ‘context’ and shape downstream processing based on context, we might only expect to see the abstract coding of the auditory context in the hierarchy task (and, perhaps, the response categories across both tasks as they encode the ’context’ for the lower-level response decision), while being invariant to lowerlevel features of the input. In our paper, we specifically contrast two accounts of lPFC coding that have emerged in the literature – one positing that the lPFC learns a representation tailored to the structure of the task, and another that the lPFC encodes a high-dimensional representation that privileges sensitivity to many task features and their non-linear mixture at the cost of generalization. Regardless of the compressibility of the tasks in question, how the lPFC encodes the two tasks is an empirical question.

In our forthcoming revision, we will clarify these points in the discussion. We will also include the results of neural network simulations alluded to above.

(4) Related to the above:

The authors have a section on p. 27: "Local structure of lPFC representational geometry of the flat task shows high separability with no evidence for abstraction" - I understand a generalization analysis can be done in the feature space, but in practice, the fact that the flat task doubles as a memorization task implies that there are no useful abstractions, so it seems to trivially follow that there would be no abstract representations. In fact, the use of task abstractions in the stimulus space would be detrimental to task performance here. I could understand the use of this analysis as a control, but the phrasing of this section seems to indicate that this is a surprising result.

As explained above, there is no need for high local separability in the flat task. The lPFC could have completely abstracted over the individual trial-types that contributed to each response category, encoding only the response categories. Indeed, as also noted above, it is possible to train a simple, three-layer feedforward artificial neural network to perform the flat task perfectly with only 2 units in the hidden layer. The two hidden layer units code for each of the two response categories. 

(5) Statistical inferences:

Throughout the manuscript, the authors appear to conflate failure to reject the null with acceptance of the null. For example, p. 24: "However, unlike left lPFC, paired t-tests showed no reliable difference in the separability of the task-relevant features vs the orthogonal, task-irrelevant features... Therefore, the overall separability of pAC representations is not shaped by either task-relevance of task structure."

We thank the reviewer for pointing these out. These sentences will be corrected in the revision. For instance, the sentence above will be modified to “Therefore, we find no evidence that the overall separability of pAC representations is shaped by either taskrelevance or task structure.”

Reviewer #3 (Public review):

Summary:

In this paper, Bhandari, Keglovits, et al. explore the representational structure of task encoding in the lateral prefrontal cortex. Through an impressive fMRI data-collection effort, they compare and contrast neural representations across tasks with different highlevel stimulus-response structures. They find that the lateral prefrontal cortex shows enhanced encoding of task-relevant information, but that most of these representations do not generalize across conditions (i.e., have low abstraction). This appears to be driven in part by the representation of task conditions being clustered by the higher-order task properties ('global' representations), with poor generalization across these clusters ('local' representations). Overall, this paper provides an interesting account of how task representations are encoded in the PFC.

Strengths:

(1) Impressive dataset, which may provide further opportunities for investigating prefrontal representations.

(2) Clever task design, allowing the authors to confound several features within a complex paradigm.

(3) Best-practice analysis for decoding, similarity analyses, and assessments of representational geometry.

(4) Extensive analyses to quantify the structure of PFC task representations.

Weaknesses:

(1) The paper would benefit from improved presentational clarity: more scaffolding of design and analysis decisions, clearer grounding to understand the high-level interpretations of the analyses (e.g., context, cluster, abstraction), and better visualizations of the key findings.

(2) The paper would benefit from stronger theoretical motivation for the experimental design, as well as a refined discussion on the implications of these findings for theories of cognitive control.

We thank the reviewer for highlighting the strengths of our paper and their feedback on the writing. We have reviewed these helpful suggestions with an eye to which we may implement in our revision to improve clarity. Our forthcoming revision will 1) provide clearer scaffolding to aid the reader in understanding our design, analyses and our interpretation of the results 2) incorporate the MDS-based visualization of the representational geometries, which is currently presented in the Supplement, as a figure panel in the main text, 3) provide a justification for the particular task structures we picked in the introduction and 4) incorporate a new paragraph in the Discussion section to highlight the implications of our findings for cognitive control.

eLife Assessment

The study introduces new tools for measuring the intracellular calcium concentration close to transmitter release sites, which may be relevant for synaptic vesicle fusion and replenishment. This approach yields important new information about the spatial and temporal profile of calcium concentrations near the site of entry at the plasma membrane. This experimental work is complemented by a coherent, open-source, computational model that successfully describes changes in calcium domains. The conclusions are solid and well supported by the data.

Reviewer #1 (Public Review):

This paper describes technically impressive measurements of calcium signals near synaptic ribbons in zebrafish bipolar cells. The data presented provides high spatial and temporal resolution information about calcium concentrations along the ribbon at various distances from the site of entry at the plasma membrane. This is important information. The experiments appear to be well-done and provide strong evidence for the main conclusions reached.

Strengths

The technical aspects of the measurements are impressive. The authors use calcium indicators bound to the ribbon and high-speed line scans to resolve changes with a spatial resolution of ~250 nm and temporal resolution of less than 10 ms. These spatial and temporal scales are much closer to those relevant for vesicle release than previous measurements. Hence the results provide a unique window onto these events.

The use of calcium indicators with very different affinities and of different intracellular calcium buffers helps provide confirmation of key results.

Reviewer #2 (Public review):

Summary:

The study introduces new tools for measuring intracellular Ca2+ concentration gradients around retinal rod bipolar cell (rbc) synaptic ribbons. This is done by comparing the Ca2+ profiles measured with mobile Ca2+ indicator dyes versus ribbon-tethered (immobile) Ca2+ indicator dyes. The Ca2+ imaging results provide a straightforward demonstration of Ca2+ gradients around the ribbon and validate their experimental strategy. This experimental work is complemented by a coherent, open-source, computational model that successfully describes changes in Ca2+ domains as a function of Ca2+ buffering. In addition, the authors try to demonstrate that there is heterogeneity among synaptic ribbons within an individual rbc terminal.

Strengths:

The study introduces a new set of tools for estimating Ca2+ concentration gradients at ribbon AZs, and the experimental results are accompanied by an open-source, computational model that nicely describes Ca2+ buffering at the rbc synaptic ribbon. In addition, the dissociated retinal preparation remains a valuable approach for studying ribbon synapses. Lastly, excellent EM.

Comments on revisions:

Several concerns were raised about the kinetic analyses, and the authors have carefully acknowledged the critiques. The ideal outcome would have been a more complete kinetic readout and analyses (in particular a better readout of risetime would have improved the results). In the absence of a suitable readout of the risetime, the authors scaled back their claims and improved on the description of the falling phase of the signals. The authors have given a reasonable response under the circumstances.

In addition, the authors provided more context to their results.

I have no further concerns.

Reviewer #3 (Public review):

Summary:

In this study, the authors have developed a new Ca indicator conjugated to the peptide, which likely recognizes synaptic ribbons and have measured microdomain Ca near synaptic ribbons at retinal bipolar cells. This interesting approach allows one to measure Ca close to transmitter release sites, which may be relevant for synaptic vesicle fusion and replenishment. Though microdomain Ca at the active zone of ribbon synapses has been measured by Hudspeth and Moser, the new study uses the peptide recognizing synaptic ribbons, potentially measuring the Ca concentration relatively proximal to the release sites.

Strengths:

The study is, in principle, technically well done, and the peptide approach is technically interesting, which allows one to image Ca near the particular protein complexes. The approach is potentially applicable to other types of imaging.

Weaknesses:

Peptides may not be entirely specific, and genetic approach tagging particular active zone proteins with fluorescent Ca indicator proteins may well be more specific. The readers should be aware of this, when interpreting the results.

Author response:

The following is the authors’ response to the previous reviews

Reviewer #1 (Public review): 

This paper describes technically-impressive measurements of calcium signals near synaptic ribbons in goldfish bipolar cells. The data presented provides high spatial and temporal resolution information about calcium concentrations along the ribbon at various distances from the site of entry at the plasma membrane. This is important information. Important gaps in the data presented mean that the evidence for the main conclusions is currently inadequate. 

Strengths 

The technical aspects of the measurements are impressive. The authors use calcium indicators bound to the ribbon and high speed line scans to resolve changes with a spatial resolution of ~250 nm and temporal resolution of less than 10 ms. These spatial and temporal scales are much closer to those relevant for vesicle release than previous measurements. 

The use of calcium indicators with very different affinities and of different intracellular calcium buffers helps provide confirmation of key results. 

Thank you very much for this positive evaluation of our work.

Weaknesses 

Multiple key points of the paper lack a statistical test or summary data from populations of cells. For example, the text states that the proximal and distal calcium kinetics in Figure 2A differ. This is not clear from the inset to Figure 2A - where the traces look like scaled versions of each other. Values for time to half-maximal peak fluorescence are given for one example cell but no statistics or summary are provided. Figure 8 shows examples from one cell with no summary data. This issue comes up in other places as well. 

Thank you for this fair and valuable feedback. Following also the suggestion by the Editor, we have now removed the rise-time kinetic fitting results from the manuscript and only retain the bi-exponential decay time constant values. Further, we explicitly detail the issues with kinetic fitting, and state that the precise quantitative conclusions should not be drawn from the differences in kinetic parameters (pages 7 and 2728). 

We have included the results of paired-t-tests to compare the amplitudes of proximal vs. distal calcium signals shown in Fig. 2A & B, Fig. 3C & D, Fig. 4C & D, Fig. 5A-D, and Fig. 8E&F. Because proximal and distal calcium signals were obtained from the same ribbons within 500-nm distances, as the Reviewer pointed out, “the traces look like scaled versions of each other”. For experiments where we make comparisons across cells or different calcium indicators, as shown in Fig. 3E & F, Fig.5E, and Fig. 8B&C, we have included the results of an unpaired t-test. We have also included the t-test statistics information in the respective figure legends in the revised version.

In Figure 8, we have shown example fluorescence traces from two different cells at the bottom of the A panel, and example traces from different ribbons of RBC a in the D, and the summary data is described in B-C and E-F, with statistics provided in the figure legends.

The rise time measurements in Figure 2 are very different for low and high affinity indicators, but no explanation is given for this difference. Similarly, the measurements of peak calcium concentration in Figure 4 are very different with the two indicators. That might suggest that the high affinity indicator is strongly saturated, which raises concerns about whether that is impacting the kinetic measurements. 

Yes, we do believe that the high-affinity indicator is partially saturated, and therefore, the measurement with the low-affinity indicator dye is a more accurate reflection of the measured Ca²⁺ signal. We now state this more explicitly in the text. Further, we note that the rise time values are no longer listed due to lack of statistical significance for such comparisons, as noted above.

Reviewer #2 (Public review): 

Summary: 

The study introduces new tools for measuring intracellular Ca2+ concentration gradients around retinal rod bipolar cell (rbc) synaptic ribbons. This is done by comparing the Ca2+ profiles measured with mobile Ca2+ indicator dyes versus ribbon-tethered (immobile) Ca2+ indicator dyes. The Ca2+ imaging results provide a straightforward demonstration of Ca2+ gradients around the ribbon and validate their experimental strategy. This experimental work is complemented by a coherent, open-source, computational model that successfully describes changes in Ca2+ domains as a function of Ca2+ buffering. In addition, the authors try to demonstrate that there is heterogeneity among synaptic ribbons within an individual rbc terminal. 

Strengths: 

The study introduces a new set of tools for estimating Ca2+ concentration gradients at ribbon AZs, and the experimental results are accompanied by an open-source, computational model that nicely describes Ca2+ buffering at the rbc synaptic ribbon. In addition, the dissociated retinal preparation remains a valuable approach for studying ribbon synapses. Lastly, excellent EM. 

Thank you very much for this positive evaluation of our work.

Comments on revisions: 

Specific minor comments: 

(1) Rewrite the final sentence of the Abstract. It is difficult to understand. 

Thank you for pointing that out. We have updated the final sentence of the Abstract.

(2) Add a definition in the Introduction (and revisit in the Discussion) that delineates between micro- and nano-domain. A practical approach would be to round up and round down. If you round up from 0.6 um, then it is microdomain which means ~ 1 um or higher. Likewise, round down from 0.3 um to nanodomain? If you are using confocal, or even STED, the resolution for Ca imaging will be in the 100 to 300 nm range. The point of your study is that your new immobile Ca2-ribbon indicator may actually be operating on a tens of nm scale: nanophysiology. The Results are clearly written in a way that acknowledges this point but maybe make such a "definition" comment in the intro/discussion in order to: 1) demonstrate the power of the new Ca2+ indicator to resolve signals at the base of the ribbon (effectively nano), and 2) (Discussion) to acknowledge that some are achieving nanoscopic resolution (50 to 100nm?) with light microscopy (as you ref'd Neef et al., 2018 Nat Comm).  

Thank you for the valuable comments. We have now provided this information in the introduction and discussion.  

(3) Suggested reference: Grabner et al. 2022 (Sci Adv, Supp video 13, and Fig S5). Here rod Cav channels are shown to be expressed on both sides the ribbon, at its base, and they are within nanometers from other AZ proteins. This agrees with the conclusions from your imaging work.  

Thank you for the valuable suggestion. We have now provided this information in the introduction and discussion.

(4) In the Discussion, add a little more context to what is known about synaptic transmission in the outer and inner retina.. First, state that the postsynaptic receptors (for example: mGluR6-OnBCs vs KARs-OffBCs, vs. AMPAR-HCs), and possibly the synaptic cleft (ground squirrel), are known to have a significant impact on signaling in the outer retina. In the inner retina, there are many more unknowns. For example, when I think of the pioneering Palmer JPhysio study, which you sight, I think of NMDAR vs AMPAR, and uncertainty in what type postsynaptic cell was patched (GC or AC....). Once you have informed the reader that the postsynapse is known to have a significant impact on signaling, then promote your experimental work that addresses presynaptic processes: "...the new tool and results allow us to explore release heterogeneity, ribbon by ribbon in dissociated preps, which we eventually plan to use at ribbon synapses within slices......to better understand how the presynapse shapes signaling......". 

Thank you for the valuable comments. We have now provided this information in the introduction and discussion.

Reviewer #3 (Public review): 

Summary: 

In this study, the authors have developed a new Ca indicator conjugated to the peptide, which likely recognizes synaptic ribbons and have measured microdomain Ca near synaptic ribbons at retinal bipolar cells. This interesting approach allows one to measure Ca close to transmitter release sites, which may be relevant for synaptic vesicle fusion and replenishment. Though microdomain Ca at the active zone of ribbon synapses has been measured by Hudspeth and Moser, the new study uses the peptide recognizing synaptic ribbons, potentially measuring the Ca concentration relatively proximal to the release sites. 

Strengths: 

The study is, in principle, technically well done, and the peptide approach is technically interesting, which allows one to image Ca near the particular protein complexes. The approach is potentially applicable to other types of imaging. 

Thank you very much for this appreciation.

Weaknesses: 

Peptides may not be entirely specific, and genetic approach tagging particular active zone proteins with fluorescent Ca indicator proteins may well be more specific. Although the authors are aware of this and the peptide approach is generally used for ribbon synapses, the authors should be aware of this, when interpreting the results. 

We acknowledge the reviewer’s point and believe the peptides and genetic approaches to measure local calcium signals have their merits, each with separate advantages and disadvantages.  

Reviewer #1 (Recommendations for the authors): 

The revisions helped with some concerns about the original paper, but some issues were not adequately addressed. I have left two primary concerns in my public review. To summarize those: 

The difference in kinetics of proximal and distal locations is emphasized and quantified in the paper, but the quantification consists of a fit to the average responses. This does not give an idea of whether the difference observed is significant or not. Without an estimate of the error across measurements the difference in kinetic quoted is not interpretable. 

Thank you for this feedback. Since the kinetics information is a minor part of the manuscript, we have followed the Editor’s advice to significantly tone down the comparison of kinetic fit parameters (completely removing the rise-time comparisons), in order to put more focus on the better-documented conclusions. We also note that we did establish statistical significance of the differences in fluorescence signal amplitudes. 

Somewhat relatedly, the difference in amplitude and kinetics of the calcium signals measured with low and high affinity indicators is quite concerning. The authors added one sentence stating that the high affinity indicator might be saturated. This is not adequate. Should we distrust the measurements using the high affinity indicator? The differences between the results using the low and high affinity indicators is in some cases large - e.g. larger than the differences cited as a key result between distal and proximal locations. This issue needs to be dealt with directly in the paper. 

Thank you for this feedback. Yes, the measurements from high-affinity indicators cannot report the Ca2+ as accurately as low-affinity indicators. However, the value of HA indicators is in their ability to detect lowamplitude signals that lower-affinity indicators may miss due to lower signal-to-noise resolution.  We added a sentence on page 12 to further stress this point.

Related to the point about statistics, it is not clear how to related the horizontal lines in Figure 8 to the actual measurements. It is critical for the evaluation of the conclusions from that figure to understand what is plotted and what the error bars are on the plotted data. 

