Reviewer #2 (Public Review):

Summary:

The goal of the authors in this study is to develop a more reliable approach for quantifying codon usage such that it is more comparable across species. Specifically, the authors wish to estimate the degree of adaptive codon usage, which is potentially a general proxy for the strength of selection at the molecular level. To this end, the authors created the Codon Adaptation Index for Species (CAIS) that controls for differences in amino acid usage and GC% across species. Using their new metric, the authors find a previously unobserved negative correlation between the overall adaptiveness of codon usage and body size across 118 vertebrates. As body size is negatively correlated with effective population size and thus the general strength of natural selection, the negative correlation between CAIS and body size is expected. The authors argue this was previously unobserved due to failures of other popular metrics such as Codon Adaptation Index (CAI) and the Effective Number of Codons (ENC) to adequately control for differences in amino acid usage and GC content across species. Most surprisingly, the authors also find a positive relationship between CAIS and the overall "disorderedness" of a species protein domains. As some of these results are unexpected, which is acknowledged by the authors, I think it would be particularly beneficial to work with some simulated datasets. I think CAIS has the potential to be a valuable tool for those interested in comparing codon adaptation across species in certain situations. However, I have certain theoretical concerns about CAIS as a direct proxy for the efficiency of selection when the mutation bias changes across species.

Strengths:

(1) I appreciate that the authors recognize the potential issues of comparing CAI when amino acid usage varies and correct for this in CAIS. I think this is sometimes an under-appreciated point in the codon usage literature, as CAI is a relative measure of codon usage bias (i.e. only considers synonyms). However, the strength of natural selection on codon usage can potentially vary across amino acids, such that comparing mean CAI between protein regions with different amino acid biases may result in spurious signals of statistical significance (see Cope et al. Biochemica et Biophysica Acta - Biomembranes 2018 for a clear example of this).

(2) The authors present numerous analysis using both ENC and mean CAI as a comparison to CAIS, helping given a sense of how CAIS corrects for some of the issues with these other metrics. I also enjoyed that they examined the previously unobserved relationship between codon usage bias and body size, which has bugged me ever since I saw Kessler and Dean 2014. The result comparing protein disorder to CAIS was particularly interesting and unexpected.

(3) The CAIS metric presented here is generally applicable to any species that has an annotated genome with protein-coding sequences.

Weaknesses:

(1) The main weakness of this work is that it lacks simulated data to confirm that it works as expected. This would be particularly useful for assessing the relationship between CAIS and the overall effect of protein structure disorder, which the authors acknowledge is an unexpected result. I think simulations could also allow the authors to assess how their metric performs in situations where mutation bias and natural selection act in the same direction vs. opposite directions. Additionally, although I appreciate their comparisons to ENC and mean CAI, the lack of comparison to other popular codon metrics for calculating the overall adaptiveness of a genome (e.g. dos Reis et al.'s statistic, which is a function of tRNA Adaptation Index (tAI) and ENC) may be more appropriate. Even if results are similar to , CAIS has a noted advantage that it doesn't require identifying tRNA gene copy numbers or abundances, which I think are generally less readily available than genomic GC% and protein-coding sequences.

The authors mention the selection-mutation-drift equilibrium model, which underlies the basic ideas of this work (e.g. higher results in stronger selection on codon usage), but a more in-depth framing of CAIS in terms of this model is not given. I think this could be valuable, particularly in addressing the question "are we really estimating what we think we're estimating?"

Let's take a closer look at the formulation for RSCUS. From here on out, subscripts will only be used to denote the codon and it will be assumed that we are only considering the case of for some species

I think what the authors are attempting to do is "divide out" the effects of mutation bias (as given by , such that only the effects of natural selection remain, i.e. deviations from the expected frequency based on mutation bias alone represent adaptive codon usage. Consider Gilchrist et al. MBE 2015, which says that the expected frequency of codon at selection-mutation-drift equilibrium in gene for an amino acid with synonymous codons is

where is the mutation bias, is the strength of selection scaled by the strength of drift, and is the gene expression level of gene \(g\). In this case, \ and reflect the strength and direction of mutation bias and natural selection relative to a reference codon, for which . Assuming the selection-mutation-drift equilibrium model is generally adequate to model the true codon usage patterns in a genome (as I do and I think the authors do, too), the could be considered the expected observed frequency codon in gene .

Let's re-write the in the form of Gilchrist et al., such that it is a function of mutation bias . For simplicity, we will consider just the two-codon case and assume the amino acid sequence is fixed. Assuming GC% is at equilibrium, the term and can be written as

where is the mutation rate from nucleotides to. As described in Gilchrist et al. MBE 2015 and Shah and Gilchrist PNAS 2011, the mutation bias . This can be expressed in terms of the equilibrium GC content by recognizing that

As we are assuming the amino acid sequence is fixed, the probability of observing a synonymous codon at an amino acid becomes just a Bernoulli process.

If we do this, then

Recall that in the Gilchrist et al. framework, the reference codon has . Thus, we have recovered the Gilchrist et al. model from the formulation of under the assumption that natural selection has no impact on codon usage and codon NNG is the pre-defined reference codon. To see this, plug in 0 for in equation (1).

We can then calculate the expected RSCUS using equation (1) (using notation and equation (6) for the two codon case. For simplicity assume, we are only considering a gene of average expression (defined as . Assume in this case that NNG is the reference codon .

This shows that the expected value of RSCUS for a two-codon amino acid is expected to increase as the strength of selection increases, which is desired. Note that in Gilchrist et al. is formulated in terms of selection against a codon relative to the reference, such that a negative value represents that a codon is favored relative to the reference. If (i.e. selection does not favor either codon), then . Also note that the expected RSCUS does not remain independent of the mutation bias. This means that even if (i.e. the strength of natural selection) does not change between species, changes to the strength and direction of mutation bias across species could impact RSCUS. Assuming my math is right, I think one needs to be cautious when interpreting CAIS as representative of the differences in the efficiency of selection across species except under very particular circumstances. One such case could be when it is known that mutation bias varies little across the species of interest. Looking at the species used in this manuscript, most of them have a GC content ranging around 0.41, so I suspect their results are okay.

Although I have not done so, I am sure this could be extended to the 4 and 6 codon amino acids.

Another minor weakness of this work is that although the method is generally applicable to any species with an annotated genome and the code is publicly available, the code itself contains hard-coded values for GC% and amino acid frequencies across the 118 vertebrates. The lack of a more flexible tool may make it difficult for less computationally-experienced researchers to take advantage of this method.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

The image analysis pipeline is tested in analysing microscopy imaging data of gastruloids of varying sizes, for which an optimised protocol for in toto image acquisition is established based on whole mount sample preparation using an optimal refractive index matched mounting media, opposing dual side imaging with two-photon microscopy for enhanced laser penetration, dual view registration, and weighted fusion for improved in toto sample data representation. For enhanced imaging speed in a two-photon microscope, parallel imaging was used, and the authors performed spectral unmixing analysis to avoid issues of signal cross-talk.

In the image analysis pipeline, different pre-treatments are done depending on the analysis to be performed (for nuclear segmentation - contrast enhancement and normalisation; for quantitative analysis of gene expression - corrections for optical artifacts inducing signal intensity variations). Stardist3D was used for the nuclear segmentation. The study analyses into properties of gastruloid nuclear density, patterns of cell division, morphology, deformation, and gene expression.

Strengths:

The methods developed are sound, well described, and well-validated, using a sample challenging for microscopy, gastruloids. Many of the established methods are very useful (e.g. registration, corrections, signal normalisation, lazy loading bioimage visualisation, spectral decomposition analysis), facilitate the development of quantitative research, and would be of interest to the wider scientific community.

We thank the reviewer for this positive feedback.

Weaknesses:

A recommendation should be added on when or under which conditions to use this pipeline.

We thank the reviewer for this valuable feedback, which will be addressed in the revision. In general, the pipeline is applicable to any tissue, but it is particularly useful for large and dense 3D samples—such as organoids, embryos, explants, spheroids, or tumors—that are typically composed of multiple cell layers and have a thickness greater than 50 µm.

The processing and analysis pipeline are compatible with any type of 3D imaging data (e.g. confocal, 2 photon, light-sheet, live or fixed).

- Spectral unmixing to remove signal cross-talk of multiple fluorescent targets is typically more relevant in two-photon imaging due to the broader excitation spectra of fluorophores compared to single-photon imaging. In confocal or light-sheet microscopy, alternating excitation wavelengths often circumvents the need for unmixing. Spectral decomposition performs even better with true spectral detectors; however, these are usually not non-descanned detectors, which are more appropriate for deep tissue imaging. Our approach demonstrates that simultaneous cross-talk-free four-color two-photon imaging can be achieved in dense 3D specimen with four non-descanned detectors and co-excitation by just two laser lines. Depending on the dispersion in optically dense samples, depth-dependent apparent emission spectra need to be considered.

- Nuclei segmentation using our trained StarDist3D model is applicable to any system under two conditions: (1) the nuclei exhibit a star-convex shape, as required by the StarDist architecture, and (2) the image resolution is sufficient in XYZ to allow resampling. The exact sampling required is object- and system-dependent, but the goal is to achieve nearly isotropic objects with diameters of approximately 15 pixels while maintaining image quality. In practice, images containing objects that are natively close to or larger than 15 pixels in diameter should segment well after resampling. Conversely, images with objects that are significantly smaller along one or more dimensions will require careful inspection of the segmentation results.

- Normalization is broadly applicable to multicolor data when at least one channel is expected to be ubiquitously expressed within its domain. Wavelength-dependent correction requires experimental calibration using either an ubiquitous signal at each wavelength. Importantly, this calibration only needs to be performed once for a given set of experimental conditions (e.g., fluorophores, tissue type, mounting medium).

- Multi-scale analysis of gene expression and morphometrics is applicable to any 3D multicolor image. This includes both the 3D visualization tools (Napari plugins) and the various analytical plots (e.g., correlation plots, radial analysis). Multi-scale analysis can be performed even with imperfect segmentation, as long as segmentation errors tend to cancel out when averaged locally at the relevant spatial scale. However, systematic errors—such as segmentation uncertainty along the Z-axis due to strong anisotropy—may accumulate and introduce bias in downstream analyses. Caution is advised when analyzing hollow structures (e.g., curved epithelial monolayers with large cavities), as the pipeline was developed primarily for 3D bulk tissues, and appropriate masking of cavities would be needed.

Reviewer #2 (Public review):

Summary:

This study presents an integrated experimental and computational pipeline for high-resolution, quantitative imaging and analysis of gastruloids. The experimental module employs dual-view two-photon spectral imaging combined with optimized clearing and mounting techniques to image whole-mount immunostained gastruloids. This approach enables the acquisition of comprehensive 3D images that capture both tissue-scale and single-cell level information.

The computational module encompasses both pre-processing of acquired images and downstream analysis, providing quantitative insights into the structural and molecular characteristics of gastruloids. The pre-processing pipeline, tailored for dual-view two-photon microscopy, includes spectral unmixing of fluorescence signals using depth-dependent spectral profiles, as well as image fusion via rigid 3D transformation based on content-based block-matching algorithms. Nuclei segmentation was performed using a custom-trained StarDist3D model, validated against 2D manual annotations, and achieving an F1 score of 85+/-3% at a 50% intersection-over-union (IoU) threshold. Another custom-trained StarDist3D model enabled accurate detection of proliferating cells and the generation of 3D spatial maps of nuclear density and proliferation probability. Moreover, the pipeline facilitates detailed morphometric analysis of cell density and nuclear deformation, revealing pronounced spatial heterogeneities during early gastruloid morphogenesis.

All computational tools developed in this study are released as open-source, Python-based software.

Strengths:

The authors applied two-photon microscopy to whole-mount deep imaging of gastruloids, achieving in toto visualization at single-cell resolution. By combining spectral imaging with an unmixing algorithm, they successfully separated four fluorescent signals, enabling spatial analysis of gene expression patterns.

The entire computational workflow, from image pre-processing to segmentation with a custom-trained StarDist3D model and subsequent quantitative analysis, is made available as open-source software. In addition, user-friendly interfaces are provided through the open-source, community-driven Napari platform, facilitating interactive exploration and analysis.

We thank the reviewer for this positive feedback.

Weaknesses:

The computational module appears promising. However, the analysis pipeline has not been validated on datasets beyond those generated by the authors, making it difficult to assess its general applicability.

We agree that applying our analysis pipeline to published datasets—particularly those acquired with different imaging systems—would be valuable. However, only a few high-resolution datasets of large organoid samples are publicly available, and most of these either lack multiple fluorescence channels or represent 3D hollow structures. Our computational pipeline consists of several independent modules: spectral filtering, dual-view registration, local contrast enhancement, 3D nuclei segmentation, image normalization based on a ubiquitous marker, and multiscale analysis of gene expression and morphometrics.

Spectral filtering has already been applied in other systems (e.g. [7] and [8]), but is here extended to account for imaging depth-dependent apparent emission spectra of the different fluorophores. In our pipeline, we provide code to run spectral filtering on multichannel images, integrated in Python. In order to apply the spectral filtering algorithm utilized here, spectral patterns of each fluorophore need to be calibrated as a function of imaging depth, which depend on the specific emission windows and detector settings of the microscope.

Image normalization using a wavelength-dependent correction also requires calibration on a given imaging setup to measure the difference in signal decay among the different fluorophores species. To our knowledge, the calibration procedures for spectral-filtering and our image-normalization approach have not been performed previously in 3D samples, which is why validation on published datasets is not readily possible. Nevertheless, they are described in detail in the Methods section, and the code used—from the calibration measurements to the corrected images—is available open-source at the Zenodo link in the manuscript.

Dual-view registration, local contrast enhancement, and multiscale analysis of gene expression and morphometrics are not limited to organoid data or our specific imaging modalities. If we identify suitable datasets to validate these modules, we will include them in the revised manuscript.

To evaluate our 3D nuclei segmentation model, we plan to test it on diverse systems, including gastruloids stained with the nuclear marker Draq5 from Moos et al. [1]; breast cancer spheroids; primary ductal adenocarcinoma organoids; human colon organoids and HCT116 monolayers from Ong et al. [2]; and zebrafish tissues imaged by confocal microscopy from Li et al [3]. These datasets were acquired using either light-sheet or confocal microscopy, with varying imaging parameters (e.g., objective lens, pixel size, staining method).

Preliminary results are promising (see Author response image 1). We will provide quantitative comparisons of our model’s performance on these datasets, using annotations or reference predictions provided by the original authors where available.

Author response image 1.

Qualitative comparison of our custom Stardist3D segmentation strategy on diverse published 3D nuclei datasets. We show one slice from the XY plane for simplicity. (a) Gastruloid stained with the nuclear marker DRAQ5 imaged with an open-top dual-view and dual-illumination LSM [1]. (b) Breast cancer spheroid [2]. (c) Primary pancreatic ductal adenocarcinoma organoids imaged with confocal microscopy[2]. (d) Human colon organoid imaged with LSM laser scanning confocal microscope [2]. (e) Monolayer HCT116 cells imaged with LSM laser scanning confocal microscope [2]. (f) Fixed zebrafish embryo stained for nuclei and imaged with a Zeiss LSM 880 confocal microscopy [3].

Besides, the nuclei segmentation component lacks benchmarking against existing methods.

We agree with the reviewer that a benchmark against existing segmentation methods would be very useful. We tried different pre-trained models:

- CellPose, which we tested in a previous paper ([4]) and which showed poor performances compared to our trained StarDist3D model.

- DeepStar3D ([2]) is only available in the software 3DCellScope. We could not benchmark the model on our data, because the free and accessible version of the software is limited to small datasets. An image of a single whole-mount gastruloid with one channel, having dimensions (347,467,477) was too large to be processed, see screenshot below. The segmentation model could not be extracted from the source code and tested externally because the trained DeepStar3D weights are encrypted.

Author response image 2.

Screenshot of the 3DCellScore software. We could not perform 3D nuclei segmentation of a whole-mount gastruloids because the image size was too large to be processed.

- AnyStar ([5]), which is a model trained from the StarDist3D architecture, was not performing well on our data because of the heterogeneous stainings. Basic pre-processing such as median and gaussian filtering did not improve the results and led to wrong segmentation of touching nuclei. AnyStar was demonstrated to segment well colon organoids in Ong et al, 2025 ([2]), but the nuclei were more homogeneously stained. Our Hoechst staining displays bright chromatin spots that are incorrectly labeled as individual nuclei.

- Cellos ([6]), another model trained from StarDist3D, was also not performing well. The objects used for training and to validate the results are sparse and not touching, so the predicted segmentation has a lot of false negatives even when lowering the probability threshold to detect more objects. Additionally, the network was trained with an anisotropy of (9,1,1), based on images with low z resolution, so it performed poorly on almost isotropic images. Adapting our images to the network’s anisotropy results in an imprecise segmentation that can not be used to measure 3D nuclei deformations.

We tried both Cellos and AnyStar predictions on a gastruloid image from Fig. S2 of our main manuscript. Author response image 3 displays the results qualitatively compared to our trained model Stardist-tapenade. For the revision of the paper, we will perform a comprehensive benchmark of these state-of-the-art routines, including quantitative assessment of the performance.

Author response image 3.

Qualitative comparison of two published segmentation models versus our model. We show one slice from the XY plane for simplicity. Segmentations are displayed with their contours only. (Top left) Gastruloid stained with Hoechst, image extracted from Fig S2 of our manuscript. (Top right) Same image overlayed with the prediction from the Cellos model, showing many false negatives. (Bottom left) Same image overlayed with the prediction from our Stardist-tapenade model. (Bottom right) Same image overlayed with the prediction from the AnyStar model, false positives are indicated with a red arrow.

Appraisal:

The authors set out to establish a quantitative imaging and analysis pipeline for gastruloids using dual-view two-photon microscopy, spectral unmixing, and a custom computational framework for 3D segmentation and gene expression analysis. This aim is largely achieved. The integration of experimental and computational modules enables high-resolution in toto imaging and robust quantitative analysis at the single-cell level. The data presented support the authors' conclusions regarding the ability to capture spatial patterns of gene expression and cellular morphology across developmental stages.

Impact and utility:

This work presents a compelling and broadly applicable methodological advance. The approach is particularly impactful for the developmental biology community, as it allows researchers to extract quantitative information from high-resolution images to better understand morphogenetic processes. The data are publicly available on Zenodo, and the software is released on GitHub, making them highly valuable resources for the community.

We thank the reviewer for these positive feedbacks.

Reviewer #3 (Public review):

Summary

The paper presents an imaging and analysis pipeline for whole-mount gastruloid imaging with two-photon microscopy. The presented pipeline includes spectral unmixing, registration, segmentation, and a wavelength-dependent intensity normalization step, followed by quantitative analysis of spatial gene expression patterns and nuclear morphometry on a tissue level. The utility of the approach is demonstrated by several experimental findings, such as establishing spatial correlations between local nuclear deformation and tissue density changes, as well as the radial distribution pattern of mesoderm markers. The pipeline is distributed as a Python package, notebooks, and multiple napari plugins.

Strengths

The paper is well-written with detailed methodological descriptions, which I think would make it a valuable reference for researchers performing similar volumetric tissue imaging experiments (gastruloids/organoids). The pipeline itself addresses many practical challenges, including resolution loss within tissue, registration of large volumes, nuclear segmentation, and intensity normalization. Especially the intensity decay measurements and wavelength-dependent intensity normalization approach using nuclear (Hoechst) signal as reference are very interesting and should be applicable to other imaging contexts. The morphometric analysis is equally well done, with the correlation between nuclear shape deformation and tissue density changes being an interesting finding. The paper is quite thorough in its technical description of the methods (which are a lot), and their experimental validation is appropriate. Finally, the provided code and napari plugins seem to be well done (I installed a selected list of the plugins and they ran without issues) and should be very helpful for the community.

We thank the reviewer for his positive feedback and appreciation of our work.

Weaknesses

I don't see any major weaknesses, and I would only have two issues that I think should be addressed in a revision:

(1) The demonstration notebooks lack accompanying sample datasets, preventing users from running them immediately and limiting the pipeline's accessibility. I would suggest to include (selective) demo data set that can be used to run the notebooks (e.g. for spectral unmixing) and or provide easily accessible demo input sample data for the napari plugins (I saw that there is some sample data for the processing plugin, so this maybe could already be used for the notebooks?).

We thank the reviewer for this relevant suggestion. The 7 notebooks were updated to automatically download sample tests. The different parts of the pipeline can now be run immediately: https://github.com/GuignardLab/tapenade/tree/chekcs_on_notebooks/src/tapenade/notebooks

(2) The results for the morphometric analysis (Figure 4) seem to be only shown in lateral (xy) views without the corresponding axial (z) views. I would suggest adding this to the figure and showing the density/strain/angle distributions for those axial views as well.

We agree with the reviewer that a morphometric analysis based on the axial views would be informative and plan to perform this analysis for the revision.

(1) Moos, F., Suppinger, S., de Medeiros, G., Oost, K.C., Boni, A., Rémy, C., Weevers, S.L., Tsiairis, C., Strnad, P. and Liberali, P., 2024. Open-top multisample dual-view light-sheet microscope for live imaging of large multicellular systems. Nature Methods, 21(5), pp.798-803.

(2) Ong, H.T., Karatas, E., Poquillon, T., Grenci, G., Furlan, A., Dilasser, F., Mohamad Raffi, S.B., Blanc, D., Drimaracci, E., Mikec, D. and Galisot, G., 2025. Digitalized organoids: integrated pipeline for high-speed 3D analysis of organoid structures using multilevel segmentation and cellular topology. Nature Methods, 22(6), pp.1343-1354.

(3) Li, L., Wu, L., Chen, A., Delp, E.J. and Umulis, D.M., 2023. 3D nuclei segmentation for multi-cellular quantification of zebrafish embryos using NISNet3D. Electronic Imaging, 35, pp.1-9.

(4) Vanaret, J., Dupuis, V., Lenne, P. F., Richard, F., Tlili, S., & Roudot, P. (2023). A detector-independent quality score for cell segmentation without ground truth in 3D live fluorescence microscopy. IEEE Journal of Selected Topics in Quantum Electronics, 29(4: Biophotonics), 1-12.

(5) Dey, N., Abulnaga, M., Billot, B., Turk, E. A., Grant, E., Dalca, A. V., & Golland, P. (2024). AnyStar: Domain randomized universal star-convex 3D instance segmentation. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (pp. 7593-7603).

(6) Mukashyaka, P., Kumar, P., Mellert, D. J., Nicholas, S., Noorbakhsh, J., Brugiolo, M., ... & Chuang, J. H. (2023). High-throughput deconvolution of 3D organoid dynamics at cellular resolution for cancer pharmacology with Cellos. Nature Communications, 14(1), 8406.

(7) Rakhymzhan, A., Leben, R., Zimmermann, H., Günther, R., Mex, P., Reismann, D., ... & Niesner, R. A. (2017). Synergistic strategy for multicolor two-photon microscopy: application to the analysis of germinal center reactions in vivo. Scientific reports, 7(1), 7101.

(8) Dunsing, V., Petrich, A., & Chiantia, S. (2021). Multicolor fluorescence fluctuation spectroscopy in living cells via spectral detection. Elife, 10, e69687.

Author Response:

Reviewer #1 (Public Review):

Summary:

This manuscript reports the substrate-bound structure of SiaQM from F. nucleatum, which is the membrane component of a Neu5Ac-specific Tripartite ATP-dependent Periplasmic (TRAP) transporter. Until recently, there was no experimentally derived structural information regarding the membrane components of the TRAP transporter, limiting our understanding of the transport mechanism. Since 2022, there have been 3 different studies reporting the structures of the membrane components of Neu5Ac-specific TRAP transporters. While it was possible to narrow down the binding site location by comparing the structures to proteins of the same fold, a structure with substrate bound has been missing. In this work, the authors report the Na+-bound state and the Na+ plus Neu5Ac state of FnSiaQM, revealing information regarding substrate coordination. In previous studies, 2 Na+ ion sites were identified. Here, the authors also tentatively assign a 3rd Na+ site. The authors reconstitute the transporter to assess the effects of mutating the binding site residues they identified in their structures. Of the 2 positions tested, only one of them appears to be critical to substrate binding.

Strengths:

The main strength of this work is the capture of the substrate-bound state of SiaQM, which provides insight into an important part of the transport cycle.

Weaknesses:

The main weakness is the lack of experimental validation of the structural findings. The authors identified the Neu5Ac binding site, but only tested 2 residues for their involvement in substrate interactions, which was very limited. The authors tentatively identified a 3rd Na+ binding site, which if true would be an impactful finding, but this site was not tested for its contribution to Na+ dependent transport, and the authors themselves report that the structural evidence is not wholly convincing. This lack of experimental validation undermines the confidence of the findings. However, the reporting of these new data is important as it will facilitate follow-up studies by the authors or other researchers.

The main concern, also mentioned by other reviewers, is the lack of mutational data and functional studies on the identified binding sites. Two other structures of TRAP transporters have been determined, one from Haemophilus influenzae (Hi) and the other from Photobacterium profundum (Pp). We will refer to the references in this paper as [1], Peter et al. as [2], and Davies et al. as [3]. The table below lists all the mutations made in the Neu5Ac binding site, including direct polar interactions between Neu5Ac and the side chains, as well as the newly identified metal sites.

The structure of Fusobacterium nucleatum (Fn) that we have reported shows a significant sequence identity with the previously reported Hi structure. When we superimpose the Pp and Fn structures, we observe that nearly all the residues that bind to the Neu5Ac and the third metal site are conserved. This suggests that mutagenesis and functional studies from other research can be related to the structure presented in our work.

The table below shows that all three residues that directly interact with Neu5Ac have been tested by site-directed mutagenesis for their role in Neu5Ac transport. Both D521 and S300 are critical for transport, while S345 is not. We do not believe that a mutation of D521A in Fn, followed by transport studies, will provide any new information.

However, Peter et al. have mutated only one of the 5 residues near the newly identified metal binding site, which resulted in no transport. The rest of the residues have not been functionally tested. We propose to mutate these residues into Ala, express and purify the proteins, and then carry out transport assays on those that show expression. We will include this information in the revised manuscript.

Reviewer #2 (Public Review):

In this exciting new paper from the Ramaswamy group at Purdue, the authors provide a new structure of the membrane domains of a tripartite ATP-independent periplasmic (TRAP) transporter for the important sugar acid, N-acetylneuraminic acid or sialic acid (Neu5Ac). While there have been a number of other structures in the last couple of years (the first for any TRAP-T) this is the first to trap the structure with Neu5Ac bound to the membrane domains. This is an important breakthrough as in this system the ligand is delivered by a substrate-binding protein (SBP), in this case, called SiaP, where Neu5Ac binding is well studied but the 'hand over' to the membrane component is not clear. The structure of the membrane domains, SiaQM, revealed strong similarities to other SBP-independent Na+-dependent carriers that use an elevator mechanism and have defined Na+ and ligand binding sites. Here they solve the cryo-EM structure of the protein from the bacterial oral pathogen Fusobacterium nucleatum and identify a potential third (and theoretically predicted) Na+ binding site but also locate for the first time the Neu5Ac binding site. While this sits in a region of the protein that one might expect it to sit, based on comparison to other transporters like VcINDY, it provides the first molecular details of the binding site architecture and identifies a key role for Ser300 in the transport process, which their structure suggests coordinates the carboxylate group of Neu5Ac. The work also uses biochemical methods to confirm the transporter from F. nucleatum is active and similar to those used by selected other human and animal pathogens and now provides a framework for the design of inhibitors of these systems.

The strengths of the paper lie in the locating of Neu5Ac bound to SiaQM, providing important new information on how TRAP transporters function. The complementary biochemical analysis also confirms that this is not an atypical system and that the results are likely true for all sialic acid-specific TRAP systems.

The main weakness is the lack of follow-up on the identified binding site in terms of structure-function analysis. While Ser300 is shown to be important, only one other residue is mutated and a much more extensive analysis of the newly identified binding site would have been useful.

Please see the comments above.

Reviewer #3 (Public Review):

The manuscript by Goyal et al reports substrate-bound and substrate-free structures of a tripartite ATP-independent periplasmic (TRAP) transporter from a previously uncharacterized homolog, F. nucleatum. This is one of the most mechanistically fascinating transporter families, by means of its QM domain (the domain reported in his manuscript) operating as a monomeric 'elevator', and its P domain functioning as a substrate-binding 'operator' that is required to deliver the substrate to the QM domain; together, this is termed an 'elevator with an operator' mechanism. Remarkably, previous structures had not demonstrated the substrate Neu5Ac bound. In addition, they confirm the previously reported Na+ binding sites and report a new metal binding site in the transporter, which seems to be mechanistically relevant. Finally, they mutate the substrate binding site and use proteoliposomal uptake assays to show the mechanistic relevance of the proposed substrate binding residues.

The structures are of good quality, the functional data is robust, the text is well-written, and the authors are appropriately careful with their interpretations. Determination of a substrate-bound structure is an important achievement and fills an important gap in the 'elevator with an operator' mechanism. Nevertheless, I have concerns with the data presentation, which in its current state does not intuitively demonstrate the discussed findings. Furthermore, the structural analysis appears limited, and even slight improvements in data processing and resulting resolution would greatly improve the authors' claims. I have several suggestions to hopefully improve the clarity and quality of the manuscript.

We appreciate your feedback and will make the necessary modifications to the manuscript incorporating most of the suggestions. We will submit the revised version once the experiments are completed. We are also working on improving the quality of the figures and have made several attempts to enhance the resolution using CryoSPARC or RELION, but without success. We will continue to explore newer methods in an effort to achieve higher resolution and to model more lipids, particularly in the binding pocket.