We apologize for the earlier ambiguity in Fig. 8. In this figure, we first compare proximal (panel B) and distal (panel C) calcium signals across several RBCs, labeled RBC-a through RBC-d. Each RBC contains multiple ribbons, and for each cell, we present the average calcium signals from multiple ribbons using box plots in panels B and C. In these box plots, the horizontal lines represent the average calcium signal for each cell, while the size of the error bars reflects the variability in proximal and distal calcium signals among the ribbons within that RBC.

For example, RBC-a had five identifiable ribbons. In panels D–F, we use RBC-a to illustrate the variability in calcium signals across individual ribbons. Specifically, we distinguished proximal and distal calcium signals from five ribbons (ribbons 1–5) within RBC-a. When feasible, we acquired multiple x–t line scans at a single ribbon, shown now as individual data points, to assess variability in calcium signals recorded from the same ribbon.

The box plots in panels E and F display the average calcium signal (horizontal lines) for each ribbon, based on multiple recordings. These plots demonstrate considerable variability between ribbons of RBC-a. Importantly, the lack of or minimal error bars for repeated measurements at the same ribbon indicates that the proximal and distal calcium signals are consistent within a ribbon. These findings emphasize that the observed variability among ribbons and among cells reflects true biological heterogeneity in local calcium domains, rather than experimental noise.

Hogescholen en universiteiten registreren een aanzienlijke hoeveelheid gegevens over hun studenten, wat een aanzienlijk potentieel biedt voor het genereren van informatie, kennis en monitoring. Zowel schooluitval als schooluitval in het hoger onderwijs vormen een obstakel voor economische groei, werkgelegenheid, concurrentievermogen en productiviteit, en hebben een directe impact op het leven van studenten en hun families, hogescholen en de samenleving als geheel. De hier beschreven dataset is het resultaat van de aggregatie van informatie uit verschillende onsamenhangende databronnen en bevat demografische, sociaaleconomische, macro-economische en academische gegevens over inschrijving en academische prestaties aan het einde van het eerste en tweede semester. De dataset wordt gebruikt om machine learning-modellen te bouwen voor het voorspellen van academische prestaties en uitval. Deze maakt deel uit van een Learning Analytic-tool die is ontwikkeld aan het Polytechnic Institute of Portalegre en die informatie verstrekt aan het tutoringteam met een schatting van het risico op uitval en falen. De dataset is nuttig voor onderzoekers die vergelijkende studies willen uitvoeren naar de academische prestaties van studenten en ook voor training op het gebied van machine learning.

I believe being sub certified so early in our educational careers will really help us a lot with having experience in the classroom.

This is an awesome thing you do, letters of recommendation can be very helpful and this gives us all an incentive to do our best on every assignment.

I think that this is really important. When in a classroom it is always best to have positive relationships and respect.

I am very curious what all the Â's mean, they are in a very important part of the syllabus.

eLife Assessment

This useful study presents a hierarchical computational model that integrates locomotion, navigation, and learning in Drosophila larvae. The evidence supporting the model is solid, as it qualitatively replicates empirical behavioral data, but the experimental data is incomplete. While some simplifications in neuromechanical representation and sensory-motor integration are limiting factors, the study could be of use to researchers interested in computational modeling of biological movement and adaptive behavior.

Reviewer #1 (Public review):

Summary:

The paper presents a three-layered hierarchical model for simulating Drosophila larva locomotion, navigation, and learning. The model consists of a basic locomotory layer that generates crawling and turning using a coupled oscillator framework, incorporating intermittency in movement through alternating runs and pauses. The intermediate layer enables navigation by allowing larvae to actively sense and respond to odor gradients, facilitating chemotaxis. The adaptive learning layer integrates a spiking neural network model of the Mushroom Body, simulating associative learning where larvae modify their behavior based on past experiences. The model is validated through simulations of free exploration, chemotaxis, and odor preference learning, demonstrating close agreement with empirical behavioral data. This modular framework provides a valuable advance for modeling larva behavior.

Strengths:

Every modeling paper requires certain assumptions and abstractions. The main strength of this paper lies in its modular and hierarchical approach to modeling behavior, making connections to influential theories of motor control in the brain. The authors also provide a convincing discussion of the experimental evidence supporting their layered behavioral architecture. This abstraction is valuable, offering researchers a useful conceptual framework and marking a significant step forward in the field. Connections to empirical larval movement are another major strength.

Weaknesses:

While the model represents a conceptual advance in the field, some of its assumptions and choices fall behind state-of-the-art approaches. One limitation is the paper's simplified representation of larval neuromechanics, in which the body is reduced to a two-segment structure with basic neural control. Another limitation is the absence of an explicit neuromuscular control system, which would better capture the role of segmental central pattern generators (CPGs) and neuronal circuits in regulating peristalsis and turning in Drosophila larvae. Many detailed neuromechanical models, as cited by the authors, have already been published. These abstractions overlook valuable experimental studies that detail segmental dynamics during crawling and the larval connectome.

The strength of the model could also be its weakness. The model follows a subsumption architecture, where low-level behaviors operate autonomously while higher layers modulate them. However, this approach may underestimate the complexity of real neural circuits, which likely exhibit more intricate feedback mechanisms between sensory input and motor execution.

Reviewer #2 (Public review):

Summary:

Sakagiannis et al. propose a hierarchically layer architecture to larval locomotion and foraging. They go from exploration to chemotaxis and odour preference test after associative learning.

Strengths:

A new locomotion model based on two oscillators that also incorporates peristaltic strides.

Weaknesses:

• The model is not always clearly or sufficiently explained (chemotaxis and odour test).

• Data analysis of the model movement is not very thorough.

• Comparisons with locomotion of behaving animals missing in chemotaxis and odour preference test after associative learning.

• Overall it is hard to judge the descriptive and predictive value of the model.

Reviewer #3 (Public review):

Summary:

This paper presents a framework for a multilevel agent-based model of the drosophila larva, using a simplified larval body and locomotor equations coupled to oscillators and sensory input. The model itself is built upon significant existing literature, particularly Wystrach, Lagogiannis, and Webb 2016 and Jürgensen et al. 2024. The aim is to generate an easily configurable, well-documented platform for organism-scale behavioral simulation in specific experiments. The authors demonstrate qualitative similarity between in vivo behavioral experiments to calibrated models.

Strengths:

The goal is excellent - a system to rapidly run computational experiments that align naturally with behavioral experiments would be well-suited to develop intuitions and cut through hypotheses. The authors provide quantitative descriptions that show that the best-fit parameters in their models produce results that agree with several properties of larval locomotion.

The description of model calibration in the appendix is clear and explains several aspects of the model better than the main text.

In addition, the code is well-organized using contemporary Python tooling and the documentation is nicely in progress (although it remains incomplete). However, see notes for difficulties with installation.

Weaknesses:

(1) As presented here the modeling itself is described in an unclear fashion and without a particular scientific question. The majority of the effort appears to be calibrating modest extensions of existing models and applying them to very simple experiments. This could be an effective first part of a paper on the software tool, but the paper needs to point to a scientific question or, if it is a tool paper, a gap in the current state of modeling tools needed to address scientific goals. While the manuscript has a good overview of larval behavioral papers, the discussion of modeling is more of an afterthought. However, the paper is a modeling paper and the contribution is to modeling and particularly with this work's minor adaptions of existing models, it is unclear what the principle contribution is intended to be.

(2) While the models presented do qualitatively agree with experimental data in specific situations, there is no effort to challenge the model assumptions or compare them to alternative models. Simply because the data is consistent in a small number of simple experiments does not mean that the models are correct. Moreover, given the highly empirical nature of the modeling, I wonder what results are largely the model putting out what was put in, particularly with regards to kinematic results like frequency and body length or the effect of learning simply changing the sensory gain constant. It is difficult to imagine how at this level of empirical modeling, it would appear quite difficult to integrate the type of cell-type-specific perturbation or functional observation that is common in larval experiments.

(3) The central framing of a "layered control architecture" does not have a significant impact on the work presented here and the paper would do better with less emphasis on it. Given the limited empirical models, there are only so many parameters where different components can influence one another, and as best as I can tell from the paper there is only chemotaxis and modulation of a chemotactic gain constant that are incorporated so far. However, since these are empirical functions it says little about how the layers are actually controlled by the nervous system - indeed, the larval nervous system appears to have many levels of local and long-range module of circuits at both the sensory and motor layers. It is not clear how this aspect would contribute beyond the well-appreciated concept of a relatively finite set of behavioral primitives in an insect brain, particularly for the fly larva. What would be a contradictory model and how would the authors differentiate between that and the one they currently propose? If focusing only on olfactory learning and chemotaxis, how does the current framing add to the existing understanding?

(4) The paper uses experimental data to calibrate the models, however, the experiments are not described at all in the text.

Author response:

We thank all three anonymous reviewers for their thoughtful evaluations of our manuscript and for recognizing the conceptual advance in combining agent-based behavioral simulations with systems neuroscience models. We are especially encouraged by the acknowledgement of the framework’s potential to support simulation of neural control of individual animal behavior in realistic sensory environments.

Below, we respond to each reviewer’s public comments in turn. Throughout, we have aimed to clarify our rationale for modeling choices, acknowledge limitations, and outline concrete steps for improvement in the revised manuscript.

Furthermore, the call for a better description of the model implementation as voiced by all three reviewers and additional requests from community members has prompted us to formulate a separate technically detailed description of the publicly available larvaworld software package as well as of the readily implemented models in form of a preprint paper (Sakagiannis et al., 2025, bioRxiv, DOI: https://doi.org/10.1101/2025.06.15.659765).

Reviewer #1:

We are happy to read that this reviewer considers the proposed behavioral architecture ‘a significant step forward in the field’, and that she/he recognizes the strengths of our work in the modular and hierarchical approach that provides connections to influential theories of motor control in the brain, in the experimental evidence it is based on, and in the valuable abstractions that we have chosen for the larval behavioral modeling.

The reviewer raises important points about the simplifications we have made, both conceptually and in the specific implementation of larval behaviors. Our main goal in this study is to introduce a conceptual framework that integrates agent-based modeling with systems neuroscience models in a modular fashion. To serve this purpose, we aimed for a minimal yet representative implementation at the motor layer of the architecture, calibrated to larval locomotion kinematics. This choice enables efficient simulation while allowing us to test top-down modulation and adaptive mechanisms in higher layers without the computational overhead of a full neuromechanical model. In addition to chemotaxis, we have recently used this simplified approach to model thermotaxis in larvae (Kafle et al., 2025, iScience, DOI: https://doi.org/10.1016/j.isci.2025.112809).

The reviewer notes the absence of explicit segmental neuromuscular control or central pattern generators (CPGs). We deliberately abstracted from these mechanisms, representing the larval body as two segments with basic kinematic control, to focus on reproducing overall locomotor patterns. This bisegmental simplification, which we illustrate in Supplemental Video “Bisegmental larva-body simplification”, retains the behavioral features relevant to our current aims. However, the modular structure of the framework means that more detailed neuromechanical models—incorporating CPG dynamics or connectome-derived circuit models—can be integrated in future work without altering the architecture as a whole.

We fully agree that real neural circuits are more complex than a strict subsumption architecture implies. In the Drosophila larva, there is clear evidence for ascending sensory feedback from the motor periphery to premotor and higher brain circuits, as well as neuromodulatory influences. These add layers of complexity beyond the predominantly descending control in our present model. At the same time, both larval and adult connectome data show that across-level descending and ascending connections are sparse compared to the dense within-layer connectivity. We see value in casting our model as a hierarchical control system precisely to make the strengths and limitations of such an abstraction explicit. The revised manuscript will include further discussion of these points.

In summary, our design choices reflect a trade-off: by limiting the biological detail in the lower layers, we gain computational efficiency and maintain a clear modular structure that can host models at different levels of abstraction. This ensures that the architecture remains both a tool for immediate behavioral simulation and a scaffold for integrating richer neural and biomechanical models as they become available.

Reviewer #2:

We thank the reviewer for recognizing the novelty of our locomotory model, particularly the implementation of peristaltic strides based on our new analyses of empirical larval tracks, and for providing constructive feedback that will help us improve the manuscript.

The reviewer highlights the need for clearer explanations of the chemotaxis and odor preference modules. We expand these sections in the revised manuscript with more explicit descriptions of model structure, parameterization, and calibration. As mentioned above, we have also prepared a separate preprint dedicated to the larvaworld Python package, which contains detailed implementation notes and hands-on tutorials that allow users to adapt or extend individual modules.

Regarding the comparison to empirical behavior in chemotaxis, our present analysis is indeed primarily qualitative. However, we would like to emphasize that the temporal profile of odor concentration at the larval head in our simulations matches that measured in Gomez-Marin et al. (Nature Comm., 2011, DOI: https://doi.org/10.1038/ncomms1455) using only one additional free parameter, while all parameters of the basic locomotory model had been fitted to a separate exploration dataset before and were kept fixed in the chemotaxis experiments. In addition to the simulation of chemotaxis in the present paper, we recently used larvaworld in a practical model application to estimate a species-specific parameter of thermotaxis from experiments across different drosophilids (Kafle et al., 2025, iScience, DOI: https://doi.org/10.1016/j.isci.2025.112809).

The preference index in our simulations was computed using the same definition as in the established experimental group assay for larval memory retention, enabling a direct quantitative comparison between simulated and empirical results. Variability in the simulated outcomes arose naturally from inter-individual differences in body length and locomotory parameters, derived from real larval measurements, as well as from the random initial orientation of each individual in the arena. These factors contributed to variation in individual tracks and ultimately produced preference index values that closely matched those observed experimentally. In the revised manuscript, we also discuss handedness, as highlighted by the reviewer, as another meaningful expression of inter-individual variability in Drosophila larvae and insects more generally.

Finally, we acknowledge the reviewer’s concern about the scalability and broader applicability of the model. While the present paper focuses on three specific behavioral paradigms (exploration, chemotaxis, odor preference), the modular structure of the architecture is designed for flexibility: modules at any layer can be exchanged for more detailed or alternative implementations, and new sensory modalities or behaviors can be integrated without redesigning the system. The larvaworld package, associated codebase, and documentation are openly available to encourage adoption and adaptation by the larval research community.

Reviewer #3:

This public review provides an excellent account of our central aim to build an easily configurable, well-documented platform for organism-scale behavioral simulation and we are happy to read that the reviewer considers this an excellent goal.

We thank the reviewer for her/his account of our well-organized code using contemporary Python tooling. We are currently further improving code readability and code documentation, and we will release a new version of the larvaworld Python package. We further agree with the reviewer’s assessment that understanding the model calibration currently requires reading of the appendix. For the revised manuscript we thus aim at improving our description of all calibration and modeling steps along the way. We will also make sure to improve the description of the experimental datasets used for calibration.

We recognize that our description of the paper’s scientific contribution could be clearer. In revision, we will sharpen the Introduction and Discussion to highlight our main contributions:

(1) Promoting a shift from isolated neural circuit modeling to integrated agent-based simulations in realistic environments.

(2) Proposing the layered behavioral architecture, adopting the subsumption paradigm for modular integration.

(3) Providing the larvaworld software as a ready-to-use, extensible modeling platform.

(4) Implementing an empirically calibrated locomotory model and demonstrating its integration with navigation and learning modules in replicated behavioral paradigms.

We agree with the reviewer that the next challenge is to integrate the empirically based behavioral simulations presented here with functional brain models capable of reproducing or predicting experimental findings at the level of cellular neurophysiology, including the effects of cell-type-specific manipulations such as gene knock-down or optogenetic activation/inhibition. However, based on our experience with systems-level modeling, we deliberately invested in behavioral simulation because functional models of the nervous system—including our own—often lack translation into simulated agent behavior. In many cases, model output is limited to one or more variables that can at best be interpreted as a behavioral bias, and most often represents an “average animal” that fails to capture inter-individual differences. By linking our spiking mushroom body model to behavioral simulations in a group of individual agents during memory retention tests (Figure 6C,D), we were able to achieve a first successful direct comparison between simulated and experimental behavior metrics—in this case, the behavioral preference index reported in Jürgensen et al. (iScience, 2024, DOI: https://doi.org/10.1016/j.isci.2023.108640).

Finally, we reiterate that the layered behavioral architecture is designed to promote a modular modeling paradigm. Our adoption of a subsumption architecture does not conflict with the concept of behavioral primitives; on the contrary, the notion that such primitives follow (semi-)autonomous motor programs and can be combined into more complex behaviors was the starting point for our implementation of the architecture in the fly larva. In our view, a genuinely contradictory paradigm for neural control of behavior would require a non-modular, strictly non-hierarchical organization of the nervous system and, by extension, of behavioral control.

so this milestone is important. It prevents the execution to go to infinity

so what should be the optimal frequency? one per week seems too sparse. And btw I read somewhere that allocate a "mind wandering" slot in everyday also improves concentration when we're in the main task

eLife Assessment

NeuroSC is an accessible and interactive tool for streamlined observation of neuronal morphology, membrane contact, and synaptic connectivity across developmental stages in the nematode C. elegans. This important tool relies on solid electron microscopy datasets. This resource will be of high interest to C. elegans researchers interested in nervous system wiring and circuit function.

Reviewer #1 (Public review):

The authors have done a terrific job and addressed the questions raised in my previous review. There are only some minor requests that I have and list below.

Reviewer #2 (Public review):

Summary

The past several years has seen publication of both new (Witvliet et al., 2021) and newly analyzed (Cook et al., 2019; Moyle et al., 2021; Brittin et al., 2021) data for the C. elegans connectome. The increase in data availability for a single species allows researchers to examine variability due to both stochastic events and due to changes over development. The quantity of these data are huge. To help the community make these data more accessible, the authors present a new online tool that allows examination of 3D models for C. elegans neurons in the central neuropil across development. In addition to visualizing the overall structure of the neuronal processes and locations of synapses, the NeuroSC tool also allows users to probe into the C-PHATE visualization results, which this group previously pioneered to describe similarities in neuron adjacency (Moyle et al., 2021).

Strengths

The ability to visualize the data from both a connectomics and contactomics perspective across developmental time has significant power. The original C. elegans connectome (White et al., 1986) presented their circuits as line drawings with chemical and electrical synapses indicated through arrows and bars. While these line drawings are incredibly useful, they were necessary simplifications for a 2D publication and lack details of the complex architecture seen within each EM image. Koonce et al takes advantage of their own and others segmented image data of each neuronal process within the nerve ring to create a web interface where users can visualize 3D models for their neuron of choice. The C-PHATE visualization is intended to allow users to explore similarities among different neurons in terms of adjacency and then go directly to the 3D model for these neurons. The 3-D models it generates are beautiful and will likely be showing up in many future presentations and publications. The tool doesn't require any additional downloading and is open source. This revision includes an option where hovering over an individual neurons, synapse, or contact will pull up a statistics panel. The addition of text to the video tutorials in the revision is very useful.

Weaknesses

There are several bugs with this tool, which make it a bit clunky to use and suggest a lack of rigorous testing. There are also issues with data availability. I was disappointed that my "recommendations for the authors", which focused on the user interface, were not addressed in the response to reviewers.

Reviewer #3 (Public review):

Summary:

This work provides graphical tools for reconstructing the detailed anatomy of a nervous system from a series of sections imaged by electron microscopy. Contact between neuronal processes can direct outgrowth and is necessary for connectivity, thus function. A bioinformatic approach is used to group neurons according to shared features (e.g., contact, synapses) in a hierarchy of "relatedness" that can be interrogated at each step. In this work, Koonze et al analyze vEM data sets for the C. elegans nerve ring (NR), a dense fascicle of processes from181 neurons. In a bioinformatic approach, the clustering algorithm Diffusion Condensation (DC) groups neurons according to similar cell biological features in iterations that remove chunks of differences in feature data with each step ultimately merging all NR neurons in one cluster. DC results are displayed with C-Phate a 3D visualization tool to produce a trajectory that can be interrogated for cell identities and other features at each iterative step. In previous work by these authors, this approach was utilized to identify subgroups of neuronal processes or "strata" in the NR that can be grouped by physical contact and connectivity. Here they expand their analysis to include a series of available vEM data sets across C. elegans larval development. This approach suggests that strata initially established during embryonic development are largely preserved in the adult. Importantly, exceptions involving stage specific-specific reorganization of neuronal placement in specific strata were also detected. A case study featured in the paper demonstrates the utility of this approach for visualizing the integration of newly generated neurons into the existing NR anatomy. Visualization tools used in this work are publicly available at NeuroSCAN.

Strengths:

A web-based app, NeuroSCAN, that individual researchers can use to interrogate the structure and organization of the C. elegans nerve ring across development.

Weaknesses:

minor revisions

Comments on Revisions:

The authors have satisfactorily addressed my critiques.

Author response:

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

Reviewer #1 (Public review)

Comment 

Koonce et al. have generated a web-based visualization tool for exploring C. elegans neuronal morphology, contact area between neurons, and synaptic connectivity data. Here, the authors integrate volumetric segmentation of neurons and visualization of contact area patterns of individual neurons generated from Diffusion Condensation and C-PHATE embedding based on previous work from adult volumetric electron microscopy (vEM) data, extended to available vEM data for earlier developmental stages, which effectively summarizes modularity within the collated C. elegans contactomes to date. Overall, NeuroSC's relative ease of use for generating visualizations, its ability to quickly toggle between developmental stages, and its integration of a concise visualization of individual neurons' contact patterns strengthen its utility.

We thank that reviewer for this positive assessment of our work.

Comment

NeuroSC provides an accessible and convenient platform. However, many of the characteristics of NeuroSC overlap with that of an existing tool for visualizing connectomics data, Neuroglancer, which is a widely-used and shared platform with data from other organisms. The authors do not make clear their motivation for generating this new tool rather than building on a system that has already collated previous connectomics data. Although the field will benefit from any tool that collates connectomics data and makes it more accessible and user-friendly, such a tool is only useful if it is kept up-to-date, and if data formatting for submitting electron microscopy data to be added to the tool is made clear. It is unclear from this manuscript whether NeuroSC will be updated with recently published and future C. elegans connectomes, or how additional datasets can be submitted to be added in the future.

We have added new language to more explicitly state the motivations for developing NeuroSC (Introduction, lines 98-111, and discussion lines 375-384). In a new discussion section, we also include comparisons of the features of NeuroSC with other existing tools, like Neuroglancer and Webknossos, (lines 393-417).

Briefly, the functional features of NeuroSC are substantially different (and do not exist) in other web-based tools for navigating EM datasets, including NeuroGlancer. This is because the intended use of NeuroSC is substantially different (and purposefully synergistic) to the intended use, and tools available, in NeuroGlancer. 

NeuroGlancer is a versatile tool designed primarily for web-based visualizations and sharing of large EM datasets. NeuroSC was not designed to enable this type of access to the primary EM data (purposefully done because these features were already available through tools like NeuroGlancer). 

Instead, the explicit goal of NeuroSC is to provide a platform specifically optimized for examining neuronal relationships across connectomic datasets. NeuroSC builds on the segmentations emerging from programs like NeuroGlancer, but the tools are tailored to explore relationships such as contact profiles in the context of neuronal morphologies and synaptic positions, and across datasets that represent different animals or different developmental stages. 

To achieve this, all datasets in NeuroSC were optimized to facilitate comparisons across different connectomes of segmented neuronal features, including: 1) alignment of the neurons that are compared upon the display of the segmentations; 2) synchronization of the 3D windows; 3) implementation of a ‘universal color code’ across datasets for each neuron and relationship for easy visual comparisons; 4) use of the specific neuronal names to label instances of the same cells across all available datasets. The use of precise neuronal names among separate data sets allows integration of these objects with other catalogued datasets, including genomic and neuronal activity profiles.

The formatting and display of the datasets used in NeuroSC was accompanied by the development of new tools including: 1) Rendering of the contact profiles of all neurons in the context of the morphology of the cell and the synapses and 2) C-PHATE diagrams to inspect multidimensional relationship hierarchies based on these contact profiles. In NeuroSC, C-PHATEs can be navigated and compared across multiple stages of development while visualizing neuronal reconstructions, allowing users to compare neuronal relationships across individual datasets.

We agree with the reviewer that these tools are most useful when integrated. With that intention in mind, we designed NeuroSC as a series of modular, open-source tools that could be integrated into other programs, including Neuroglancer. In that sense our intent was not to produce another free-standing tool, but a set of tools that, if useful, could be integrated to other existing web-based connectomic resources to enhance the user experience of navigating complex EM datasets and draw biological meaning from the relationships between the neurons. Additionally, we intentionally designed NeuroSC to enable the ability to integrate new methods of understanding neuron relationships as they arise. We have dedicated a more detailed section to the discussion (lines 369- 417) to better convey this intention and directly address the unique abilities of NeuroSC as a complementary tool to the powerful existing tools, including Neuroglancer.

Comment

The interface for visualizing contacts and synapses would be improved with better user access to the quantitative underlying data. When contact areas or synapses are added to the viewer, adding statistics on the magnitude of the contact area, the number of synapses, and the rank of these values among the neuron's top connections, would make the viewer more useful for hypothesis generation. Furthermore, synapses are currently listed individually, with names that are not very legible to the web user. Grouping them by pre- and postsynaptic neurons and linking these groups across developmental stages would also be an improvement.

[what do they even mean by linking?]

We thank the reviewer for this insightful comment and have implemented several improvements to address these suggestions. Specifically, we have added new features to enhance user access to quantitative data within the NeuroDevSCAN viewer:

Cell, Patch, and Synapse Statistics: Users can now see a statistics panel when clicking on a rendered neuron, contact patch, or a synapse. These panels provide the following information, respectively, and are highlighted in lines 303-315):

Cell Stats: Click on a cell rendering to show cell stats which displays the total volume and surface area of the selected neuron within the defined neuropil area of our datasets (see Methods). 

Contact Stats: Click on a patch rendering to show ‘contact stats’. This pop up displays quantifications of the selected contact relationship. Rank compares the summed surface area of contacts ("patches") between these two neurons relative to all other contact relationships for the primary neuron for the cell and the whole nerve ring. A rank of 1, for example, means this neuron pair shares the largest contact surface area of the examined relationship. “Total surface area” is displayed in nanometers, and is the summed surface area of all patches of this identity. Contact percentages are presented in two ways: (1) as the proportion of the primary cell's total surface area occupied by the contact in question, and (2) as the proportion of the total surface area of the nerve ring occupied by that same contact. (Showcased in figure S5). 

Synapse Stats: A click on a synapse rendering now shows ‘synapse stats’, which displays the number of synapses of the selected identity within the primary neuron, including any polyadic synapse combinations involving the primary neurons. (Showcased in figure S7).

(1) Grouping and Readability Improvements: While individual synapses are still visualized, their display has been improved for legibility. We have condensed the lengthy naming scheme to improve clarity and codified the synapse type by using superscript letters C, E, U to represent chemical, electrical and undefined synapses, respectively. This is explained and shown in figure S7, we added arrows to indicate the directionality of presumed information flow at each synapse. 

(2) Developmental Linkage: We can link objects across datasets via cellular identity, but each synapse in the dataset does not yet have an identity attributed to its spatial coordinates, preventing us from linking specific synapses across development beyond their connectivity (ie, that a given synapses connects cell X to cell Y, for instance), also addressed in R1.11.  

Together, these improvements substantially enhance the utility of the viewer for hypothesis generation by making key quantitative data readily accessible.

Comment

While the DC/C-PHATE visualizations are a useful tool for the user, it is difficult to understand when grouping or splitting of cell contact patterns is biologically significant. DC is a deterministic algorithm applied to a contactome from a single organism, and the authors do not provide quantitative metrics of distances between individual neurons or a number of DC iterations on the C-PHATE plot, nor is the selection process for the threshold for DC described in this manuscript. In the application of DC/C-PHATE to larval stage nerve ring strata organization shown by the authors, qualitative observations of C-PHATE plots colored based on adult data seem to be the only evidence shown for persistent strata during development (Figure 3) or changing architectural motifs across stages (Figure 4). Quantitation of differences in neuron position within the DC hierarchy, or differences in modularity across stages, is needed to support these conclusions. Furthermore, illustrating the quantitative differences in C-PHATE plots used to make these conclusions will provide a more instructive guide for users of NeuroSC in generating future hypotheses.

There are several ways to visualize DC outputs, and one way to quantitatively compare DC clustering events of neurons is via Sankey diagrams. To make the inclusion of these resources more clear, we have highlighted them in lines 175-178 (Supplemental Tables 3-6). ‘DC outputs for each strata across animals can also be inspected using Sankey diagrams (Supplemental Tables 3-6). These spreadsheets detail the neuron members at each iteration of DC, allowing the user to derive quantitative comparisons of clustering events.’

As the reviewer points out, DC is a deterministic algorithm that will iteratively cluster neurons based on the similarity of their contact profiles. To better explain the selection process for the threshold, the number of DC iterations and the quantitative metrics between the neurons, we have added new text in the Diffusion Condensation methods section.  Briefly:

Number of DC iterations: During diffusion Condensation (DC) we track the modularity of the resulting clusters at each iteration and select the iteration with the highest modularity to define the clusters that represent the strata  (Moyle et al., 2021), (Brugnone et al., 2019). Mathematically, modularity is calculated by comparing the actual number of edges within clusters to the expected number of such edges in a randomized network with the same degree distribution (Newman et al., 2006). A higher modularity value implies that nodes within the same cluster are more densely connected to each other than to nodes in other clusters. We now better explain this in lines 562-567.

Threshold for merging points: The threshold (epsilon) used to merge data points in each iteration is set as a small fraction of the spatial extent of the data: for each coordinate dimension (x, y, z), we compute the range (maximum minus minimum), take the maximum of these three values, and divide it by 10,000. This process is performed iteratively for each round of clustering until all data points cluster into a single point. We have updated the manuscript to clarify this threshold selection and included this information in the revised algorithm description and pseudocode. We now better explain this in lines 556-559.

Distances between neurons in DC C-PHATE: In our previous description in Box 1 algorithm 1, we had provided a general algorithm for DC for any high dimensional dataset. We have now revised the algorithm to indicate how we used DC for these EM datasets. 

Distances between neurons are determined by the pixel overlap between their segmented shapes in the EM dataset. We use these distances to build a graph with weighted edges, in which the weight of the edge represents the pixel overlap (the adjacency in the actual EM segmentation). Affinities between neurons, which are a proxy for their distance in the graph, are then computed as now revised in Box 1, Algorithm 1. This process is done iteratively as neurons cluster. To better communicate this, we have changed the text in lines 533-538.  

Comment

R1.5. While the case studies presented by the authors help to highlight the utility of the different visualizations offered by the NeuroSC platform, the authors need to be more careful with the claims they make from these correlative observations. For example, in Figure 4, the authors use C-PHATE clustering patterns to make conclusions about changes in clustering patterns of individual neurons across development based on single animal datasets. In this and many other cases presented in this study with the limited existing datasets, it is difficult to differentiate between developmental changes and individual variability between the neurite positions, contacts, and synapse differences within these data. This caveat needs to be clearly addressed.

We now better explain in the manuscript that the selected case study, of the AVF neuron outgrowth, is not one of just correlation based solely on an EM dataset. Instead, the case study represents the NeuroSC-driven exploration of a biologically significant event supported by several independent datasets, as now explained in lines 257-276.

Briefly, we agree with the reviewer that examining differences across individual EM datasets is insufficient evidence to make conclusions about developmental changes. But the strength of NeuroSC is in its ability to combine and compare multiple datasets, bolstering observations that are not possible by looking at just one dataset, and providing new insights on the way to new hypotheses. We now better explain that we are not looking at single connectomes in isolation and then deriving conclusions, but instead using NeuroSC to compare across 9 EM datasets. We better explain how the tools in NeuroSC, including C-PHATE, enabled comparisons across these multiple connectomes to identify apparent differences in neuronal relationships. We then explain that by using NeuroSC, we could examine these variations in neuronal relationships at the level of individual, cell biological differences of neuronal morphologies between the developmental datasets. This could be due, as pointed by the reviewer, to differences due to development, or just differences between individual animals. In the case of AVF, that features are absent in all early specimens, then arise and persist in all specimens after a certain time point, which lead us to hypothesize they result from a developmental event. Because the segmented objects in NeuroSC are linked to neuronal identities, we are also able to cross reference our observations from the EM datasets with information in other datasets and the literature. In the specific case of postembryonic development of AVF outgrowth, we can now tie the knowledge, from developmental lineage information and molecular profiles, that AVF is a postembryonically born neuron (Sulston et al. 1977, Sun et al 2022, Poole et al 2024, wormatlas.org) to the outgrowth dynamics of its neurites using the postembryonic EM datasets. Our findings using  NeuroSC provide a proof of concept of the utility of the resource and extended our understanding of how the outgrowth of this neuron affects the relationships between the neural circuits in the nerve ring.

Comment

R1.6. Given that recent studies have also quantified contact area between neurons across multiple connectomes (Cook et al., Current Biology, 2023; Yim et al., Nature Communications, 2024), and that the authors use a slightly different approach to quantify contact area, a direct comparison between contact area values obtained in this study with prior studies seems appropriate.

We acknowledge that there are multiple different approaches to calculate adjacencies. In the papers cited above, there are 3 different algorithms used:

(1) Brittin 2019 (python parse Track EM, boundary thresholds), used in Cook et al 2023, Moyle 2021, and this study).