Reviewer #3 (Public Review):

The manuscript presents an intriguing explanation for why grid cell firing fields do {\em not} lie on a lattice whose axes aligned to the walls of a square arena. This observation, by itself, merits the manuscript's dissemination to the journals audience.

The presentation is quirky (but keep the quirkiness!).

But let me recast the problem presented by the authors as one of combinatorics. Given repeating, spatially separated firing fields across cells, one obtains temporal sequences of grid cells firing. Label these cells by integers from $[n]$. Any two cells firing in succession should uniquely identify one of six directions (from the hexagonal lattice) in which the agent is currently moving.

Now, take the symmetric group $\Sigma$ of cyclic permutations on $n$ elements.<br /> We ask whether there are cyclic permutations of $[n]$ such that

So, for instance, $(4,2,3,1)$ would not be counted as a valid permutation of $(1,2,3,4)$, as $(2,3)$ and $(1,4)$ are adjacent.

Furthermore, given $[n]$, are there two distinct cyclic permutations such that {\em no} adjacencies are preserved when considering any pair of permutations (among the triple of the original ordered sequence and the two permutations)? In other words, if we consider the permutation required to take the first permutation into the second, that permutation should not preserve any adjacencies.

{\bf Key question}: is there any difference between the solution to the combinatorics problem sketched above and the result in the manuscript? Specifically, the text argues that for $n=7$ there is only {\em one} solution.

Ideally, one would strive to obtain a closed-form solution for the number of such permutations as a function of $n$.

Author Response

Joint Public Review

Strengths

Overall, the idea that the PAG interacts with the BLA via the midline thalamus during a predator vs. foraging test is new and quite interesting. The authors have used appropriate tools to address their questions. The major impact in the field would be to add evidence to claims that the BLA can be downstream of the dPAG to evoke defensive behaviors. The study also adds to a body of evidence that the PAG mediates primal fear responses.

Weaknesses

(Anatomical concerns)

1) The authors claim that the recordings were performed in the dorsal PAG (dPAG), but the histological images in Fig. 1B and Supplementary S2 for example show the tip of the electrode in a different subregion of PAG (ventral/lateral). They should perform a more careful histological analysis of the recording sites and explain the histological inclusion and exclusion criteria. Diagrams showing the sites of all PAG and BLA recordings, as well as all fiber optics, would be helpful.

The PAG is composed of dorsomedial (dm), dorsolateral (dl), lateral (l), and ventrolateral (vl) columns that extend along the rostro-caudal axis of the aqueduct. The term “dorsal PAG” (dPAG) generally encompasses dmPAG, dlPAG, and lPAG, as substantiated by track-tracing, neurochemical, and immunohistochemical techniques (e.g., Bandler et al., 1991; Bandler & Keay, 1996; Carrive, 1993). As Bandler and Shipley (1994) summarized, “These findings suggest that what has been traditionally called the 'dorsal PAG' (a collective term for regions dorsal and lateral to the aqueduct), consists of three anatomically distinct longitudinal columns: dorsomedial and lateral columns…and a dorsolateral column…" Similarly, Schenberg et al. (2005) clarified in their review that, “According to this parcellation...the defensive behaviors (freezing, flight or fight) and aversion-related responses (switchoff behavior) were ascribed to the DMPAG, DLPAG, and LPAG (usually named the ‘dorsal’ PAG).” In our study, all recordings were conducted within the dPAG. Also, Figures 1B and S2 in our manuscript correspond to the -6.04 mm template from Paxinos & Watson’s atlas (1998), which is shown in the left panel in Author response image 1 and is considerably anterior to the location where the vlPAG emerges, as shown in the right panel. In our revised manuscript, we will provide a detailed definition of the dPAG, inclusive of dmPAG, dlPAG, and lPAG, and support this with the referenced literature.

Author response image 1.

2) Prior studies investigating the role of BLA neurons during a foraging vs. robot test similar to the one used in this study should be also cited and discussed (e.g., Amir et al 2019; Amir et al 2015). These two studies demonstrated that most neurons in the basal portion of the BLA exhibit inhibitory activity during foraging behavior and only a small fraction of neurons (~4%) display excitatory activity in response to the robot (in contrast to the 25% reported in the present study). A very accurate histological analysis of BLA recording sites should be performed to clarify whether distinct subregions of the BLA encode foraging and predator-related information, as previously shown in the two described studies.

In the revised manuscript, we will discuss papers by Amir et al. (2015) and Amir et al. (2019) that utilized a similar 'approach food-avoid predator' paradigm. These studies found a correlation between the neuronal activities in the basolateral amygdala (BL) and the velocity of animal movement during foraging, regardless of the presence or absence of predators. Specifically, the majority of BL neurons were inhibited in both conditions, with only 4.5% being responsive to predators. Consequently, Amir et al. posited that amygdala activity predominantly aligns with behavioral output such as foraging, rather than with responses to threats.

In contrast, our body of work (Kim et al., 2018; Kong et al., 2021; the present study) reveals that the majority of neurons in the BA/BLA displayed distinct responses in pre-robot and robot sessions. Kong et al. (2021) discussed in depth several factors that may account for this discrepancy, given that both Amir et al. and our research used similar behavioral paradigms. Differences in apparatus features, experimental procedures, and data analysis methodologies (refer to Amir et al., 2019) could be contributing to the conflicting results and interpretations concerning the significance of amygdalar neuronal activities.

Additionally, our studies uniquely monitored the same set of amygdalar neurons during pre-robot and robot sessions, affording us the opportunity for a direct comparison of neuronal activities under different threat conditions.

Another salient difference lines in the foraging success rates, which were markedly higher in Amir et al (~80%) compared to our studies (<3-4%). We hypothesize that there may be an inverse relationship between the pellet procurement rate and the intensity of fear. The high foraging success rate in Amir et al., which correlates with subdued amygdalar activity, stands in contrast to our findings of heightened amygdalar activity associated with a lower foraging success rate. Supporting this notion, optogeneticallyinduced amygdalar activity led naïve rats to abandon foraging and escape to the nest (Kong et al., 2021, the present study).

3) An important claim of this study that the PAG sends predator-related signals to BLA via the PVT (Fig. 4). The authors stated that PVT neurons labeled by intra-BLA injection of the retrograde tracer CTB were activated by the predator, but a proper immunohistochemical quantification with a control group was not provided to support this claim. To provide better support for their claim, the authors should quantify the doublelabeled PVT neurons (cFos plus CTB positive neurons) during the robot test.

As recommended, we will include a revised Fig. 4 in the manuscript to present the quantification of neurons that are double-labeled with c-Fos and CTB in the PVT. This updated figure will provide a more rigorous analysis and visual representation of the data.

4) The AVV anterograde tracer deposit spread to a large part of the PAG, including dorsolateral and lateral PAG, and supraoculomotor regions (Fig. 4B). Is the projection to the PVT from the dPAG or other regions of the PAG?

As previously addressed in response to Comment #1, the dPAG comprises the dmPAG, dlPAG, and lPAG. In the revised manuscript, we will acknowledge the diffusion of the AAV to the adjacent deep gray layer of the superior colliculus. Additionally, we are considering conducting more restricted AAV injections into the dPAG to verify terminal expressions in the PVT.

(Concerns about the strength of the evidence supporting a role for the PVT)

5) The authors conclude in the discussion section that the dPAG-amygdala pathway is involved in generating antipredatory defensive behavior. However, the current results are entirely based on correlational analyses of neural firing rate and there is no direct demonstration that the PAG provides information about the robot to the BLA. Therefore, the authors should tone down their interpretation or provide more evidence to support it by performing experiments applying inhibitory tools in the dPAG > PVT > BLA pathway and examining the impact on behavior and downstream neural firing.

As suggested, we will moderate the assertions about the functional implications of the PVT, based on the data from anterograde and retrograde tracers, to present a more measured interpretation in the manuscript.

(Other concerns)

6) One of the main findings of this study is the observation that BLA neurons that are responsive to PAG photostimulation are preferentially recruited during the foraging vs. robot test (Fig. 3). However, the experimental design used to address this question is problematic because the laser photostimulation of PAG neurons preceded the foraging vs. robot test. Prior photoactivation of PAG may have caused indirect shortterm synaptic plasticity in BLA cells, which would favor the response of these cells to the robot. Please see Oishi et al, 2019 PMID: 30621738, which demonstrated that 10 trains of 20Hz photoactivation (300 pulses each) was sufficient to induce LTP in brain slices.

After approximately eight photostimulation trials of the dPAG, with 40 pulses each, the animals entered a post-photostimulation testing phase (referred to as "Post"; Fig. 3C), lasting 10-15 minutes over an average of eight trials before robot testing. Although the PAG does not directly project to the BLA, the remote possibility of trans-synaptic plasticity in the BLA cannot be completely excluded and will be acknowledged. Additionally, it is noteworthy that Oishi et al's (2019) study applied a total of 3,000 pulses (i.e., 10 15-s trains of 20-Hz pulses) and investigated CA3-CA3 synaptic plasticity, as opposed to a total of 320 pulses (i.e., 8 2-s trains of 20-Hz pulses) in our study.

7) The authors should perform a longitudinal analysis of the behavioral responses of the rats across the trials to clarify whether the animals habituate to the robot or not. In Figure 1E, it appears that PAG neurons fire less across the trials, which could be associated with behavioral habituation to the predator robot. If that is the case, the activity of many other PAG and BLA neurons will also most likely vary according to the trial number, which would impact the current interpretation of the results.

In Figure 1E, the y-axis represents the Z scores of individual dPAG neurons, instead of representing repeated tests of the same neuron across multiple trials. The raster plot in Figure 1F clearly depicts that the same dPAG neurons consistently display heightened neural activity in response to the approaching robot across successive trials.

8) In Figure 1, it is unclear why the authors compared the activity of neurons that respond to the robot activation against the activity of the neurons during the retrieval of the food pellets in the pre-robot and postrobot sessions. The best comparison would be aligning the cells that were responsive to the activation of the robot with the moment in which the animals run back to the nest after consuming the pellets during the prerobot or post-robot sessions. This would enable the authors to demonstrate that the PAG responses are directly associated with the expression of escaping behavior in the presence of the robot rather than associated with the onset of goal-directed movement in direction to the next during the pre- and post-robot sessions. A graphic showing the correlation between PAG firing rate and escape response would be also informative.

Figure 1E compares the dPAG neural activity when animals enter a designated pellet zone (time-stamped by camera tracking) during both pre-robot and post-robot trials to the dPAG neural activity when entering the robot trigger zone (time-stamped by robot activation). We wish to clarify that rats carry the large (0.5 g) pellet back to the nest for consumption rather than consume it in the open arena before returning to the nest.

In our study, we aimed to investigate the direct response of dPAG neurons to the looming predator and explore the communication between dPAG and BLA in relation to antipredatory defensive responses. To build upon our previous research that suggests a potential role of dPAG in conveying such responses to the BLA (Kim et al., 2013) and the immediate firing of BLA neurons in response to predatory threats (Kim et al., 2018; Kong et al., 2021), we chose to narrow our testing window to a short latency period (< 500 ms) following robot activations. This specific time window allowed us to focus on the initial stages of the threat stimulus processing and minimize potential confounding factors such as the presence of residual firing activity triggered by the robot during the animals’ escape or any activity changes induced by the animals' behavior.

Furthermore, Figure S1C clearly demonstrates that (i) increased activity of dPAG robot cells preceded the animals’ actual turning and fleeing behavior toward the nest, as indicated by the peak values of movement speed (dark yellow), and (ii) the presence of pellets did not affect activity changes of the robot cells during pre- and post-robot sessions. These observations suggest that the heightened activity of dPAG robot cells was not due to movement changes or pellet motivation.

Lastly, as stated in the original manuscript, the vast majority of robot cells (90.9%) did not show significant correlations between movement speed and firing rates, lending further support to the interpretation that the dPAG activity observed was not merely a reflection of movement changes.

References

Bandler, R., Carrive, P., & Depaulis, A. (1991). Emerging principles of organization of the midbrain periaqueductal gray matter. The midbrain periaqueductal gray matter: functional, anatomical, and neurochemical organization, 1-8.

Bandler, R. & Keay, K. A. (1996). Columnar organization in the midbrain periaqueductal gray and the integration of emotional expression. Progress in brain research, 107, 285-300.

Bandler, R. & Shipley, M. T. (1994) Columnar organization in the midbrain periaqueductal gray: modules for emotional expression? Trends in Neurosciences, 17(9), 379-89.

Carrive, P. (1993). The periaqueductal gray and defensive behavior: functional representation and neuronal organization. Behavioural brain research, 58(1-2), 27-47.

Oishi, N., Nomoto, M., Ohkawa, N., Saitoh, Y., Sano, Y., Tsujimura, S., ... & Inokuchi, K. (2019). Artificial association of memory events by optogenetic stimulation of hippocampal CA3 cell ensembles. Molecular brain, 12, 1-10.

Paxinos, G. & Watson, C. (1998). The Rat Brain in Stereotaxic Coordinates. Academic Press, San Diego. Schenberg, L. C., Póvoa, R. M. F., Costa, A. L. P., Caldellas, A. V., Tufik, S., & Bittencourt, A. S. (2005). Functional specializations within the tectum defense systems of the rat. Neuroscience & Biobehavioral Reviews, 29(8), 1279-1298.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

This work by Ding et al uses agent-based simulations to explore the role of the structure of molecular motor myosin filaments in force generation in cytoskeletal structures. The focus of the study is on disordered actin bundles which can occur in the cell cytoskeleton and have also been investigated with in vitro purified protein experiments.

Strengths:

The key finding is that cooperative effects between multiple myosin filaments can enhance both total force and the efficiency of force generation (force per myosin). These trends were possible to obtain only because the detailed structure of the motor filaments with multiple heads is represented in the model.

We appreciate your comments about the strength of our study.

Weaknesses:

It is not clearly described what scientific/biological questions about cellular force production the work answers. There should be more discussion of how their simulation results compare with existing experiments or can be tested in future experiments.

Thank you for the comment. First, our study explains why non-muscle myosin II in stress fibers shows focal distributions rather than uniform distributions; if they stay closely, they can generate much larger forces in the stress fibers via the cooperative overlap. Our study also predicts a difference between bipolar structures (found in skeletal muscle myosins and non-muscle myosins) and side polar structures (found in smooth muscle myosins) in terms of the likelihood of the cooperative overlap. As shown below, myosin filaments with the bipolar structure can add up their forces better than those with the side polar structure when their overlap level is the same. We will add discussion about these in the revised manuscript.

Author response image 1.

As the reviewer noticed, our results were briefly compared with prior observations in Ref. 4 (Thoresen et al., Biophys J, 2013) where different myosin isoforms were used for in vitro actin bundles. We will add more quantitative comparisons between the in vitro study and our results.

In addition, at the end of the conclusion section, we suggested future experiments that can be used for verifying our results. In particular, experiments with synthetic myosin filaments with tunable geometry seem to be suitable for verifying our computational predictions and observations.

The model assumptions and scientific context need to be described better.

We apologize for the insufficient descriptions about the model. We will revise those parts to better explain model assumptions and scientific context.

The network contractility seems to be a mere appendix to the bundle contractility which is presented in much more detail.

We included some cases run with the two-dimensional network in this study to prove the generality of our conclusions. We included minimal preliminary results in this study because we are currently working on a follow-up study with network structures. I hope that the reviewer would understand our intention and situation.

Reviewer #2 (Public review):

Summary:

In this study, the authors use a mechanical model to investigate how the geometry and deformations of myosin II filaments influence their force generation. They introduce a force generation efficiency that is defined as the ratio of the total generated force and the maximal force that the motors can generate. By changing the architecture of the myosin II filaments, they study the force generation efficiency in different systems: two filaments, a disorganized bundle, and a 2D network. In the simple two-filament systems, they found that in the presence of actin cross-linking proteins motors cannot add up their force because of steric hindrances. In the disorganized bundle, the authors identified a critical overlap of motors for cooperative force generation. This overlap is also influenced by the arrangement of the motor on the filaments and influenced by the length of the bare zone between the motor heads.

Strengths:

The strength of the study is the identification of organizational principles in myosin II filaments that influence force generation. It provides a complementary mechanistic perspective on the operation of these motor filaments. The force generation efficiency and the cooperative overlap number are quantitative ways to characterize the force generation of molecular motors in clusters and between filaments. These quantities and their conceptual implications are most likely also applicable in other systems.

Thank you for the comments about the strength of our study.

Weaknesses:

The detailed model that the authors present relies on over 20 numerical parameters that are listed in the supplement. Because of this vast amount of parameters, it is not clear how general the findings are. On the other hand, it was not obvious how specific the model is to myosin II, meaning how well it can describe experimental findings or make measurable predictions. The model seems to be quantitative, but the interpretation and connection to real experiments are rather qualitative in my point of view.

As the reviewer mentioned, all agent-based computational models for simulating the actin cytoskeleton are inevitably involved with such a large number of parameters. Some of the parameter values are not known well, so we have tuned our parameter values carefully by comparing our results with experimental observations in our previous studies since 2009. 

We were aware of the importance of rigorous representation of unbinding and walking rates of myosin motors, so we implemented the parallel cluster model, which can predict those rates with consideration of the mechanochemical rates of myosin II, into our model. Thus, we are convincing that our motors represent myosin II.

In our manuscript, our results were compared with prior observations in Ref. 4 (Thoresen et al., Biophys J, 2013) several times. In particular, larger force generation with more myosin heads per thick filament was consistent between the experiment and our simulations.

Our study can make various predictions. First, our study explains why non-muscle myosin II in stress fibers shows focal distributions rather than uniform distributions; if they stay closely, they can generate much larger forces in the stress fibers via the cooperative overlap. Our study also predicts a difference between bipolar structures (found in skeletal muscle myosins and non-muscle myosins) and side polar structures (found in smooth muscle myosins) in terms of the likelihood of the cooperative overlap. As shown in Author response image 1, myosin filaments with the bipolar structure can add up their forces better than those with the side polar structure when their overlap level is the same. We will add discussion about these in the revised manuscript.

We will add more discussion about these in the revised manuscript.

It was often difficult for me to follow what parameters were changed and what parameters were set to what numerical values when inspecting the curve shown in the figures. The manuscript could be more specific by explicitly giving numbers. For example, in the caption for Figure 6, instead of saying "is varied by changing the number of motor arms, the bare zone length, the spacing between motor arms", the authors could be more specific and give the ranges: ""is varied by changing the number of motor arms form ... to .., the bare zone length from .. to..., and the spacing between motor arms from .. to ..".

This unspecificity is also reflected in the text: "We ran simulations with a variation in either L_sp or L_bz" What is the range of this variation? "When L_M was similar" similar to what? "despite different N_M." What are the different values for N_M? These are only a few examples that show that the text could be way more specific and quantitative instead of qualitative descriptions.

We appreciate the comment. We will specify the range of the variation in each parameter in the revised manuscript.

In the text, after equation (2) the authors discuss assumptions about the binding of the motor to the actin filament. I think these model-related assumptions and explanations should be discussed not in the results section but rather in the "model overview" section.

Thank you for pointing this out. We will reorganize the text in the revised manuscript.

The lines with different colors in Figure 2A are not explained. What systems and parameters do they represent?

The different colors used in Fig. 2A were used for distinguishing 20 cases. We will add explanation about the colors in the figure caption in the revised manuscript.

Author response:

We thank the reviewers for their support of this work and insightful recommendations for how to improve it. We have provided specific responses to each reviewer comment below. To summarize how we intend to address the requested revisions:

Many of the reviewers’ comments requested additional technical or quality details about the DMS libraries or assays (e.g., number of cells tested, number of sequencing reads, assay replication, assay sensitivity, library balance), and we provide additional information and analyses that we can incorporate into the relevant portions of the text, supplementary tables, and supplementary figures to address these questions.

Some comments asked to clarify nomenclature/wording or provide additional labels to images, and we will make these changes as requested.

A few questions would require additional experimental data to address. Where experiments have already been performed, we will incorporate those results or cite relevant work previously reported in the literature.

Reviewer 1:

Summary

Howard et al. performed deep mutational scanning on the MC4R gene, using a reporter assay to investigate two distinct downstream pathways across multiple experimental conditions. They validated their findings with ClinVar data and previous studies. Additionally, they provided insights into the application of DMS results for personalized drug therapy and differential ligand responses across variant types.

Strengths

They captured over 99% of variants with robust signals and investigated subtle functionalities, such as pathway-specific activities and interactions with different ligands, by refining both the experimental design and analytical methods.

Weaknesses

While the study generated informative results, it lacks a detailed explanation regarding the input library, replicate correlation, and sequencing depth for a given number of cells.

Additionally, there are several questions that it would be helpful for authors to clarify.

(1) It would be helpful to clarify the information regarding the quality of the input library and experimental replicates. Are variants evenly represented in the library? Additionally, have the authors considered using long-read sequencing to confirm the presence of a single intended variant per construct? Finally, could the authors provide details on the correlation between experimental replicates under each condition?

Are variants evenly represented in the library?

We strive to achieve as evenly balanced library as possible at every stage of the DMS process (e.g., initial cloning in E. coli through integration into human cells). Below is a representative plot showing the number of barcodes per amino acid variant at each position in a given ~60 amino acid subregion of MC4R, which highlights how evenly variants are represented at the E. coli cloning stage.

Author response image 1.

We also make similar measurements after the library is integrated into HEK293T cell lines, and see similarly even coverage across all variants, as shown in the plot below.

Author response image 2.

Additionally, have the authors considered using long-read sequencing to confirm the presence of a single intended variant per construct?

We agree long-read sequencing would be an excellent way to confirm that our constructs contain a single intended variant. However, we elected for an alternate method (outlined in more detail in Jones et al. 2020) that leverages multiple layers of validation. First, the oligo chip-synthesized portions of the protein containing the variants are cloned into a sequence-verified plasmid backbone, which greatly decreases the chances of spuriously generating a mutation in a different portion of the protein. We then sequence both the oligo portion and random barcode using overlapping paired end reads during barcode mapping to avoid sequencing errors and to help detect DNA synthesis errors. At this stage, we computationally reject any constructs that have more than one variant. Given this, the vast majority of remaining unintended variants would come from somatic mutations introduced by the E. coli cloning or replication process, which should be low frequency. We have used our in-house full plasmid sequencing method, OCTOPUS, to sample and spot check this for several other DMS libraries we have generated using the same cloning methods. We have found variants in the plasmid backbone in only ~1% of plasmids in these libraries. Our statistical model also helps correct for this by accounting for barcode-specific variation. Finally we believe this provides further motivation for having multiple barcodes per variant, which dilutes the effect of any unintended additional variants.

Finally, could the authors provide details on the correlation between experimental replicates under each condition?

Certainly! In general, the Gs reporter had higher correlation between replicates than the Gq system (r ~ 0.5 vs r ~ 0.4). The plots below show two representative correlations at the RNA-seq stage of read counts for barcodes between the low a-MSH conditions. One important advantage of our statistical model is that it’s able to leverage information from barcodes regardless of the number of replicates they appear in.

Author response image 3.

Since the functional readout of variants is conducted through RNA sequencing, it seems crucial to sequence a sufficient number of cells with adequate sequencing saturation. Could the authors clarify the coverage depth used for each RNA-seq experiment and how this depth was determined? Additionally, how many cells were sequenced in each experiment?

This will be addressed by incorporating the following details into the manuscript:

We seeded 17 million cells per replicate at the start of each assay and, with a doubling of ~1.5x over the course of the assay, harvested ~25.5 million cells per replicate for RNA extraction and sequencing. We found this sufficient to get at least ~30-60x cellular coverage per amino acid variant.

Total mapped reads per replicate at RNA-seq stage

- Gs/CRE: 9.1-18.2 million mapped reads, median=12.3

- Gq/UAS: 8.6-24.1 million mapped reads, median=14.5

- Gs/CRE+Chaperone: 6.4-9.5 million mapped reads, median=7.5

Reads per barcode distribution

- Median read counts of 8, 10, and 6 reads per sample per barcode for Gs/CRE, Gq/UAS, and Gs/CRE+Chaperone assays, respectively.

Barcodes per variant distribution

- As reported, the median number of barcodes per variant across samples (the “median of medians”) is 56 for Gs/CRE and 28 for Gq/UAS

- Additionally, it is 44 for Gs/CRE+Chaperone

It appears that the frequencies of individual RNA-seq barcode variants were used as a proxy for MR4C activity. Would it be important to also normalize for heterogeneity in RNA-seq coverage across different cells in the experiment? Variability in cell representation (i.e., the distribution of variants across cells) could lead to misinterpretation of variant effects. For example, suppose barcode_a1 represents variant A and barcode_b1 represents variant B. If the RNA-seq results show 6 reads for barcode_a1 and 7 reads for barcode_b1, it might initially appear that both variants have similar effect sizes. However, if these reads correspond to 6 separate cells each containing 1 copy of barcode_a1, and only 1 cell containing 7 copies of barcode_b1, the interpretation changes significantly. Additionally, if certain variants occupy a larger proportion of the cell population, they are more likely to be overrepresented in RNA sequencing.

We account for this heterogeneity in several ways. First, as shown above (Response to Reviewer 1, Question 1), we aim to have even representation of variants within our libraries. Second, we utilize compositional control conditions like forskolin or unstimulated conditions to obtain treatment-independent measurements of barcode abundance and, consequently, of mutant-vs-WT effects that are due to compositional rather than biological variability. We expect that variability observed under these controls is due to subtle effects of molecular cloning, gene expression, and stochasticity. Using these controls, we observe that mutant-vs-WT effects are generally close to zero in these normalization conditions (e.g., in untreated Gq, see Supplementary Figure 3) as compared to drug-treated conditions. For example, pre-mature stops behave similar to WT in normalization conditions. This indicates that mutant abundance is relatively homogenous. Where there are barcode-dependent effects on abundance, we can use information from these conditions to normalize that effect. Finally, our mixed-effect model accounts for barcode-specific deviations from the expected mutant effect (e.g. a “high count” barcode consistently being high relative to the mean).

Although the assay system appears to effectively represent MC4R functionality at the molecular level, we are curious about the potential disparity between the DMS score system and physiological relevance. How do variants reported in gnomAD distribute within the DMS scoring system?

Figure 2D shows DMS scores (variant effect on Gs signaling) relative to human population frequency for all MC4R variants reported in gnomAD as of January 8, 2024.

To measure Gq signaling, the authors used the GAL4-VPR relay system. Is there additional experimental data to support that this relay system accurately represents Gq signaling?

The full Gq reporter uses an NFAT response element from the IL-2 promoter to regulate the expression of the GAL4-VPR relay. In this system, the activation of Gq signaling results in the activation of the NFAT response element, and this signal is then amplified by the GAL4-VPR relay. The NFAT response element has been previously well-validated to respond to the activation of Gq signaling (e.g., PMID: 8631834). We will add this reference to the text to further support the use of the Gq assay.

Identifying the variants responsive to the corrector was impressive. However, we are curious about how the authors confirmed that the restoration of MC4R activity was due to the correction of the MC4R protein itself. Is there a possibility that the observed effect could be influenced by other factors affected by the corrector? When the corrector was applied to the cells, were any expected or unexpected differential gene expression changes observed?

While we do not directly measure whether Ipsen-17 has effects on other signaling processes, previous work has shown that Ipsen-17 treatment does not indirectly alter signaling kinetics such as receptor internalization (Wang et al., 2014). Furthermore, our analysis methods inherently account for this by normalizing variant effects to WT signaling levels. Any observed rescue of a given variant inherently means that the variant is specifically more responsive to Ipsen-17 than WT, and the fact that different variants exhibit different levels of rescue is reassuring that the mechanism is on target to MC4R. Lastly, Ipsen-17 is known to be an antagonist of alpha-MSH activity and is thought to bind directly to the same site on MC4R (Wang et al., 2014).

As mentioned in the introduction, gain-of-function (GoF) variants are known to be protective against obesity. It would be interesting to see further studies on the observed GoF variants. Do the authors have any plans for additional research on these variants?

We agree this would be an excellent line of inquiry, but due to changes in company priorities we unfortunately do not have any plans for additional research on these variants.

Reviewer 2:

Overview

In this manuscript, the authors use deep mutational scanning to assess the effect of ~6,600 protein-coding variants in MC4R, a G protein-coupled receptor associated with obesity. Reasoning that current deep mutational scanning approaches are insufficiently precise for some drug development applications, they focus on articulating new, more precise approaches. These approaches, which include a new statistical model and innovative reporter assay, enable them to probe molecular phenotypes directly relevant to the development of drugs that target this receptor with high precision and statistical rigor.

They use the resulting data for a variety of purposes, including probing the relationship between MC4R's sequence and structure, analyzing the effect of clinically important variants, identifying variants that disrupt downstream MC4R signaling via one but not both pathways, identifying loss of function variants are amenable to a corrector drug and exploring how deep mutational scanning data could guide small molecule drug optimization.

Strengths

The analysis and statistical framework developed by the authors represent a significant advance. In particular, the study makes use of barcode-level internally replicated measurements to more accurately estimate measurement noise.

The framework allows variant effects to be compared across experimental conditions, a task that is currently hard to do with rigor. Thus, this framework will be applicable to a large number of existing and future deep mutational scanning experiments.

The authors refine their existing barcode transcription-based assay for GPCR signaling, and develop a clever "relay" new reporter system to boost signaling in a particular pathway. They show that these reporters can be used to measure both gain of function and loss of function effects, which many deep mutational scanning approaches cannot do.

The use of systematic approaches to integrate and then interrogate high-dimensional deep mutational scanning data is a big strength. For example, the authors applied PCA to the variant effect results from reporters for two different MC4R signaling pathways and were able to discover variants that biased signaling through one or the other pathway. This approach paves the way for analyses of higher dimensional deep mutational scans.