(2) Witvliet 2021 (Matlab 2D masks), used in Cook et al 2023.

(3) Yim 2024 (3D masks), used in Yim et al 2024.

To briefly describe the different approaches, and the methods we chose for this paper:

Algorithm 1 (used in this study) defines adjacency based on distances between boundary points in TrakEM2 segmentations, allowing threshold tuning to accommodate differences in resolution and image quality across datasets—an important feature for consistent cross-dataset comparisons.

Algorithm 2 infers contact via morphological dilation of VAST segmentations, identifying adjacency through overlapping expanded boundaries. 

Algorithm 3 uses voxelwise contact detection with directional surface area measurements and normalization to account for dataset size differences. 

In NeuroSC, we use algorithm 1, mostly because we had tested the rigor of this method in (Moyle et al. 2021), where we have shown that results were robust across a range of thresholds. This flexibility enables tailored application across datasets of varying quality and scale, critical for NeuroSC’s mission of curating data sets across differing methodologies to allow for direct relationship comparisons. We detail the methodology for defining thresholds for each dataset in methods section lines 492-521, defined in Supplementary table 1. Another difference between our analysis and the previously cited work is that for our analysis we also chose to include all individually resolved neurons, including post-embryonic cells, without collapsing them into left/right or dorsal/ventral symmetry classes. In this way our approach retains the full cellular resolution of the nervous system. 

Comment

Neuroglancer is not mentioned at all in the manuscript, despite it being a very similar and widely accepted platform for vEM data visualization across model organisms. An explicit comparison of NeuroSC and Neuroglancer would be appropriate, given the similarity of the tools. Currently, published C. elegans data (Witvliet et al., 2021; Yim et al., 2024) use Neuroglancer-based viewers, and directly comparing NeuroSC and highlighting its strengths relative to Neuroglancer would strengthen the paper.

In the original manuscript we had not mentioned tools like Neuroglancer because we envisioned them as distinct, in intended use and output, from NeuroSC. But, as explained in R1.2 comment, in the revised version we have included a section in the Introduction lines 98-108 and in the Discussion (lines 369- 417) that compares these types of web-based tools and highlights synergies. 

Comment

Assigning shorthand names to strata, such as "shallow reflex circuit" (page 4, line 172), may oversimplify this group of neurons. Either more detailed support for shorthand names of C-PHATE modules should be included, or less speculative names for strata should be used.

We appreciate this comment and understand that the original language used in the manuscript to describe strata categorizations may run the risk of oversimplification. We have now clarified the text to communicate that: 1) Strata are labeled by numbers (Strata 1, Strata 2, Strata 3 and Strata 4), rather than functional features of the neurons forming part of the strata, and that 2) the assignment of ‘strata’ is just one level of classification available via DC/CPHATE (as explained below). 

To be sure, we have observed and published (Moyle et. al. Nature 2021) that within a given stratum, many neurons share the functional identities that we have used as summary descriptors for the strata (eg, shallow reflex circuits for Stratum 1; sensory and integrative circuits in Strata 3 and Strata 4; command interneurons in Strata 2, etc). However, those cell types are not the only members of the strata. We have adjusted the language in lines 197-204 to reflect this more clearly. “Stratum 1, which contains most neurons contributing to shallow reflex circuits that control aversive head movements in response to noxious stimuli, displayed the fewest changes among the developmental connectomes (Figure 3B–F; Supplementary Table 3). In contrast, C. elegans exhibit tractable behaviors that adapt to changing environmental conditions (Flavell et al., 2020). Strata 3 and 4 contain most neurons involved in circuits associated with such learned behaviors, including mechano- and thermo-sensation. This is reflected in Strata 3 and 4 showing the most change in neuronal relationships across postembryonic development.“

Comment

The authors state that NeuroSC can be applied to other model organisms. Since model organisms with greater neuron numbers include more individual neurons per cell class, the authors should support this by quantitatively demonstrating how DC/C-PHATE relationships correlate with shared functional roles among C. elegans neurons.

We now clarify in the manuscript that, like in other organisms, C. elegans neurons are also grouped into functional classes with shared characteristics. In the context of the cylindrical nerve ring of the animal, these neuronal classes are sometimes bilaterally symmetric (forming left-right pairs), four-fold symmetric and six-fold symmetric. We now explain in the discussion that the DC/CPHATE analyses group these neuron classes and their relationships (lines 442-451). In the specific section mentioned by the reviewer, we now also add new text to contextualize this concept and how it might relate to the possible use of these tools in organisms with larger nervous systems: ‘However, our previous work has demonstrated that DC/CPHATE clustering of C. elegans neurons consistently pulls out clusters of shared neuron classes and shared functional roles Moyle et al. (2021). Building on this foundation, we envision applying similar clustering approaches to larger connectomes, aiming to identify classes and functionally related neuronal groups in more complex nervous systems. We suggest that contact profiles, along with neuron morphologies and synaptic partners, can act as ‘fingerprints’ for individual neurons and neuron classes. These ‘fingerprints’ can be aligned across animals of the same species to create identities for neurons. Frameworks for systematic connectomics analysis in tractable model systems such as C. elegans are critical in laying a foundation for future analyses in other organisms with up to a billion-fold increase in neurons (Toga et al., 2012).’

Comment

Lack of surface smoothing in NeuroSC leads to processes sometimes appearing to have gaps, which could be remedied by smoothing with a surface mesh. 

We thank the reviewer for the suggestion, and understand the visibility of gaps in certain neuron processes can be distracting. But this was an intentional choice, with our main goal being to show the most accurate representation of the available data segmentation and avoid any rendering interpretations. In this way, we render the data with the highest fidelity we can and as close as possible to the ground truth of the EM segmentation. We have added language to describe this in the methods, lines 490-491, and in Figure legend 5b.

Comment

Toggling between time points while maintaining the same neurons and contact area in NeuroSC is a really valuable feature. The tool would be improved even more by extending this feature to synapses, specifically by allowing the user to add an entire group of synapses to the viewer at once (e.g. "all synapses between AIM and PVQ"), and to keep this synapse group invariant when toggling between developmental stages.

We thank the reviewer for this suggestion. In response we have now implemented a new feature to ‘clone’ a rendered scene across time while preserving the original elements to ease comparisons. Once the user has rendered a scene, they can use the in-viewer developmental slider to clone the renderings and assigned colors, but display the renderings of the newly selected timepoint. These renderings populate a new window tab which can be dragged to align developmental stage windows side by side. We have added a sentence to account for this in lines 315-317 and to the legend of supplemental Figure S11. 

Reviewer #2 (Public review)

Comment

The ability to visualize the data from both a connectomics and contactomics perspective across developmental time has significant power. The original C. elegans connectome (White et al., 1986) presented their circuits as line drawings with chemical and electrical synapses indicated through arrows and bars. While these line drawings remain incredibly useful, they were also necessary simplifications for a 2D publication and they lack details of the complex architecture seen within each EM image. Koonce et al take advantage of segmented image data of each neuronal process within the nerve ring to create a web interface where users can visualize 3D models for their neuron of choice. The C-PHATE visualization allows users to explore similarities among different neurons in terms of adjacency and then go directly to the 3D model for these neurons. The 3D models it generates are beautiful and will likely be showing up in many future presentations and publications. The tool doesn't require any additional downloading and is open source.

We thank that reviewer for this positive assessment of our work.

Comment

While it's impossible to create one tool that will satisfy all potential users, I found myself wanting to have numbers associated with the data. For example, knowing the number of connections or the total surface area of contacts between individual neurons wasn't possible through the viewer, which limits the utility of taking deep analytical dives. While connectivity data is readily accessible through other interfaces such as Nemanode and WormWiring, a more thorough integration may be helpful to some users.

We thank the reviewer for this feedback and in response have now implemented displays with quantitative information in NeuroSC. Now, upon hovering over a contact patch or synapse, the user will see the quantitative data of the relationship. For contact patches, you will see the total area shared between two neurons in that dataset. On hovering over a synapse, you will see how many synapses there are in total with the same members and throughout the dataset. We agree that this improves user analyses, (see also R1.3 response).

Comment

There were several issues with the user interface that made it a bit clunky to use. For example, as I added additional neurons to the filter search box, the loading time got longer and longer. I ran an experiment uploading all of the amphid neurons, one pair at a time. Each additional neuron pair added an additional 5-10 seconds to the loading. By the time I got to the last pair, it took over a minute to load. Issues like these, some of which may be unavoidable given the size of the data, could be conveyed through better documentation. I did not find the tutorial very helpful and the supplementary movies lacked any voiceover, so it wasn't always clear what they were trying to show.

We appreciate that some of the more complex models can take a while to load. One of our core goals is to keep the high resolution of our models to most accurately represent the EM data, so we had to compromise between resolution and loading times. But to address this concern we have now added a ‘loading’ prompt that reassures the user when there is a wait. We also added, as suggested, text guidance throughout all of the supplemental videos (Supplemental Videos 1-4).

Reviewer #3 (Public review)

Comment

A web-based app, NeuroSC, that individual researchers can use to interrogate the structure and organization of the C. elegans nerve ring across development In the opinion of this reviewer, only minor revisions are required.

We thank that reviewer for this positive assessment of our work.

Comment

Contact is defined by length, why not contact area? How are these normalized for changes in the overall dimensions of neurons during development?

To clarify our methodology: the adjacency algorithm that we use generates a 2D adjacency profile by summing the number of adjacent boundary points per EM section, which are then summed across all EM z slices.

Contact area can be derived by multiplying the adjacency length in each slice by pixel resolution and z-thickness. Prompted by the reviewer we have now also calculated and display contact surface areas, along with their ranks among all contact relationships for a given neuron. These can be inspected directly via the interface by clicking on a rendered cell or contact patch (Figure S5 and lines 308-312). We believe these additional surface area metrics enhance the interpretability and utility of the viewer.

We apply normalization at the level of the adjacency threshold to account for dataset-specific differences such as contrast, boundary definition, and age-related changes in neuropil packing density. This normalization is applied before running the adjacency algorithm. We do not normalize by individual neuron size, as the contact data are intended to reflect relational differences between neurons, rather than absolute morphological scaling. In fact, our addition of a scale-spheroid within each rendered model emphasizes the large increase in spatial scale that the nerve ring experiences during larval growth.  

Comment

Figure 1, C&D, explanation unclear for how the adjacency matrix is correlated with C-Phate schematic in D.

We thank the reviewer for the comment and have clarified this section by adding greater detail to the explanation of how an adjacency matrix is computed (lines 149-155), as well as a description now in the figure legend 1C. Additionally, we revised Figure 1C and D to simplify neuron representations/colors and to simplify the adjacency heat map gradient. We also extended the area of contact between neurons on Figure 1C to better reflect what would be considered a “contact”. Lastly, in the figure, we changed the color and placement for the z plane arrow and label from black to white, to make it more visible, to highlight the method of computing adjacency for each z slice. 

Comment

Figure 4, panels F & G, unclear why AVF is shown in panel G (L3) but not panel F (L1). Explanation (see below) should be provided earlier, i.e., AVF is not generated until the end of the L1.

We have now clarified this important point by adding labels to Figure 4 panels F and G, ‘Pre-AVF outgrowth’ and ‘Post-AVF outgrowth’ respectively. Briefly, the point is that AVF grows into the nerve ring after the L2 stage, and that is why it is absent in panel F (L1 stage, now with the label ‘Pre-AVF outgrowth’).  

Comment

Line 146 What is the justification for the statement: "By end of Larval Stage 1 (L1), neuronal differentiation has concluded...."? This statement is confusing since this sentence also states that "90% of neurons in the neuropil...have entered the nerve ring..." which would suggest that at least 10% additional NR neurons have NOT fully differentiated.

We have fixed this sentence in the text. Now the sentence reads ‘By Larval stage 1 (L1) 90% of the neurons in the neuropil (161 neurons out of the 181 neurons) have grown into the nerve ring and adopted characteristic morphologies and positions. 

Lines 171-175 What is meant by the statement that "degree of these changes mapped onto...plasticity? What are examples of "behavioral plasticity?"

We have added the following new lines of text (lines 200-204) and now additionally cite a review discussing C. elegans behaviors to clarify and give context to behavioral plasticity. ‘C. elegans exhibit tractable behaviors which can adapt due to changing environmental conditions  (Flavell et. al. Genetics 2020). Strata 3 and 4 contain most neurons belonging to circuits associated with such learned behaviors, including chemo, mechano and thermo sensation. This is seemingly reflected by strata 3 and 4 harboring the most readily recognized set of changes in neuronal relationships across postembryonic development.’  

Comment

Lines 189-190 The meaning of this sentence is unclear, "The logic in....merge events."

This sentence has been deleted and we have instead refocused our descriptions of C-PHATES comparisons by neuronal clustering trajectories and cluster members (rather than iterations).

Comment

Lines 193-208 This section reports varying levels of convergence across larval development in C-Phate maps for the interneurons AIML and PVQL. Iterations leading to convergence varied: 16 (L1), 14 (L2), 22 (L3), 20 (l4), 14 (adult). The authors suggest that these differences are biologically significant and reflect the reorganization of AIML and PVQL contact relationships especially between the L4 and adult. Are these differences in iterations significant?

We agree this could be confusing and instead of focusing on comparing the iteration at which each merging event occurs, we now focus on examining the differences in members of clusters, before and after the merge event. Cluster membership is easier to interpret than the differences in the number of DC iterations (lines 224-229).

Lines 240-241 States that AVF neurons "terminally differentiate in the embryo" which is not correct. AVF neurons are generated from neuronal precursors (P0 and P1) at the end of the L1 stage which accounts for their outgrowth into the NR during the L2 stage. 

We thank the reviewer for the correction and have edited the text to read: ‘AVF neurons are generated from neuronal precursors (P0 and P1) at the end of the L1 stage (Sulston et al. (1983); Sun and Hobert (2023); Poole et al. (2024); Hall and Altun (2008); Sulston and Horvitz (1977). AVF neurons do not grow into the nerve ring until the L2 stage, and continue to grow until the Adult stage (lines 261-266).’

Comment

Lines 289-315. A detailed and highly technical description of website architecture would seem more appropriate for the Methods section.

We agree and have moved this section to the methods as suggested (lines 663-690).

Comment

Line 307 "source data is" should be "source data are"

Thank you- we have fixed this grammatical error.

Comment

Line 324 "circuits identities" should be "circuit identity".

Thank you- we have fixed this grammatical error.

Comment

Trademark/copyright conflict with these sites? https://compumedicsneuroscan.com/about/ https://www.neuroscanai.com/

We thank the reviewer for drawing our attention to this. To avoid potential conflicts, we have proactively altered the name to NeuroSC throughout the paper.

Plumbing pipes Guide: Understanding which pipe is better & is PPR pipe the one to consider.

Learning to understand!

DOI: 10.24272/j.issn.2095-8137.2024.397

Resource: (Jianzhen Li lab The Chinese University of Hong Kong Cat# lhb, RRID:AB_2651067)

Curator: @scibot

SciCrunch record: RRID:AB_2651067

What is this?

DOI: 10.24272/j.issn.2095-8137.2024.397

Resource: (Jianzhen Li lab The Chinese University of Hong Kong Cat# fshb, RRID:AB_2651068)

Curator: @scibot

SciCrunch record: RRID:AB_2651068

What is this?

DOI: 10.1080/2162402X.2025.2543620

Resource: None

Curator: @dhovakimyan1

SciCrunch record: RRID:IMSR_JAX:000664

What is this?

I think I saw an experiment like this somewhere on Instagram. I noticed how easy it was to be influenced by words of another. I wonder what would have happened if some of the students saw another experiment similar prior to theirs, would their results be different?

Once for a key-club community service event, we were teaching a group 3rd graders how to make some type of craft and art. After the students were done they had all of their art presented, but then I had ignorantly said "I really like this one" which is a big no no. I was then later told that I cannot say things like that because it might put another student down and keep them from trying other things in the future.

I was introduced to the concept of Growth vs Fixed mindset very early on in my education. Even though I was taught this I would say I am fairly fixed on a lot of things and that I just cannot do something. As of now I would say I have a growth mindset, so I wonder if it can be as easy to change how you think with the flick of a switch?

This reminds me of what Dr Alberto Gutierrez mentioned, in the story with one of his cousins. How his cousin was hyped up in highs-school, but when something challenging occurred he sort of crumbled and hid from the problem. How much is too much affirmation?

Mindsets and Achievement

• "The students with a fixed mindset believed that if you worked hard it meant that you didn't have the ability, and that things would just come naturally to you if you did."
• This is the exact mindset I remember having coming out of high school during the peak of covid. The discomfort of forcing the students to study at home and not live the actual college lifestyle.

How Do Students Learn These Mindsets ?

• "Instead of giving them confidence, it made them fragile, so much so that a brush with difficulty erased their confidence, their enjoyment, and their good performance, and made them ashamed of their work."
• While we are in school, teachers tend to sweeten things up, so that we feel better about ourselves. When in reality, all it is doing is deteriorating our brains and starting bad habits. We should be told how things really are to prepare us for the real world.

** Brainology**

• "Their studies and ours also found that negatively stereotyped students (such as girls in math, or African-American and Hispanic students in math and verbal areas) showed substantial benefits from being in a growth mindset workshop."
• Being stereotyped at a young age can be damaging to the brain, almost limiting the mindset because of your ethnicity sounds insane.

What Do We Value?

• "Most of all, teach them that by taking on challenges, making mistakes, and putting forth effort, they are making themselves smarter."
• We often think taking the easy route is the best, but in reality, when we work on more difficult situations it helps develop our brain and allows us to think thoroughly.

Having a fixed mindset is deciding to limit your brain to endless ideas. I believe all students should grow out of a fixed mindset and find their purpose. Being able to grow out of the fixed mindset might even show you how strong your brain really is or how easy it maybe to rewire it.

I felt this annotation was significant to those that have inspirational idols because they don't realize that those idols constantly worked hard for people to notice their achievements. When viewing our idols through their hard work and effort, it helps us to learn how to use our talents in developing our success.

I believe this annotation was important because school is a place where you can grow and learn so it's great that the session workshop was able to redirect a group of student's minds in realizing that they are part of a learning community.

I found this information important because most parents don't understand that there's a difference between giving your kid confidence and helping your kid gain the confidence in their learning ability. I liked the way Dweck, Mueller, and Kamins studied this test to improve their theory on the fixed and growth mindsets as well.

I believe this annotation is important because many fixed mindset students tend to compare themselves to others such as, Einstein, but when students have a growth mindset, they believe that they can be just as smart as Einstein through hard work and dedication.

I found this annotation important because there are many students that give up when things get hard and don't strive to their excellence or is afraid of taking risks and challenges, so it is good that we learn the different mindsets, so that students don't get any demotivation.

I can't find the Appendix?

A lesson learnt from Haskell

Debating between professional and DIY appliance repair in Sydney? Learn the key differences, risks, and benefits to help you make the best decision for your home appliances.

I see full sentences copied verbatim from this reference into the current article.

Is these also being written by AI?

• The virtues highlighted—self-control, generosity and other higher principles of life.
• Hospitality emphasized through the Persian understanding of giving gifts—communicating the ideas of returning honor.
• Patriarchal ideas of male lineage formulate cultural alliances while women are seen as diplomatic instruments.

• the ethical virtues of pietas are displayed in Siawosh’s devotion. Ritual respect understood from the perfumed letters to show adherence to the cultural order.
• Yet again, loyalty and honesty are highlighted, along with diplomacy
• “Tender green of spring”—refers to a metaphor for the joyand prospering of righteousness. This is a shift into the emotionaspects of the character.

• There is divine disfavor here—kingship is seen with Iran’s moral victory. Zoroastrian ideals are embedded in the religious perspective—truth and what it means to be righteous over ideas of evil. Courage and discipline are placed over Garsivaz’s loyalty to Afrasiyab because it supports the wrongdoing.
• There are geographical and symbolic Persians rhymthmic prose utilized in words like— balkh and Zabolestan.
• In regards to the patriarchal status, kings are the center of all power. There are emphasis on authority and succession.

• This reinforces the central tragic theme of the entire play—It is a cosmic casualty. All disaster is referred to the divine hand of the gods bringing about judgement upon the pride of the people. The belief are purely Greek in their expressions, especially in how they understood pride to be dealt with the most shameful judgement.

• The scale of the slaughter conveyed. This is seen as an act of divine judgement for the Persians.

• The hyperbolic opening to set the scene of the destress and mood.

• Grendel is seen as the divine wrath upon the people, a curse thing. The Christian ideas of punishment and the permission of human trials by God. Grendel’s evils are condemned for killing innocent warriors. This is also emphasized by the Gummere rendition (https://studylib.net/doc/9435792/file?utm_source=chatgpt.com). Both capture the theological imagery and how the medieval scribes would have understood or edited the story to fit their culture.

• “undarkly” an old English diction meaning—clear. This provides and archaic hue by the author.

Here he is basically saying that the lower class (famers and laborers) are not deserving of wealth because they did not “work for it”

Donggyun used to mention with me about this. It's possible that the improvement in performance is just due to noise/artifact (e.g randomness in seed), but might not be real improvement.

this is gold

Yes different methods break in different ways but they might share some failure modes (e.g sample inefficiency). We want to find a big bit to flip, i.e the one that is shared by as much existing methods as possible.

need elaboration

With the learning in this class, how are student's that has a harder time to grasp and understand the class's lecture, assignments, the modules, and training? Especially if English is their 2nd language or have a harder time understanding and if they were in ESL/ ELL classes in their high school years and was not able to/ were never able to pass/ get out of the Special Education Class?

yes and it is also more than that isn't it, or it can be. you can see yourself, the way you act and behave and speak in an extraordinary level of detail, whilst at the same time knowing what is in the mind of the main character i.e. you.

Stephen Grosz psychoanalyst and author of The Examined Life...

a bit cheesy (it's the times uk so) but still

the third annotation

I never knew what the term "open pedagogy" was. Discovering what it fulfills is very interesting because it makes education more flexible, showcasing that students can not only learn from information that is given but also make other students learn and be welcomed to creative thoughts and ideas.

Television news is today as political as it was back in the 1950s. One could say that Fox aligns more with Republican views and that NBC favors more Democratic views.

It is worth noting the difference in technology here. TV has come a long way from the days of being offered via UHF (Ultra-high Definition) to now being offered in HD (high definition), and what about the many types of TVs that exist today? LCD (Liquid Crystal Display) and OLED (Organic Light-Emitting) are just a couple. Back in the old days, at most we could only choose from black and white or colored TV.

Before talking pictures, the big thing was radio. In 1930, talking pictures were still relatively new. It would be interesting to see how much business radio stations lost due to the creation of talking pictures. One movie that illustrates this change is "Singing in the Rain"

The only current broadcasting that comes to mind is PBS, which is even losing funding. While most streaming services and cable TV providers are privately owned, these companies still need to obtain federal licensing. However, it seems like the standards to get a license are much lower today.

Something that has not changed from the old days and now is that television entertainment is all about making money.

I find it so fascinating to read in detail about the different processes of communication. In my own life, it feels like such an ingrained ability and people communicate without seeming to have a second thought about it. I do not usually comprehend the scale with which communication has grown and changed over the centuries. Thinking back on the history of communication, shifting from the "talking era" to the "manuscript era" and comparing it to where we are now, with social media and global mass communication is a staggering thought.

Good references 👍

this is fact, gifts can only take someone so far in life, but to work hard and put effort into something can take someone even farther and give them more opportunities.

Giving a sterotype to certain groups and telling them that they either "got it or they don't" can be harmful in growing their brains. They can assume that others will excel at a task because they "have it" while they will sya that they themselves "Don't have it" because they aren't part of that group. It can also be harmful to the group as they can feel like they also can't "have it" because they are part of a group of people.

Effort is the key to success, not only talent. And "talent is nothing without hard work" - Cristiano Ronaldo. Anyone has the potential to do anything, the only thing holding someone back is their mindset.

The students given appraisal for intelligence suffered the consequences of having a big ego and when faced with difficulty they would've rather avoided it and thus learned a "fixed mindset", however the student praised for their efforts were able to maintain their confidence and were able to improve as they ddint back away from challenging problems and faced them to learn more giving them a "growth mindset"

This is important because it proves that learning about growth mindset can actually improve grades. And it also proves that there's no such thing as being "math person." Putting effort will allow anyone to be anything they want if they work hard enough for it.

These students had a growth mindset, they realized that through effort they could expand their brains and become smarter through effort. Even if they don't believe they can be as smart as Einstein they believe tat people can still grow and improve te abilities that they have.

This shows how praising someone for their efforts rather than their "intelligence" then they'd choose to actually go through the efforts of failing to get better. Further increasing their confidence and motivation to push forward and grow.

The idea that even just thinking that intelligence is something that can grow or is fixed actually can affect the ability to grow the mind is quite an astonishing fact, It affects the " motivation, learning, and school achievement" of the person.

This shows that mindset of people can hold them back as usually a person with a growth mindset tends to be on the correct side. They are usually also straightforward and believe if you work hard for something you will accomplish it. While on the other hand someone with a fixed mindset believes that no matter how hard you work for something then it means you didn't have the ability to do it, and that it should come naturally.

Intrapersonal communication (the way you talk to yourself in your mind) is a form of communication that I have always thought to be incredibly important. Ever sense I was a child, I have always had a "think before I do" mentality and it has carried through to adulthood. It has helped me in many situation to reason through problems in my head and avoid conflict and confusion with others. it also plays a huge role in how we see ourselves and the world around us. Learning to communicate with yourself kindly and productively, plays a key role in how we communicate to others.

Mass communication, especially on the scale that we are witnessing it today, is such a double edged sword. The ability to project information into society rapidly and in easily accessible ways can be a great benefit, as is the case with amber alerts, which help to increase public vigilance on a large scale. A major drawback with mass communication as I see it, comes with people that have little regard for its ethical use. The ability to spread highly palatable and sometimes subtle information that can sway the way in which we think, act, and feel, perhaps without even being aware, means that society as a whole must be hyper aware of the media they are consuming.

My dad has connections with the local newspaper from Star, and they print magazines now instead of only newspapers. Currently they continue to grow, which is interesting in this mostly digital world today. I find it interesting especially because of how you only really find newspapers in small town areas such as Star, and usually at the cafes. Newspapers have their niche places that you find them

Essentially, Intrapersonal is thoughts withing your head/personal self Interpersonal communication is within friends/family/ to one person, Group communication is generally with multiple others than yourself, sometimes towards a shared goal Public communication is the X post, facebook post, or instagram pic you posted last week, Mass communication is used in situations like newspapers, PSAs, and school rallies

Goal-oriented communication could also happen intrapersonally, could it not? it is a form of group communication, but could be considered a form of intrapersonal communication due to the way that most people plan within their brains. For instance, I have a goal, so I plan, and work on that goal by myself, but use intrapersonal goal-oriented communication.

I'm curious if people who have been diagnosed have more intrapersonal communication than neurotypical people. Is it as significant as people who are neurotypical? Are there as many internal discussions? What if people who have ADHD have more significant questions?

This reminds me of how my father and one of his employees communicate. There is a language barrier between them, so most of the time when they work together he's using sound affects and grand gestures. It's quite funny to watch, so if he needs to talk about blending something, he generally uses the "SZHHZZZZ" (idk blender sounds) and twirling a finger to simulate the blender spinning. She does the same with him.

hello! you're on hypothes.is... it has such potential. if you find something interesting on the internet please @everythingness!... @noone

What measures were taken during the Riots? How many cities were under Republican control?

In this excerpt, "I'm whacked" is an informal way of saying that the speaker is extremely tired or exhausted. The surrounding conversation reflects on activities and feelings of fatigue related to playing golf and other casual interactions. The phrase suggests a sense of drain, either physical or emotional, after engaging in a lengthy or challenging experience.

中文解释： 在这一摘录中，“我累坏了”是一种非正式的表达方式，意思是说说话者感到非常疲惫或精疲力尽。周围的对话回顾了与打高尔夫有关的活动和疲惫感。这句话暗示了在经历了一段漫长或具有挑战性的经历后所感受到的身体或情感上的疲惫。

English Explanation:

In the excerpt, "feisty" describes an energetic and spirited attitude, particularly regarding a female professional golfer who makes the sport seem enjoyable and engaging. The conversation highlights how her feistiness brings excitement to golf, indicating that her lively nature enhances the experience. The context also includes humorous banter about golf, suggesting that the enjoyment of the game can depend on the people you are playing with.

Chinese Explanation:

在这个摘录中，“feisty”形容一种充满活力和精神的态度，特别是指一位女性职业高尔夫球手，她让这项运动看起来既有趣又吸引人。对话强调她的活泼使高尔夫变得更加刺激，表明她的生动个性提升了这项运动的体验。上下文还包含了关于高尔夫的幽默对话，暗示打球的乐趣可能取决于与你一起打球的人。

whacking

English Explanation: In this excerpt, the term "whacking" refers to the act of hitting a golf ball with force, often used informally to describe swinging a golf club. The conversation involves two individuals discussing golf, and one mentions following a female professional golfer online who is skilled and energetic. This suggests that "whacking" can also imply enjoying the game and making it look exciting. The mention of playing with a partner, Martin Gibson, indicates that one's experience can vary based on the playing companion.

Chinese Explanation: 在这段摘录中，“whacking”一词指的是用力击打高尔夫球，通常以非正式的方式形容挥动高尔夫球杆。对话中，两个人在讨论高尔夫，其中一人提到在线关注一位女性职业高尔夫球手，她技艺高超且充满活力。这表明“whacking”也意味着享受这项运动并使其看起来充满乐趣。提到与合作伙伴马丁·吉布森一起打球，说明与打球伙伴的配合会影响个人的游戏体验。

compare traditional value of rhetoric?

Note: This response was posted by the corresponding author to Review Commons. The content has not been altered except for formatting.

Learn more at Review Commons

Reply to the reviewers

We are very grateful for the positive feedback from all three reviewers. Below, we address each point in detail and outline proposed experiments and revision plans, with changes indicated by an underscore.

__Reviewer #1 (Evidence, reproducibility and clarity (Required)):

In this paper "Magnesium depletion unleashes two unusual modes of colistin resistance with different fitness costs," the authors examine how Pseudomonas aeruginosa evolves resistance to colistin, a last-resort antibiotic for multidrug-resistant Gram-negative infections. Although colistin resistance is a major clinical challenge, its underlying mechanisms, particularly under nutrient-limited conditions typical of infections, are not fully understood. The study shows that under low magnesium (Mg²_⁺_) conditions-mimicking infection or biofilm stress-P. aeruginosa can develop colistin resistance via two distinct genetic pathways, each with unique fitness costs. The first involves mutations in genes such as htrB2 and lpxO2, granting strong resistance but compromising the outer membrane and increasing susceptibility to other antibiotics. The second involves regulatory mutations (e.g., in the oprH/phoP/phoQ promoter) that confer resistance with minimal membrane defects and generally lower fitness costs. These resistance strategies lead to different trade-offs: membrane-compromising mutations reduce bacterial fitness without colistin, while regulatory mutations typically avoid these penalties, with context-dependent effects. The study underscores clinical relevance, noting that in infections-such as in cystic fibrosis-other microbes like Candida albicans may deplete magnesium, indirectly promoting resistance evolution. Overall, this work offers important insights into antibiotic resistance in nutrient-stressed, polymicrobial environments, highlighting how magnesium availability shapes resistance evolution and fitness costs. The findings suggest new avenues for therapeutic intervention and call for a reevaluation of antibiotic strategies in nutrient-competitive infection settings.

Work is timely and important. Colistin resistance represents an urgent threat as colistin is a last-resort antibiotic used against multidrug-resistant Gram-negative pathogens. Insights into mechanisms evolving under nutrient limitation are highly relevant given the prevalence of such environmental conditions during infection and microbial biofilm growth. The study reveals two previously uncharacterized pathways to colistin resistance in P. aeruginosa triggered by magnesium (Mg²_⁺_) depletion, each with distinct genetic signatures and trade-offs. This finding directly impacts the understanding of polymicrobial infection dynamics, especially where magnesium sequestration by fungi/ or other microbes may occur. The identification of fitness costs and pleiotropic effects associated with specific resistance mutations provides crucial guidance for clinicians considering antibiotic stewardship and combination therapy strategies.

__

We thank the reviewer for their summary of our study and its potential impact.

__Drawbacks

• Experimental scope: While the study is comprehensive for P. aeruginosa, the broader applicability to other Gram-negative pathogens is not directly tested.__

In our revision, we now explicitly point out that the magnesium limitation we have observed broadly applies to Gram-negative bacteria, as we demonstrated in our previous PLOS Biology paper. Therefore, we expect the same themes (and even genes, which are broadly conserved) to apply to Gram-negative bacteria in general. However, a full-fledged experimental study of other Gram-negative pathogens is outside the scope of our current study, which required a 90-day experimental evolution.

__Strengths

• Experimental evolution: This work uses laboratory evolution under controlled Mg²_⁺_-limited conditions to simulate selection pressures relevant to infection microenvironments. • Genetics: Systematic identification and functional validation of key mutations-particularly in htrB2, lpxO2, and the oprH/phoP/phoQ promoter-give mechanistic depth to the findings. • Two distinct resistance modes: Evidence for (i) one pathway leading to colistin resistance via htrB2 mutations, resulting in high resistance but significant membrane integrity loss and increased susceptibility to other antibiotics. (ii) a second pathway providing resistance without compromising membrane integrity, highlighting evolutionary flexibility and ecological implications. • Fitness assessments: measurement of the costs associated with each resistance strategy, both in terms of membrane integrity and susceptibility to other agents. • Relevance: Connection to natural scenarios, such as magnesium sequestration by fungi (e.g. Candida albicans) in polymicrobial environments, underscores the ecological and clinical significance. • This manuscript is well written with clearly logical hypothesis testing__

We thank the reviewer for their appraisal, especially for recognizing the rigor and broader biological implications of our study.