The authors use the deep mutational scanning data they collect to map how different variants impact small molecule agonists activate MC4R signaling. This is an exciting idea, because developing small-molecule protein-targeting therapeutics is difficult, and this manuscript suggests a new way to map small-molecule-protein interactions.

Weaknesses

The authors derive insights into the relationship between MC4R signaling through different pathways and its structure. While these make sense based on what is already known, the manuscript would be stronger if some of these insights were validated using methods other than deep mutational scanning.

Likewise, the authors use their data to identify positions where variants disrupt MC4R activation by one small molecule agonist but not another. They hypothesize these effects point to positions that are more or less important for the binding of different small molecule agonists. The manuscript would be stronger if some of these insights were explored further.

Impact

In this manuscript, the authors present new methods, including a statistical framework for analyzing deep mutational scanning data that will have a broad impact. They also generate MC4R variant effect data that is of interest to the GPCR community.

Author response:

Reviewer 1:

There are no significant weaknesses to signal in the manuscript. However, in order to fully conclude that there is no obvious advantage for the linguistic dimension in neonates, it would have been most useful to test a third condition in which the two dimensions were pitted against each other, that is, in which they provide conflicting information as to the boundaries of the words comprised in the artificial language. This last condition would have allowed us to determine whether statistical learning weighs linguistic and non-linguistic features equally, or whether phonetic content is preferentially processed.

We appreciate the reviewers' suggestion that a stream with conflicting information would provide valuable insights. In the present study, we started with a simpler case involving two orthogonal features (i.e., phonemes and voices), with one feature being informative and the other uninformative, and we found similar learning capacities for both. Future work should explore whether infants—and humans more broadly—can simultaneously track regularities in multiple speech features. However, creating a stream with two conflicting statistical structures is challenging. To use neural entrainment, the two features must lead to segmentation at different chunk sizes so that their effects lead to changes in power/PLV at different frequencies—for instance, using duplets for the voice dimension and triplets for the linguistic dimension  (or vice versa). Consequently, the two dimensions would not be directly comparable within the same participant in terms of the number of distinguishable syllables/voices, memory demand, or SNR given the 1/F decrease in amplitude of background EEG activity. This would involve comparisons between two distinct groups counter-balancing chunk size and linguistic non-linguistic dimension. Considering the test phase, words for one dimension would have been part-words for the other dimension. As we are measuring differences and not preferences, interpreting the results would also have been difficult. Additionally, it may be difficult to find a sufficient number of clearly discriminable voices for such a design (triplets imply 12 voices). Therefore, an entirely different experimental paradigm would need to be developed.

If such a design were tested, one possibility is that the regularities for the two dimensions are calculated in parallel, in line with the idea that the calculation of statistical regularities is a ubiquitous implicit mechanism (see Benjamin et al., 2024, for a proposed neural mechanism). Yet, similar to our present study, possibly only phonetic features would be used as word candidates. Another possibility is that only one informative feature would be explicitly processed at a time due to the serial nature of perceptual awareness, which may prioritise one feature over the other.

Note: The reviewer’s summary contains a typo: syllabic rate (4 Hz) –not 2 Hz, and word rate (2 Hz) –not 4 Hz.

Reviewer 2:

N400: I am skeptical regarding the interpretation of the phoneme-specific ERP effect as a precursor of the N400 and would suggest toning it down. While the authors are correct in that infant ERP components are typically slower and more posterior compared to adult components, and the observed pattern is hence consistent with an adult N400, at the same time, it could also be a lot of other things. On a functional level, I can't follow the author's argument as to why a violation in phoneme regularity should elicit an N400, since there is no evidence for any semantic processing involved. In sum, I think there is just not enough evidence from the present paradigm to confidently call it an N400.

The reviewer is correct that we cannot definitively determine the type of processing reflected by the ERP component that appears when neonates hear a triplet after exposure to a stream with phonetic regularities. We interpreted this component as a precursor to the N400, based on prior findings in speech segmentation tasks without semantic content, where a ~400 ms component emerged when adult participants recognised pseudowords (Sander et al., 2002) or during structured streams of syllables (Cunillera et al., 2006, 2009). Additionally, the component we observed had a similar topography and timing to those labelled as N400 in infant studies, where semantic processing was involved (Parise et al., 2010; Friedrich & Friederici, 2011).

Given our experimental design, the difference we observed must be related to the type of regularity during familiarisation (either phonemes or voices). Thus, we interpreted this component as reflecting lexical search— a process which could be triggered by a linguistic structure but which would not be relevant to a non-linguistic regularity such as voices. However, we are open to alternative interpretations. In any case, this difference between the two streams reveals that computing regularities based on phonemes versus voices does not lead to the same processes. We will revise and tone down the corresponding part of the discussion to clarify that it is just a possible interpretation of the results.  

Female and male voices: Why did the authors choose to include male and female voices? While using both female and male stimuli of course leads to a higher generalizability, it also introduces a second dimension for one feature that is not present for this other (i.e., phoneme for Experiment 1 and voice identity plus gender for Experiment 2). Hence, couldn't it also be that the infants extracted the regularity with which one gender voice followed the other? For instance, in List B, in the words, one gender is always followed by the other (M-F or F-M), while in 2/3 of the part-words, the gender is repeated (F-F and M-M). Wouldn't you expect the same pattern of results if infants learned regularities based on gender rather than identity?

We used three female and three male voices to maximise acoustic variability. The streams were synthesised using MBROLA, which provides a limited set of artificial voices. Indeed, there were not enough French voices of acceptable quality, so we also used two Italian voices (the phonemes used existed in both Italian and French).

Voices differ in timbre, and female voices tend to be higher pitched. However, it is sometimes difficult to categorise low-pitched female voices and high-pitched male voices. Given that gender may be an important factor in infants' speech perception (newborns, for instance, prefer female voices at birth), we conducted tests to assess whether this dimension could have influenced our results.  

We first quantified the transitional probabilities matrices during the structured stream of Experiment 2, considering that there are only two types of voices: Female and Male.  

For List A, all transition probabilities are equal to 0.5 (P(M|F), P(F|M), P(M|M), P(F|F)), resulting in flat TPs throughout the stream (see Author response image 1, top). Therefore, we would not expect neural entrainment at the word rate (2 Hz), nor would we anticipate ERP differences between the presented duplets in the test phase.

For List B, P(M|F)=P(F|M)=0.66 while P(M|M)=P(F|F)=0.33. However, this does not produce a regular pattern of TP drops throughout the stream (see Author response image 1, bottom). As a result, strong neural entrainment at 2 Hz was unlikely, although some degree of entrainment might have occasionally occurred due to some drops occurring at a 2 Hz frequency. Regarding the test phase, all three Words and only one Part-word presented alternating patterns (TP=0.6). Therefore, the difference in the ERPs between Words and Partwords in List B might be attributed to gender alternation.  

However, it seems unlikely that gender alternation alone explains the entire pattern of results, as the effect is inconsistent and appears in only one of the lists. To rule out this possibility, we analysed the effects in each list separately.

Author response image 1.

Transition probabilities (TPs) across the structured stream in Experiment 2, considering voices processed by gender (Female or Male). Top: List A. Bottom: List B.

We computed the mean activation within the time windows and electrodes of interest and compared the effects of word type and list using a two-way ANOVA. For the difference between Words and Part-words over the positive cluster, we observed a main effect of word type (F(1,31) = 5.902, p = 0.021), with no effects of list or interactions (p > 0.1). Over the negative cluster, we again observed a main effect of word type (F(1,31) = 10.916, p = 0.0016), with no effects of list or interactions (p > 0.1). See Author response image 2.  

Author response image 2.

Difference in ERP voltage (Words – Part-words) for the two lists (A and B); W=Words; P=Part-Words, 

We conducted a similar analysis for neural entrainment during the structured stream on voices. A comparison of entrainment at 2 Hz between participants who completed List A and List B showed no significant differences (t(30) = -0.27, p = 0.79). A test against zero for each list indicated significant entrainment in both cases (List A: t(17) = 4.44, p = 0.00036; List B: t(13) = 3.16, p = 0.0075). See Author response image 3.

Author response image 3.

Neural entrainment at 2Hz during the structured stream of Experiment 2 for Lists A and B.

Words entrainment over occipital electrodes: Do you have any idea why the duplet entrainment effect occurs over the electrodes it does, in particular over the occipital electrodes (which seems a bit unintuitive given that this is a purely auditory experiment with sleeping neonates).

Neural entrainment might be considered as a succession of evoked response induced by the stream. After applying an average reference in high-density EEG recordings, the auditory ERP in neonates typically consists of a central positivity and a posterior negativity with a source located at the electrical zero in a single-dipole model (i.e. approximately in the superior temporal region (Dehaene-Lambertz & Dehaene, 1994). In adults, because of the average reference (i.e. the sum of voltages is equal to zero at each time point) and because the electrodes cannot capture the negative pole of the auditory response, the negativity is distributed around the head. In infants, however, the brain is higher within the skull, allowing for a more accurate recording of the negative pole of the auditory ERP (see Author response image 4 for the location of electrodes in an infant head model).  

Besides the posterior electrodes, we can see some entrainment on more anterior electrodes that probably corresponds to the positive pole of the auditory ERP.

Author response image 4.

International 10–20 sensors' location on the skull of an infant template, with the underlying 3-D reconstruction of the grey-white matter interface and projection of each electrode to the cortex. Computed across 16 infants (from Kabdebon et al, Neuroimage, 2014). The O1, O2, T5, and T6 electrodes project lower than in adults.

Reviewer 3:

(1) While it's true that voice is not essential for language (i.e., sign languages are implemented over gestures; the use of voices to produce non-linguistic sounds, like laughter), it is a feature of spoken languages. Thus I'm not sure if we can really consider this study as a comparison between linguistic and non-linguistic dimensions. In turn, I'm not sure that these results show that statistical learning at birth operates on non-linguistic features, being voices a linguistic dimension at least in spoken languages. I'd like to hear the authors' opinions on this.

On one hand, it has been shown that statistical learning (SL) operates across multiple modalities and domains in human adults and animals. On the other hand, SL is considered essential for infants to begin parsing speech. Therefore, we aimed to investigate whether SL capacities at birth are more effective on linguistic dimensions of speech, potentially as a way to promote language learning.

We agree with the reviewer that voices play an important role in communication (e.g., for identifying who is speaking); however, they do not contribute to language structure or meaning, and listeners are expected to normalize across voices to accurately perceive phonemes and words. Thus, voices are speech features but not linguistic features. Additionally, in natural speech, there are no abrupt voice changes within a word as in our experiment; instead, voice changes typically occur on a longer timescale and involve only a limited number of voices, such as in a dialogue. Therefore, computing regularities based on voice changes would not be useful in real-life language learning. We considered that contrasting syllables and voices was an elegant way to test SL beyond its linguistic dimension, as the experimental paradigm is identical in both experiments.  

Along the same line, in the Discussion section, the present results are interpreted within a theoretical framework showing statistical learning in auditory non-linguistic (string of tones, music) and visual domains as well as visual and other animal species. I'm not sure if that theoretical framework is the right fit for the present results.

(2) I'm not sure whether the fact that we see parallel and independent tracking of statistics in the two dimensions of speech at birth indicates that newborns would be able to do so in all the other dimensions of the speech. If so, what other dimensions are the authors referring to?

The reviewer is correct that demonstrating the universality of SL requires testing additional modalities and acoustic dimensions. However, we postulate that SL is grounded in a basic mechanism of long-term associative learning, as proposed in Benjamin et al. (2024), which relies on a slow decay in the representation of a given event. This simple mechanism, capable of operating on any representational output, accounts for many types of sequence learning reported in the literature (Benjamin et al., in preparation). We will revise the discussion section to clarify this theoretical framework.

(3) Lines 341-345: Statistical learning is an evolutionary ancient learning mechanism but I do not think that the present results are showing it. This is a study on human neonates and adults, there are no other animal species involved therefore I do not see a connection with the evolutionary history of statistical learning. It would be much more interesting to make claims on the ontogeny (rather than philogeny) of statistical learning, and what regularities newborns are able to detect right after birth. I believe that this is one of the strengths of this work.

We did not intend to make claims about the phylogeny of SL. Since SL appears to be a learning mechanism shared across species, we use it as a framework to suggest that SL may arise from general operational principles applicable to diverse neural networks. Thus, while it is highly useful for language acquisition, it is not specific to it. We will revise this section to tone down our claims.  

(4) The description of the stimuli in Lines 110-113 is a bit confusing. In Experiment 1, e.g., "pe" and "tu" are both uttered by the same voice, correct? ("random voice each time" is confusing). Whereas in Experiment 2, e.g., "pe" and "tu" are uttered by different voices, for example, "pe" by yellow voice and "tu" by red voice. If this is correct, then I recommend the authors to rephrase this section to make it more clear.

To clarify, in Experiment 1, the voices were randomly assigned to each syllable, with the constraint that no voice was repeated consecutively. This means that syllables within the same word were spoken by different voices, and each syllable was heard with various voices throughout the stream. As a result, neonates had to retrieve the words based solely on syllabic patterns, without relying on consistent voice associations or specific voice relationships.

In Experiment 2, the design was orthogonal: while the syllables were presented in a random order, the voices followed a structured pattern. Similar to Experiment 1, each syllable (e.g., “pe” and “tu”) was spoken by different voices. The key difference is that in Experiment 2, the structured regularities were applied to the voices rather than the syllables. In other words, the “green” voice was always followed by the “red” voice for example but uttered different syllables.

We will revise the methods section to clarify these important points.

(5) Line 114: the sentence "they should compute a 36 x 36 TPs matrix relating each acoustic signal, with TPs alternating between 1/6 within words and 1/12 between words" is confusing as it seems like there are different acoustic signals. Can the authors clarify this point?

Thank you for highlighting this point. To clarify, our suggestion is that neonates might not track regularities between phonemes and voices as separate features. Instead, they may treat each syllable-voice combination as a distinct item—for example, "pe" spoken by the "yellow" voice is one item, while "pe" spoken by the "red" voice is another. Under this scenario, there would be a total of 36 unique items (6 syllables × 6 voices), and infants would need to track regularities between these 36 combinations.

We will rephrase this sentence in the manuscript to make it clearer.

eLife Assessment

In their study, Neiswender et al. provide important insights into how BicD2 variants linked to spinal muscular atrophy alter dynein activity and cargo specificity. The authors present convincing evidence that disease-associated mutations lead to interactome changes, supported by additional validation of the BicD2/HOPS complex and discussion of their functional implications. This well-executed study offers invaluable datasets and a strong foundation for future exploration of disease mechanisms.

Reviewer #1 (Public review):

In this work, Neiswender and colleagues test the hypothesis that mutations in BicD2 that are associated with SMALED alter BicD2-cargo interactions. To do this, they first establish the WT BicD2 cargo interactome (using a proximity-dependent biotin ligase screen with Turbo-ID on the BicD2 C-terminus). In addition to known cargo interactors, they also identified many proteins in the HOPs complex. Interestingly, they find that the HOPs complex may interact with BicD2 in a different manner than other known cargos. The authors also show that while BicD2 is required for the HOPs complex localization, on average, depletion of BicD2 from HeLa and Cos7 cells causes HOPs and Lysosome mislocalization that is consistent with Kinesin-1 trafficking defects, rather than dynein. The authors also use proximity biotin ligase approaches to define the cargo interactome of three BicD2 variants associated with SMALED. One variant (R747C) has the most altered cargo interactome. The authors highlight one protein, in particular, GRAMD1A, that is only found in the R747C dataset and mislocalizes specifically when R747C is expressed.

The work in this manuscript is of a very high quality and contributes important findings to the field.

Comments on revisions:

The authors did a great job addressing the points I brought up!

Reviewer #2 (Public review):

Neiswender et al. investigated the interactomes between wild-type BICD2 and BICD2 mutants that are associated with Spinal Muscular Atrophy with Lower Extremity Predominance (SMALED2). Although BICD2 has previously been implicated in SMALED2, it is unclear how mutations in BICD2 may contribute to disease symptoms. In this study, the authors characterize the interactome of wild-type BICD2 and identify potential new cargos including the HOPS complex. The authors then chose three SMALED2-associated BICD2 mutants and compared each mutant interactome to that of wild-type BICD2. Each mutant had a change in the interactome, with the most drastic being BICD2_R747C, a mutation in the cargo binding domain of BICD2. This mutant displayed less interaction with a potential new BICD2 cargo, the HOPS complex. Additionally, it displayed more interaction with an ER protein, GRAMD1A.

The data in the paper is generally strong but the major conclusions of this paper need more evidence to be better supported.

(1) The authors use cells that have been engineered to express the different BICD2 constructs. As shown in Figure 4B, the authors see wide expression of BICD2_WT throughout the cell. However, WT BICD2 usually localizes to the TGN. This widespread localization introduces some uncertainty about the interactome data. The authors should either try to verify the interaction data (specifically with the HOPS complex and GRAMD1A) by immunoprecipitating endogenous BICD2 or by repeating their interactome experiment in Figure 1 using BICD2 knockout cells that express the BICD2_WT construct. This should also be done to verify the immunoprecipitation and microscopy data shown in Figure 7.

(2) The authors conclude that cargo transport defects resulting from BICD2 mutations may contribute to SMALED2 symptoms. However, the authors are unable to determine if BICD2 directly binds to the potential new cargo, the HOPS complex. To address this, the authors could purify full-length WT BICD2 and perform in vitro experiments. Furthermore, the authors were unable to identify the minimal region of BICD2 needed for HOPS interaction. The authors could expand on the experiment attempted with the extended BICD2 C-terminal using a deltaCC1 construct, which could also be used for in vitro experiments.

(3) Again, the authors conclude that BICD2 mutants cause cargo transport defects that are likely to lead to SMALED2 symptoms. This would be better supported if the authors are able to find a protein relevant to SMALED2 and examine if/how its localization is changed under expression of the BICD2 mutants. The authors currently use the HOPS complex and GRAMD1A as indicators of cargo transport defects, but it is unclear if these are relevant to SMALED2 symptoms.

Comments on revisions:

The investigators did a good job in responding to our initial concerns (see below). We appreciate that they used siRNA to address our first comment because they do not have a BICD2 KO cell line. We appreciated that they added a new section in the Discussion to address the limitations of the study.

In regards to our first comment about the BICD2 WT construct localization, since they use KD to validate the interaction between their BICD2 WT construct and VPS41, it would be nice to see localization of this construct under the KD condition. However, the binding they presented in Sup. Fig 1B does look convincing, so this may not be necessary.

Overall, I believe this revision has satisfied our previous concerns.

Reviewer #3 (Public review):

Summary:

BicD2 is a motor adapter protein that facilitates cellular transport pathways, which are impacted by human disease mutations of BicD2 causing spinal muscular atrophy with lower extremity dominance (SMALED2). The authors provide evidence that some of these mutations result in interactome changes, which may be the underlying cause of the disease. This is supported by proximity biotin ligation screens, immunoprecipitation and cell biology assays. The authors identify several novel BicD2 interactions such as the HOPS complex that participates in the fusion of late endosomes and autophagosomes with lysosomes, which could have important functions. Three BicD2 disease mutants studied had changes in the interactome, which could be an underlying cause for SMALED2. The study extends our understanding of the BicD2 interactome under physiological conditions, as well as of the changes of cellular transport pathways that result in SMALED2. It will be of great interest for the BicD2 and dynein fields.

Strengths:

Extensive interactomes are presented for both WT BicD2 as well as the disease mutants, which will be valuable for the community. The HOPS complex was identified as a novel interactor of BicD2, which is important for fusion of late endosomes and lysosomes, which is of interest, since some of the BicD2 disease mutations result in Golgi-fragmentation phenotypes. The interaction with the HOPS complex is affected by the R747C mutation, which also results in a gain of function interaction with GRAMD1A.

Weaknesses:

The manuscript should be strengthened by further evidence of the BicD2/HOPS complex interaction and the functional implications for spinal muscular atrophy by changes in the interactome through mutations. Which functional implications does the loss of the BicD2/HOPS complex interaction and the gain of function interaction with GRAMD1A have in the context of the R747C mutant?

Major points:

(1) In the biotin proximity ligation assay, a large number of targets were identified, but it is not clear why only the HOPS complex was chosen for further verification. Immunoprecipitation was used for target verification, but due to the very high number of targets identified in the screen, and the fact that the HOPS complex is a membrane protein that could potentially be immunoprecipitated along with lysosomes or dynein, additional experiments to verify the interaction of BicD2 with the HOPS complex (reconstitution of a complex in vitro, GST-pull down of a complex from cell extracts or other approaches) are needed to strengthen the manuscript.<br /> (2) In the biotin proximity ligation assay, a large number of BicD2 interactions were identified that are distinct between the mutant and the WT, but it was not clear why particularly GRAMD1A was chosen as gain of function interaction, and what the functional role of a BicD2/GRAMD1A interaction may be. A Western blot shows a strengthened interaction with the R747C mutant but GRAMD1A also interacts with WT BicD2.<br /> (3) Furthermore, functional implications of changed interactions with HOPS and GRAMD1A in the R747C mutant are unclear. Additional experiments are needed to establish the functional implication of the loss of the BicD2/HOPS interaction in the BicD2/R747C mutant. For the GRAMD1A gain of function interaction, according to the authors a significant amount of the protein localized with BicD2/R747C at the centrosomal region. This changed localization is not very clear from the presented images (no centrosomal or other markers were used, and the changed localization could also be an effect of dynein hyper activation in the mutant). Furthermore, the functional implication of a changed localization of GRAMD1A is unclear from the presented data.

Comments on revisions:

After a major revision, the manuscript is much improved. Additional evidence for the HOPS complex/BicD2 interaction was provided (the interaction was identified in multiple independent screens), and while the authors unfortunately were not able to confirm a direct interaction between BicD2 and the HOPS complex, additional caveats were added in the result section, which clearly state these limitations. The authors also included a very nice discussion of potential physiological roles of the GRAMD1A mislocalization in the disease mutant, which could potentially affect cholesterol transport and homostatis. Limitations of the presented approaches were clearly described as caveats.

Author response:

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

Reviewer #1 (Public review):

(1) I was surprised at the effect of BicD2 knockdown on LAMP (and VPS41) localization, which really suggests that in HeLa and Cos7 cells, BicD2 regulation of Kinesin-1 (rather than dynein) is the primary driver of lysosome localization. The KIF5B-knockout rescue of the BicD2overexpression phenotype was a very powerful result that supports this conclusion. Have the authors looked at other cargos, eg, Golgi or centrosomes in G2? Can the authors include more discussion about what this result means or how they imagine dynein and kinesin-1's interaction with BicD2 is regulated? 

We have performed this experiment as requested by the reviewer. The BICD2 siRNA also resulted in Golgi fragmentation and localization defects of the centrosome in cells that are in G2 phase of the cell cycle (Supplemental Fig. 2E-H).

We have also added additional discussion related to how BICD2 might couple cargos to opposite polarity motors (lines 440-447). Interestingly, the lysosome motility defect we observe upon BICD2 knock down has similarity to the RAB6A trafficking phenotype. In both cases, what one sees is a sharp reduction in the number of motile particles rather than a reversal in the direction of motility. This suggests that both motors are involved in the steady state distribution of these cargoes.

(2) Have the authors examined if the SMALED mutants show diminished or increased binding to KIF5B? While the authors are correct that the mutations could hyperactivate dynein because they reduce BicD2 autoinhibition, it is possible that the SMALED mutants hyperactivate dynein because they no longer bind kinesin. This would be particularly interesting, given the complex relationship between BicD2 regulation of dynein and kinesin that the authors show in Figure 3. 

Thank you for this suggestion. We had not considered this. We have added this experiment in the revised manuscript (Supplemental Fig. 3H, I). We find that the interaction between wild-type BICD2 and KIF5B is only slightly above the control. This is consistent with published findings that indicate that although the isolated CC2 domain of BICD2 is able to interact with KIF5B, the binding is lower for the full-length protein. This is most likely due to the intramolecular interaction between the N and C-termini of BICD2 partially blocking the binding site. Interestingly, however, all three mutants display a reduced interaction with KIF5B, with the reduction being most severe for the cargo domain binding mutants. Thus, as we discuss in the revised manuscript, dynein hyperactivity likely results from increased binding to dynein and a concurrent reduction in binding to KIF5B.

(3) What is already known about the protein GRAMD1A? Did the authors choose to focus on GRAMD1A because it was the only novel interaction found in the SMALED mutant interactomes, or was this protein interesting for a different reason? Does the known function of GRAMD1A explain the potential dysfunction of cells expressing BICD2_R747C or patients who have this mutation? More discussion of this protein and why the authors focused on it would really strengthen the manuscript. 

We chose to focus on GRAMD1A for a few reasons. The protein that displayed the highest gain of function interaction with BICD2_R747C in our proteomic analysis was Plastin. However, using at least one antibody against Plastin, we were not able to validate this result. In addition, we had previously performed a proteomic screen using a BICD2_R747A (arginine to alanine) mutation and had compared the interactome of this mutant to the wild-type protein. Plastin was not recovered in that screen but the top hit was GRAMD1A. Given that we isolated GRAMD1A in two separate screens as a gain of function interaction, we believed the result was worth focusing on for followup studies. 

GRAMD1A (as well as its paralogs GRAMD1B and C) function in non-vesicle transport of accessible cholesterol from the plasma membrane to the ER. We have added additional discussion on GRAMD1A (lines 484-495). While we observe a relocalization of GRAMD1A in mutant expressing cells, we do not know whether this is sufficient to result in cholesterol transport defects. There are several routes for cholesterol uptake, with the GRAMD1A pathway representing just one these routes. 

Reviewer #2 (Public review):

(1) The authors use cells that have been engineered to express the different BICD2 constructs. As shown in Figure 4B, the authors see wide expression of BICD2_WT throughout the cell. However, WT BICD2 usually localizes to the TGN. This widespread localization introduces some uncertainty about the interactome data. The authors should either try to verify the interaction data (specifically with the HOPS complex and GRAMD1A) by immunoprecipitating endogenous BICD2 or by repeating their interactome experiment in Figure 1 using BICD2 knockout cells that express the BICD2_WT construct. This should also be done to verify the immunoprecipitation and microscopy data shown in Figure 7. 

The localization of our exogenous BICD2-mNeon constructs is similar to what others have seen using GFP tagged versions of the protein (for example Peeters et al., 2013). In addition, in the experiment shown in the initial version of the paper, we were focusing on the centrosomal localization of BICD2. However, our BICD2-mNeon construct is also observed at the Golgi, in addition to its localization throughout the cell (Supplemental Fig. 3C). 

We attempted to perform a co-immunoprecipitation experiment using endogenous proteins as suggested by the reviewer. Although a rabbit polyclonal antibody was able to coimmunoprecipitate RANBP2 with BICD2, the antibody complex of heavy and light chains comigrated with the VPS41 band and was abundantly detected by the secondary antibody used in the western blot. Thus, we were not able to make a conclusion regarding whether or not VPS41 was present in the co-immunoprecipitate. We attempted the experiment using a mouse monoclonal antibody against BICD2. However, this antibody failed in the immunoprecipitation experiment and we could not detect either RANBP2 (a validated cargo) or VPS41. Although the VPS41 antibody we used in the paper works for western blot, it does not recognize the native protein. Thus, despite our best efforts, we are not able to draw a valid conclusion from these coip experiments.

It is beyond the scope of the revision to perform the entire experiment in a BICD2 KO cell line.  A BICD2 KO cell line does not exist and it would take several months to make such a knock out in the FLP IN HEK cells that were used in this manuscript. However, we have validated the interaction between BICD2 and VPS41 in cells that have been depleted of endogenous BICD2 (Supplemental Fig. 1B). The transgenic constructs contain silent mutations that make them refractory to bicD2 siRNA1. Thus, although endogenous BICD2 is depleted by the siRNA treatment, wild-type and mutant BICD2_TurboID is not. A similar approach was also used to demonstrate the gain of function interaction between BICD2_R747C and GRAMD1A in cells depleted of endogenous BICD2 (Supplemental Fig. 5A).

(2) The authors conclude that cargo transport defects resulting from BICD2 mutations may contribute to SMALED2 symptoms. However, the authors are unable to determine if BICD2 directly binds to the potential new cargo, the HOPS complex. To address this, the authors could purify full-length WT BICD2 and perform in vitro experiments. Furthermore, the authors were unable to identify the minimal region of BICD2 needed for HOPS interaction. The authors could expand on the experiment attempted with the extended BICD2 C-terminal using a deltaCC1 construct, which could also be used for in vitro experiments. 

We have not been successful in purifying full length BICD2 in bacteria, perhaps due to solubility issues. However, we have added several experiments to further examine the nature of the BICD2-HOPS complex interaction.

We have performed the experiment as requested. We find that BICD2_delCC1 is able to bind VPS41, but not as efficiently as the full length protein. However, unlike the CC3 cargo binding construct, the BICD2_delCC1 construct also displays reduced binding to RANBP2 (Supplemental Fig. 1D). We attribute this defect to either the intramolecular BICD2 interaction blocking cargo binding or potentially to a folding defect in the BICD2_delCC1 construct. Thus, although we performed this experiment as suggested by the reviewer, we are not able to make a solid conclusion.

Based on the fact that VPS41 was the most abundantly detected HOPS component in the BICD2 interactome, we hypothesized that it was the point of direct contact between BICD2 and the HOPS complex. However, contrary to our hypothesis, depletion of VPS41 did not compromise the association between BICD2 and VPS16 and VPS18 (Supplemental Fig. 1E). Thus, we conclude that there are multiple points of contact between BICD2 and the HOPS complex, with BICD2 perhaps recognizing a common motif or domain present in these proteins.

We next attempted to map the interaction site using Alphafold2 multimer. Although we were able to use this platform to predict a high confidence interaction between BICD2 and RAB6A (consistent with published results), this did not yield a high confidence prediction for the BICD2HOPS complex interaction.

Ultimately although we added several new experiments, we were not able to determine the minimal region for binding, nor whether the interaction is direct or indirect. These caveats are clearly stated in the revised manuscript. Regardless of whether the interaction is direct or indirect however, it is noteworthy that the association between BICD2 and the HOPS complex is reduced by the R747C SMALED2 mutation.

(3) Again, the authors conclude that BICD2 mutants cause cargo transport defects that are likely to lead to SMALED2 symptoms. This would be better supported if the authors are able to find a protein relevant to SMALED2 and examine if/how its localization is changed under expression of the BICD2 mutants. The authors currently use the HOPS complex and GRAMD1A as indicators of cargo transport defects, but it is unclear if these are relevant to SMALED2 symptoms. 

This point was addressed in the general discussion. Given the complexity of SMALED2 (autosomal dominant disorder; variable phenotypic severity; adult onset disorder in many instances, etc.) it is very hard to model in a cell line. One of the reasons we focused our studies on the HOPS complex and VPS41 in particular was because mutations in VPS41 are associated with spinocerebellar ataxia, a neurodevelopment disorder. However, we cannot conclude whether the reduction/loss of interaction of BICD2 with the HOPS complex is causative for disease symptoms. We also cannot conclude at present whether the mis-targeting of GRAMD1A is causative for disease symptoms. We have discussed these caveats in the revised manuscript and have included a section in the discussion that specifically lists the limitations of our study (lines 511-530).

With that said, we can conclude that mutations in the cargo binding domain of BICD2 result in dynein hyperactivity, altered BICD2 localization in hippocampal neurons, and reduced neurite growth. Given that we observe interactome changes in HEK cells, it is plausible that interactome changes also exist in motor neurons. However, even in the absence of interactome changes, hyperactivation of dynein alone can result in cargo trafficking defects; the same cargos can be excessively localized in the soma vs the axon. As noted previously, however, a thorough examination of these points will require the use of genetically engineered motor neurons and is beyond the scope of the current study.

Reviewer #3 (Public review):

Strengths: 

Extensive interactomes are presented for both WT BicD2 as well as the disease mutants, which will be valuable for the community. The HOPS complex was identified as a novel interactor of BicD2, which is important for fusion of late endosomes and lysosomes, which is of interest, since some of the BicD2 disease mutations result in Golgi-fragmentation phenotypes. The interaction with the HOPS complex is affected by the R747C mutation, which also results in a gain-of-function interaction with GRAMD1A. 

Weaknesses: 

The manuscript should be strengthened by further evidence of the BicD2/HOPS complex interaction and the functional implications for spinal muscular atrophy by changes in the interactome through mutations. Which functional implications does the loss of the BicD2/HOPS complex interaction and the gain of function interaction with GRAMD1A have in the context of the R747C mutant? 

(1) In the biotin proximity ligation assay, a large number of targets were identified, but it is not clear why only the HOPS complex was chosen for further verification. Immunoprecipitation was used for target verification, but due to the very high number of targets identified in the screen, and the fact that the HOPS complex is a membrane protein that could potentially be immunoprecipitated along with lysosomes or dynein, additional experiments to verify the interaction of BicD2 with the HOPS complex (reconstitution of a complex in vitro, GST-pull down of a complex from cell extracts or other approaches) are needed to strengthen the manuscript. 

As discussed for reviewer 2 (point 2), we have added several experiments to better characterize the BICD2-HOPS complex interaction.

We chose to focus on the HOPS complex for a few reasons. The list of interactions that displayed a >2 fold enrichment vs control was actually not that large (66 proteins). Within this list, we identified 4 out of 6 HOPS components and VPS41 was the 5th most enriched protein in the BICD2 interactome (RANBP2 by contrast was #16 on this list). Furthermore, the BICD2_R747C mutation resulted in greatly reduced interaction of BICD2 with the HOPS complex, whereas its interaction with dynein was increased. These results indicate that these proteins are not simply immunoprecipitating with the BICD2/dynein complex. Apart from the HOPS complex, lysosomal proteins were not present in the interactome, making it unlikely that they were identified due to non-specific interactions between BICD2 and co-precipitating lysosomes.

(2) In the biotin proximity ligation assay, a large number of Bi cD2 interactions were identified that are distinct between the mutant and the WT, but it was not clear why, particularly GRAMD1A was chosen as a gain-of-function interaction, and what the functional role of a BicD2/GRAMD1A interaction may be. A Western blot shows a strengthened interaction with the R747C mutant, but GRAMD1A also interacts with WT BicD2. 

Please see the above discussion on GRAMD1A (reviewer 1, point 3). GRAMD1A comes down non-specifically with the binding control as well as BICD2_wt. We therefore conclude that wildtype BICD2 does not specifically interact with GRAMD1A above background levels (Fig. 7, compare the control lane vs BICD2-wt).

(3) Furthermore, the functional implications of changed interactions with HOPS and GRAMD1A in the R747C mutant are unclear. Additional experiments are needed to establish the functional implication of the loss of the BicD2/HOPS interaction in the BicD2/R747C mutant. For the GRAMD1A gain of function interaction, according to the authors, a significant amount of the protein localized with BicD2/R747C at the centrosomal region. This changed localization is not very clear from the presented images (no centrosomal or other markers were used, and the changed localization could also be an effect of dynein hyperactivation in the mutant). Furthermore, the functional implication of a changed localization of GRAMD1A is unclear from the presented data. 

We have performed the experiment as requested by the reviewer. The re-localized GRAMD1A localizes adjacent to Pericentrin, a centrosomal marker (Supplemental Fig. 5B-F). GRAMD1A and BICD2 appear to co-localize in a ring around the Pericentrin marked centrosome.

The re-localization of GRAMD1A to the centrosomal area by BICD2_R747C appears to be unique to this mutant, and not simply an issue of dynein hyperactivity. The other two mutants tested, BICD2_N188T and BICD2_R694C also hyperactivate dynein. However, they do not result in the same type of dramatic re-localization of GRAMD1A as we observe with the BICD2_R747C mutant. We conclude that this altered localization results from a gain of function interaction with BICD2_R747C as well as dynein hyperactivity.

Reviewer #1 (Recommendations for the authors): 

Please add a discussion about how the authors calculated the Cell Body enrichment shown in 5E. Is this a ratio of the BicD2 intensity in the cell body:axon? Did the authors normalize for potential differences in BicD2 variant expression? 

Yes, it is a ratio of the intensity between the cell body and axon. This is described in the Methods section under quantification (lines 725-728). We attempted to image cells expressing similar amounts of protein.  

Reviewer #2 (Recommendations for the authors): 

(1) The paper would benefit from an explanation of why the authors chose to follow up on the HOPS complex out of all proteins identified in the interactome experiment. 

This discussion has been included in the revised manuscript.  

(2) In panel B of Supplementary Figure 1, RFP mTurbo has a significant amount of non-specific binding to VPS18. The authors note that in the initial interactome experiment, there was a twofold enrichment of this protein in BICD2 pulldown versus control. Do the authors have a co-IP that has a similar enrichment?

VPS18 occasionally comes down non-specifically with our RFP-TurboID control. However, the interaction is specific, because very little VPS18 comes down with the BICD2 construct lacking the cargo binding domain (Fig. 2B). An additional example of the VPS18 binding result is shown in Supplemental Fig. 1E.

(3) In Figure 2B, there seems to be less Vps18 in the input for BICD2 delCC3-mTrbo. Do the authors have a blot where there is equal input across all conditions? This may increase the slight signal seen in the pulldown.

The blot shown in Supplemental Fig. 1C has equivalent load for VPS18 across all lanes. Minimal binding of VPS18 is observed with the BICD2_delCC3 sample.

(4) In Figure 3, can the authors show representative images of GFP-VPS-41 and LAMP1 localization that are at the same magnification? It currently looks as if the localization pattern differs between the two under control siRNA. Alternatively, the authors should show colocalization of the two, as the authors note both are localized to late endosomes/lysosomes. 

We have provided additional images that are at the same magnification (Supplemental Fig. 2IK). Co-localization between GFP-VPS41 (rabbit polyclonal antibody against GFP) and LAMP1 (rabbit polyclonal antibody) is not possible. However, published studies have shown that a subset of V5 tagged VPS41 vesicles are positive for LAMP1. We have cited this study.

(5) In Supplementary Figure 2, the authors should show the knockdown efficiency of both BICD2 siRNAs. The VPS41 staining in panel B looks like there is less perinuclear localization than with BICD2 siRNA 1. Is the because of knockdown efficiency? 

We have included this data (Supplemental Fig. 2B). Both siRNAs are capable of depleting BICD2. However, we do see slightly more effective knock down with siRNA-1.

(6) The data in Figure 4A would be more striking with quantification. 

Quantifications have been provided (Supplemental Fig. 3A,B). Using a one-way Anova analysis, BICD2_R747C is the only mutant that shows significance. Variability in the binding experiment resulted in the other two mutants not showing a statistically significant change. However, the additional assays that are provided (centrosomal enrichment of BICD2 and peroxisome tethering) clearly demonstrate that the R694C mutant also results in dynein hyperactivation. It should be noted that the analysis done by Huynh et al., 2017 also showed a binding increase between BICD2 disease mutants and dynein. However, due to binding variability, their results were not not statistically significant.

(7) Can the authors explain how centrosome enrichment is calculated in Figure 4F? The intensity of colocalization with the centrosome between mutant constructs visually does not look significantly different. Is this a ratio of centrosome localization to cell body localization? 

We apologize for this omission. This has been added to the quantification section of the Methods (lines 721-723). Yes, it is a ratio of mean signal at the centrosome vs mean signal in the rest of the cell.

(8) The current input blot in Supplementary Figure 4A shows increasing amounts of importin beta across the lanes. Do the authors have a blot of panel A in which the input level of importin beta is the same between constructs? Does this change the level of importin beta that is pulled down?

Another replicate of this experiment has been shown. We have retained the original experiment as well (Supplemental Figs. 4A, B).

Reviewer #3 (Recommendations for the authors): 

Minor points: 

(1) In the .pdf version of the supplemental tables, the text is often cropped. It is recommended to delete the .pdf versions and just retain the Excel versions of the tables. 

We are not sure why this occurred. Excel files were provided. In addition, the raw data from the mass spectrometry experiments will also be included with the final version of the manuscript.

(2) Line 367: For transport of Rab6, kinesin-1 is the dominant motor, but dynein is still active and engaging in a tug of war (Serra Marquez et al 2022). 

Thank you. We have revised our text to include this discussion. In this regard, LAMP1 vesicles are similar. Loss of BICD2 results in a greater number of stationary vesicles rather than vesicles that are excessively targeted towards the microtubules minus end.

(3) Line 371: BicD2 is required for the transport of RanBP2 from annulate lamellae to nuclear pore complexes.

Thank you. We have modified our text. 

(4) Yi et al., 2023 have previously shown changed interactions of the BicD2/R747C mutant, such as decreased binding to Nup358 and increased binding to Nesprin-2, as well as functional implications for the associated brain developmental pathways, which should be acknowledged.

We apologize for leaving this out. In the original version of the manuscript, we were attempting to keep the discussion more concise. We have added a discussion of these findings in the revised manuscript (lines 496-507).

Author response:

Reviewer #1 (Public review):

The study examines how pyruvate, a key product of glycolysis that influences TCA metabolism and gluconeogenesis, impacts cellular metabolism and cell size. It primarily utilizes the Drosophila liver-like fat body, which is composed of large post-mitotic cells that are metabolically very active. The study focuses on the key observations that over-expression of the pyruvate importer MPC complex (which imports pyruvate from the cytoplasm into mitochondria) can reduce cell size in a cell-autonomous manner. They find this is by metabolic rewiring that shunts pyruvate away from TCA metabolism and into gluconeogenesis. Surprisingly, mTORC and Myc pathways are also hyper-active in this background, despite the decreased cell size, suggesting a non-canonical cell size regulation signaling pathway. They also show a similar cell size reduction in HepG2 organoids. Metabolic analysis reveals that enhanced gluconeogenesis suppresses protein synthesis. Their working model is that elevated pyruvate mitochondrial import drives oxaloacetate production and fuels gluconeogenesis during late larval development, thus reducing amino acid production and thus reducing protein synthesis.

Strengths:

The study is significant because stem cells and many cancers exhibit metabolic rewiring of pyruvate metabolism. It provides new insights into how the fate of pyruvate can be tuned to influence Drosophila biomass accrual, and how pyruvate pools can influence the balance between carbohydrate and protein biosynthesis. Strengths include its rigorous dissection of metabolic rewiring and use of Drosophila and mammalian cell systems to dissect carbohydrate:protein crosstalk.

Weaknesses:

However, questions on how these two pathways crosstalk, and how this interfaces with canonical Myc and mTORC machinery remain. There are also questions related to how this protein:carbohydrate crosstalk interfaces with lipid biosynthesis. Addressing these will increase the overall impact of the study.

We thank the reviewer for recognizing the significance of our work and for providing constructive feedback. Our findings indicate that elevated pyruvate transport into mitochondria acts independently of canonical pathways, such as mTORC1 or Myc signaling, to regulate cell size. To investigate these pathways, we utilized immunofluorescence with well-validated surrogate measures (p-S6 and p-4EBP1) in clonal analyses of MPC expression, as well as RNA-seq analyses in whole fat body tissues expressing MPC. These methods revealed hyperactivation of mTORC1 and Myc signaling in fat body cells expressing MPC in Drosophila, which are dramatically smaller than control cells. One explanation of these seemingly contradictory observations could be an excess of nutrients that activate mTORC1 or Myc pathways. However, our data is inconsistent with a nutrient surplus that could explain this hyperactivation. Instead, we observed reduced amino acid abundance upon MPC expression, which is very surprising given the observed hyperactivation of mTORC1. This led us to hypothesize the existence of a feedback mechanism that senses inappropriate reductions in cell size and activates signaling pathways to promote cell growth. The best characterized “sizer” pathway for mammalian cells is the CycD/CDK4 complex which has been well studied in the context of cell size regulation of the cell cycle (PMID 10970848, 34022133). However, the mechanisms that sense cell size in post-mitotic cells, such as fat body cells and hepatocytes, remain poorly understood. Investigating the hypothesized size-sensing mechanisms at play here is a fascinating direction for future research.

For the current study, we conducted epistatic analyses with mTOR pathway members by overexpressing PI3K and knocking down the TORC1 inhibitor Tuberous Sclerosis Complex 1 (Tsc1). These manipulations increased the size of control fat body cells but not those over-expressing the MPC (Supplementary Fig. 3c, 3d). Regarding Myc, its overexpression increased the size of both control and MPC+ clones (Supplementary Fig. 3e), but Myc knockdown had no additional effect on cell size in MPC+ clones (Supplementary Fig. 3f). These results suggest that neither mTORC1, PI3K, nor Myc are epistatic to the cell size effects of MPC expression. Consequently, we shifted our focus to metabolic mechanisms regulating biomass production and cell size.

When analyzing cellular biomolecules contributing to biomass, we observed a significant impact on protein levels in Drosophila fat body cells and mammalian MPC-expressing HepG2 spheroids. TAG abundance in MPC-expressing HepG2 spheroids and whole fat body cells showed a statistically insignificant decrease compared to controls. Furthermore, lipid droplets in fat body cells were comparable in MPC-expressing clones when normalized to cell size.

Interestingly, RNA-seq analysis revealed increased expression of fatty acid and cholesterol biosynthesis pathways in MPC-expressing fat body cells. Upregulated genes included major SREBP targets, such as ATPCL (2.08-fold), FASN1 (1.15-fold), FASN2 (1.07-fold), and ACC (1.26-fold). Since mTOR promotes SREBP activation and MPC-expressing cells showed elevated mTOR activity and upregulation of SREBP targets, we hypothesize that SREBP is activated in these cells. Nonetheless, our data on amino acid abundance and its impact on protein synthesis activity suggest that protein abundance, rather than lipids, is likely to play a larger causal role in regulating cell size in response to increased pyruvate transport into mitochondria.

Reviewer #2 (Public review):

In this manuscript, the authors leverage multiple cellular models including the drosophila fat body and cultured hepatocytes to investigate the metabolic programs governing cell size. By profiling gene programs in the larval fat body during the third instar stage - in which cells cease proliferation and initiate a period of cell growth - the authors uncover a coordinated downregulation of genes involved in mitochondrial pyruvate import and metabolism. Enforced expression of the mitochondrial pyruvate carrier restrains cell size, despite active signaling of mTORC1 and other pathways viewed as traditional determinants of cell size. Mechanistically, the authors find that mitochondrial pyruvate import restrains cell size by fueling gluconeogenesis through the combined action of pyruvate carboxylase and phosphoenolpyruvate carboxykinase. Pyruvate conversion to oxaloacetate and use as a gluconeogenic substrate restrains cell growth by siphoning oxaloacetate away from aspartate and other amino acid biosynthesis, revealing a tradeoff between gluconeogenesis and provision of amino acids required to sustain protein biosynthesis. Overall, this manuscript is extremely rigorous, with each point interrogated through a variety of genetic and pharmacologic assays. The major conceptual advance is uncovering the regulation of cell size as a consequence of compartmentalized metabolism, which is dominant even over traditional signaling inputs. The work has implications for understanding cell size control in cell types that engage in gluconeogenesis but more broadly raise the possibility that metabolic tradeoffs determine cell size control in a variety of contexts.

We thank the reviewer for their thoughtful recognition of our efforts, and we are honored by the enthusiasm the reviewer expressed for the findings and the significance of our research. We share the reviewer’s opinion that our work might help to unravel metabolic mechanisms that regulate biomass gain independent of the well-known signaling pathways.

Reviewer #3 (Public review):

Summary:

In this article, Toshniwal et al. investigate the role of pyruvate metabolism in controlling cell growth. They find that elevated expression of the mitochondrial pyruvate carrier (MPC) leads to decreased cell size in the Drosophila fat body, a transformed human hepatocyte cell line (HepG2), and primary rat hepatocytes. Using genetic approaches and metabolic assays, the authors find that elevated pyruvate import into cells with forced expression of MPC increases the cellular NADH/NAD+ ratio, which drives the production of oxaloacetate via pyruvate carboxylase. Genetic, pharmacological, and metabolic approaches suggest that oxaloacetate is used to support gluconeogenesis rather than amino acid synthesis in cells over-expressing MPC. The reduction in cellular amino acids impairs protein synthesis, leading to impaired cell growth.

Strengths:

This study shows that the metabolic program of a cell, and especially its NADH/NAD+ ratio, can play a dominant role in regulating cell growth.

The combination of complementary approaches, ranging from Drosophila genetics to metabolic flux measurements in mammalian cells, strengthens the findings of the paper and shows a conservation of MPC effects across evolution.

Weaknesses:

In general, the strengths of this paper outweigh its weaknesses. However, some areas of inconsistency and rigor deserve further attention.

Thank you for reviewing our manuscript and offering constructive feedback. We appreciate your recognition of the significance of our work and your acknowledgment of the compelling evidence we have presented. We will carefully revise the manuscript in line with the reviewers' recommendations.

The authors comment that MPC overrides hormonal controls on gluconeogenesis and cell size (Discussion, paragraph 3). Such a claim cannot be made for mammalian experiments that are conducted with immortalized cell lines or primary hepatocytes.

We appreciate the reviewer’s insightful comment. Pyruvate is a primary substrate for gluconeogenesis, and our findings suggest that increased pyruvate transport into mitochondria increases the NADH-to-NAD+ ratio, and thereby elevates gluconeogenesis. Notably, we did not observe any changes in the expression of key glucagon targets, such as PC, PEPCK2, and G6PC, suggesting that the glucagon response is not activated upon MPC expression. By the statement referenced by the reviewer, we intended to highlight that excess pyruvate import into mitochondria drives gluconeogenesis independently of hormonal and physiological regulation.

It seems the reviewer might also have been expressing the sentiment that our in vitro models may not fully reflect the in vivo situation, and we completely agree.  Moving forward, we plan to perform similar analyses in mammalian models to test the in vivo relevance of this mechanism. For now, we will refine the language in the manuscript to clarify this point.

Nuclear size looks to be decreased in fat body cells with elevated MPC levels, consistent with reduced endoreplication, a process that drives growth in these cells. However, acute, ex vivo EdU labeling and measures of tissue DNA content are equivalent in wild-type and MPC+ fat body cells. This is surprising - how do the authors interpret these apparently contradictory phenotypes?

We thank the reviewer for raising this important issue. The size of the nucleus is regulated by DNA content and various factors, including the physical properties of DNA, chromatin condensation, the nuclear lamina, and other structural components (PMID 32997613). Additionally, cytoplasmic and cellular volume also impacts nuclear size, as extensively documented during development (PMID 17998401, PMID 32473090).

In MPC-expressing cells, it is plausible that the reduced cellular volume impacts chromatin condensation or the nuclear lamina in a way that slightly decreases nuclear size without altering DNA content. Specifically, in our whole fat body experiments using CG-Gal4 (as shown in Supplementary Figure 2a-c), we noted that after 12 hours of MPC expression, cell size was significantly reduced (Supplementary Figure 2c and Author response image 1A). However, the reduction in nuclear size became significant only after 36 hours of MPC expression (Author response image 1B), suggesting that the reduction in cell size is a more acute response to MPC expression, followed only later by effects on nuclear size.

In clonal analyses, this relationship was further clarified. MPC-expressing cells with a size greater than 1000 µm² displayed nuclear sizes comparable to control cells, whereas those with a drastic reduction in cell size (less than 1000 µm²) exhibited smaller nuclei (Author response image 1C and D). These observations collectively suggest that changes in nuclear size are more likely to be downstream rather than upstream of cell size reduction. Given that DNA content remains unaffected, we focused on investigating the rate of protein synthesis. Our findings suggest that protein synthesis might play a causal role in regulating cell size, thereby reinforcing the connection between cellular and nuclear size in this context.

Author response image 1.

Cell Size vs. Nuclear Size in MPC-Expressing Fat Body Cells. A. Cell size comparison between control (blue, ay-GFP) and MPC+ (red, ay-MPC) fat body cells over time, measured in hours after MPC expression induction. B. Nuclear area measurements from the same fat body cells in ay-GFP and ay-MPC groups. C. Scatter plot of nuclear area vs. cell area for control (ay-GFP) cells, including the corresponding R^² value. D. Scatter plot of nuclear area vs. cell area for MPC-expressing (ay-MPC) cells, with the respective R^² value.

This image highlights the relationship between nuclear and cell size in MPC-expressing fat body cells, emphasizing the distinct cellular responses observed following MPC induction.

In Figure 4d, oxygen consumption rates are measured in control cells and those over-expressing MPC. Values are normalized to protein levels, but protein is reduced in MPC+ cells. Is oxygen consumption changed by MPC expression on a per-cell basis?

As described in the manuscript, MPC-expressing cells are smaller in size. In this context, we felt that it was most appropriate to normalize oxygen consumption rates (OCR) to cellular mass to enable an accurate interpretation of metabolic activity. Therefore, we normalized OCR with protein content to account for variations in cellular size and (probably) mitochondrial mass.

Trehalose is the main circulating sugar in Drosophila and should be measured in addition to hemolymph glucose. Additionally, the units in Figure 4h should be related to hemolymph volume - it is not clear that they are.

We appreciate this valuable suggestion. In the revised manuscript, we will quantify trehalose abundance in circulation and within fat bodies. As described in the Methods section, following the approach outlined in Ugrankar-Banerjee et al., 2023, we bled 10 larvae (either control or MPC-expressing) using forceps onto parafilm. From this, 2 microliters of hemolymph were collected for glucose measurement. We will apply this methodology to include the trehalose measurements as part of our updated analysis.

Measurements of NADH/NAD ratios in conditions where these are manipulated genetically and pharmacologically (Figure 5) would strengthen the findings of the paper. Along the same lines, expression of manipulated genes - whether by RT-qPCR or Western blotting - would be helpful to assess the degree of knockdown/knockout in a cell population (for example, Got2 manipulations in Figures 6 and S8).

We appreciate this suggestion, which will provide additional rigor to our study. We have already quantified NADH/NAD+ ratios in HepG2 cells under UK5099, NMN, and Asp supplementation, as presented in Figure 6k. As suggested, we will quantify the expression of Got2 manipulations mentioned in Figure 6j using RT-qPCR and validate the corresponding data in Supplementary Figure 8f through western blot analysis.

Additionally, we will assess the efficiency of pcb, pdha, dlat, pepck2, and Got2 manipulations used to modulate the expression of these genes. These validations will ensure the robustness of our findings and strengthen the conclusions of our study.

Author response:

Reviewer #1:

Weaknesses:

(1) The crystal structure of HsIFT172c reveals a single globular domain formed by the last three TPR repeats and C-terminal residues of IFT172. However, the authors subdivide this globular domain into TPR, linker, and U-box-like regions that they treat as separate entities throughout the manuscript. This is potentially misleading as the U-box surface that is proposed to bind ubiquitin or E2 is not surface accessible but instead interacts with the TPR motifs. They justify this approach by speculating that the presented IFT172c structure represents an autoinhibited state and that the U-box-like domain can become accessible following phosphorylation. However, additional evidence supporting the proposed autoinhibited state and the potential accessibility of the U-box surface following phosphorylation is needed, as it is not tested or supported by the current data.

We thank the reviewer for this comment. IFT172C contains TPR region and Ubox-like region which are admittedly tightly bound to each other. While there is a possibility that this region functions and exists as one domain, below are the reasons why we chose to classify these regions as two different domains.

(1) TPR and Ubox-like regions are two different structural classes

(2) TPR region is linked to Ubox-like region via a long linker which seems poised to regulate the relative movement between these regions.

(3) Many ciliopathy mutations are mapped to the interface of TPR region and the Ubox region hinting at a regulatory mechanism governed by this interface.

(2) While in vitro ubiquitination of IFT172 has been demonstrated, in vivo evidence of this process is necessary to support its physiological relevance.

We thank the reviewer for this comment. We are currently working on identifying the substrates of IF172 to reveal the physiological relevant of its ubiquitination activity.

(3) The authors describe IFT172 as being autoubiquitinated. However, the identified E2 enzymes UBCH5A and UBCH5B can both function in E3-independent ubiquitination (as pointed out by the authors) and mediate ubiquitin chain formation in an E3-independent manner in vitro (see ubiquitin chain ladder formation in Figure 3A). In addition, point mutation of known E3-binding sites in UBCH5A or TPR/U-box interface residues in IFT172 has no effect on the mono-ubiquitination of IFT172c1. Together, these data suggest that IFT172 is an E3-independent substrate of UBCH5A in vitro. The authors should state this possibility more clearly and avoid terminology such as "autoubiquitination" as it implies that IFT172 is an E3 ligase, which is misleading. Similarly, statements on page 10 and elsewhere are not supported by the data (e.g. "the low in vitro ubiquitination activity exhibited by IFT172" and "ubiquitin conjugation occurring on HsIFT172C1 in the presence of UBCH5A, possibly in coordination with the IFT172 U-box domain").

We now consider this possibility and tone down our statements about the autoubiquitination activity of IFT172 in a revised version of the manuscript.

(4) Related to the above point, the conclusion on page 11, that mono-ubiquitination of IFT172 is U-box-independent while polyubiquitination of IFT172 is U-box-dependent appears implausible. The authors should consider that UBCH5A is known to form free ubiquitin chains in vitro and structural rearrangements in F1715A/C1725R variants could render additional ubiquitination sites or the monoubiquitinated form of IFT172 inaccessible/unfavorable for further processing by UBCH5A.

We now consider this possibility and tone down our statements about the autoubiquitination activity of IFT172 in the conclusion on pg. 11.

(5) Identification of the specific ubiquitination site(s) within IFT172 would be valuable as it would allow targeted mutation to determine whether the ubiquitination of IFT172 is physiologically relevant. Ubiquitination of the C1 but not the C2 or C3 constructs suggests that the ubiquitination site is located in TPRs ranging from residues 969-1470. Could this region of TPR repeats (lacking the IFT172C3 part) suffice as a substrate for UBCH5A in ubiquitination assays?

We thank the reviewer for raising this important point about ubiquitination site identification. While not included in our manuscript, we did perform mass spectrometry analysis of ubiquitination sites using wild-type IFT172 and several mutants (P1725A, C1727R, and F1715A). As shown in the figure below, we detected multiple ubiquitination sites across these constructs. The wild-type protein showed ubiquitination at positions K1022, K1237, K1271, and K1551, while the mutants displayed slightly different patterns of modification. However, we should note that the MS intensity signals for these ubiquitinated peptides were relatively low compared to unmodified peptides, making it difficult to draw strong conclusions about site specificity or physiological relevance.

Author response image 1.

These results align with the reviewer's suggestion that ubiquitination occurs within the TPR-containing region. However, given the technical limitations of the MS analysis and the potential for E3-independent ubiquitination by UBCH5A, we have taken a conservative approach in interpreting these findings.

(6) The discrepancy between the molecular weight shifts observed in anti-ubiquitin Western blots and Coomassie-stained gels is noteworthy. The authors show the appearance of a mono-ubiquitinated protein of ~108 kDa in anti-ubiquitin Western blots. However, this molecular weight shift is not observed for total IFT172 in the corresponding Coomassie-stained gels (Figures 3B, D, F). Surprisingly, this MW shift is visible in an anti-His Western blot of a ubiquitination assay (Fig 3C). Together, this raises the concern that only a small fraction of IFT172 is being modified with ubiquitin. Quantification of the percentage of ubiquitinated IFT172 in the in vitro experiments could provide helpful context.