__Drawbacks

• Experimental scope: While the study is comprehensive for P. aeruginosa, the broader applicability to other Gram-negative pathogens is not directly tested.__

We agree with the reviewer's point about broader applicability in other Gram-negative bacteria, as many of the lipid A biosynthesis genes are conserved among diverse bacterial lineages. We will include this point in our revised Discussion to suggest relevance to other Gram-negative bacteria:

"We previously showed that magnesium sequestration by fungi applies not only to P. aeruginosa but to other Gram-negative bacteria as well (ref). Our current study lays a foundation for developing evolution-guided strategies to combat multidrug-resistant P. aeruginosa and other Gram-negative bacteria that can also acquire colistin resistance. Since many other antibiotic mechanisms are similarly dependent on metal ions (refs), our work suggests that nutritional competition for metal ions may alter initial antibiotic resistance in Gram-negative bacteria and potentiate new evolutionary pathways of antibiotic resistance."

• __ Mechanistic depth: Some inferred mechanisms (e.g., the precise molecular impact of late-occurring adaptive mutations) merit deeper biochemical analysis.__ We will emphasize in our Revision that the MS data of endpoint clones and triple mutants reveal that their lipid A structures are identical. This suggests that the role of other late-occurring mutations in enhancing resistance is likely through lipid A-independent pathways.

• __ Results Lines 414- 423: While correlation is most what makes sense for some drugs, causality is implied (membrane defects increase susceptibility), but could be strengthened by directly measuring antibiotic uptake (e.g., fluorescence) or membrane permeability for these 3 antibiotics.__ We thank the reviewer for highlighting the issue of causality. For the three antibiotics tested, the most direct way to measure their effect is by measuring their impact on bacterial growth directly, which is what we have done. Our membrane permeability assay using NpN uptake operates under the same conditions suggested by the reviewer and directly measures molecular uptake. Moreover, only fluorescently labeled vancomycin is commercially available among the three antibiotics tested. Since it binds to the cell wall, its utility to measure membrane defects is more limited than the NpN assay we have already used. However, in response to this comment, we will make clear in our revision that we infer that increased susceptibility to other antibiotics is due to their increased membrane permeability.

__ o Effect is mild and mostly not significant. It is also not clear whether authors only tested a handful of mutants shown in Fig. 7B-D or whether other clones were also tested. The sample of endpoints (P2, P5, P8) covers well-characterized lineages, but additional evolved clones or a broader panel could boost generality about other antibiotics. The authors note "significantly lower MICs" statistical treatment is implied; explicit statistical values and replicate numbers should be given in the text or figures.__

We slightly disagree with the reviewer that the results are not significant. Even two-to-three-fold differences in MICs translate to large differences in microbial competition. These three endpoint clones are representative of all eight evolved strains after 90-day evolution experiments. Moreover, we will emphasize in the Revision that we have tested all the mutations found in the endpoint clones; we know what these are from whole genome sequencing of multiple endpoint clones. In addition, we will explicitly state the p-value in the legend of Figure 7.

• __ The structural or physiological nature of "mild" vs. "severe" membrane defects could be better defined/quantified.__ Although we agree with the reviewer's suggestion, the variability of the SEM assay makes the classification of membrane defects based on cell morphology hard to quantify. We therefore only use the SEM images as representative of the various defects observed. For a more quantitative assay of the membrane defects, we instead rely on the standard NpN uptake assay to quantify membrane permeability as a quantifiable readout for membrane defects.

• __ Quantitative limits: Authors should add in the discussion that statistical robustness could be strengthened-for example, by including longer-term evolutionary predictions.__ We are not sure what the reviewer means and so cannot address this point completely. We ask the reviewer to rephrase this point, and we will address it to the best of our abilities.

• __ in vivo relevance: While the ecological context is discussed, direct in vivo confirmation (e.g., in animal infection models) of the observed resistance trajectories would increase translational impact and relevance.__ We agree with the reviewer's point. However, it is not trivial to directly perform evolution experiments of microbes in animal models. There are only a handful of labs worldwide that have working CF-relevant animal models. However, the colistin resistance mutations we identified provide a tool to look deeper into how colistin-resistant P. aeruginosa can evolve in vivo.

• __ Some sections are repetitive or overly detailed; condense where possible (especially on mutation lists and background for each claim).__

We will condense our manuscript as the reviewer suggested in our revision. Adding a graphical summary as suggested will also allow us to be more succinct in our description.

__Other comments

• Authors should provide clarification on how the Mg²_⁺_ concentrations used in vitro compare to those found in clinically relevant infection settings. This would be helpful to enhance significance.__

We thank the reviewer for raising this good point. Based on our previous work, we know the Mg2+ levels in our model (0.3-0.45mM) are within the physiological range of Mg2+ in infection settings (0.1-0.8mM). We will highlight this point in the introduction.

• __ Authors should explicitly report statistical methods (e.g., types of tests, adjustments for multiple comparisons) in figure legends for reproducibility.__

We will include the details of our statistical tests in each panel of figures both in the main text and the supplement.

• __ Nomenclature for key mutations and their position within the genetic context (e.g., htrB2 mutation specifics) could be more detailed in figures or supplemental materials.__

We will name each of the particular mutations tested to be specific about the nature of all the evolved mutations in our figure legends.

• __ The manuscript could benefit from a graphical summary illustrating the two distinct evolutionary pathways and their respective fitness landscapes.__ We thank the reviewer for this suggestion to enhance the clarity of our work. We will make a new graphical summary highlighting two different evolutionary pathways as a new figure.

• __ A brief discussion of therapeutic implications-such as combining colistin with agents that target membrane integrity-would help bridge the gap from mechanism to clinical management.__ In our discussion, we have suggested that collateral sensitivity (line 446-453) and PhoPQ kinase inhibitors (line 512-515) could be exploited to combat colistin resistance. To make this point more clearly, we will slightly expand our Discussion to include the therapeutic implications of our study.

• __ Additional discussion on whether the fitness costs are reversible or can be compensated by further adaptation would be valuable for long-term dynamics.__ We thank the reviewer for raising this interesting point. The evolution trajectory of P8 suggests that fitness costs can be compensated by later-occurring mutations during evolution. We will further discuss this point to highlight the importance of understanding the mutational dynamics of antibiotic resistance evolution.

• __ It would be valuable for the authors to comment on, or further analyze, whether there is a direct association between specific fitness costs and sensitivity to other antibiotics. Such information could inform on evolutionary constraints and possible trade-offs relevant to clinical settings.__

We will include a supplemental figure showing the correlation between fitness costs and antibiotic susceptibility for P2, P5, and P8.

__ Main figures and support for claims

The main and supplementary figures comprehensively illustrate the evolutionary trajectories, genetic bases, and phenotypic outcomes associated with colistin resistance under magnesium depletion in P. aeruginosa. The figures effectively detail: • Genetic pathways involved including the experimental evolution design (colistin selection under Mg²_⁺_ depletion), whole-genome sequencing results, and timelines of observed mutations (e.g., in htrB2, lpxO2, oprH/phoP/phoQ promoter, PA4824). • Phenotypes and biochemical analyses such as lipid A structure (via mass spectrometry), minimum inhibitory concentration (MIC) assays, and epistasis analyses between mutations are depicted. • Fitness trade-offs are demonstrated using bacterial survival, membrane integrity (e.g., scanning electron microscopy images), membrane permeability assays (NPN uptake), and competitive fitness assays. • Mechanistic claims about the necessity of early mutations, the requirement of the PhoPQ pathway at different evolutionary stages, and the fitness cost imposed by certain resistance mutations. To further enhance the rigor and clarity of the manuscript, the authors should implement the following improvements: • Labelling consistency: In some instances, figure legends could provide more granular detail about specific mutations (e.g., positions of amino acid changes). • Graphical summary: A schematic summary figure that visually integrates the three main evolutionary resistance trajectories, the mutational order, corresponding lipid A changes, and fitness costs, would enhance readability. • Replicates: Plots should more thoroughly indicate the number of replicates and show individual data points (not just means {plus minus} SD), add number of replicates in each experiment. • Supplementary: figures referenced in the text (e.g., lipid A structures or mutation reversion outcomes) should be made more prominent or better cross-referenced from the main results section. Authors should highlight when supplementary data provide critical functional confirmation (e.g., confirming mutation function or fitness reversal).__

We thank the reviewer for their appreciation of our work and constructive feedback.

__Statistics

The authors have appropriately incorporated statistical analyses throughout the figures. To enhance the robustness and credibility of their findings, authors should also cross-check • Tests in legends: Every figure and supplementary figure should clearly state the type of statistical test used, how many biological replicates, and any corrections for multiple comparisons.__

As mentioned above, we will provide more details about the statistical tests of each panel.

• __ Effect sizes: Where appropriate, reporting effect sizes-rather than just p values-would contextualize the biological impact.__ We agree with the reviewer; we will mention the magnitude of MIC changes in the corresponding figure legends.

• __ Raw data accessibility: For full transparency, consider sharing underlying raw data and analysis scripts.

__ We will provide the raw data of each panel.

__Overall, the main and supplementary figures effectively illustrate and substantiate the key claims-particularly the alternative molecular pathways, phenotypic trade-offs, and the role of environmental magnesium in mediating colistin resistance. Statistical analysis is generally robust and appropriately presented throughout, though improvements could include more explicit reporting, additional controls, and accessible raw data. The visual and quantitative data in the figures provide support for the authors' conclusions about the evolution of antibiotic resistance under nutrient limitation in microbial environments. Understanding these alternative pathways is important for designing better treatment strategies and for predicting how resistance might evolve under varying clinical and environmental conditions.

__

We thank the reviewer for their positive assessment.

__ Reviewer #1 (Significance (Required)):

Overall, this work offers important insights into antibiotic resistance in nutrient-stressed, polymicrobial environments, highlighting how magnesium availability shapes resistance evolution and fitness costs. The findings suggest new avenues for therapeutic intervention and call for a reevaluation of antibiotic strategies in nutrient-competitive infection settings.__

We sincerely thank the reviewer for constructive and thoughtful feedback and the acknowledgement of our figure presentation and experimental design. We feel very encouraged by the reviewer's perspective that our study provides unique insights into resistance evolution in polymicrobial environments and may inform therapeutic strategies.

__My expertise: Gut microbiome, gut microbiota resilience, ecology, and evolution in microbial communities, antimicrobial resistance, high-throughput drug-bacteria interactions

Reviewer #2 (Evidence, reproducibility and clarity (Required)):

Summary: The paper by Hsieh and colleagues unravels the molecular basis of colistin resistance in Pseudomonas aeruginosa under low magnesium (Mg2+) conditions. Colistin is a last resort antibiotic that compromises bacterial cell wall integrity. Bacteria can respond (phenotypically and genotypically) to colistin by modifying membrane-anchored lipopolysaccharides. Mg2+ depletion can trigger similar responses. In their study, Hsieh et al. find that Mg2+ depletion (induced by a co-infecting fungal pathogen, Candida albicans) leads to evolutionary trajectories and resistance mechanisms that differ from those observed under Mg-rich conditions. The authors conducted a series of detailed genetic, chemical and fitness-based experiments to elucidate the molecular, physiological and evolutionary basis of these new resistance mechanisms.__

We thank the reviewer for their summary of our study.__

Major comments: __ 1. The authors reconstituted key mutations observed during experimental evolution in the ancestral background. Moreover, they took clones from the final stage of the evolution experiment and restored the ancestral state of the mutated genes. This dual approach is extremely strong and allows to decipher the causal effects of colistin resistance. I like to applaud the authors for this rigorous approach.

We thank the reviewer's appreciation about the rigor and comprehensive analyses of our study.

2. I understand that this work focusses on evolved mutants isolated from a previous experiment. The focus is on Mg2+ limitation. However, it would still have been nice to include a characterised colistin resistent strain featuring more standard resistance mechanisms. How different would such a strain be in the analyses shown in Fig. 3? Would morphological changes (Fig. 5A), fitness trade-offs (Fig. 6) and collateral sensitivity (Fig. 7) also occur in such a mutant. I do not regard it as imperative to include data from such a strain. But putting the new data into context (at least in the discussion) would clearly increase the overall impact of this work.

We thank the reviewer for raising this fascinating and vital point. We will address the point in our Revision using the monoculture (high Mg2+) evolved strains, which acquired many known mutations for colistin resistance, as our reference. We will provide a supplemental figure about the membrane permeability, fitness costs, and collateral sensitivity of monoculture evolved strains. We will also contrast their difference from co-culture evolved strains in the revised Discussion.__

1. I recommend to discuss the findings in the context of the work conducted by Jochumsen et al. 2016 Nature Communications https://doi.org/10.1038/ncomms13002. To me, this is one of the most insightful papers on the genetic basis and epistasis of colistin resistance.__

We thank the reviewer for pointing out this important reference. We will include this reference and its findings in the Discussion.

__Minor comments:

1. First section of results and Fig. 1. It is unclear what parts are repetition from the ref. 37 and what is new. Please clarify.__

We thank the reviewer for this suggestion. Figures 1A and 1B summarize the previous paper; all other panels are new data. We will make this clear in the revised text and figure legend.

5. MIC-data (e.g. Fig. 2) come in discrete categories (based on the underlying dilution series). This comes with some challenges for statistical analysis. First, linear models like ANOVAs are based on normally distributed residuals. This is violated with discrete data distributions. Second, there is often no within-treatment variation (e.g., Fig. 2B), which makes statistical analyses obsolete. These points need to be addressed. Moreover, how is it possible to have subtle variations in MIC (e.g., Fig. 2A, P2 endpoint clone) with classic dilution series (as indicated on the y-axis, 128, 256, 512)? Please explain.

We agree with the reviewer that statistical analysis of MIC data is not straightforward. ANOVAs are not well-suited for this type of discrete data, and the lack of variation within replicates reduces the power of non-parametric tests such as the Mann-Whitney U test. To improve the statistical reporting of MIC data, we will apply non-parametric tests and include effect size measurements, as recommended by Reviewer 1.

Moreover, the design of dilution series may underestimate the true nature of antibiotic susceptibility. To address these issues, we have also performed survival assays to assess colistin resistance in both the endpoint and reversion strains; we will also include statistics to assess the significance of their different survival frequencies.

We thank the reviewer for highlighting the point about subtle variations in a classical dilution series. Our endpoint strains grew robustly in media containing 192 μg/mL colistin-the highest concentration used in our evolution experiment. To more accurately determine and compare their maximum MICs, we expanded the colistin concentration range using finer fold increases (1.5×, 2×, 2.5×, 3×, 3.5×, and 4×) from 192 to 768 μg/mL. We will update these details in the Materials & Methods.

__ Lines 264-269. This analysis focusses on enzyme impairment. However, mutations could also change enzyme activity. Could any of these mutations have such an effect?__

The answer is "yes". As evolved strains with lpxA mutation still have lipid A, we suspect this mutation does not altogether abolish lipid A synthesis. However, this mutation could affect the amount of lipid A or change enzyme specificity. These are interesting ideas for further investigation, but they fall beyond the scope of our current study. We will, however, include the requested detail in the discussion.

__ Figure 5A. Some arrows seem to be out of place and point at void spaces. Please check.__

We thank the reviewer for pointing out this error, which we will correct.

8. The use of polymyxin B is not well justified (Fig. 5 and Fig. S13). Did the authors aim to test whether there is cross-resistance to other antimicrobial peptides?

We will more clearly justify our choice of using polymyxin B for directly assaying binding of polymyxin antibiotics to bacterial cells using fluorescence-labeled polymyxins, since no such reagents exist for colistin and since previous studies (including ours) have shown similarity of susceptibility to colistin and polymyxin B:

"Although P2 and P5 endpoint clones have more permeable membranes, they exhibited greater resistance to polymyxin antibiotics, including colistin (polymyxin E) (Fig. 5D), and polymyxin B (Fig. S13A) than WT cells. To investigate how membrane-compromised cells gain increased resistance to antibiotics that target the outer membrane, we used dansyl-labeled polymyxin B [51] to quantify the binding of polymyxins to P. aeruginosa; dansyl-labeled polymyxin fluoresces upon binding the hydrophobic portion of bacterial membranes. We used polymyxin B binding as a surrogate for how bacterial cells bind to all polymyxin antibiotics, including colistin."

__ Line 564. Please indicate the dilution factor used.__

Thank you for pointing out this inadvertent omission. We will update our Materials & Methods accordingly, as in response to the Reviewer 2's comment 5.