We do acknowledge in the manuscript is that the conjugation of ubiquitins to IFT172C is weak (Page 16). Future experiments of identification of potential substrates and its implications in ciliary regulation will provide further context to our in vitro ubiquitination experiments.

(7) The authors propose that IFT172 binds ubiquitin and demonstrate that GST-tagged HsIFT172C2 or HsIFT172C3 can pull down tetra-ubiquitin chains. However, ubiquitin is known to be "sticky" and to have a tendency for weak, nonspecific interactions with exposed hydrophobic surfaces. Given that only a small proportion of the ubiquitin chains bind in the pull-down, specific point mutations that identify the ubiquitin-binding site are required to convincingly show the ubiquitin binding of IFT172.

(8) The authors generated structure-guided mutations based on the predicted Ub-interface and on the TPR/U-box interface and used these for the ubiquitination assays in Fig 3. These same mutations could provide valuable insights into ubiquitin binding assays as they may disrupt or enhance ubiquitin binding (by relieving "autoinhibition"), respectively. Surprisingly, two of these sites are highlighted in the predicted ubiquitin-binding interface (F1715, I1688; Figure 4E) but not analyzed in the accompanying ubiquitin-binding assays in Figure 4.

We agree that these mutations could provide insights into ubiquitin binding by IFT172. We are currently pursuing further mutagenesis studies on the IFT172-Ub interface based on the AF model. We however have evaluated the ubiquitin binding activity of the mutant F1715A using similar pulldowns, which showed no significant impact for the mutation on the ubiquitin binding activity of IFT172. We are yet to evaluate the impact of alternate amino acid substitutions at these positions. The I1688 mutants we cloned could not be expressed in soluble form, thus could not be used for testing in ubiquitination activity or ubiquitin binding assays.

(9) If IFT172 is a ubiquitin-binding protein, it might be expected that the pull-down experiments in Figure S1 would identify ubiquitin, ubiquitinated proteins, or E2 enzymes. These were not observed, raising doubt that IFT172 is a ubiquitin-binding protein.

It is likely that IFT172 only binds ubiquitin with low affinity as indicated by our in vitro pulldowns and the AF interface. In our pull down experiment performed using the Chlamy flagella extracts, we have used extensive washes to remove non-specific interactors. This might have also excluded the identification of weak but bona fide interactors of IFT172. Additionally, we have not used any ubiquitination preserving reagents such as NEM in our pulldown buffers, exposing the cellular ubiquitinated proteins to DUB mediated proteolysis further preventing their identification in our pulldown/MS experiment.

(10) The cell-based experiments demonstrate that the U-box-like region is important for the stability of IFT172 but does not demonstrate that the effect on the TGFb pathway is due to the loss of ubiquitin-binding or ubiquitination activity of IFT172.

We acknowledge that our current data cannot distinguish whether the TGFβ pathway defects arise from general protein instability or from specific loss of ubiquitin-related functions. Our experiments demonstrate that the U-box-like region is required for both IFT172 stability and proper TGFβ signaling, but we agree that establishing a direct mechanistic link between these phenomena would require additional evidence. We will revise our discussion to more clearly acknowledge this limitation in our current understanding of the relationship between IFT172's U-box region and TGFβ pathway regulation.

(11) The challenges in experimentally validating the interaction between IFT172 and the UBX-domain-containing protein are understandable. Alternative approaches, such as using single domains from the UBX protein, implementing solubilizing tags, or disrupting the predicted binding interface in Chlamydomonas flagella pull-downs, could be considered. In this context, the conclusion on page 7 that "The uncharacterized UBX-domain-containing protein was validated by AF-M as a direct IFT172 interactor" is incorrect as a prediction of an interaction interface with AF-M does not validate a direct interaction per se.

We agree with the reviewer that our AlphaFold-Multimer (AF-M) predictions alone do not constitute experimental validation of a direct interaction. We appreciate the reviewer's understanding of the technical challenges in validating this interaction experimentally. We will revise our text to more precisely state that "The uncharacterized UBX-domain-containing protein was validated by AF-M as a potential direct IFT172 interactor" and will discuss the AF-M predictions as computational evidence that suggests, but does not prove, a direct interaction. This more accurately reflects the current state of our understanding of this potential interaction.

Reviewer #3:

Weaknesses:

(1) Interaction studies were carried out by pulldown experiments, which identified more IFT172 interaction partners. Whether these interactions can be seen in living cells remains to be elucidated in subsequent studies.

We agree with the reviewer that validation of protein-protein interactions in living cells provides important physiological context. While our pulldown experiments have identified several promising interaction partners and the AF-M predictions provide computational support for these interactions, we acknowledge that demonstrating these interactions in vivo would strengthen our findings. However, we believe our current biochemical and structural analyses provide valuable insights into the molecular basis of IFT172's interactions, laying important groundwork for future cell-based studies.

(2) The cell culture-based experiments in the IFT172 mutants are exciting and show that the U-box domain is important for protein stability and point towards involvement of the U-box domain in cellular signaling processes. However, the characterization of the generated cell lines falls behind the very rigorous analysis of other aspects of this work.

We thank the reviewer for noting that the characterization of our cell lines could be more rigorous. In the revised manuscript, we will provide additional characterization of the cell lines, including detailed sequencing information and validation data for the IFT172 mutants. This will bring the documentation of our cell-based experiments up to the same standard as other aspects of our work.

Author response:

We thank the reviewers for their help and their suggestions to make this manuscript more rigorous. We would like to post provisional author responses when eLife publish the reviewed preprint, and the more detailed responses will be supplemented with the revised manuscript.

• There are questions about choices made in the computational approach (architecture and type of generative model, training set).

We will train a new generator model based on the current GAN architecture, but with ‘hybrid’ AMP/AVP training sets (Reviewer 1 and 3). Hence, we can directly compare the performances of two generators. Based on our preliminary data, providing GAN with more AVP sequences during training helped the designed peptides pass the AVP filter, at the cost of reducing the average AMPredicgtor scores. The new generator also elevated the diversity of designed sequences.

We also perturbed the detailed architecture of our deep learning models, including fully-connected graph edge encodings and different versions of ESM (e.g. esm1b_t33_650M_UR50S, esm2_t48_15B_UR50D, Reviewer 2). In the revised manuscript, we will report the effects of these modifications and suggest the overall construct of GCN and GAN are suitable for a light-weight sequence label model, as demonstrated in Author response table 1 and 2. For the generator, we suggest that using our approach, we may have reached a plateau for the GAN sampling (Author response table 3).

Author response table 1.

Results of AMPredictor with different graph edge encodings

Author response table 2.

Results of AMPredictor with different ESM versions

Author response table 3.

Evaluation of generated sequences with different sampling numbers

• There is an important concern about the small number of antimicrobial peptides tested, compared to other studies, and the origin of antiviral activities.

We will address this concern by increasing the number of peptides tested in anti-microbial and anti-viral experiments. As reported in current version of our manuscript, the first generation of GAN generated 128 unique designs and the top 2% (3 designs) was tested experimentally. The second generation of GAN will produce ~1024 designs (1-2 weeks) and the top 2% (~ 20 new sequences) will be tested. We are in the process of synthesize (2-3 weeks) and MIC measurement (1 week). The overall size of tested sample will reach 20-30 sequences. We will focus on sequences with low similarity (< 30%) to any known AMPs, thus expanding the universe functional peptides. We estimated the collection of these new data in 6 weeks.

Author response:

Reviewer #1 (Public Review):

(1) Figure 3: it is unclear what is the efficiency of Msi2 deletion shRNA - could you demonstrate it by at least two independent methods? (QPCR, Western, or IHC?) please quantitate the data.

In Figure 3, we did not delete Msi2 via shRNA. Instead, we utilized a genetic model in which the Msi2 gene was disrupted via gene trap mutagenesis. We have also used this model in previous publications to define the impact of Msi2 loss in other systems1.

(2) In Figure 4, similarly, it is unclear if Msi2 depletion was effective- and what is shRNA efficiency. Please test this by at least two independent methods (QPCR, Western, or IHC) and also please quantitate the data

We demonstrated that the efficiency of Msi2 depletion was ~83% (Figures 4A and 4C) via qPCR analysis for our in vitro and in vivo experiments, respectively, and verified the knockdown via bulk RNA-seq analysis. The shRNA hairpin used was previously validated and published by our lab2.

(3) the reason for impairment of cell growth demonstrated in Figs 3 and 4 is not clear: is it apoptosis? Necrosis? Cell cycle defects? Autophagy? Senescence? Please probe 2-3 possibilities and provide the data.

The basis of the cell growth impairment after Msi2 deletion/knockdown in this paper is certainly an important question, and future experiments will be performed to better delineate this. In previous publications loss of Msi2 in leukemia cells has been shown to inhibit growth via arrested cell cycle progression by increasing the expression of p213. Further, loss of Msi2 was also shown to promote apoptosis in part by upregulating Bax3. These data suggest that Msi2 can have an impact via multiple distinct mechanisms including by mediating cell cycle arrest and blocking apoptosis. While these specific genes were not detectably changed after loss of Msi2 in lung cancer cells, other genes in these and other pathways will be important to study in the future.

(4) Since Musashi-1 is a Musashi-2 paralogue that could compensate for Musashi-2 loss, please test Msi1 expression levels in matching Fig 3 and Fig 4 sections (in cells/ tumors with Msi2 deletion and in KP cells with Msi2 shRNA). One method could suffice here.

In our RNA-seq of cells following Msi2 knockdown, Msi1 expression was undetectable. The TPM values for Msi1 in control and knockdown cells were less than 0.01, suggesting that it did not compensate for the loss of Msi2.

(5) It is not exactly clear why RNA-seq (as opposed to proteomics) was done to investigate downstream Msi2 targets (since Msi2 is in first place, translational and not transcriptional regulator)- . RNA effects in Fig 5J are quite modest, 2-fold or so. It would be useful (if antibodies available) to test four targets in Fig 5J by Western blot, to see any impact of musashi-2 depletion on those target protein levels. Indeed, several papers - including Kudinov et al PNAS, PMID: 27274057, Makhov P et al PMID: 33723247 and PMID: 37173995 - used proteomics/ RIP approaches and found direct Musashi-2 targets in lung cancer, including EGFR, and others.

Previous published work from the lab showed that expression of Msi2 in the context of myeloid leukemia1can not only repress NUMB protein (I believe protein should be all caps?) (as has been previously demonstrated in the nervous system) but also Numb RNA. This indicated that as an RNA binding protein, Msi2 also can bind and destabilize direct binding targets such as Numb; this was the reason for pursuing transcriptomic analysis.  However as the reviewer suggests, proteomic studies are certainly very important to develop a complete picture of the impact of Musashi to determine which targets are controlled by Msi2 at the protein level.

Reviewer #2 (Public Review):

(1) It will be interesting to determine whether Msi2+ cells are a relatively stable subset or rather the Msi2+ cells in lung is a dynamic concept that is transient or interconvertible. This is relevant to the interpretation of what Msi2 positivity really means.

In previous unpublished work from our lab, we have found that Msi2+ cells from a GFP reporter KPf/fC mouse are readily able to become GFP negative (Msi2-), but the inverse is not true. Specifically, when Msi2+ KPf/fC pancreatic cells were transplanted into the flanks of NSG mice, Msi2+ cells formed tumors in all recipients; these tumors contained both GFP+ and GFP- cells (over 80%)  recapitulating the original heterogeneity and suggesting GFP+ cells can give rise to both GFP+ and GFP- cells (Lytle and Reya, unpublished observations). In contrast only a small subset of GFP- transplanted mice formed tumors. One of the rare GFP- derived tumors was isolated and found to contain largely GFP- cells, with ~0.1% GFP+ cells. The small frequency of GFP expression could be from contaminating cells or may suggest that GFP- cells retain some ability to switch on Msi under selective pressure, and that although they pose a lower risk of driving tumorigenesis than Msi+ cells, they may nonetheless bear latent potential to become higher risk. These data may offer a possible model for projecting the potential of Msi2+ cells in the lung, but is something that needs to be further studied in this tissue.

(2) Does Kras mutation and/or p53 loss upregulate Msi2? This point and the point above are related to whether Msi2+ cells are truly more susceptible to tumorigenesis, as the authors suggested.

In unpublished work from our lab, we have found that Kras mutation upregulates Msi2 over baseline and subsequent p53 loss upregulates Msi2 further in the context of pancreatic cells (Lytle and Reya unpublished results), therefore it is possible that the same is true for the lung. Specifically, we have observed that Msi2 increased from normal acinar cells to Kras-mutated acinar (e.g. pancreatic intraepithelial neoplasia (PanIN)).

To address whether Msi2+ cells are more susceptible to tumorigenesis, we have recently published data showing that the stabilization of the oncogenic MYC protein in lung Msi2+ cells drive the formation of small-cell lung cancer in a new inducible Msi2-CreERT2; CAG-LSL-MycT58A mice (Msi2-Myc)4 model. More importantly, this data provides the first evidence that normal Msi2+ cells are primed and highly sensitive to MYC-driven transformation across many organs and not just the lung4.

(3) The KO of Msi2 reducing tumor number and burden in the lung cancer initiation model is interesting. However, there are two alternative interpretations. First, it is possible that the Msi2 KO mice (without Kras activation and p53 loss) has reduced total lung cell numbers or altered percentage of stem cells. There is currently only one sentence citing data not shown on line 125, commenting that there is no difference in BASC and AT2 cell populations. It will be helpful that such data are shown and the effect of KO on overall lung mass or cellularity is clarified. Second, the phenotype may also be due to a difference in the efficiencies of cre on Kras and p53 in the Msi2 WT and KO mice.

We isolated the lungs of three Msi2 WT and three Msi2 KO mice and used immunofluorescence staining to stain for CC10 (BASC) and SPC (AT2) to determine if these cell populations were reduced after Msi2 loss alone. Below are representative images showing that the Msi2 KO mice did not have lower numbers of both BASC and AT2 cell populations. 

Author response image 1.

(4) All shRNA experiments (for both Msi2 KD and the KD of candidate genes) utilized a single shRNA. This approach cannot exclude off-target effects of the shRNA.

The shRNA hairpin used for Msi2 was previously validated and published by our lab2. Additionally, in this work we did develop and use a Msi2 genetic knockout mouse model that validates our shRNA knockdown data showing the specific impact of Msi2 on lung tumor growth.

(5) The technical details of the PDX experiment (Figure 4F) are not fully explained.

Due to space considerations, we were unable not put the specifics in the legend, but the details are in the methods section (Flank Transplant Assays). In brief, 500,000 cells/well were plated in a 6-well plate coated with Matrigel and 83,000 cells/well were plated in a 24-well plate coated with Matrigel for subsequent determination of transduction efficiency via FACS. 24 hours after transduction, media from the cells was collected and placed on ice. 1mL of 2mg/mL collagenase/dispase was then added to the well and incubated for 45 minutes at 37ºC to dissociate the remaining cells from Matrigel followed by subsequent washes. Cells were pelleted by centrifugation and an equivalent number of shControl and shMsi2 transduced cells were resuspended in full media, mixed at a 1:1 ratio with growth factor reduced Matrigel at a final volume of 100 μL, and transplanted subcutaneously into the flanks of NSG recipient mice.

Reviewer #3 (Public Review):

- In Figure 1, characterization of Msi2 expression in the normal mouse lung was carried out by using a Msi2-GFP Knock-in reporter and analyzed by flow cytometry followed by cytospins and immunostaining. Additional characterization of Msi2 expression by co-immunostaining with well-known markers of airway and alveolar cell types in intact lung tissue will strengthen the existing data and provide more specific information about Msi2 expression and abundancy in relevant cell types. It will be also interesting to know whether Msi2 is expressed or not in other abundant lung cell types such as ciliated and AT1 cells.

We performed co-staining of Msi2 and CC10 as well as Msi2 and SPC in Figure 1C. In the future we can include additional markers as well as markers for airway and other alveolar cell types.

- While this set of experiments provide strong evidence that Msi2 is required for tumor progression and growth in lung adenocarcinoma, it is unclear whether normal Msi2+ lung cells are more responsive to transformation or whether Msi2 is upregulated early during the process of tumorigenesis. Future lineage tracing experiments using Msi2-CreER and mouse models of chemically-induced lung carcinogenesis will provide additional data that will fully support this claim.

Recently, we published data showing that Msi2 is expressed in Clara cells at the bronchoalveolar junction in the lung of our new Msi2-CreERT2 knock-in mouse model4. Furthermore, stabilization of the oncogenic MYC protein in these specific cells to model Myc amplification was sufficient to drive the formation of small-cell lung cancer4. These data excitingly demonstrate that Msi2+ cells are more responsive to transformation after Myc stabilization.

- In Figure 4F, Patient-derived xenograft (PDX) assays were conducted in 2 patients only and the percentage of cells infected by shRNA-Msi2 is low in both PDX (30% and 10% for patient 1 and 2 respectively). It is surprising that Msi2 downregulation in a small percentage of tumor cells has such a dramatic effect on tumor growth and expansion. Confirmation of this finding with additional patient samples would suggest an important non-cell autonomous role for Msi2 in lung adenocarcinoma.

In the future we hope to collect more patient samples to further validate the data presented with the first 2 patients shown here. We are not certain about the reason behind the large impact of Msi2 inhibition, but as cancer stem cells drive the formation of the rest of the tumor and also drive the stromal microenvironment, it is possible that when Msi2 is deleted, Msi2- cells no longer form tumors? and also the ability to build the stromal microenvironment is impacted. This possibility needs to be further tested in future experiments.

References

(1) Ito, T. Kwon, H. Y., Zimdahl, B., Congdon, K. L., Blum, J., Lento, W. E., Zhao, C., Lagoo, A., Gerrard, G., Foroni, L., Goldman, J., Goh, H., Kim, S. H., Kim, D. W., Chuah, C., Oehler, V. G., Radich, J. P., Jordan, C. T., & Reya, T. Regulation of myeloid leukaemia by the cell-fate determinant Musashi. Nature 466, 765–768 (2010).

(2) Fox, R. G. Lytle, N. K., Jaquish, D. V., Park, F. D., Ito, T., Bajaj, J., Koechlein, C. S., Zimdahl, B., Yano, M., Kopp, J. L., Kritzik, M., Sicklick, J. K., Sander, M., Grandgenett, P. M., Hollingsworth, M. A., Shibata, S., Pizzo, D., Valasek, M. A., Sasik, R., Scadeng, M., Okano, H., Kim, Y., MacLeod, A. R., Lowy, A. M., & Reya, T. Image-based detection and targeting of therapy resistance in pancreatic adenocarcinoma. Nature 534, 407–411 (2016).

(3) Zhang, H. Tan, S., Wang, J., Chen, S., Quan, J., Xian, J., Zhang, Ss., He, J., & Zhang, L. Musashi2 modulates K562 leukemic cell proliferation and apoptosis involving the MAPK pathway. Exp Cell Res 320, 119-27 (2014).

(4) Rajbhandari, N., Hamilton, M., Quintero, C.M., Ferguson, L.P., Fox, R., Schürch, C.M., Wang, J., Nakamura, M., Lytle, N.K., McDermott, M., Diaz, E., Pettit, H., Kritzik, M., Han, H., Cridebring, D., Wen, K.W., Tsai, S., Goggins, M.G., Lowy, A.M., Wechsler-Reya, R.J., Von Hoff, D.D., Newman, A.M., & Reya, T. Single-cell mapping identifies MSI+ cells as a common origin for diverse subtypes of pancreatic cancer. Cancer Cell 41(11):1989-2005.e9 (2023).

Author Response

Reviewer #1 (Public Review):

1) “It is unclear whether new in vivo experiments were conducted for this study”.

All in vivo experiments shown were conducted independently by new researchers in the lab, using the original fly stocks. This will be more clearly stated in the revised supplement. The aim of repeating the experiments was to directly compare the consequences of impaired N- and C-terminal shedding side-by-side in two Hh-dependent developmental systems.

2) “A critical shortcoming of the study is that experiments showing Shh secretion/export do not include a Shh(-) control condition. Without demonstration that the bands analyzed are specific for Shh(+) conditions, these experiments cannot be appropriately evaluated”.

C9C5 antibody reactivity and specificity is shown below, and this control will be added to the revised manuscript. We established the C9C5 immunoblotting protocol – and generated the blot shown in Author Response Image 1 - before any of the experiments in the manuscript were started. The immunoblot clearly shows Shh specificity similar to that of R&D AF464 anti-Shh antibodies that were previously used in the lab. The immunoblot also shows that both antibodies detect the same Shh signals in media, that C9C5 is more sensitive, and that AF464 and C9C5 detect 5E1-IP’d dual-lipidated and monolipidated soluble Shh equally well. Also note that, in our hands, C9C5 is highly specific: this antibody detects N-truncated C25S;Δ26-35Shh of increased electrophoretic mobility, but does not cause unspecific signals above or below, even if the blot is strongly overexposed (as shown here). Specific Shh detection by C9C5 is also discussed in our response to editor’s comments below.

Cells were transfected with constructs encoding full-length C25SShh or truncated C25S;Δ26-35Shh, and proteins in serum-containing media were 5E1 immunoprecipitated or concentrated by heparin-sepharose pulldown. Dual-lipidated R&D 8908-SH was dissolved in the same medium and subjected to the same 5E1 immunoprecipitation or heparin pulldown. The blot was incubated with antibody AF464 and (after stripping) with antibody C9C5. Immunoblot analysis revealed high specificity of both antibodies and also revealed poor interactions of dual-lipidated 8908-SH with highly charged heparin.

3) “A stably expressing Shh/Hhat cell line would reduce condition to condition and experiment to experiment variability”.

We fully agree with this reviewer and therefore aimed to establish stable Hhat expressing cell lines several years ago. However, stable Hhat expression eliminated transfected cells after several passages, or cells gradually ceased to express Hhat, preventing us to establish a stable line despite several attempts and tried strategies. For this reason, we established transient co-expression of Shh/Hhat from the same mRNA to at least eliminate variability between relative Shh/Hhat expression levels and to assure complete Shh palmitoylation in our assays.

4) “Unusual normalization strategies are used for many experiments, and quantification/statistical analyses are missing for several experiments”.

This comment refers to data shown in Fig. 3 (here, no quantification of Scube2 function in Disp-/- cells had been conducted) and to qPCR data shown in Fig. 4 (here, Shh and C25AShh were compared only indirectly via dual-lipidated R&D 8908-SH, but not directly in a side-by-side experiment, and Shh variants with an N-terminal alanine or a serine were directly compared). We agree with the reviewer and therefore currently repeat qPCR assays and quantify blots to eliminate these technical shortcomings from the final manuscript.

5) “The study provides a modest advance in the understanding of the complex issue of Shh membrane extraction”

Our investigation identified unexpected links between Disp as a furin-activated Hh exporter, sheddase-mediated Shh release, Scube2-mediated Shh release and lipoprotein-mediated Hh transport – established modes indeed but with no previously established direct connections – that increase their relevance. We also identified a previously unknown N-processed Shh variant attached to lipoproteins and show that Disp/Scube2 function absolutely requires lipoproteins. Therefore, although we do agree that our findings are confirmatory for the above modes, they also provide new mechanistic insight and challenge the currently dominating model of Disp-mediated hand-over of dual-lipidated Hh to Scube2 chaperones (this model does not predict a role for lipoprotein particles but for both Shh lipids in signaling, for a recent discussion, see PMID 36932157). Our findings suggest an answer to the intensely debated question of whether Disp/Ptch extract cholesterol from the outer or inner plasma membrane leaflet, and suggest that N-palmitate is dispensable for signaling of lipoprotein-associated Shh to Ptch receptors. Finally, we note that previous in vivo studies in flies often relied on Hh overexpression in the fat body, raising questions on their physiological relevance. Our in vivo analyses of Hh function in wing- and eye discs are more physiologically relevant and can explain the previously reported presence of non-lipidated bioactive Hh in disc tissue (PMID: 23554573).

Reviewer #2 (Public Review):

1) “However, the results concerning the roles of lipoproteins and Shh lipid modifications are largely confirmatory of previous results, and molecular identity/physiological relevance of the newly identified Shh variant remain unclear”.

Regarding the confirmatory aspects of our work, please also refer to our response to reviewer 1. In addition, we would like to reply that our unbiased experimental approach was designed to challenge the model of Shh shedding by testing whether established Shh release regulators affect it (e.g. support it) or not. As described in our work, Disp, Scube2 and lipoproteins all contribute to increased shedding (which is new), that Disp function depends on lipoprotein presence (also new), and that lipoproteins modify the outcome of Shh shedding (dual Shh shedding versus N-shedding and lipoprotein association), which is also new.

Regarding physiological relevance, we would like to reply that our finding that artificially generated monolipidated variants (C25SShh and ShhN) solubilize in uncontrolled manner from producing cells can explain previously observed, highly variable gain-of-function or loss-of-function phenotypes upon their overexpression in vivo 1, 2, 3, 4, 5. Our data is also supported by the observed presence of variably lipidated Shh/Hh variants in vivo 6, and the in vivo observation that complete removal of Scube activity in zebrafish embryos phenocopies a complete loss of Hh function that is bypassed by increased ligand expression - and even results in wild-type-like ectopic Shh target gene expression 7. The in vivo observations are compatible with our data but are incompatible with proposed alternative models of Scube-mediated dual-lipidated Shh extraction and continued Shh/Scube association to allow for morphogen transport.

2) “Thus, it would be important to demonstrate key findings in cells that secrete Shh endogenously”.

Experimental data shown in Fig. S8B demonstrates that en-controlled expression of sheddase-resistant Hh variants blocks endogenous Hh function in the same wing disc compartment. To our knowledge, this assay is the most physiologically relevant test of the mechanism of Disp-mediated Hh release. Still, we have now started to analyze Hh from Drosophila disc tissue biochemically and hope that we can include our findings in the final manuscript.

3) “The authors could use an orthogonal approach, optimally a demonstration of physical interaction, or at least fractionation by a different parameter”.

We agree with this reviewer’s assessment and are currently in the process to establish co-IP and density gradient conditions to test physical HDL/Shh interactions. The results will be included in the final version of record.

Author response:

Reviewer #1 (Public review):

Summary:

In this study, the authors attempt to devise general rules for aptamer design based on structure and sequence features. The main system they are testing is an aptamer targeting a viral sequence.

Strengths:

The method combines a series of well-established protocols, including docking, MD, and a lot of system-specific knowledge, to design several new versions of the Ta aptamer with improved binding affinity.

We thank the reviewer for this accurate summary and for recognizing the strength of our integrated computational–experimental workflow in improving aptamer affinity. We will emphasize this contribution more clearly in the revised Introduction.

Weaknesses:

The approach requires a lot of existing knowledge and, impo rtantly, an already known aptamer, which presumably was found with SELEX. In addition, although the aptamer may have a stronger binding affinity, it is not clear if any of it has any additional useful properties such as stability, etc.

Thanks for these critical comments.

(1) On the reliance on a known aptamer: We agree that our CAAMO framework is designed as a post-SELEX optimization platform rather than a tool for de novo discovery. Its primary utility lies in rationally enhancing the affinity of existing aptamers that may not yet be sequence-optimal, thereby complementing experimental technologies such as SELEX. In the revised manuscript, we plan to clarify this point more explicitly in both the Introduction and Discussion sections, emphasizing that the propose CAAMO framework is intended to serve as a complementary strategy that accelerates the iterative optimization of lead aptamers.

(2) On stability and developability: We also appreciate the reviewer’s important reminder that affinity alone is not sufficient for therapeutic development. We acknowledge that the present study has focused mainly on affinity optimization, and properties such as nuclease resistance, structural stability, and overall developability were not evaluated. In the revised manuscript, we will add a dedicated section highlighting the critical importance of these characteristics and outlining them as key priorities for our future research efforts.

Reviewer #2 (Public review):

Summary:

This manuscript proposes a workflow for discovering and optimizing RNA aptamers, with application in the optimization of a SARS-CoV-2 RBD. The authors took a previously identified RNA aptamer, computationally docked it into one specific RBD structure, and searched for variants with higher predicted affinity. The variants were subsequently tested for RBD binding using gel retardation assays and competition with antibodies, and one was found to be a stronger binder by about three-fold than the founding aptamer. Overall, this would be an interesting study if it were performed with truly high-affinity aptamers, and specificity was shown for RBD or several RBD variants.

Strengths:

The computational workflow appears to mostly correctly find stronger binders, though not de novo binders.

We thank the reviewer for the clear summary and for acknowledging that our workflow effectively prioritizes stronger binders.

Weaknesses:

(1) Antibody competition assays are reported with RBD at 40 µM, aptamer at 5 µM, and a titration of antibody between 0 and 1.2 µg. This approach does not make sense. The antibody concentration should be reported in µM. An estimation of the concentration is 0-8 pmol (from 0-1.2 µg), but that's not a concentration, so it is unknown whether enough antibody molecules were present to saturate all RBD molecules, let alone whether they could have displaced all aptamers.

Thanks for your insightful comment. We have calculated that 0–1.2 µg antibody corresponds to a final concentration range of 0–1.6 µM (see Author response image 1). In practice, 1.2 µg was the maximum amount of commercial antibody that could be added under the conditions of our assay. In the revised manuscript, we plan to report all antibody quantities in molar concentrations in the Materials and Methods section for clarity and rigor.

Author response image 1.<br /> Estimation of antibody concentration. Assuming a molecular weight of 150 kDa, dissolving 1.2 µg of antibody in a 5 µL reaction volume results in a final concentration of 1.6 µM.<br />

As shown in Figure 5D of the main text, the purpose of the antibody–aptamer competition assay was not to achieve full saturation but rather to compare the relative competitive binding of the optimized aptamer (Ta^G34C) versus the parental aptamer (Ta). Molecular interactions at this scale represent a dynamic equilibrium of binding and dissociation. While the antibody concentration may not have been sufficient to saturate all available RBD molecules, the experimental results clearly reveal the competitive binding behavior that distinguishes the two aptamers. Specifically, two consistent trends emerged:

(1) Across all antibody concentrations, the free RNA band for Ta was stronger than that of Ta^G34C, while the RBD–RNA complex band of the latter was significantly stronger, indicating that Ta^G34Cbound more strongly to RBD.

(2) For Ta, increasing antibody concentration progressively reduced the RBD–RNA complex band, consistent with antibody displacing the aptamer. In contrast, for Ta^G34C, the RBD–RNA complex band remained largely unchanged across all tested antibody concentrations, suggesting that the antibody was insufficient to displace Ta^G34C from the complex.

Together, these observations support the conclusion that Ta^G34C exhibits markedly stronger binding to RBD than the parental Ta aptamer, in line with the predictions and objectives of our CAAMO optimization framework.

(2) These are not by any means high-affinity aptamers. The starting sequence has an estimated (not measured, since the titration is incomplete) KD of 110 µM. That's really the same as non-specific binding for an interaction between an RNA and a protein. This makes the title of the manuscript misleading. No high-affinity aptamer is presented in this study. If the docking truly presented a bound conformation of an aptamer to a protein, a sub-micromolar Kd would be expected, based on the number of interactions that they make.

In fact, our starting sequence (Ta) is a high-affinity aptamer, and then the optimized sequences (such as Ta^G34C) with enhanced affinity are undoubtedly also high-affinity aptamers. See descriptions below:

(1) Origin and prior characterization of Ta. The starting aptamer Ta (referred to as RBD-PB6-Ta in the original publication by Valero et al., PNAS 2021, doi:10.1073/pnas.2112942118) was selected through multiple positive rounds of SELEX against SARS-CoV-2 RBD, together with counter-selection steps to eliminate non-specific binders. In that study, Ta was reported to bind RBD with an IC₅₀ of ~200 nM as measured by biolayer interferometry (BLI), supporting its high affinity and specificity.

(2) Methodological differences between EMSA and BLI measurements. We acknowledge that the discrepancy between our obtained binding affinity (K_d = 110 µM) and the previously reported one (IC₅₀ ~ 200 nM) for the same Ta sequence arises primarily from methodological and experimental differences between EMSA and BLI. Namely, different experimental measurement methods can yield varied binding affinity values. While EMSA may have relatively low measurement precision, its relatively simple procedures were the primary reason for its selection in this study. Particularly, our framework (CAAMO) is designed not as a tool for absolute affinity determination, but as a post-SELEX optimization platform that prioritizes relative changes in binding affinity under a consistent experimental setup. Thus, the central aim of our work is to demonstrate that CAAMO can reliably identify variants, such as Ta^G34C, that bind more strongly than the parental sequence under identical assay conditions.

(3) Evidence of specific binding in our assays. We emphasize that the binding observed in our EMSA experiments reflects genuine aptamer–protein interactions. As shown in Figure 2G of the main text, a control RNA (Tc) exhibited no detectable binding to RBD, whereas Ta produced a clear binding curve, confirming that the interaction is specific rather than non-specific.

(3) The binding energies estimated from calculations and those obtained from the gel-shift experiments are vastly different, as calculated from the Kd measurements, making them useless for comparison, except for estimating relative affinities.

We thank the reviewer for raising this important point. CAAMO was developed as a post-SELEX optimization tool with the explicit goal of predicting relative affinity changes (ΔΔG) rather than absolute binding free energies (ΔG). Empirically, CAAMO correctly predicted the direction of affinity change for 5 out of 6 designed variants (e.g., ΔΔG < 0 indicates enhanced binding free energy relative to WT); such predictive power for relative ranking is highly valuable for prioritizing candidates for experimental testing. Our prior work on RNA–protein interactions likewise supports the reliability of relative affinity predictions (see: Nat Commun 2023, doi:10.1038/s41467-023-39410-8). In the revised manuscript we will explicitly state that the primary utility of CAAMO is to accurately predict affinity trends and to rank variants for follow-up, and we will moderate any statements that could be interpreted as claims about precise absolute ΔΔG values.

Author Response

eLife assessment

This study presents potentially valuable results on glutamine-rich motifs in relation to protein expression and alternative genetic codes. The author's interpretation of the results is so far only supported by incomplete evidence, due to a lack of acknowledgment of alternative explanations, missing controls and statistical analysis and writing unclear to non experts in the field. These shortcomings could be at least partially overcome by additional experiments, thorough rewriting, or both.

We thank both the Reviewing Editor and Senior Editor for handling this manuscript and will submit our revised manuscript after the reviewed preprint is published by eLife.  

Reviewer #1 (Public Review):

Summary

This work contains 3 sections. The first section describes how protein domains with SQ motifs can increase the abundance of a lacZ reporter in yeast. The authors call this phenomenon autonomous protein expression-enhancing activity, and this finding is well supported. The authors show evidence that this increase in protein abundance and enzymatic activity is not due to changes in plasmid copy number or mRNA abundance, and that this phenomenon is not affected by mutants in translational quality control. It was not completely clear whether the increased protein abundance is due to increased translation or to increased protein stability.

In section 2, the authors performed mutagenesis of three N-terminal domains to study how protein sequence changes protein stability and enzymatic activity of the fusions. These data are very interesting, but this section needs more interpretation. It is not clear if the effect is due to the number of S/T/Q/N amino acids or due to the number of phosphorylation sites.

In section 3, the authors undertake an extensive computational analysis of amino acid runs in 27 species. Many aspects of this section are fascinating to an expert reader. They identify regions with poly-X tracks. These data were not normalized correctly: I think that a null expectation for how often poly-X track occur should be built for each species based on the underlying prevalence of amino acids in that species. As a result, I believe that the claim is not well supported by the data.

Strengths

This work is about an interesting topic and contains stimulating bioinformatics analysis. The first two sections, where the authors investigate how S/T/Q/N abundance modulates protein expression level, is well supported by the data. The bioinformatics analysis of Q abundance in ciliate proteomes is fascinating. There are some ciliates that have repurposed stop codons to code for Q. The authors find that in these proteomes, Q-runs are greatly expanded. They offer interesting speculations on how this expansion might impact protein function.

Weakness

At this time, the manuscript is disorganized and difficult to read. An expert in the field, who will not be distracted by the disorganization, will find some very interesting results included. In particular, the order of the introduction does not match the rest of the paper.

In the first and second sections, where the authors investigate how S/T/Q/N abundance modulates protein expression levels, it is unclear if the effect is due to the number of phosphorylation sites or the number of S/T/Q/N residues.

There are three reasons why the number of phosphorylation sites in the Q-rich motifs is not relevant to their autonomous protein expression-enhancing (PEE) activities:

First, we have reported previously that phosphorylation-defective Rad51-NTD (Rad51-3SA) and wild-type Rad51-NTD exhibit similar autonomous PEE activity. Mec1/Tel1-dependent phosphorylation of Rad51-NTD antagonizes the proteasomal degradation pathway, increasing the half-life of Rad51 from ∼30 min to ≥180 min (Ref 27; Woo, T. T. et al. 2020).

1. T. T. Woo, C. N. Chuang, M. Higashide, A. Shinohara, T. F. Wang, Dual roles of yeast Rad51 N-terminal domain in repairing DNA double-strand breaks. Nucleic Acids Res 48, 8474-8489 (2020).

Second, in our preprint manuscript, we have also shown that phosphorylation-defective Rad53-SCD1 (Rad51-SCD1-5STA) also exhibits autonomous PEE activity similar to that of wild-type Rad53-SCD (Figure 2D, Figure 4A and Figure 4C).

Third, as revealed by the results of our preprint manuscript (Figure 4), it is the percentages, and not the numbers, of S/T/Q/N residues that are correlated with the PEE activities of Q-rich motifs.

The authors also do not discuss if the N-end rule for protein stability applies to the lacZ reporter or the fusion proteins.

The autonomous PEE function of S/T/Q-rich NTDs is unlikely to be relevant to the N-end rule. The N-end rule links the in vivo half-life of a protein to the identity of its N-terminal residues. In S. cerevisiae, the N-end rule operates as part of the ubiquitin system and comprises two pathways. First, the Arg/N-end rule pathway, involving a single N-terminal amidohydrolase Nta1, mediates deamidation of N-terminal asparagine (N) and glutamine (Q) into aspartate (D) and glutamate (E), which in turn are arginylated by a single Ate1 R-transferase, generating the Arg/N degron. N-terminal R and other primary degrons are recognized by a single N-recognin Ubr1 in concert with ubiquitin-conjugating Ubc2/Rad6. Ubr1 can also recognize several other N-terminal residues, including lysine (K), histidine (H), phenylalanine (F), tryptophan (W), leucine (L) and isoleucine (I) (Bachmair, A. et al. 1986; Tasaki, T. et al. 2012; Varshavshy, A. et al. 2019). Second, the Ac/N-end rule pathway targets proteins containing N-terminally acetylated (Ac) residues. Prior to acetylation, the first amino acid methionine (M) is catalytically removed by Met-aminopeptides, unless a residue at position 2 is non-permissive (too large) for MetAPs. If a retained N-terminal M or otherwise a valine (V), cysteine (C), alanine (A), serine (S) or threonine (T) residue is followed by residues that allow N-terminal acetylation, the proteins containing these AcN degrons are targeted for ubiquitylation and proteasome-mediated degradation by the Doa10 E3 ligase (Hwang, C. S., 2019).

A. Bachmair, D. Finley, A. Varshavsky, In vivo half-life of a protein is a function of its amino-terminal residue. Science 234, 179-186 (1986).

T. Tasaki, S. M. Sriram, K. S. Park, Y. T. Kwon, The N-end rule pathway. Annu Rev Biochem 81, 261-289 (2012).

A. Varshavsky, N-degron and C-degron pathways of protein degradation. Proc Natl Acad Sci 116, 358-366 (2019).

C. S. Hwang, A. Shemorry, D. Auerbach, A. Varshavsky, The N-end rule pathway is mediated by a complex of the RING-type Ubr1 and HECT-type Ufd4 ubiquitin ligases. Nat Cell Biol 12, 1177-1185 (2010).

The PEE activities of these S/T/Q-rich domains are unlikely to arise from counteracting the N-end rule for two reasons. First, the first two amino acid residues of Rad51-NTD, Hop1-SCD, Rad53-SCD1, Sup35-PND, Rad51-ΔN, and LacZ-NVH are MS, ME, ME, MS, ME, and MI, respectively, where M is methionine, S is serine, E is glutamic acid and I is isoleucine. Second, Sml1-NTD behaves similarly to these N-terminal fusion tags, despite its methionine and glutamine (MQ) amino acid signature at the N-terminus.

The most interesting part of the paper is an exploration of S/T/Q/N-rich regions and other repetitive AA runs in 27 proteomes, particularly ciliates. However, this analysis is missing a critical control that makes it nearly impossible to evaluate the importance of the findings. The authors find the abundance of different amino acid runs in various proteomes. They also report the background abundance of each amino acid. They do not use this background abundance to normalize the runs of amino acids to create a null expectation from each proteome. For example, it has been clear for some time (Ruff, 2017; Ruff et al., 2016) that Drosophila contains a very high background of Q's in the proteome and it is necessary to control for this background abundance when finding runs of Q's.

We apologize for not explaining sufficiently well the topic eliciting this reviewer’s concern in our preprint manuscript. In the second paragraph of page 14, we cite six references to highlight that SCDs are overrepresented in yeast and human proteins involved in several biological processes (32, 74), and that polyX prevalence differs among species (43, 75-77).

1. Cheung HC, San Lucas FA, Hicks S, Chang K, Bertuch AA, Ribes-Zamora A. An S/T-Q cluster domain census unveils new putative targets under Tel1/Mec1 control. BMC Genomics. 2012;13:664.

2. Mier P, Elena-Real C, Urbanek A, Bernado P, Andrade-Navarro MA. The importance of definitions in the study of polyQ regions: A tale of thresholds, impurities and sequence context. Comput Struct Biotechnol J. 2020;18:306-13.

3. Cara L, Baitemirova M, Follis J, Larios-Sanz M, Ribes-Zamora A. The ATM- and ATR-related SCD domain is over-represented in proteins involved in nervous system development. Sci Rep. 2016;6:19050.

4. Kuspa A, Loomis WF. The genome of Dictyostelium discoideum. Methods Mol Biol. 2006;346:15-30.

5. Davies HM, Nofal SD, McLaughlin EJ, Osborne AR. Repetitive sequences in malaria parasite proteins. FEMS Microbiol Rev. 2017;41(6):923-40.

6. Mier P, Alanis-Lobato G, Andrade-Navarro MA. Context characterization of amino acid homorepeats using evolution, position, and order. Proteins. 2017;85(4):709-19.

We will cite the two references by Kiersten M. Ruff in our revised manuscript.

K. M. Ruff and R. V. Pappu, (2015) Multiscale simulation provides mechanistic insights into the effects of sequence contexts of early-stage polyglutamine-mediated aggregation. Biophysical Journal 108, 495a.

K. M. Ruff, J. B. Warner, A. Posey and P. S. Tan (2017) Polyglutamine length dependent structural properties and phase behavior of huntingtin exon1. Biophysical Journal 112, 511a.

The authors could easily address this problem with the data and analysis they have already collected. However, at this time, without this normalization, I am hesitant to trust the lists of proteins with long runs of amino acid and the ensuing GO enrichment analysis.

Ruff KM. 2017. Washington University in St.

Ruff KM, Holehouse AS, Richardson MGO, Pappu RV. 2016. Proteomic and Biophysical Analysis of Polar Tracts. Biophys J 110:556a.

We thank Reviewer #1 for this helpful suggestion and now address this issue by means of a different approach described below.

Based on a previous study (43; Palo Mier et al. 2020), we applied seven different thresholds to seek both short and long, as well as pure and impure, polyX strings in 20 different representative near-complete proteomes, including 4X (4/4), 5X (4/5-5/5), 6X (4/6-6/6), 7X (4/7-7/7), 8-10X (≥50%X), 11-10X (≥50%X) and ≥21X (≥50%X).

To normalize the runs of amino acids and create a null expectation from each proteome, we determined the ratios of the overall number of X residues for each of the seven polyX motifs relative to those in the entire proteome of each species, respectively. The results of four different polyX motifs are shown below, i.e., polyQ (Author response image 1), polyN (Author response image 2), polyS (Author response image 3) and polyT (Author response image 4).

Author response image 1.

Q contents in 7 different types of polyQ motifs in 20 near-complete proteomes. The five ciliates with reassigned stops codon (TAAQ and TAGQ) are indicated in red. Stentor coeruleus, a ciliate with standard stop codons, is indicated in green.  

Author response image 2.

N contents in 7 different types of polyN motifs in 20 near-complete proteomes. The five ciliates with reassigned stops codon (TAAQ and TAGQ) are indicated in red. Stentor coeruleus, a ciliate with standard stop codons, is indicated in green.

Author response image 3.

S contents in 7 different types of polyS motifs in 20 near-complete proteomes. The five ciliates with reassigned stops codon (TAAQ and TAGQ) are indicated in red. Stentor coeruleus, a ciliate with standard stop codons, is indicated in green.  

Author response image 4.

T contents in 7 different types of polyT motifs in 20 near-complete proteomes. The five ciliates with reassigned stops codon (TAAQ and TAGQ) are indicated in red. Stentor coeruleus, a ciliate with standard stop codons, is indicated in green.

The results summarized in these four new figures support that polyX prevalence differs among species and that the overall X contents of polyX motifs often but not always correlate with the X usage frequency in entire proteomes (43; Palo Mier et al. 2020).

Most importantly, our results reveal that, compared to Stentor coeruleus or several non-ciliate eukaryotic organisms (e.g., Plasmodium falciparum, Caenorhabditis elegans, Danio rerio, Mus musculus and Homo sapiens), the five ciliates with reassigned TAAQ and TAGQ codons not only have higher Q usage frequencies, but also more polyQ motifs in their proteomes (Figure 1). In contrast, polyQ motifs prevail in Candida albicans, Candida tropicalis, Dictyostelium discoideum, Chlamydomonas reinhardtii, Drosophila melanogaster and Aedes aegypti, though the Q usage frequencies in their entire proteomes are not significantly higher than those of other eukaryotes (Figure 1). Due to their higher N usage frequencies, Dictyostelium discoideum, Plasmodium falciparum and Pseudocohnilembus persalinus have more polyN motifs than the other 23 eukaryotes we examined here (Figure 2). Generally speaking, all 26 eukaryotes we assessed have similar S usage frequencies and percentages of S contents in polyS motifs (Figure 3). Among these 26 eukaryotes, Dictyostelium discoideum possesses many more polyT motifs, though its T usage frequency is similar to that of the other 25 eukaryotes (Figure 4).

In conclusion, these new normalized results confirm that the reassignment of stop codons to Q indeed results in both higher Q usage frequencies and more polyQ motifs in ciliates.  

Reviewer #2 (Public Review):

Summary:

This study seeks to understand the connection between protein sequence and function in disordered regions enriched in polar amino acids (specifically Q, N, S and T). While the authors suggest that specific motifs facilitate protein-enhancing activities, their findings are correlative, and the evidence is incomplete. Similarly, the authors propose that the re-assignment of stop codons to glutamine-encoding codons underlies the greater user of glutamine in a subset of ciliates, but again, the conclusions here are, at best, correlative. The authors perform extensive bioinformatic analysis, with detailed (albeit somewhat ad hoc) discussion on a number of proteins. Overall, the results presented here are interesting, but are unable to exclude competing hypotheses.

Strengths:

Following up on previous work, the authors wish to uncover a mechanism associated with poly-Q and SCD motifs explaining proposed protein expression-enhancing activities. They note that these motifs often occur IDRs and hypothesize that structural plasticity could be capitalized upon as a mechanism of diversification in evolution. To investigate this further, they employ bioinformatics to investigate the sequence features of proteomes of 27 eukaryotes. They deepen their sequence space exploration uncovering sub-phylum-specific features associated with species in which a stop-codon substitution has occurred. The authors propose this stop-codon substitution underlies an expansion of ploy-Q repeats and increased glutamine distribution.

Weaknesses:

The preprint provides extensive, detailed, and entirely unnecessary background information throughout, hampering reading and making it difficult to understand the ideas being proposed. The introduction provides a large amount of detailed background that appears entirely irrelevant for the paper. Many places detailed discussions on specific proteins that are likely of interest to the authors occur, yet without context, this does not enhance the paper for the reader.

The paper uses many unnecessary, new, or redefined acronyms which makes reading difficult. As examples:

(1) Prion forming domains (PFDs). Do the authors mean prion-like domains (PLDs), an established term with an empirical definition from the PLAAC algorithm? If yes, they should say this. If not, they must define what a prion-forming domain is formally.

The N-terminal domain (1-123 amino acids) of S. cerevisiae Sup35 was already referred to as a “prion forming domain (PFD)” in 2006 (Tuite, M. F. 2006). Since then, PFD has also been employed as an acronym in other yeast prion papers (Cox, B.S. et al. 2007; Toombs, T. et al. 2011).

M. F., Tuite, Yeast prions and their prion forming domain. Cell 27, 397-407 (2005).

B. S. Cox, L. Byrne, M. F., Tuite, Protein Stability. Prion 1, 170-178 (2007).

J. A. Toombs, N. M. Liss, K. R. Cobble, Z. Ben-Musa, E. D. Ross, [PSI+] maintenance is dependent on the composition, not primary sequence, of the oligopeptide repeat domain. PLoS One 6, e21953 (2011).

(2) SCD is already an acronym in the IDP field (meaning sequence charge decoration) - the authors should avoid this as their chosen acronym for Serine(S) / threonine (T)-glutamine (Q) cluster domains. Moreover, do we really need another acronym here (we do not).

SCD was first used in 2005 as an acronym for the Serine (S)/threonine (T)-glutamine (Q) cluster domain in the DNA damage checkpoint field (Traven, A. and Heierhorst, J. 2005). Almost a decade later, SCD became an acronym for “sequence charge decoration” (Sawle, L. et al. 2015; Firman, T. et al. 2018).

A. Traven and J, Heierhorst, SQ/TQ cluster domains: concentrated ATM/ATR kinase phosphorylation site regions in DNA-damage-response proteins. Bioessays. 27, 397-407 (2005).

L. Sawle and K, Ghosh, A theoretical method to compute sequence dependent configurational properties in charged polymers and proteins. J. Chem Phys. 143, 085101(2015).

T. Firman and Ghosh, K. Sequence charge decoration dictates coil-globule transition in intrinsically disordered proteins. J. Chem Phys. 148, 123305 (2018).

(3) Protein expression-enhancing (PEE) - just say expression-enhancing, there is no need for an acronym here.

Thank you. Since we have shown that addition of Q-rich motifs to LacZ affects protein expression rather than transcription, we think it is better to use the “PEE” acronym.

The results suggest autonomous protein expression-enhancing activities of regions of multiple proteins containing Q-rich and SCD motifs. Their definition of expression-enhancing activities is vague and the evidence they provide to support the claim is weak. While their previous work may support their claim with more evidence, it should be explained in more detail. The assay they choose is a fusion reporter measuring beta-galactosidase activity and tracking expression levels. Given the presented data they have shown that they can drive the expression of their reporters and that beta gal remains active, in addition to the increase in expression of fusion reporter during the stress response. They have not detailed what their control and mock treatment is, which makes complete understanding of their experimental approach difficult. Furthermore, their nuclear localization signal on the tag could be influencing the degradation kinetics or sequestering the reporter, leading to its accumulation and the appearance of enhanced expression. Their evidence refuting ubiquitin-mediated degradation does not have a convincing control.

Based on the experimental results, the authors then go on to perform bioinformatic analysis of SCD proteins and polyX proteins. Unfortunately, there is no clear hypothesis for what is being tested; there is a vague sense of investigating polyX/SCD regions, but I did not find the connection between the first and section compelling (especially given polar-rich regions have been shown to engage in many different functions). As such, this bioinformatic analysis largely presents as many lists of percentages without any meaningful interpretation. The bioinformatics analysis lacks any kind of rigorous statistical tests, making it difficult to evaluate the conclusions drawn. The methods section is severely lacking. Specifically, many of the methods require the reader to read many other papers. While referencing prior work is of course, important, the authors should ensure the methods in this paper provide the details needed to allow a reader to evaluate the work being presented. As it stands, this is not the case.

Thank you. As described in detail below, we have now performed rigorous statistical testing using the GofuncR package.

Overall, my major concern with this work is that the authors make two central claims in this paper (as per the Discussion). The authors claim that Q-rich motifs enhance protein expression. The implication here is that Q-rich motif IDRs are special, but this is not tested. As such, they cannot exclude the competing hypothesis ("N-terminal disordered regions enhance expression").

In fact, “N-terminal disordered regions enhance expression” exactly summarizes our hypothesis.

On pages 12-13 and Figure 4 of our preprint manuscript, we explained our hypothesis in the paragraph entitled “The relationship between PEE function, amino acid contents, and structural flexibility”.

The authors also do not explore the possibility that this effect is in part/entirely driven by mRNA-level effects (see Verma Na Comms 2019).

As pointed out by the first reviewer, we show evidence that the increase in protein abundance and enzymatic activity is not due to changes in plasmid copy number or mRNA abundance (Figure 2), and that this phenomenon is not affected by translational quality control mutants (Figure 3).

As such, while these observations are interesting, they feel preliminary and, in my opinion, cannot be used to draw hard conclusions on how N-terminal IDR sequence features influence protein expression. This does not mean the authors are necessarily wrong, but from the data presented here, I do not believe strong conclusions can be drawn. That re-assignment of stop codons to Q increases proteome-wide Q usage. I was unable to understand what result led the authors to this conclusion.

My reading of the results is that a subset of ciliates has re-assigned UAA and UAG from the stop codon to Q. Those ciliates have more polyQ-containing proteins. However, they also have more polyN-containing proteins and proteins enriched in S/T-Q clusters. Surely if this were a stop-codon-dependent effect, we'd ONLY see an enhancement in Q-richness, not a corresponding enhancement in all polar-rich IDR frequencies? It seems the better working hypothesis is that free-floating climate proteomes are enriched in polar amino acids compared to sessile ciliates.

Thank you. These comments are not supported by the results in Figure 1.

Regardless, the absence of any kind of statistical analysis makes it hard to draw strong conclusions here.

We apologize for not explaining more clearly the results of Tables 5-7 in our preprint manuscript.

To address the concerns about our GO enrichment analysis by both reviewers, we have now performed rigorous statistical testing for SCD and polyQ protein overrepresentation using the GOfuncR package (https://bioconductor.org/packages/release/bioc/html/GOfuncR.html). GOfuncR is an R package program that conducts standard candidate vs. background enrichment analysis by means of the hypergeometric test. We then adjusted the raw p-values according to the Family-wise error rate (FWER). The same method had been applied to GO enrichment analysis of human genomes (Huttenhower, C., et al. 2009).

Curtis Huttenhower, C., Haley, E. M., Hibbs, M., A., Dumeaux, V., Barrett, D. R., Hilary A. Coller, H. A., and Olga G. Troyanskaya, O., G. Exploring the human genome with functional maps, Genome Research 19, 1093-1106 (2009).

The results presented in Author response image 5 and Author response image 6 support our hypothesis that Q-rich motifs prevail in proteins involved in specialized biological processes, including Saccharomyces cerevisiae RNA-mediated transposition, Candida albicans filamentous growth, peptidyl-glutamic acid modification in ciliates with reassigned stop codons (TAAQ and TAGQ), Tetrahymena thermophila xylan catabolism, Dictyostelium discoideum sexual reproduction, Plasmodium falciparum infection, as well as the nervous systems of Drosophila melanogaster, Mus musculus, and Homo sapiens (74). In contrast, peptidyl-glutamic acid modification and microtubule-based movement are not overrepresented with Q-rich proteins in Stentor coeruleus, a ciliate with standard stop codons.

1. Cara L, Baitemirova M, Follis J, Larios-Sanz M, Ribes-Zamora A. The ATM- and ATR-related SCD domain is over-represented in proteins involved in nervous system development. Sci Rep. 2016;6:19050.

Author response image 5.

Selection of biological processes with overrepresented SCD-containing proteins in different eukaryotes. The percentages and number of SCD-containing proteins in our search that belong to each indicated Gene Ontology (GO) group are shown. GOfuncR (Huttenhower, C., et al. 2009) was applied for GO enrichment and statistical analysis. The p values adjusted according to the Family-wise error rate (FWER) are shown. The five ciliates with reassigned stop codons (TAAQ and TAGQ) are indicated in red. Stentor coeruleus, a ciliate with standard stop codons, is indicated in green.

Author response image 6.

Selection of biological processes with overrepresented polyQ-containing proteins in different eukaryotes. The percentages and numbers of polyQ-containing proteins in our search that belong to each indicated Gene Ontology (GO) group are shown. GOfuncR (Huttenhower, C., et al. 2009) was applied for GO enrichment and statistical analysis. The p values adjusted according to the Family-wise error rate (FWER) are shown. The five ciliates with reassigned stops codons (TAAQ and TAGQ) are indicated in red. Stentor coeruleus, a ciliate with standard stop codons, is indicated in green.

Author Response

Reviewer #1 (Public Review):

Wang and all present an interesting body of work focused on the effects of high altitude and hypoxia on erythropoiesis, resulting in erythrocytosis. This work is specifically focused on the spleen, identifying splenic macrophages as central cells in this effect. This is logical since these cells are involved in erythrophagocytosis and iron recycling. The results suggest that hypoxia induces splenomegaly with decreased number of splenic macrophages. There is also evidence that ferroptosis is induced in these macrophages, leading to cell destruction. Finally, the data suggest that ferroptosis in splenic red pulp macrophages causes the decrease in RBC clearance, resulting in erythrocytosis aka lengthening the RBC lifespan. However, there are many issues with the presented results, with somewhat superficial data, meaning the conclusions are overstated and there is decreased confidence that the hypotheses and observed results are directly causally related to hypoxia.

Major points:

1) The spleen is a relatively poorly understood organ but what is known about its role in erythropoiesis especially in mice is that it functions both to clear as well as to generate RBCs. The later process is termed extramedullary hematopoiesis and can occur in other bones beyond the pelvis, liver, and spleen. In mice, the spleen is the main organ of extramedullary erythropoiesis. The finding of transiently decreased spleen size prior to splenomegaly under hypoxic conditions is interesting but not well developed in the manuscript. This is a shortcoming as this is an opportunity to evaluate the immediate effect of hypoxia separately from its more chronic effect. Based just on spleen size, no conclusions can be drawn about what happens in the spleen in response to hypoxia.

Thank you for your insightful comments and questions. The spleen is instrumental in both immune response and the clearance of erythrocytes, as well as serving as a significant reservoir of blood in the body. This organ, characterized by its high perfusion rate and pliability, constricts under conditions of intense stress, such as during peak physical exertion, the diving reflex, or protracted periods of apnea. This contraction can trigger an immediate release of red blood cells (RBCs) into the bloodstream in instances of substantial blood loss or significant reduction of RBCs. Moreover, elevated oxygen consumption rates in certain animal species can be partially attributed to splenic contractions, which augment hematocrit levels and the overall volume of circulating blood, thereby enhancing venous return and oxygen delivery (Dane et al. J Appl Physiol, 2006, 101:289-97; Longhurst et al. Am J Physiol, 1986, 251: H502-9). In our investigation, we noted a significant contraction of the spleen following exposure to hypoxia for a period of one day. We hypothesized that the body, under such conditions, is incapable of generating sufficient RBCs promptly enough to facilitate enhanced oxygen delivery. Consequently, the spleen reacts by releasing its stored RBCs through splenic constriction, leading to a measurable reduction in spleen size.