__Reviewer #2 (Significance (Required)):

This is a very strong and well designed study. It provides novel and relevant insights into the resistance mechanisms against an important last resort antibiotic.__

We sincerely thank the reviewer for their thoughtful summary and generous evaluation of our work.

__Reviewer #3 (Evidence, reproducibility and clarity (Required)):

This manuscript reports on biologically interesting and clinically-relevant findings, that upon passaging in the presence of spent media from C. albicans, P. aeruginosa develops resistance to colistin through lipid A modifications. The authors thoroughly characterize novel lipid A structures seen in their resistant mutants, and test a variety of genetically constructed mutants to determine the contributions of specific mutant alleles to resistance.__

We thank the reviewer for the appreciation of our experimental design and comprehensive genetic and biochemical analyses of our evolved strains.

However, additional experiments are needed to demonstrate the specific role and necessity of the lipid modifications for colistin resistance.

We are also grateful for the reviewer's feedback and constructive criticisms to improve the clarity and impact of our manuscript. We have listed detailed responses to the reviewer below.

1. __ Evidence that the lipid A mutations are causal for colistin resistance is sparse:
2. Both the htrB2 mutations (in P2 and P5) are posited to be loss-of-function alleles. However, the phenotypes of the individual alleles are different (shown in Fig 2A and 2B). While the mutation in P2 shows a ~2x increase in resistance, the mutation in P5 does not. Thus it is not clear that the specific lipid A modifications seen in the htrB2 mutants are sufficient to confer colistin resistance. Can the authors test a clean deletion mutant of htrB2? Further, reversion of the htrB2 mutation in P2 has only a mild effect on colistin resistance, while reversion in P5 leads to a ~3-4x reduction in colistin resistance (Fig. S3), once again making it hard to parse out the exact effect of the lipid A modifications seen in the htrB2 mutants.
3. Similarly, a single lpxO2 mutation does not have any effect on colistin resistance (in P5), indicating that the modifications seen in this mutant are not sufficient to lead to resistance.__ We thank the reviewer for making this suggestion. The reviewer is correct that a clean deletion will directly assess the effects of htrB2 mutations. We will make htrB2 deletion in WT and the triple mutants and endpoint clones of P2 and P5 to check the effect of htrB2 deletion on colistin resistance.

Additionally, as Reviewer 2 pointed out, both mutation reconstruction and reversion experiments are required for understanding the roles of each mutation and interactions among different mutations in contributing to resistance. Combining all the results of htrB2 and lpxO2 mutations in these two orthogonal genetic experiments, it is the synergistic interactions among these mutations that lead to enhanced resistance after evolution. This explains why we saw genetic background effects of htrB2 mutation (P2 vs P5) and why each single mutation is required for resistance but doesn't contribute to resistance significantly by itself.

- In P8, the effect of a single lpxA mutation is not tested. Further, the resistance of a P-oprH + lpxA mutant is the same as that of just the P-oprH mutant, indicating that the lpxA mutation likely does not directly alter colistin resistance. It is possible that mutations in lpxA were selected to compensate for fitness defects resulting from the other mutations, or for adaptation to some other component of the media conditions.

This is an excellent suggestion. We will assess the MIC and fitness of reconstructed strains with the lpxA mutation to update the role of this mutation.

- While reversion of the htrB2 and lpxO2 mutations do lead to ~3-4x reduced resistance in P5 indicating some contribution of these mutations, it is specific to this population, and thus not clear whether it is due to the specific lipid A modifications (some of which are seen in the other populations too). A specific combination of lipid A modifications may confer colistin resistance, but this needs to be demonstrated by generating just those clean deletion mutants and showing an effect on resistance.

In response to this comment and comment 1, we will make lpxO2 deletions in WT, the triple mutant and the endpoint clone of P5 to test colistin resistance. However, our results of reverting single htrB2 or lpxO2 mutation to WT are robust and use two independent assays, including the standard MIC test and colistin survival assay. So, we are confident that each mutation is necessary for enhancing colistin resistance.

__ Overall, given the high levels of colistin resistance still exhibited by single mutant revertants (Fig. S3) and the absence of double or triple revertants, it is hard to come to any conclusions regarding causality. This is especially the case for P8 but also true of P2 and P5. What are the other mutations in these populations, and what role do they play in colistin resistance?__

We respectfully disagree with the reviewer on this point. One point that we have made and will re-emphasize in our Revision is that we have assayed all the mutations in these populations; this is one of the advantages of our experimental evolution and genome sequencing strategy. All the mutations that could play a role in colistin resistance have therefore been tested. Furthermore, due to genetic epistasis of mutations in different evolutionary lineages, we do not necessarily expect that a single revertant would altogether abolish colistin resistance, as has been demonstrated in several previous studies. As Reviewer 2 pointed out, combining mutation reconstruction and reversion is the best way to establish causality, and we have done so. Therefore, it is not correct to say that we cannot come to 'any conclusions regarding causality'.

__ Figure 4 is titled "The PhoPQ pathway synergizes with early-arising mutations to confer colistin resistance.", but instead what this figure shows is that the mutation upstream of oprH increases PhoP activity. I'm not sure what the synergy here is. The same is true for the section starting on line 276. Further, the first sentence of that section states "We next investigated why the mutations conferring robust colistin resistance in low Mg2+ conditions are not observed in Mg2+ replete conditions.". However, there are no experiments there testing whether the mutations conferred resistance in Mg2+ conditions, instead the authors just test whether the mutations they are studying increase PhoP activity, and require PhoPQ to confer resistance.__

We thank the reviewer for raising this point. We apologize for the unclear writing. We will use this opportunity to improve the clarity of this section by rewriting it to focus on two points: 1. Evolved resistance is PhoPQ-dependent, instead of PmrAB-dependent. 2. Two lineages evolved enhanced resistance by boosting PhoPQ activity in both high and low Mg2+ conditions. We will also remove the statement highlighted by the reviewer from this section that obfuscates the motivation of this section. We feel this approach will more clearly show how lipid A-related mutations contribute to resistance in low Mg2+.

__ The authors claim that the identified mutations did not appear in the high magnesium conditions because they had a fitness cost under those conditions, but figure 6A shows that the evolved strains have fitness costs in low magnesium conditions as well. Further, the authors suggest that because the studied mutations act via increased PhoPQ activity, they do not lead to resistance under high magnesium conditions (lines 376-379). However, the increased PhoPQ activity is mediated by the P-oprH mutation in the isolates which likely increases PhoPQ activity even in high magnesium conditions. Overall, it is not clear why the mutations in the low magnesium condition were not selected for under high magnesium conditions.__

The reviewer is correct about the fitness cost in high Mg2+ and low Mg2+ conditions. These fitness experiments were carried out in the absence of colistin, which explains the finding that there are fitness defects in both conditions. As is well known, evolution for antibiotic resistance will ultimately select for resistant mutants, despite their fitness costs. In contrast, colistin MIC of these endpoint strains in high Mg2+ conditions was still much lower than the colistin concentration we applied during evolution (Fig. S15), indicating it is much less likely for these mutations to be selected for in high Mg2+. We will clarify this point in our revised Results and Discussion.

We agree with the reviewer about the P-oprH mutations (PhoPQ expression) and will note that, unlike the other mutations, it is not clear why these emerge only in the low Mg2+ condition.

__ The authors used C. albicans spent BHI media as their low magnesium condition, but this condition has a lot of other C. albicans metabolites that may be affecting the results. It is possible that what the authors are observing is not related to magnesium at all, and the authors should test the phenotypes in normal BHI medium depleted for magnesium or some defined medium where magnesium levels can be controlled.__

We thank the reviewer for mentioning this important point. In our prior PLOS Biology paper (https://doi.org/10.1371/journal.pbio.3002694.g005), we demonstrated that supplementing Mg2+ in evolved co-culture populations reduces colistin resistance, suggesting this evolved resistance is Mg2+ dependent. We also know that the MIC of our endpoint strains in C. albicans-spent BHI with supplemented Mg2+ (MIC of all three endpoint clones is less than 48 mg/mL colistin) is much lower than in C. albicans-spent BHI. We will mention this detail in the paper and include the data in our revision if the reviewer and editor require it.

Other comments: - The authors use MIC assays as well as % survival to measure resistance against colistin, and sometimes use both in the same figure (e.g. Figure 2). This makes direct comparisons difficult. It would be better to consistently use one assay, preferably the MIC, at least in all the main figures. If the survival data needs to be included, it could go in the supplementary figures.

We thank the reviewer for this suggestion. We will move the MIC data of mutation-reversion strains to the main Fig. 2D-F.

- While the mutations seen in the low and high magnesium conditions were shown in the previous manuscript, given the extensive dissection here, it would be useful for readers if the authors gave some details about the serial passaging and evolution experiment, identification of mutations, and some mention of what mutations were seen in high Mg populations.

We will add these details in the introduction.

- Given that oprH is present in an operon, it would be more accurate to call that mutation as being in the promoter of the oprH-phoP-phoQ operon rather than it being an oprH mutation (at least in the text, e.g. lines 127-129).

We agree. We will change this as the reviewer requested.

- Unlike what is stated on lines 287-290, deletion of oprH in P2 leads to a greater than 2x reduction in colistin MIC, suggesting that OprH is playing a role (albeit a smaller role than phoP) - Line 50 has a typo, remove "160". - Line 122: Specify which Pa and Ca strain backgrounds were used. - Line 132: Were representative isolates derived from terminal passages? This should be defined.

We will change these points according to the reviewer's suggestions; we thank them for these suggestions.

- Line 215-219: It is interesting that Pa WT grown in spent medium additionally results in lipid A that is hexa-acylated. Is this sufficient to alter colistin resistance on its own?

We find that WT PAO1 in low Mg2+ conditions has PagP-mediated acylation, which can slightly increase colistin resistance, but not to the extent of resistance as our evolved strains.

- It would be useful to see a PCA plot for the samples shown in figures S6 and S7.

We will include such a plot in Figures S6 and S7

- Fig. S11: What are the colistin MICs of pmrA and phoP deletions in the WT background?

MIC of pmrA and phoP deletions in WT is 1.5ug/mL. We will include these data in the Revision.

- Instead of qualitative data, can the authors quantify cell length and perhaps some measure of cell shape (instead of just showing images in Fig. 5A and S12).

We thank the reviewer for raising this point. A similar comment was raised by Reviewer 1. As it's challenging to quantify membrane changes from the morphological data obtained through SEM (a point which we will now clarify in our Revision), we used a quantifiable NpN uptake assay to quantify membrane defects of our evolved strains.

- What is the WT MIC in high magnesium conditions? Please show that in Fig. S15.

We will include this detail in Fig. S15

- I am not an expert in lipid modifications and structures, but in figure S5, P2 and P4 show high peaks with lower m/z that seem specific to low magnesium conditions, but they are not labeled or discussed. What are these peaks?

We thank the reviewer for bringing up this concern. The unlabeled lipids in these spectra are cardiolipin, not lipid A. These peaks are present in all the samples, and the reason they appear larger in the P1 and P4 low magnesium conditions is that both spectra are scaled to the relative intensity of one another. It is important to note that MALDI-TOF MS is not a quantitative technique, and the relative intensity of the peak heights between two samples should not be used to compare the amounts of lipids in one sample versus another. Therefore, we cannot say that these lipids are present in greater quantities in low magnesium conditions versus high magnesium conditions.

- Lines 357-358 state that "mutant cells minimally bind polymyxin B (Fig. S13B)", but the figure shows increased binding compared to the WT. The legend of the figure also says something similar. Are the phoP pmrA mutants expected to bind more polymyxin B because they can't modify lipid A?

We thank the reviewer for pointing out this substantial error. We will change 'minimally bind' to 'demonstrate increased binding'.

- Given the fitness defects in just regular medium, is the data shown in Figure 7 specific collateral sensitivity to the antibiotics tested? Are there other conditions where P2 and P5 do not show increased sensitivity?

These are all the antibiotics we have tested. It is conceivable that P2 and P5 might not show increased sensitivity to other antibiotics that use the same mode of action as colistin or polymyxin B.

__Reviewer #3 (Significance (Required)):

This study aims to dissect novel mechanisms of colistin resistance in P. aeruginosa that arise upon passaging in C. albicans spent media. While the authors identify novel lipid A modifications associated with the evolved strains, the significance of the modifications for resistance, and the mechanisms for why these evolutionary trajectories were not selected for in high magnesium are not clear from the data presented.__

We thank the reviewer for recognizing the integrity of our work and for the constructive feedback on improving the clarity of our writing. We understand that some concerns may stem from a lack of clarity in our original submission, but that additional genetic experiments are necessary. We have already identified all mutations that arose independently across different lineages and characterized their contributions to resistance, which we believe supports a robust inference of causality. To strengthen our conclusions, we will incorporate additional experiments, including htrB2 deletion, lpxO2 deletion, and lpxA mutation, to better dissect the roles of these genes and mutations in colistin resistance. We hope this revision plan will ameliorate the reviewer's concerns.

Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

Referee #3

Evidence, reproducibility and clarity

This manuscript reports on biologically interesting and clinically-relevant findings, that upon passaging in the presence of spent media from C. albicans, P. aeruginosa develops resistance to colistin through lipid A modifications. The authors thoroughly characterize novel lipid A structures seen in their resistant mutants, and test a variety of genetically constructed mutants to determine the contributions of specific mutant alleles to resistance. However, additional experiments are needed to demonstrate the specific role and necessity of the lipid modifications for colistin resistance.

1. Evidence that the lipid A mutations are causal for colistin resistance is sparse:
   • Both the htrB2 mutations (in P2 and P5) are posited to be loss-of-function alleles. However, the phenotypes of the individual alleles are different (shown in Fig 2A and 2B). While the mutation in P2 shows a ~2x increase in resistance, the mutation in P5 does not. Thus it is not clear that the specific lipid A modifications seen in the htrB2 mutants are sufficient to confer colistin resistance. Can the authors test a clean deletion mutant of htrB2? Further, reversion of the htrB2 mutation in P2 has only a mild effect on colistin resistance, while reversion in P5 leads to a ~3-4x reduction in colistin resistance (Fig. S3), once again making it hard to parse out the exact effect of the lipid A modifications seen in the htrB2 mutants.
   • Similarly, a single lpxO2 mutation does not have any effect on colistin resistance (in P5), indicating that the modifications seen in this mutant are not sufficient to lead to resistance.
   • In P8, the effect of a single lpxA mutation is not tested. Further, the resistance of a P-oprH + lpxA mutant is the same as that of just the P-oprH mutant, indicating that the lpxA mutation likely does not directly alter colistin resistance. It is possible that mutations in lpxA were selected to compensate for fitness defects resulting from the other mutations, or for adaptation to some other component of the media conditions.
   • While reversion of the htrB2 and lpxO2 mutations do lead to ~3-4x reduced resistance in P5 indicating some contribution of these mutations, it is specific to this population, and thus not clear whether it is due to the specific lipid A modifications (some of which are seen in the other populations too). It is possible that a specific combination of lipid A modifications confers colistin resistance, but this needs to be demonstrated by generating just those clean deletion mutants and showing an effect on resistance.
2. Overall, given the high levels of colistin resistance still exhibited by single mutant revertants (Fig. S3) and the absence of double or triple revertants, it is hard to come to any conclusions regarding causality. This is especially the case for P8 but also true of P2 and P5. What are the other mutations in these populations, and what role do they play in colistin resistance?
3. Figure 4 is titled "The PhoPQ pathway synergizes with early-arising mutations to confer colistin resistance.", but instead what this figure shows is that the mutation upstream of oprH increases PhoP activity. I'm not sure what the synergy here is. The same is true for the section starting on line 276. Further, the first sentence of that section states "We next investigated why the mutations conferring robust colistin resistance in low Mg2+ conditions are not observed in Mg2+ replete conditions.". However, there are no experiments there testing whether the mutations conferred resistance in Mg2+ conditions, instead the authors just test whether the mutations they are studying increase PhoP activity, and require PhoPQ to confer resistance.
4. The authors claim that the identified mutations did not appear in the high magnesium conditions because they had a fitness cost under those conditions, but figure 6A shows that the evolved strains have fitness costs in low magnesium conditions as well. Further, the authors suggest that because the studied mutations act via increased PhoPQ activity, they do not lead to resistance under high magnesium conditions (lines 376-379). However, the increased PhoPQ activity is mediated by the P-oprH mutation in the isolates which likely increases PhoPQ activity even in high magnesium conditions. Overall, it is not clear why the mutations in the low magnesium condition were not selected for under high magnesium conditions.
5. The authors used C. albicans spent BHI media as their low magnesium condition, but this condition has a lot of other C. albicans metabolites that may be affecting the results. It is possible that what the authors are observing is not related to magnesium at all, and the authors should test the phenotypes in normal BHI medium depleted for magnesium or some defined medium where magnesium levels can be controlled.