However, we agree with you that further investigation is required to fully understand the implications of these changes. Considering the comments, we propose to extend our research by incorporating more detailed examinations of spleen morphology and function during hypoxia, including the potential impact on extramedullary hematopoiesis. We anticipate that such an expanded analysis would not only help elucidate the initial response to hypoxia but also provide insights into the more chronic effects of this condition on spleen function and erythropoiesis.

2) Monocyte repopulation of tissue resident macrophages is a minor component of the process being described and it is surprising that monocytes in the bone marrow and spleen are also decreased. Can the authors conjecture why this is happening? Typically, the expectation would be that a decrease in tissue resident macrophages would be accompanied by an increase in monocyte migration into the organ in a compensatory manner.

We appreciate your insightful query regarding the observed decrease in monocytes in the bone marrow and spleen, particularly considering the typical compensatory increase in monocyte migration into organs following a decrease in tissue resident macrophages.

The observed decrease in monocytes within the bone marrow is likely attributable to the fact that monocytes and precursor cells for red blood cells (RBCs) both originate from the same hematopoietic stem cells within the bone marrow. It is well established that exposure to hypobaric hypoxia (HH) induces erythroid differentiation specifically within the bone marrow, originating from these hematopoietic stem cells. As such, we postulate that the differentiation into monocytes is reduced under hypoxic conditions, which may subsequently cause a decrease in migration to the spleen.

Furthermore, we hypothesize that an increased migration of monocytes to other tissues under HH exposure may also contribute to the decreased migration to the spleen. The liver, which partially contributes to the clearance of RBCs, may play a role in this process. Our investigations to date have indeed identified an increased monocyte migration to the liver. We were pleased to discover an elevation in CSF1 expression in the liver following HH exposure for both 7 and 14 days. This finding was corroborated through flow cytometry, which confirmed an increase in monocyte migration to the liver.

Consequently, we propose that under HH conditions, the liver requires an increased influx of monocytes, which in turn leads to a decrease in monocyte migration to the spleen. However, it is important to note that these findings will be discussed more comprehensively in our forthcoming publication, and as such, the data pertaining to these results have not been included in the current manuscript.

3) Figure 3 does not definitively provide evidence that cell death is specifically occurring in splenic macrophages and the fraction of Cd11b+ cells is not changed in NN vs HH. Furthermore, the IHC of F4/80 in Fig 3U is not definitive as cells can express F4/80 more or less brightly and no negative/positive controls are shown for this panel.

We appreciate your insightful comments and critiques regarding Figure 3. We acknowledge that the figure, as presented, does not definitively demonstrate that cell death is specifically occurring in splenic macrophages. While it is challenging to definitively determine the occurrence of cell death in macrophages based solely on Figure 3D-F, our single-cell analysis provides strong evidence that such an event occurs. We initially observed cell death within the spleen under hypobaric hypoxia (HH) conditions, and to discern the precise cell type involved, we conducted single-cell analyses. Regrettably, we did not articulate this clearly in our preliminary manuscript. In the revised version, we have modified the sequence of Figure 3A-C and Figure 3D-F for better clarity. Besides, we observed a significant decrease in the fraction of F4/80hiCD11bhi macrophages under HH conditions compared to NN. To make the changes more evident in CD86 and CD206, we have transformed these scatter plots into histograms in our revised manuscript.

Considering the limitations of F4/80 as a conclusive macrophage identifier, we have concurrently presented the immunohistochemical (IHC) analyses of heme oxygenase-1 (HO-1). Functioning as a macrophage marker, particularly in cells involved in iron metabolism, HO-1 offers additional diagnostic accuracy. Observations from both F4/80 and HO-1 staining suggested a primary localization of positively stained cells within the splenic red pulp. Following exposure to hypoxia-hyperoxia (HH) conditions, a decrease was noted in the expression of both F4/80 and HO-1. This decrease implies that HH conditions contribute to a reduction in macrophage population and impede the iron metabolism process. In the revised version of our manuscript, we have enhanced the clarity of Figure 3U to illustrate the presence of positive staining, with an emphasis on HO-1 staining, which is predominantly observed in the red pulp.

4) The phagocytic function of splenic red pulp macrophages relative to infection cannot be used directly to understand erythrophagocytosis. The standard approach is to use opsonized RBCs in vitro. Furthermore, RBC survival is a standard method to assess erythrophagocytosis function. In this method, biotin is injected via tail vein directly and small blood samples are collected to measure the clearance of biotinilation by flow; kits are available to accomplish this. Because the method is standard, Fig 4D is not necessary and Fig 4E needs to be performed only in blood by sampling mice repeatedly and comparing the rate of biotin decline in HH with NN (not comparing 7 d with 14 d).

We appreciate your insightful comments and suggestions. We concur that the phagocytic function of splenic red pulp macrophages in the context of infection may not be directly translatable to understanding erythrophagocytosis. Given our assessment that the use of cy5.5-labeled E.coli alone may not be sufficient to accurately evaluate the phagocytic function of macrophages, we extended our study to include the use of NHS-biotin-labeled RBCs to assess phagocytic capabilities. While the presence of biotin-labeled RBCs in the blood could provide an indication of RBC clearance, this measure does not exclusively reflect the spleen's role in the process, as it fails to account for the clearance activities of other organs.

Consequently, we propose that the remaining biotin-labeled RBCs in the spleen may provide a more direct representation of the organ's function in RBC clearance and sequestration. Our observations of diminished erythrophagocytosis at both 7 and 14 days following exposure to HH guided our subsequent efforts to quantify biotin-labeled RBCs in both the circulatory system and spleen. These measurements were conducted during the 7 to 14-day span following the confirmation of impaired erythrophagocytosis. Comparative evaluation of RBC clearance rates under NN and HH conditions provided further evidence supporting our preliminary observations, with the data revealing a decrease in the RBC clearance rate in the context of HH conditions. In response to feedback from other reviewers, we have elected to exclude the phagocytic results and the diagram of the erythrocyte labeling assay. These amendments will be incorporated into the revised manuscript. The reviewers' constructive feedback has played a crucial role in refining the methodological precision and coherence of our investigation.

5) It is unclear whether Tuftsin has a specific effect on phagocytosis of RBCs without other potential confounding effects. Furthermore, quantifying iron in red pulp splenic macrophages requires alternative readily available more quantitative methods (e.g. sorted red pulp macrophages non-heme iron concentration).

We appreciate your comments and questions regarding the potential effect of Tuftsin on the phagocytosis of RBCs and the quantification of iron in red pulp splenic macrophages. Regarding the role of Tuftsin, we concur that the literature directly associating Tuftsin with erythrophagocytosis is scant. The work of Gino Roberto Corazza et al. does suggest a link between Tuftsin and general phagocytic capacity, but it does not specifically address erythrophagocytosis (Am J Gastroenterol, 1999;94:391-397). We agree that further investigations are required to elucidate the potential confounding effects and to ascertain whether Tuftsin has a specific impact on the phagocytosis of RBCs. Concerning the quantification of iron in red pulp splenic macrophages, we acknowledge your suggestion to employ readily available and more quantitative methods. We have incorporated additional Fe2+ staining in the spleen at two time points: 7 and 14 days subsequent to HH exposure (refer to the following Figure). The resultant data reveal an escalated deposition of Fe2+ within the red pulp, as evidenced in Figures 5 (panels L and M) and Figure 7 (panels L and M).

6) In Fig 5, PBMCs are not thought to represent splenic macrophages and although of some interest, does not contribute significantly to the conclusions regarding splenic macrophages at the heart of the current work. The data is also in the wrong direction, namely providing evidence that PBMCs are relatively iron poor which is not consistent with ferroptosis which would increase cellular iron.

We appreciate your insightful critique regarding Figure 5 and the interpretation of our data on peripheral blood mononuclear cells (PBMCs) in relation to splenic macrophages. We understand that PBMCs do not directly represent splenic macrophages, and we agree that any conclusions drawn from PBMCs must be considered with caution when discussing the behavior of splenic macrophages.

The primary rationale for incorporating PBMCs into our study was to investigate the potential correspondence between their gene expression changes and those observed in the spleen after HH exposure. This was posited as a working hypothesis for further exploration rather than a conclusive statement. The gene expression in PBMCs was congruous with changes in the spleen's gene expression, demonstrating an iron deficiency phenotype, ostensibly due to the mobilization of intracellular iron for hemoglobin synthesis. Thus, it is plausible that NCOA4 may facilitate iron mobilization through the degradation of ferritin to store iron.

It remains ambiguous whether ferroptosis was initiated in the PBMCs during our study. Ferroptosis primarily occurs as a response to an increase in Fe2+ rather than an overall increase in intracellular iron. Our preliminary proposition was that relative changes in gene expression in PBMCs could potentially mirror corresponding changes in protein expression in the spleen, thereby potentially indicating alterations in iron processing capacity post-HH exposure. However, we fully acknowledge that this is a conjecture requiring further empirical substantiation or clinical validation.

7) Tfr1 increase is typically correlated with cellular iron deficiency while ferroptosis consistent with iron loading. The direction of the changes in multiple elements relevant to iron trafficking is somewhat confusing and without additional evidence, there is little confidence that the authors have reached the correct conclusion. Furthermore, the results here are analyses of total spleen samples rather than specific cells in the spleen.

We appreciate your astute comments and agree that the observed increase in transferrin receptor (TfR) expression, typically associated with cellular iron deficiency, appears contradictory to the expected iron-loading state associated with ferroptosis. We understand that this apparent contradiction might engender some uncertainty about our conclusions.

In our investigation, we evaluated total spleen samples as opposed to distinct cell types within the spleen, a factor that could have contributed to the seemingly discordant findings. An integral element to bear in mind is the existence of immature RBCs in the spleen, particularly within the hematopoietic island where these immature RBCs cluster around nurse macrophages. These immature RBCs contain abundant TfR which was needed for iron uptake and hemoglobin synthesis. These cells, which prove challenging to eliminate via perfusion, might have played a role in the observed upregulation in TfR expression, especially in the aftermath of HH exposure. Our further research revealed that the expression of TfR in macrophages diminished following hypoxic conditions, thereby suggesting that the elevated TfR expression in tissue samples may predominantly originate from other cell types, especially immature RBCs (refer to subsequent Figure).

Reviewer #2 (Public Review):

The authors aimed at elucidating the development of high altitude polycythemia which affects mice and men staying in the hypoxic atmosphere at high altitude (hypobaric hypoxia; HH). HH causes increased erythropoietin production which stimulates the production of red blood cells. The authors hypothesize that increased production is only partially responsible for exaggerated red blood cell production, i.e. polycythemia, but that decreased erythrophagocytosis in the spleen contributes to high red blood cells counts.

The main strength of the study is the use of a mouse model exposed to HH in a hypobaric chamber. However, not all of the reported results are convincing due to some smaller effects which one may doubt to result in the overall increase in red blood cells as claimed by the authors. Moreover, direct proof for reduced erythrophagocytosis is compromised due to a strong spontaneous loss of labelled red blood cells, although effects of labelled E. coli phagocytosis are shown. Their discussion addresses some of the unexpected results, such as the reduced expression of HO-1 under hypoxia but due to the above-mentioned limitations much of the discussion remains hypothetical.

Thank you for your valuable feedback and insight. We appreciate the recognition of the strength of our study model, the exposure of mice to hypobaric hypoxia (HH) in a hypobaric animal chamber. We also understand your concerns about the smaller effects and their potential impact on the overall increase in red blood cells (RBCs), as well as the apparent reduced erythrophagocytosis due to the loss of labelled RBCs.

Erythropoiesis has been predominantly attributed to the amplified production of RBCs under conditions of HH. The focus of our research was to underscore the potential acceleration of hypoxia-associated polycythemia (HAPC) as a result of compromised erythrophagocytosis. Considering the spontaneous loss of labelled RBCs in vivo, we assessed the clearance rate of RBCs at the stages of 7 and 14 days within the HH environment, and subsequently compared this rate within the period from 7 to 14 days following the clear manifestation of erythrophagocytosis impairment at the two aforementioned points identified in our study. This approach was designed to negate the effects of spontaneous loss of labelled RBCs in both NN and HH conditions. Correspondingly, the results derived from blood and spleen analyses corroborated a decline in the RBC clearance rate under HH when juxtaposed with NN conditions.

Apart from the E. coli phagocytosis and the labeled RBCs experiment (this part of the results was removed in the revision), the injection of Tuftsin further substantiated the impairment of erythrophagocytosis in the HH spleen, as evidenced by the observed decrease in iron within the red pulp of the spleen post-perfusion. Furthermore, to validate our findings, we incorporated RBCs staining in splenic cells at 7 and 14 days of HH exposure, which provided concrete confirmation of impaired erythrophagocytosis (new Figure 4E).

As for the reduced expression of heme oxygenase-1 (HO-1) under hypoxia, we agree that this was an unexpected result, and we are in the process of further exploring the underlying mechanisms. It is possible that there are other regulatory pathways at play that are yet to be identified. However, we believe that by offering possible interpretations of our data and potential directions for future research, we contribute to the ongoing scientific discourse in this area.

Reviewer #3 (Public Review):

The manuscript by Yang et al. investigated in mice how hypobaric hypoxia can modify the RBC clearance function of the spleen, a concept that is of interest. Via interpretation of their data, the authors proposed a model that hypoxia causes an increase in cellular iron levels, possibly in RPMs, leading to ferroptosis, and downregulates their erythrophagocytic capacity. However, most of the data is generated on total splenocytes/total spleen, and the conclusions are not always supported by the presented data. The model of the authors could be questioned by the paper by Youssef et al. (which the authors cite, but in an unclear context) that the ferroptosis in RPMs could be mediated by augmented erythrophagocytosis. As such, the loss of RPMs in vivo which is indeed clear in the histological section shown (and is a strong and interesting finding) can be not directly caused by hypoxia, but by enhanced RBC clearance. Such a possibility should be taken into account.

Thank you for your insightful comments and constructive feedback. In their research, Youssef et al. (2018) discerned that elevated erythrophagocytosis of stressed red blood cells (RBCs) instigates ferroptosis in red pulp macrophages (RPMs) within the spleen, as evidenced in a mouse model of transfusion. This augmentation of erythrophagocytosis was conspicuous five hours post-injection of RBCs. Conversely, our study elucidated the decrease in erythrophagocytosis in the spleen after both 7 and 14 days.

Typically, macrophages exhibit an enhanced phagocytic capacity in the immediate aftermath of stress or stimulation. Nonetheless, the temporal points of observation in our study were considerably extended (seven and fourteen days). It remains uncertain whether phagocytic capability was amplified during the acute phase of HH exposure—particularly within the first day, considering that splenoconstriction under HH for one day results in the release of stored RBCs into the bloodstream—and whether this initial response could precipitate ferroptosis and subsequently diminished erythrophagocytosis at the 7 or 14 day marks under continued HH conditions.

Major points:

1) The authors present data from total splenocytes and then relate the obtained data to RPMs, which are quantitatively a minor population in the spleen. Eg, labile iron is increased in the splenocytes upon HH, but the manuscript does not show that this occurs in the red pulp or RPMs. They also measure gene/protein expression changes in the total spleen and connect them to changes in macrophages, as indicated in the model Figure (Fig. 7). HO-1 and levels of Ferritin (L and H) can be attributed to the drop in RPMs in the spleen. Are any of these changes preserved cell-intrinsically in cultured macrophages? This should be shown to support the model (relates also to lines 487-88, where the authors again speculate that hypoxia decreases HO-1 which was not demonstrated). In the current stage, for example, we do not know if the labile iron increase in cultured cells and in the spleen in vivo upon hypoxia is the same phenomenon, and why labile iron is increased. To improve the manuscript, the authors should study specifically RPMs.

We express our gratitude for your perceptive remarks. In our initial manuscript, we did not evaluate labile iron within the red pulp and red pulp macrophages (RPMs). To address this oversight, we utilized the Lillie staining method, in accordance with the protocol outlined by Liu et al., (Chemosphere, 2021, 264(Pt 1):128413), to discern Fe2+ presence within these regions. The outcomes were consistent with our antecedent Western blot and flow cytometry findings in the spleen, corroborating an increment in labile iron specifically within the red pulp of the spleen.

However, we acknowledge the necessity for other supplementary experimental efforts to further validate these findings. Additionally, we scrutinized the expression of heme oxygenase-1 (HO-1) and iron-related proteins, including transferrin receptor (TfR), ferroportin (Fpn), ferritin (Ft), and nuclear receptor coactivator 4 (NCOA4) in primary macrophages subjected to 1% hypoxic conditions, both with and without hemoglobin treatment. Our results indicated that the expression of ferroptosis-related proteins was consistent with in vivo studies, however the expression of iron related proteins was not similar in vitro and in vivo. It suggesting that the increase in labile iron in cultured cells and the spleen in vivo upon hypoxia are not identical phenomena. However, the precise mechanism remains elusive.

In our study, we observed a decrease in HO-1 protein expression following 7 and 14 days of HH exposure, as shown in Figure 3U, 5A, and S1A. This finding contradicts previous research that identified HO-1 as a hypoxia-inducible factor (HIF) target under hypoxic conditions (P J Lee et al., 1997). Our discussion, therefore, addressed the potential discrepancy in HO-1 expression under HH. According to our findings, HO-1 regulation under HH appears to be predominantly influenced by macrophage numbers and the RBCs to be processed in the spleen or macrophages, rather than by hypoxia alone.

It is challenging to discern whether the increased labile iron observed in vitro accurately reflects the in vivo phenomenon, as replicating the iron requirements for RBCs production induced by HH in vitro is inherently difficult. However, by integrating our in vivo and in vitro studies, we determined that the elevated Fe2+ levels were not dependent on HO-1 protein expression, as HO-1 levels was increased in vitro while decreasing in vivo under hypoxic/HH exposure.

2) The paper uses flow cytometry, but how this method was applied is suboptimal: there are no gating strategies, no indication if single events were determined, and how cell viability was assessed, which are the parent populations when % of cells is shown on the graphs. How RBCs in the spleen could be analyzed without dedicated cell surface markers? A drop in splenic RPMs is presented as the key finding of the manuscript but Fig. 3M shows gating (suboptimal) for monocytes, not RPMs. RPMs are typically F4/80-high, CD11-low (again no gating strategy is shown for RPMs). Also, the authors used single-cell RNAseq to detect a drop in splenic macrophages upon HH, but they do not indicate in Fig. A-C which cluster of cells relates to macrophages. Cell clusters are not identified in these panels, hence the data is not interpretable).

Thank you for your comments and constructive critique regarding our flow cytometry methodology and presentation. We understand the need for greater transparency and detailed explanation of our procedures, and we acknowledge that the lack of gating strategies and other pertinent information in our initial manuscript may have affected the clarity of our findings.

In our initial report, we provided an overview of the decline in migrated macrophages (F4/80hiCD11bhi), including both M1 and M2 expression in migrated macrophages, as illustrated in Figure 3, but did not specifically address the changes in red pulp macrophages (RPMs). Based on previous results, it is difficult to identify CD11b- and CD11blo cells. We will repeat the results and attempt to identify F4/80hiCD11blo cells in the revised manuscript. The results of the reanalysis are now included (Figure 3M). However, single-cell in vivo analysis studies may more accurately identify specific cell types that decrease after exposure to HH.

Furthermore, we substantiated the reduction in red pulp, as evidenced by Figure 4J, given that iron processing primarily occurs within the red pulp. In Figure 3, our initial objective was merely to illustrate the reduction in total macrophages in the spleen following HH exposure.

To further clarify the characterization of various cell types, we conducted a single-cell analysis. Our findings indicated that clusters 0,1,3,4,14,18, and 29 represented B cells, clusters 2, 10, 12, and 28 represented T cells, clusters 15 and 22 corresponded to NK cells, clusters 5, 11, 13, and 19 represented NKT cells, clusters 6, 9, and 24 represented cell cycle cells, clusters 26 and 17 represented plasma cells, clusters 21 and 23 represented neutrophils, cluster 30 represented erythrocytes, and clusters 7, 8, 16, 20, 24, and 27 represented dendritic cells (DCs) and macrophages, as depicted in Figure 3E.

3) The authors draw conclusions that are not supported by the data, some examples: a) they cannot exclude eg the compensatory involvement of the liver in the RBCs clearance (the differences between HH sham and HH splenectomy is mild in Fig. 2 E, F and G).

Thank you for your insightful comments and for pointing out the potential involvement of other organs, such as the liver, in the RBC clearance under HH conditions. We concur with your observation that the differences between the HH sham and HH splenectomy conditions in Fig. 2 E, F, and G are modest. This could indeed suggest a compensatory role of other organs in RBC clearance when splenectomy is performed. Our intent, however, was to underscore the primary role of the spleen in this process under HH exposure.

In fact, after our initial investigations, we conducted a more extensive study examining the role of the liver in RBC clearance under HH conditions. Our findings, as illustrated in the figures submitted with this response, indeed support a compensatory role for the liver. Specifically, we observed an increase in macrophage numbers and phagocytic activity in the liver under HH conditions. Although the differences in RBC count between the HH sham and HH splenectomy conditions may seem minor, it is essential to consider the unit of this measurement, which is value*1012/ml. Even a small numerical difference can represent a significant biological variation at this scale.

b) splenomegaly is typically caused by increased extramedullary erythropoiesis, not RBC retention. Why do the authors support the second possibility? Related to this, why do the authors conclude that data in Fig. 4 G,H support the model of RBC retention? A significant drop in splenic RBCs (poorly gated) was observed at 7 days, between NN and HH groups, which could actually indicate increased RBC clearance capacity = less retention.

Prior investigations have predominantly suggested that spleen enlargement under hypoxic conditions stems from the spleen's extramedullary hematopoiesis. Nevertheless, an intriguing study conducted in 1994 by the General Hospital of Xizang Military Region reported substantial exaggeration and congestion of splenic sinuses in high altitude polycythemia (HAPC) patients. This finding was based on the dissection of spleens from 12 patients with HAPC (Zou Xunda, et al., Southwest Defense Medicine, 1994;5:294-296). Moreover, a recent study indicated that extramedullary erythropoiesis reaches its zenith between 3 to 7 days (Wang H et al., 2021).

Considering these findings, the present study postulates that hypoxia-induced inhibition of erythrophagocytosis may lead to RBC retention. However, we acknowledge that the manuscript in its current preprint form does not offer conclusive evidence to substantiate this hypothesis. To bridge this gap, we further conducted experiments where the spleen was perfused, and total cells were collected post HH exposure. These cells were then smeared onto slides and subjected to Wright staining. Our results unequivocally demonstrate an evident increase in deformation and retention of RBCs in the spleen following 7 and 14 days of HH exposure. This finding strengthens our initial hypothesis and contributes a novel perspective to the understanding of splenic responses under hypoxic conditions.

c) lines 452-54: there is no data for decreased phagocytosis in vivo, especially in the context of erythrophagocytosis. This should be done with stressed RBCs transfusion assays, very good examples, like from Youssef et al. or Threul et al. are available in the literature.

Thanks. In their seminal work, Youssef and colleagues demonstrated that the transfusion of stressed RBCs triggers erythrophagocytosis and subsequently incites ferroptosis in red pulp macrophages (RPMs) within a span of five hours. Given these observations, the applicability of this model to evaluate macrophage phagocytosis in the spleen or RPMs under HH conditions may be limited, as HH has already induced erythropoiesis in vivo. In addition, it was unclear whether the membrane characteristics of stress induced RBCs were similar to those of HH induced RBCs, as this is an important signal for in vivo phagocytosis. The ambiguity arises from the fact that we currently lack sufficient knowledge to discern whether the changes in phagocytosis are instigated by the presence of stressed RBCs or by changes of macrophages induced by HH in vivo. Nonetheless, we appreciate the potential value of this approach and intend to explore its utility in our future investigations. The prospect of distinguishing the effects of stressed RBCs from those of HH on macrophage phagocytosis is an intriguing line of inquiry that could yield significant insights into the mechanisms governing these physiological processes. We will investigate this issue in our further study.

d) Line 475 - ferritinophagy was not shown in response to hypoxia by the manuscript, especially that NCOA4 is decreased, at least in the total spleen.

Drawing on the research published in eLife in 2015, it was unequivocally established that ferritinophagy, facilitated by Nuclear Receptor Coactivator 4 (NCOA4), is indispensable for erythropoiesis. This process is modulated by iron-dependent HECT and RLD domain containing E3 ubiquitin protein ligase 2 (HERC2)-mediated proteolysis (Joseph D Mancias et al., eLife. 2015; 4: e10308). As is widely recognized, NCOA4 plays a critical role in directing ferritin (Ft) to the lysosome, where both NCOA4 and Ft undergo coordinated degradation.

In our study, we provide evidence that exposure to HH stimulates erythropoiesis (Figure 1). We propose that this, in turn, could promote ferritinophagy via NCOA4, resulting in a decrease in NCOA4 protein levels post-HH exposure. We will further increase experiments to verify this concern. This finding not only aligns with the established understanding of ferritinophagy and erythropoiesis but also adds a novel dimension to the understanding of cellular responses to hypoxic conditions.

4) In a few cases, the authors show only representative dot plots or histograms, without quantification for n>1. In Fig. 4B the authors write about a significant decrease (although with n=1 no statistics could be applied here; of note, it is not clear what kind of samples were analyzed here). Another example is Fig. 6I. In this case, it is even more important as the data are conflicting the cited article and the new one: PMCID: PMC9908853 which shows that hypoxia stimulates efferocytosis. Sometimes the manuscript claim that some changes are observed, although they are not visible in representative figures (eg for M1 and M2 macrophages in Fig. 3M)

We recognize that our initial portrayal of Figure 4B was lacking in precision, given that it did not include the corresponding statistical graph. While our results demonstrated a significant reduction in the ability to phagocytose E. coli, in line with the recommendations of other reviewers, we have opted to remove the results pertaining to E. coli phagocytosis in this revision, as they primarily reflected immune function. In relation to PMC9908853, which reported metabolic adaptation facilitating enhanced macrophage efferocytosis in limited-oxygen environments, it is worth noting that the macrophages investigated in this study were derived from ER-Hoxb8 macrophage progenitors following the removal of β-estradiol. Consequently, questions arise regarding the comparability between these cultured macrophages and primary macrophages obtained fresh from the spleen post HH exposure. The characteristics and functions of these two different macrophage sources may not align precisely, and this distinction necessitates further investigation.

5) There are several unclear issues in methodology:

• what is the purity of primary RPMs in the culture? RPMs are quantitatively poorly represented in splenocyte single-cell suspensions. This reviewer is quite skeptical that the processing of splenocytes from approx 1 mm3 of tissue was sufficient to establish primary RPM cultures. The authors should prove that the cultured cells were indeed RPMs, not monocyte-derived macrophages or other splenic macrophage subtypes.

Thank you for your thoughtful comments and inquiries. Firstly, I apologize if we did not make it clear in the original manuscript. The purity of the primary RPMs in our culture was found to be approximately 40%, as identified by F4/80hiCD11blo markers using flow cytometry. We recognize that RPMs are typically underrepresented in splenocyte single-cell suspensions, and the concern you raise about the potential for contamination by other cell types is valid.

We apologize for any ambiguities in the methodological description that may have led to misunderstandings during the review. Indeed, the entirety of the spleen is typically employed for splenic macrophage culture. The size of the spleen can vary dependent on the species and age of the animal, but in mice, it is commonly approximately 1 cm in length. The spleen is then dissected into minuscule fragments, each approximately 1 mm3 in volume, to aid in enzymatic digestion. This procedure does not merely utilize a single 1 mm3 tissue fragment for RPMs cultures. Although the isolation and culture of spleen macrophages can present considerable challenges, our method has been optimized to enhance the yield of this specific cell population.

• (around line 183) In the description of flow cytometry, there are several missing issues. In 1) it is unclear which type of samples were analyzed. In 2) it is not clear how splenocyte cell suspension was prepared.

1) Whole blood was extracted from the mice and collected into an anticoagulant tube, which was then set aside for subsequent thiazole orange (TO) staining. 2) Splenic tissue was procured from the mice and subsequently processed into a single-cell suspension using a 40 μm filter. The erythrocytes within the entire sample were subsequently lysed and eliminated, and the remaining cell suspension was resuspended in phosphate-buffered saline (PBS) in preparation for ensuing analyses.

We have meticulously revised these methodological details in the corresponding section of the manuscript to ensure clarity and precision.

• In line 192: what does it mean: 'This step can be omitted from cell samples'?

The methodology employed for the quantification of intracellular divalent iron content and lipid peroxidation level was executed as follows: Splenic tissue was first processed into a single cell suspension, subsequently followed by the lysis of RBCs. It should be noted that this particular stage is superfluous when dealing with isolated cell samples. Subsequently, a total of 1 × 106 cells were incubated with 100 μL of BioTracker Far-red Labile Fe2+ Dye (1 mM, Sigma, SCT037, USA) for a duration of 1 hour, or alternatively, C11-Bodipy 581/591 (10 μM, Thermo Fisher, D3861, USA) for a span of 30 minutes. Post incubation, cells were thoroughly washed twice with PBS. Flow cytometric analysis was subsequently performed, utilizing the FL6 (638 nm/660 nm) channel for the determination of intracellular divalent iron content, and the FL1 (488 nm/525 nm) channel for the quantification of the lipid peroxidation level.

• 'TO method' is not commonly used anymore and hence it was unclear to this Reviewer. Reticulocytes should be analyzed with proper gating, using cell surface markers.