Other comments:

• The authors use MIC assays as well as % survival to measure resistance against colistin, and sometimes use both in the same figure (e.g. Figure 2). This makes direct comparisons difficult. It would be better to consistently use one assay, preferably the MIC, at least in all the main figures. If the survival data needs to be included, it could go in the supplementary figures.
• While the mutations seen in the low and high magnesium conditions were shown in the previous manuscript, given the extensive dissection here, it would be useful for readers if the authors gave some details about the serial passaging and evolution experiment, identification of mutations, and some mention of what mutations were seen in high Mg populations.
• Given that oprH is present in an operon, it would be more accurate to call that mutation as being in the promoter of the oprH-phoP-phoQ operon rather than it being an oprH mutation (at least in the text, e.g. lines 127-129).
• Unlike what is stated on lines 287-290, deletion of oprH in P2 leads to a greater than 2x reduction in colistin MIC, suggesting that OprH is playing a role (albeit a smaller role than phoP)
• Line 50 has a typo, remove "160".
• Line 122: Specify which Pa and Ca strain backgrounds were used.
• Line 132: Were representative isolates derived from terminal passages? This should be defined.
• Line 215-219: It is interesting that Pa WT grown in spent medium additionally results in lipid A that is hexa-acylated. Is this sufficient to alter colistin resistance on its own?
• It would be useful to see a PCA plot for the samples shown in figures S6 and S7.
• Fig. S11: What are the colistin MICs of pmrA and phoP deletions in the WT background?
• Instead of qualitative data, can the authors quantify cell length and perhaps some measure of cell shape (instead of just showing images in Fig. 5A and S12).
• What is the WT MIC in high magnesium conditions? Please show that in Fig. S15.
• I am not an expert in lipid modifications and structures, but in figure S5, P2 and P4 show high peaks with lower m/z that seem specific to low magnesium conditions, but they are not labeled or discussed. What are these peaks?
• Lines 357-358 state that "mutant cells minimally bind polymyxin B (Fig. S13B)", but the figure shows increased binding compared to the WT. The legend of the figure also says something similar. Are the phoP pmrA mutants expected to bind more polymyxin B because they can't modify lipid A?
• Given the fitness defects in just regular medium, is the data shown in Figure 7 specific collateral sensitivity to the antibiotics tested? Are there other conditions where P2 and P5 do not show increased sensitivity?

Significance

This study aims to dissect novel mechanisms of colistin resistance in P. aeruginosa that arise upon passaging in C. albicans spent media. While the authors identify novel lipid A modifications associated with the evolved strains, the significance of the modifications for resistance, and the mechanisms for why these evolutionary trajectories were not selected for in high magnesium are not clear from the data presented.

Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

Referee #2

Evidence, reproducibility and clarity

Summary: The paper by Hsieh and colleagues unravels the molecular basis of colistin resistance in Pseudomonas aeruginosa under low magnesium (Mg2+) conditions. Colistin is a last resort antibiotic that compromises bacterial cell wall integrity. Bacteria can respond (phenotypically and genotypically) to colistin by modifying membrane-anchored lipopolysaccharides. Mg2+ depletion can trigger similar responses. In their study, Hsieh et al. find that Mg2+ depletion (induced by a co-infecting fungal pathogen, Candida albicans) leads to evolutionary trajectories and resistance mechanisms that differ from those observed under Mg-rich conditions. The authors conducted a series of detailed genetic, chemical and fitness-based experiments to elucidate the molecular, physiological and evolutionary basis of these new resistance mechanisms.

Major comments:

1. The authors reconstituted key mutations observed during experimental evolution in the ancestral background. Moreover, they took clones from the final stage of the evolution experiment and restored the ancestral state of the mutated genes. This dual approach is extremely strong and allows to decipher the causal effects of colistin resistance. I like to applaud the authors for this rigorous approach.
2. I understand that this work focusses on evolved mutants isolated from a previous experiment. The focus is on Mg2+ limitation. However, it would still have been nice to include a characterised colistin resistent strain featuring more standard resistance mechanisms. How different would such a strain be in the analyses shown in Fig. 3? Would morphological changes (Fig. 5A), fitness trade-offs (Fig. 6) and collateral sensitivity (Fig. 7) also occur in such a mutant. I do not regard it as imperative to include data from such a strain. But putting the new data into context (at least in the discussion) would clearly increase the overall impact of this work.
3. I recommend to discuss the findings in the context of the work conducted by Jochumsen et al. 2016 Nature Communications https://doi.org/10.1038/ncomms13002. To me, this is one of the most insightful papers on the genetic basis and epistasis of colistin resistance.

Minor comments:

1. First section of results and Fig. 1. It is unclear what parts are repetition from the ref. 37 and what is new. Please clarify.
2. MIC-data (e.g. Fig. 2) come in discrete categories (based on the underlying dilution series). This comes with some challenges for statistical analysis. First, linear models like ANOVAs are based on normally distributed residuals. This is violated with discrete data distributions. Second, there is often no within-treatment variation (e.g., Fig. 2B), which makes statistical analyses obsolete. These points need to be addressed. Moreover, how is it possible to have subtle variations in MIC (e.g., Fig. 2A, P2 endpoint clone) with classic dilution series (as indicated on the y-axis, 128, 256, 512)? Please explain.
3. Lines 264-269. This analysis focusses on enzyme impairment. However, mutations could also change enzyme activity. Could any of these mutations have such an effect?
4. Figure 5A. Some arrows seem to be out of place and point at void spaces. Please check.
5. The use of polymyxin B is not well justified (Fig. 5 and Fig. S13). Did the authors aim to test whether there is cross-resistance to other antimicrobial peptides?
6. Line 564. Please indicate the dilution factor used.

Significance

This is a very strong and well designed study. It provides novel and relevant insights into the resistance mechanisms against an important last resort antibiotic.

Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

Referee #1

Evidence, reproducibility and clarity

In this paper "Magnesium depletion unleashes two unusual modes of colistin resistance with different fitness costs," the authors examine how Pseudomonas aeruginosa evolves resistance to colistin, a last-resort antibiotic for multidrug-resistant Gram-negative infections. Although colistin resistance is a major clinical challenge, its underlying mechanisms, particularly under nutrient-limited conditions typical of infections, are not fully understood.

The study shows that under low magnesium (Mg²⁺) conditions-mimicking infection or biofilm stress-P. aeruginosa can develop colistin resistance via two distinct genetic pathways, each with unique fitness costs. The first involves mutations in genes such as htrB2 and lpxO2, granting strong resistance but compromising the outer membrane and increasing susceptibility to other antibiotics. The second involves regulatory mutations (e.g., in the oprH/phoP/phoQ promoter) that confer resistance with minimal membrane defects and generally lower fitness costs. These resistance strategies lead to different trade-offs: membrane-compromising mutations reduce bacterial fitness without colistin, while regulatory mutations typically avoid these penalties, with context-dependent effects. The study underscores clinical relevance, noting that in infections-such as in cystic fibrosis-other microbes like Candida albicans may deplete magnesium, indirectly promoting resistance evolution. Overall, this work offers important insights into antibiotic resistance in nutrient-stressed, polymicrobial environments, highlighting how magnesium availability shapes resistance evolution and fitness costs. The findings suggest new avenues for therapeutic intervention and call for a reevaluation of antibiotic strategies in nutrient-competitive infection settings.

Work is timely and important. Colistin resistance represents an urgent threat as colistin is a last-resort antibiotic used against multidrug-resistant Gram-negative pathogens. Insights into mechanisms evolving under nutrient limitation are highly relevant given the prevalence of such environmental conditions during infection and microbial biofilm growth. The study reveals two previously uncharacterized pathways to colistin resistance in P. aeruginosa triggered by magnesium (Mg²⁺) depletion, each with distinct genetic signatures and trade-offs. This finding directly impacts the understanding of polymicrobial infection dynamics, especially where magnesium sequestration by fungi/ or other microbes may occur. The identification of fitness costs and pleiotropic effects associated with specific resistance mutations provides crucial guidance for clinicians considering antibiotic stewardship and combination therapy strategies.

Strengths

• Experimental evolution: This work uses laboratory evolution under controlled Mg²⁺-limited conditions to simulate selection pressures relevant to infection microenvironments.
• Genetics: Systematic identification and functional validation of key mutations-particularly in htrB2, lpxO2, and the oprH/phoP/phoQ promoter-give mechanistic depth to the findings.
• Two distinct resistance modes: Evidence for (i) one pathway leading to colistin resistance via htrB2 mutations, resulting in high resistance but significant membrane integrity loss and increased susceptibility to other antibiotics. (ii) a second pathway providing resistance without compromising membrane integrity, highlighting evolutionary flexibility and ecological implications.
• Fitness assessments: measurement of the costs associated with each resistance strategy, both in terms of membrane integrity and susceptibility to other agents.
• Relevance: Connection to natural scenarios, such as magnesium sequestration by fungi (e.g. Candida albicans) in polymicrobial environments, underscores the ecological and clinical significance.
• This manuscript is well written with clearly logical hypothesis testing

Drawbacks

• Experimental scope: While the study is comprehensive for P. aeruginosa, the broader applicability to other Gram-negative pathogens is not directly tested.
• Mechanistic depth: Some inferred mechanisms (e.g., the precise molecular impact of late-occurring adaptive mutations) merit deeper biochemical analysis.
• Results Lines 414- 423: While correlation is most what makes sense for some drugs, causality is implied (membrane defects increase susceptibility), but could be strengthened by directly measuring antibiotic uptake (e.g., fluorescence) or membrane permeability for these 3 antibiotics.
  • Effect is mild and mostly not significant. It is also not clear whether authors only tested a handful of mutants shown in Fig. 7B-D or whether other clones were also tested. The sample of endpoints (P2, P5, P8) covers well-characterized lineages, but additional evolved clones or a broader panel could boost generality about other antibiotics. The authors note "significantly lower MICs" statistical treatment is implied; explicit statistical values and replicate numbers should be given in the text or figures.
• The structural or physiological nature of "mild" vs. "severe" membrane defects could be better defined/quantified.
• Quantitative limits: Authors should add in the discussion that statistical robustness could be strengthened-for example, by including longer-term evolutionary predictions.
• in vivo relevance: While the ecological context is discussed, direct in vivo confirmation (e.g., in animal infection models) of the observed resistance trajectories would increase translational impact and relevance.
• Some sections are repetitive or overly detailed; condense where possible (especially on mutation lists and background for each claim).

Other comments

• Authors should provide clarification on how the Mg²⁺ concentrations used in vitro compare to those found in clinically relevant infection settings. This would be helpfu to enhance significance.
• Authors should explicitly report statistical methods (e.g., types of tests, adjustments for multiple comparisons) in figure legends for reproducibility.
• Nomenclature for key mutations and their position within the genetic context (e.g., htrB2 mutation specifics) could be more detailed in figures or supplemental materials.
• The manuscript could benefit from a graphical summary illustrating the two distinct evolutionary pathways and their respective fitness landscapes.
• A brief discussion of therapeutic implications-such as combining colistin with agents that target membrane integrity-would help bridge the gap from mechanism to clinical management.
• Additional discussion on whether the fitness costs are reversible or can be compensated by further adaptation would be valuable for long-term dynamics.
• It would be valuable for the authors to comment on, or further analyze, whether there is a direct association between specific fitness costs and sensitivity to other antibiotics. Such information could inform on evolutionary constraints and possible trade-offs relevant to clinical settings.

Main figures and support for claims

The main and supplementary figures comprehensively illustrate the evolutionary trajectories, genetic bases, and phenotypic outcomes associated with colistin resistance under magnesium depletion in P. aeruginosa. The figures effectively detail: - Genetic pathways involved including the experimental evolution design (colistin selection under Mg²⁺ depletion), whole-genome sequencing results, and timelines of observed mutations (e.g., in htrB2, lpxO2, oprH/phoP/phoQ promoter, PA4824). - Phenotypes and biochemical analyses such as lipid A structure (via mass spectrometry), minimum inhibitory concentration (MIC) assays, and epistasis analyses between mutations are depicted. - Fitness trade-offs are demonstrated using bacterial survival, membrane integrity (e.g., scanning electron microscopy images), membrane permeability assays (NPN uptake), and competitive fitness assays. - Mechanistic claims about the necessity of early mutations, the requirement of the PhoPQ pathway at different evolutionary stages, and the fitness cost imposed by certain resistance mutations. To further enhance the rigor and clarity of the manuscript, the authors should implement the following improvements: - Labelling consistency: In some instances, figure legends could provide more granular detail about specific mutations (e.g., positions of amino acid changes). - Graphical summary: A schematic summary figure that visually integrates the three main evolutionary resistance trajectories, the mutational order, corresponding lipid A changes, and fitness costs, would enhance readability. - Replicates: Plots should more thoroughly indicate the number of replicates and show individual data points (not just means {plus minus} SD), add number of replicates in each experiment. - Supplementary: figures referenced in the text (e.g., lipid A structures or mutation reversion outcomes) should be made more prominent or better cross-referenced from the main results section. Authors should highlight when supplementary data provide critical functional confirmation (e.g., confirming mutation function or fitness reversal).

Statistics

The authors have appropriately incorporated statistical analyses throughout the figures. To enhance the robustness and credibility of their findings, authors should also cross-check - Tests in legends: Every figure and supplementary figure should clearly state the type of statistical test used, how many biological replicates, and any corrections for multiple comparisons. - Effect sizes: Where appropriate, reporting effect sizes-rather than just p values-would contextualize the biological impact. - Raw data accessibility: For full transparency, consider sharing underlying raw data and analysis scripts.

Overall, the main and supplementary figures effectively illustrate and substantiate the key claims-particularly the alternative molecular pathways, phenotypic trade-offs, and the role of environmental magnesium in mediating colistin resistance. Statistical analysis is generally robust and appropriately presented throughout, though improvements could include more explicit reporting, additional controls, and accessible raw data. The visual and quantitative data in the figures provide support for the authors' conclusions about the evolution of antibiotic resistance under nutrient limitation in microbial environments. Understanding these alternative pathways is important for designing better treatment strategies and for predicting how resistance might evolve under varying clinical and environmental conditions.

Significance

Overall, this work offers important insights into antibiotic resistance in nutrient-stressed, polymicrobial environments, highlighting how magnesium availability shapes resistance evolution and fitness costs. The findings suggest new avenues for therapeutic intervention and call for a reevaluation of antibiotic strategies in nutrient-competitive infection settings.

My expertise:

Gut microbiome, gut microbiota resilience, ecology, and evolution in microbial communities, antimicrobial resistance, high-throughput drug-bacteria interactions

This is very important because often times we expect very different things from what the college expects from us, so being able to find the right blend between what we expect and what the college expects is very helpful in our success as students.

Even if they did see it they had their own worries and duties to do.

This is important because this is another reason why it was weird that no one cared about the Boy falling from the sky.

This is what the Icarus is telling us, that life must go on, and people have things to do, places to go. That while some big event happened, it might not affect most people, and therefore, they ignore.

Nobody really cares about how a boy falls out of the sky.

Who are these "old masters"?

this is the test

As explained below, System.out.println is the subroutine. This means a command is a subroutine with an inputted data (in this case "Hello World!").

What the poet said here is sarcasm, because uncleanliness is portrayed as the opposite of beautiful, and he says that a girl is beautiful even though she had made stains in the book.

How he knew it was a beautiful girl stood out to me cus all she wrote was "pardon the egg salad stains"

What? Maybe he meant that the ride into the future on a vessel more lasting then themselves being the present day??

to me this is saying that when we read poetry we write in the margins to almost prove to ourselves that we are really thinking deep and trying to better understand the poem we read rather than appearing to have just skimmed the poem giving it no thought

The notes are just simply writing in the margins of the book.

The students would be more subtle about how they would write their "marks on the book".

The slave trade started in the 1400s and ended in the 1800s. Why didn’t African people take effective preventive measures to avoid being taken by those human traffickers?

“Those people were barbaric. Why did they take even the children?”

According to history, by 1839 the slave trade should have already been eradicated. Why were people still being shipped as slaves?”

Alan!

All our lives, we are taught that the economy was this big complicated system that is hard for just anyone to understand. We are taught to trust the economists because they know best and truly understand the economy better than anyone. While this may be true, Stanford poses real world questions that make the economy easier to understand. This way we can make an economic profile of my community and therefore understand how the economy works around us.

In this paragraph, the Stanford critiques the field of economics by pointing out how deeply embedded it is in systems of power and wealth. Stanford argues that many economists, both in academia and in the professional world, are influenced—consciously or subconsciously—by the interests of the institutions and systems that fund and support them. These institutions, such as banks, corporations, schools and governments, benefit from maintaining the current economic system. Because of this, economists tend to view inequality, competition, and wealth accumulation as natural and inevitable or even good for society.

yes

Testing.