We are appreciative of your astute observation pertaining to the methodology we employed to analyze reticulocytes in our study. We value your recommendation to utilize cell surface markers for effective gating, which indeed represents a more modern and accurate approach. However, as reticulocyte identification is not the central focus of our investigation, we opted for the TO staining method—due to its simplicity and credibility of results. In our initial exploration, we adopted the TO staining method in accordance with the protocol outlined (Sci Rep, 2018, 8(1):12793), primarily owing to its established use and demonstrated efficacy in reticulocyte identification.

• The description of 'phagocytosis of E. coli and RBCs' in the Methods section is unclear and incomplete. The Results section suggests that for the biotinylated RBCs, phagocytosis? or retention? Of RBCs was quantified in vivo, upon transfusion. However, the Methods section suggests either in vitro/ex vivo approach. It is vague what was indeed performed and how in detail. If RBC transfusion was done, this should be properly described. Of note, biotinylation of RBCs is typically done in vivo only, being a first step in RBC lifespan assay. The such assay is missing in the manuscript. Also, it is not clear if the detection of biotinylated RBCs was performed in permeablized cells (this would be required).

Thanks for the comments. In our initial methodology, we employed Cy5.5-labeled Escherichia coli to probe phagocytic function, albeit with the understanding that this may not constitute the most ideal model for phagocytosis detection within this context (in light of recommendations from other reviewers, we have removed the E. coli phagocytosis results from this revision, as they predominantly mirror immune function). Our fundamental aim was to ascertain whether HH compromises the erythrophagocytic potential of splenic macrophages. In pursuit of this, we subsequently analyzed the clearance of biotinylated RBCs in both the bloodstream and spleen to assess phagocytic functionality in vivo.

In the present study, instead of transfusing biotinylated RBCs into mice, we opted to inject N-Hydroxysuccinimide (NHS)-biotin into the bloodstream. NHS-biotin is capable of binding with cell membranes in vivo and can be recognized by streptavidin-fluorescein isothiocyanate (FITC) after cells are extracted from the blood or spleen in vitro. Consequently, biotin-labeled RBCs were detectable in both the blood and spleen following NHS-biotin injection for a duration of 21 days.

Ultimately, we employed flow cytometry to analyze the NHS-biotin labeled RBCs in the blood or spleen. This method facilitates the detection of live cells and is not applicable to permeabilized cells. We believe this approach better aligns with our investigative goals and offers a more robust evaluation of erythrophagocytic function under hypoxic conditions.

Author response:

Reviewer #1 (Public review):

Summary:

Jocher, Janssen, et al examine the robustness of comparative functional genomics studies in primates that make use of induced pluripotent stem cell-derived cells. Comparative studies in primates, especially amongst the great apes, are generally hindered by the very limited availability of samples, and iPSCs, which can be maintained in the laboratory indefinitely and defined into other cell types, have emerged as promising model systems because they allow the generation of data from tissues and cells that would otherwise be unobservable.

Undirected differentiation of iPSCs into many cell types at once, using a method known as embryoid body differentiation, requires researchers to manually assign all cell types in the dataset so they can be correctly analysed. Typically, this is done using marker genes associated with a specific cell type. These are defined a priori, and have historically tended to be characterised in mice and humans and then employed to annotate other species. Jocher, Janssen, et al ask if the marker genes and features used to define a given cell type in one species are suitable for use in a second species, and then quantify the degree of usefulness of these markers. They find that genes that are informative and cell type specific in a given species are less valuable for cell type identification in other species, and that this value, or transferability, drops off as the evolutionary distance between species increases.

This paper will help guide future comparative studies of gene expression in primates (and more broadly) as well as add to the growing literature on the broader challenges of selecting powerful and reliable marker genes for use in single-cell transcriptomics.

Strengths:

Marker gene selection and cell type annotation is a challenging problem in scRNA studies, and successful classification of cells often requires manual expert input. This can be hard to reproduce across studies, as, despite general agreement on the identity of many cell types, different methods for identifying marker genes will return different sets of genes. The rise of comparative functional genomics complicates this even further, as a robust marker gene in one species need not always be as useful in a different taxon. The finding that so many marker genes have poor transferability is striking, and by interrogating the assumption of transferability in a thorough and systematic fashion, this paper reminds us of the importance of systematically validating analytical choices. The focus on identifying how transferability varies across different types of marker genes (especially when comparing TFs to lncRNAs), and on exploring different methods to identify marker genes, also suggests additional criteria by which future researchers could select robust marker genes in their own data.

The paper is built on a substantial amount of clearly reported and thoroughly considered data, including EBs and cells from four different primate species - humans, orangutans, and two macaque species. The authors go to great lengths to ensure the EBs are as comparable as possible across species, and take similar care with their computational analyses, always erring on the side of drawing conservative conclusions that are robustly supported by their data over more tenuously supported ones that could be impacted by data processing artefacts such as differences in mappability, etc. For example, I like the approach of using liftoff to robustly identify genes in non-human species that can be mapped to and compared across species confidently, rather than relying on the likely incomplete annotation of the non-human primate genomes. The authors also provide an interactive data visualisation website that allows users to explore the dataset in depth, examine expression patterns of their own favourite marker genes and perform the same kinds of analyses on their own data if desired, facilitating consistency between comparative primate studies.

We thank the Reviewer for their kind assessment of our work.

Weaknesses and recommendations:

(1) Embryoid body generation is known to be highly variable from one replicate to the next for both technical and biological reasons, and the authors do their best to account for this, both by their testing of different ways of generating EBs, and by including multiple technical replicates/clones per species. However, there is still some variability that could be worth exploring in more depth. For example, the orangutan seems to have differentiated preferentially towards cardiac mesoderm whereas the other species seemed to prefer ectoderm fates, as shown in Figure 2C. Likewise, Supplementary Figure 2C suggests a significant unbalance in the contributions across replicates within a species, which is not surprising given the nature of EBs, while Supplementary Figure 6 suggests that despite including three different clones from a single rhesus macaque, most of the data came from a single clone. The manuscript would be strengthened by a more thorough exploration of the intra-species patterns of variability, especially for the taxa with multiple biological replicates, and how they impact the number of cell types detected across taxa, etc.

You are absolutely correct in pointing out that the large clonal variability in cell type composition is a challenge for our analysis. We also noted the odd behavior of the orangutan EBs, and their underrepresentation of ectoderm. There are many possible sources for these variable differentiation propensities: clone, sample origin (in this case urine) and individual. However, unfortunately for the orangutan, we have only one individual and one sample origin and thus cannot say whether this germ layer preference says something about the species or is due to our specific sample.

Because of this high variability from multiple sources, getting enough cell types with an appreciable overlap between species was limiting to analyses. In order to be able to derive meaningful conclusions from intra-species analyses and the impact of different sources of variation on cell type propensity, we would need to sequence many more EBs with an experimental design that balances possible sources of variation. This would go beyond the scope of this study.

Instead, here we control for intra-species variation in our analyses as much as possible: For the analysis of cell type specificity and conservation the comparison is relative for the different specificity degrees (Figure 3C).  For the analysis of marker gene conservation, we explicitly take intra-species variation into account (Figure 4D).

The same holds for the temporal aspect of the data, which is not really discussed in depth despite being a strength of the design. Instead, days 8 and 16 are analysed jointly, without much attention being paid to the possible differences between them.

Concerning the temporal aspect, indeed we knowingly omitted to include an explicit comparison of day 8 and day 16 EBs, because we felt that it was not directly relevant to our main message. Our pseudotime analysis showed that the differences of the two time points were indeed a matter of degree and not so much of quality. All major lineages were already present at day 8 and even though day 8 cells had on average earlier pseudotimes, there was a large overlap in the pseudotime distributions between the two sampling time points (Author response image 1). That is why we decided to analyse the data together.

Are EBs at day 16 more variable between species than at day 8? Is day 8 too soon to do these kinds of analyses?

When we started the experiment, we simply did not know what to expect. We were worried that cell types at day 8 might be too transient, but longer culture can also introduce biases. That is why we wanted to look at two time points, however as mentioned above the differences are in degree.

Concerning the cell type composition: yes, day 16 EBs are more heterogeneous than day 8 EBs. Firstly, older EBs have more distinguishable cell types and hence even if all EBs had identical composition, the sampling variance would be higher given that we sampled a similar number of cells from both time points. Secondly, in order to grow EBs for a longer time, we moved them from floating to attached culture on day 8 and it is unclear how much variance is added by this extra handling step.

Are markers for earlier developmental progenitors better/more transferable than those for more derived cell types?

We did not see any differences in the marker conservation between early and late cell types, but we have too little data to say whether this carries biological meaning.

Author response image 1.

Pseudotime analysis for a differentiation trajectory towards neurons. Single cells were first aggregated into metacells per species using SEACells (Persad et al. 2023). Pluripotent and ectoderm metacells were then integrated across all four species using Harmony and a combined pseudotime was inferred with Slingshot (Street et al. 2018), specifying iPSCs as the starting cluster. Here, lineage 3 is shown, illustrating a differentiation towards neurons. (A) PHATE embedding colored by pseudotime (Moon et al. 2019). (B) PHATE embedding colored by celltype. (C) Pseudotime distribution across the sampling timepoints (day 8 and day 16) in different species.

(2) Closely tied to the point above, by necessity the authors collapse their data into seven fairly coarse cell types and then examine the performance of canonical marker genes (as well as those discovered de novo) across the species. However some of the clusters they use are somewhat broad, and so it is worth asking whether the lack of specificity exhibited by some marker genes and driving their conclusions is driven by inter-species heterogeneity within a given cluster.

Author response image 2.

UMAP visualization for the Harmony-integrated dataset across all four species for the seven shared cell types, colored by cell type identity (A) and species (B).

Good point, if we understand correctly, the concern is that in our relatively broadly defined cell types, species are not well mixed and that this in turn is partly responsible for marker gene divergence. This problem is indeed difficult to address, because most approaches to evaluate this require integration across species which might lead to questionable results (see our Discussion).

Nevertheless, we attempted an integration across all four species. To this end, we subset the cells for the 7 cell types that we found in all four species and visualized cell types and species in the UMAPs above (Author response image 2).

We see that cardiac fibroblasts appear poorly integrated in the UMAP, but they still have very transferable marker genes across species. We quantified integration quality using the cell-specific mixing score (cms) (Lütge et al. 2021) and indeed found that the proportion of well integrated cells is lowest for cardiac fibroblasts (Author response image 3A). On the other end of the cms spectrum, neural crest cells appear to have the best integration across species, but their marker transferability between species is rather worse than for cardiac fibroblasts (Supplementary Figure 9). Cell-type wise calculated rank-biased overlap scores that we use for marker gene conservation show the same trends (Author response image 3B) as the F1 scores for marker gene transferability.  Hence, given our current dataset we do not see any indication that the low marker gene conservation is a result of too broadly defined cell types.

Author response image 3.

(A) Evaluation of species mixing per cell type in the Harmony-integrated dataset, quantified by the fraction of cells with an adjusted cell-specific mixing score (cms) above 0.05. (B) Summary of rank-biased overlap (RBO) scores per cell type to assess concordance of marker gene rankings for all species pairs.

Reviewer #2 (Public review):

Summary:

The authors present an important study on identifying and comparing orthologous cell types across multiple species. This manuscript focuses on characterizing cell types in embryoid bodies (EBs) derived from induced pluripotent stem cells (iPSCs) of four primate species, humans, orangutans, cynomolgus macaques, and rhesus macaques, providing valuable insights into cross-species comparisons.

Strengths:

To achieve this, the authors developed a semi-automated computational pipeline that integrates classification and marker-based cluster annotation to identify orthologous cell types across primates. This study makes a significant contribution to the field by advancing cross-species cell type identification.

We thank the reviewer for their positive and thoughtful feedback.

Weaknesses:

However, several critical points need to be addressed.

(1) Use of Liftoff for GTF Annotation

The authors used Liftoff to generate GTF files for Pongo abelii, Macaca fascicularis, and Macaca mulatta by transferring the hg38 annotation to the corresponding primate genomes. However, it is unclear why they did not use species-specific GTF files, as all these genomes have existing annotations. Why did the authors choose not to follow this approach?

As Reviewer 1 also points out, also we have observed that the annotation of non-human primates often has truncated 3’UTRs. This is especially problematic for 3’ UMI transcriptome data as the ones in the 10x dataset that we present here. To illustrate this we compared the Liftoff annotation derived from Gencode v32,  that we also used throughout our manuscript to the Ensembl gene annotation Macaca_fascicularis_6.0.111. We used transcriptomes from human and cynomolgus iPSC bulk RNAseq  (Kliesmete et al. 2024) using the Prime-seq protocol (Janjic et al. 2022) which is very similar to 10x in that it also uses 3’ UMIs. On average using Liftoff produces higher counts than the Ensembl annotation (Author response image 4A). Moreover, when comparing across species, using Ensembl for the macaque leads to an asymmetry in differentially expressed genes, with apparently many more up-regulated genes in humans. In contrast, when we use the Liftoff annotation, we detect fewer DE-genes and a similar number of genes is up-regulated in macaques as in humans (Author response image 4B). We think that the many more DE-genes are artifacts due to mismatched annotation in human and cynomolgus macaques. We illustrate this for the case of the transcription factor SALL4 in Author response image 4 C,D.  The Ensembl annotation reports 2 transcripts, while Liftoff from Gencode v32 suggests 5 transcripts, one of which has a longer 3’UTR. This longer transcript is also supported by Nanopore data from macaque iPSCs. The truncation of the 3’UTR in this case leads to underestimation of the expression of SALL4 in macaques and hence SALL4 is detected as up-regulated in humans (DESeq2: LFC= 1.34, p-adj<2e-9). In contrast, when using the Liftoff annotation SALL4 does not appear to be DE between humans and macaques (LFC=0.33, p.adj=0.20).

Author response image 4. 

(A) UMI-counts/ gene for the same cynomolgus macaque iPSC samples. On the x-axis the gtf file from Ensembl Macaca_fascicularis_6.0.111 was used to count and on the y-axis we used our filtered Liftoff annotation that transferred the human gene models from Gencode v32. (B) The # of DE-genes between human  and cynomolgus iPSCs detected with DESeq2. In Liftoff, we counted human samples using Gencode v32 and compared it to the Liftoff annotation of the same human gene models to macFas6. In Ensembl, we use Gencode v32 for the human and  Ensembl Macaca_fascicularis_6.0.111 for the Macaque. For both comparisons we subset the genes to only contain one to one orthologues as annotated in biomart. Up and down regulation is relative to human expression. C) Read counts for one example gene SALL4. Here we used in addition to the Liftoff and Ensembl annotation also transcripts derived from Nanopore cDNA sequencing of cynomolgus iPSCs. D) Gene models for SALL4 in the space of MacFas6 and a coverage for iPSC-Prime-seq bulk RNA-sequencing.

(2) Transcript Filtering and Potential Biases

The authors excluded transcripts with partial mapping (<50%), low sequence identity (<50%), or excessive length differences (>100 bp and >2× length ratio). Such filtering may introduce biases in read alignment. Did the authors evaluate the impact of these filtering choices on alignment rates?

We excluded those transcripts from analysis in both species, because they present a convolution of sequence-annotation differences and expression. The focus in our study is on regulatory evolution and we knowingly omit marker differences that are due to a marker being mutated away, we will make this clearer in the text of a revised version.

(3) Data Integration with Harmony

The methods section does not specify the parameters used for data integration with Harmony. Including these details would clarify how cross-species integration was performed.

We want to stress  that none of our conservation and marker gene analyses relies on cross-species integration. We only used the Harmony integrated data for visualisation in Figure 1 and the rough germ-layer check up in Supplementary Figure S3.  We will add a better description in the revised version.

References

Janjic, Aleksandar, Lucas E. Wange, Johannes W. Bagnoli, Johanna Geuder, Phong Nguyen, Daniel Richter, Beate Vieth, et al. 2022. “Prime-Seq, Efficient and Powerful Bulk RNA Sequencing.” Genome Biology 23 (1): 88.

Kliesmete, Zane, Peter Orchard, Victor Yan Kin Lee, Johanna Geuder, Simon M. Krauß, Mari Ohnuki, Jessica Jocher, Beate Vieth, Wolfgang Enard, and Ines Hellmann. 2024. “Evidence for Compensatory Evolution within Pleiotropic Regulatory Elements.” Genome Research 34 (10): 1528–39.

Lütge, Almut, Joanna Zyprych-Walczak, Urszula Brykczynska Kunzmann, Helena L. Crowell, Daniela Calini, Dheeraj Malhotra, Charlotte Soneson, and Mark D. Robinson. 2021. “CellMixS: Quantifying and Visualizing Batch Effects in Single-Cell RNA-Seq Data.” Life Science Alliance 4 (6): e202001004.

Moon, Kevin R., David van Dijk, Zheng Wang, Scott Gigante, Daniel B. Burkhardt, William S. Chen, Kristina Yim, et al. 2019. “Visualizing Structure and Transitions in High-Dimensional Biological Data.” Nature Biotechnology 37 (12): 1482–92.

Persad, Sitara, Zi-Ning Choo, Christine Dien, Noor Sohail, Ignas Masilionis, Ronan Chaligné, Tal Nawy, et al. 2023. “SEACells Infers Transcriptional and Epigenomic Cellular States from Single-Cell Genomics Data.” Nature Biotechnology 41 (12): 1746–57.

Street, Kelly, Davide Risso, Russell B. Fletcher, Diya Das, John Ngai, Nir Yosef, Elizabeth Purdom, and Sandrine Dudoit. 2018. “Slingshot: Cell Lineage and Pseudotime Inference for Single-Cell Transcriptomics.” BMC Genomics 19 (1): 477.

Author Response

We would like to thank the senior editor, reviewing editor and all the reviewers for taking out precious time to review our manuscript and appreciating our study. We are excited that all of you have found strength in our work and have provided comments to strengthen it further. We sincerely appreciate the valuable comments and suggestions, which we believe will help us to further improve the quality of our work.

Reviewer 1

The manuscript by Dubey et al. examines the function of the acetyltransferase Tip60. The authors show that (auto)acetylation of a lysine residue in Tip60 is important for its nuclear localization and liquid-liquid-phase-separation (LLPS). The main observations are: (i) Tip60 is localized to the nucleus, where it typically forms punctate foci. (ii) An intrinsically disordered region (IDR) within Tip60 is critical for the normal distribution of Tip60. (iii) Within the IDR the authors show that a lysine residue (K187), that is auto-acetylated, is critical. Mutation of that lysine residue to a non-acetylable arginine abolishes the behavior. (iv) biochemical experiments show that the formation of the punctate foci may be consistent with LLPS.

On balance, this is an interesting study that describes the role of acetylation of Tip60 in controlling its biochemical behavior as well as its localization and function in cells. The authors mention in their Discussion section other examples showing that acetylation can change the behavior of proteins with respect to LLPS; depending on the specific context, acetylation can promote (as here for Tip60) or impair LLPS.

Strengths:

The experiments are largely convincing and appear to be well executed.

Weaknesses:

The main concern I have is that all in vivo (i.e. in cells) experiments are done with overexpression in Cos-1 cells, in the presence of the endogenous protein. No attempt is made to use e.g. cells that would be KO for Tip60 in order to have a cleaner system or to look at the endogenous protein. It would be reassuring to know that what the authors observe with highly overexpressed proteins also takes place with endogenous proteins.

Response: The main reason to perform these experiments with overexpression system was to generate different point mutants and deletion mutants of TIP60 and analyse their effect on its properties and functions. To validate our observations with overexpression system, we also examined localization pattern of endogenous TIP60 by IFA and results depict similar kind of foci pattern within the nucleus as observed with overexpressed TIP60 protein (Figure 4A). However, we understand the reviewers concern and agree to repeat some of the overexpression experiments under endogenous TIP60 knockdown conditions using siRNA or shRNA against 3’ UTR region.

Also, it is not clear how often the experiments have been repeated and additional quantifications (e.g. of western blots) would be useful.

Response: The experiments were performed as independent biological replicates (n=3) and this is mentioned in the figure legends. Regarding the suggestion for quantifying Western blots, we want to bring into the notice that where ever required (for blots such as Figure 2F, 6H) that require quantitative estimation, graph representing quantitated value with p-value had already been added. However as suggested, in addition, quantitation for Figure 6D will be performed and added in the revised version.

In addition, regarding the LLPS description (Figure 1), it would be important to show the wetting behaviour and the temperature-dependent reversibility of the droplet formation.

Response: We appreciate the suggestion, and we will perform these assays and include the results in the revised version.

In Fig 3C the mutant (K187R) Tip60 is cytoplasmic, but still appears to form foci. Is this still reflecting phase separation, or some form of aggregation?

Response: TIP60 (K187R) mutant remains cytosolic with homogenous distribution as shown in Figure 2E. Also with TIP60 partners like PXR or p53, this mutant protein remains homogenously distributed in the cytosol. However, when co-expressed with TIP60 (Wild-type) protein, this mutant protein although still remain cytosolic some foci-like pattern is also observed at the nuclear periphery which we believe could be accumulated aggregates.

Reviewer 2

The manuscript "Autoacetylation-mediated phase separation of TIP60 is critical for its functions" by Dubey S. et al reported that the acetyltransferase TIP60 undergoes phase separation in vitro and cell nuclei. The intrinsically disordered region (IDR) of TIP60, particularly K187 within the IDR, is critical for phase separation and nuclear import. The authors showed that K187 is autoacetylated, which is important for TIP60 nuclear localization and activity on histone H4. The authors did several experiments to examine the function of K187R mutants including chromatin binding, oligomerization, phase separation, and nuclear foci formation. However, the physiological relevance of these experiments is not clear since TIP60 K187R mutants do not get into nuclei. The authors also functionally tested the cancer-derived R188P mutant, which mimics K187R in nuclear localization, disruption of wound healing, and DNA damage repair. However, similar to K187R, the R188P mutant is also deficient in nuclear import, and therefore, its defects cannot be directly attributed to the disruption of the phase separation property of TIP60. The main deficiency of the manuscript is the lack of support for the conclusion that "autoacetylation-mediated phase separation of TIP60 is critical for its functions".

This study offers some intriguing observations. However, the evidence supporting the primary conclusion, specifically regarding the necessity of the intrinsically disordered region (IDR) and K187ac of TIP60 for its phase separation and function in cells, lacks sufficient support and warrants more scrutiny. Additionally, certain aspects of the experimental design are perplexing and lack controls to exclude alternative interpretations. The manuscript can benefit from additional editing and proofreading to improve clarity.

Response: We understand the point raised by the reviewer, however we would like to draw his attention to the data where we clearly demonstrated that acetylation of lysine 187 within the IDR of TIP60 is required for its phase separation (Figure 2J). We would like to draw reviewer’s attention to other TIP60 mutants within IDR (R177H, R188H, K189R) which all enters the nucleus and make phase separated foci. Cancer-associated mutation at R188 behaves similarly because it also hampers TIP60 acetylation at the adjacent K187 residue. Our in vitro and in cellulo results clearly demonstrate that autoacetylation of TIP60 at K187 within its IDR is critical for multiple functions including its translocation inside the nucleus, its protein-protein interaction and oligomerization which are prerequisite for phase separation of TIP60.

There are two putative NLS sequences (NLS #1 from aa145; NLS #2 from aa184) in TIP60, both of which are within the IDR. Deletion of the whole IDR is therefore expected to abolish the nuclear localization of TIP60. Since K187 is within NLS #2, the cytoplasmic localization of the IDR and K187R mutants may not be related to the ability of TIP60 to phase separation.

Response: We are not disputing the presence of putative NLS within IDR region of TIP60, however our results through different mutations within IDR region (K76, K80, K148, K150, R177, R178, R188, K189) clearly demonstrate that only K187 residue acetylation is critical to shuttle TIP60 inside the nucleus while all other lysine mutants located within these putative NLS region exhibited no impact on TIP60’s nuclear shuttling. We have mentioned this in our discussion, that autoacetylation of TIP60’s K187 may induce local structural modifications in its IDR which is critical for translocating TIP60 inside the nucleus where it undergoes phase separation critical for its functions. A previous example of similar kind shows, acetylation of lysine within the NLS region of TyrRS by PCAF promote its nuclear localization (Cao X et al 2017, PNAS). IDR region (which also contains K187 site) is important for phase separation once the protein enters inside the nucleus. This could be the cell’s mechanism to prevent unwarranted action of TIP60 until it enters the nucleus and phase separate on chromatin at appropriate locations.

The chromatin-binding activity of TIP60 depends on HAT activity, but not phase-separation (Fig 1I), (Fig 2B). How do the authors reconcile the fact that the K187R mutant is able to bind to chromatin with lower activity than the HAT mutant (Fig 2F, 2I)?

Response: K187 acetylation is required for TIP60’s nuclear translocation but not critical for chromatin binding. When soluble fraction is prepared in fractionation experiment, nuclear membrane is disrupted and TIP60 (K187R) mutant has no longer hindrance in accessing the chromatin and thus can load on the chromatin (although not as efficient as Wild-type protein). For efficient chromatin binding auto-acetylation of other lysine residues in TIP60 is required which might be hampered due to reduced catalytic activity or not sufficient enough to maintain equilibrium with HDAC’s activity inside the nucleus. In case of K187R, the reduced auto-acetylation is captured when protein is the cytosol. During fractionation, once this mutant has access to chromatin, it might auto-acetylate other lysine residues critical for chromatin loading (remember catalytic domain is intact in this mutant). This is evident due to hyper auto-acetylation of Wild-type protein compared to K187R or HAT mutant proteins. We want to bring into notice that phase-separation occurs only after efficient chromatin loading of TIP60 that is the reason that under in-cellulo conditions, both K187R (which cannot enter the nucleus) and HAT mutant (which enters the nucleus but fails to efficiently binds onto the chromatin) fails to form phase separated nuclear punctate foci.

The DIC images of phase separation in Fig 2I need to be improved. The image for K187R showed the irregular shape of the condensates, which suggests particles in solution or on the slide. The authors may need to use fluorescent-tagged TIP60 in the in vitro LLPS experiments.

Response: We believe this comment is for figure 2J. The irregularly shaped condensates observed for TIP60 K187R are unique to the mutant protein and are not caused by particles on the slide. We would like to draw reviewer’s attention to supplementary figure S2A, where DIC images for TIP60 (Wild-type) protein tested under different protein and PEG8000 conditions are completely clear where protein did not made phase separated droplets ruling out the probability of particles in solution or slides.

The authors mentioned that the HAT mutant of TIP60 does not phase separate, which needs to be included.

Response: We have already added the image of RFP-TIP60 (HAT mutant) in supplementary Fig S4A (panel 2) in the manuscript.

Related to Point 3, the HAT mutant that doesn't form punctate foci by itself, can incorporate into WT TIP60 (Fig 5A). In vitro LLPS assay for WT, HAT, and K187R mutants with or without acetylation should be included. WT and mutant TIP can be labelled with GFP and RFP, respectively.

Response: We would like to draw reviewer’s attention towards our co-expression experiments performed in Figure 5 where Wild-type protein (both tagged and untagged condition) is able to phase separate and make punctate foci with co-expressed HAT mutant protein (with depleted autoacetylation capacity). We believe these in cellulo experiments are already able to answer the queries what reviewer is suggesting to acheive by in vitro experiments.

Fig 3A and 3B showed that neither K187 mutant nor HAT mutant could oligomerize. If both experiments were conducted in the absence of in vitro acetylation, how do the authors reconcile these results?

Response: We thank the reviewer for highlighting our oversight in omitting the mention of acetyl coenzyme A here. To induce acetylation under in vitro conditions, we have added 10 µM acetyl CoA into the reactions depicted in Figure 3A and 3B. The information for acetyl CoA for Figure 3B was already included in the GST-pull down assay (material and methods section). We will add the same in the oligomerization assay of material and methods in the revised manuscript.

In Fig 4, the colocalization images showed little overlap between TIP60 and nuclear speckle (NS) marker SC35, indicating that the majority of TIP60 localized in the nuclear structure other than NS. Have the authors tried to perturbate the NS by depleting the NS scaffold protein and examining TIP60 foci formation? Do PXR and TP53 localize to NS?

Response: Under normal conditions majority of TIP60 is not localized in nuclear speckles (NS) so we believe that perturbing NS will not have significant effect on TIP60 foci formation. Interestingly, recently a study by Shelly Burger group (Alexander KA et al Mol Cell. 2021 15;81(8):1666-1681) had shown that p53 localizes to NS to regulate subset of its targeted genes. We have mentioned about it in our discussion section. No information is available about localization of PXR in NS.

Were TIP60 substrates, H4 (or NCP), PXR, TP53, present inTIP60 condensates in vitro? It's interesting to see both PXR and TP53 had homogenous nuclear signals when expressed together with K187R, R188P (Fig 6E, 6G), or HAT (Suppl Fig S4A) mutants. Are PXR or TP53 nuclear foci dependent on their acetylation by TIP60? This can and should be tested.

Response: Both p53 and PXR are known to be acetylated by TIP60. In case of PXR, TIP60 acetylate PXR at lysine 170 and this TIP60-mediated acetylation of PXR at K170 is important for TIP60-PXR foci which now we know are formed by phase separation (Bakshi K et al Sci Rep. 2017 Jun 16;7(1):3635).

Since R188P mutant, like K187R, does not get into the nuclei, it is not suitable to use this mutant to examine the functional relevance of phase separation for TIP60. The authors need to find another mutant in IDR that retains nuclear localization and overall HAT activity but specifically disrupts phase separation. Otherwise, the conclusion needs to be restated. All cancer-derived mutants need to be tested for LLPS in vitro.

Response: We appreciate the reviewer’s point here, but it is important to note that the objective of these experiments is to understand the impact of K187R (critical in multiple aspects of TIP60 including phase separation) and R188P (a naturally occurring cancer-associated mutation and behaving similarly to K187R) on TIP60’s activities to determine their functional relevance. As suggested by the reviewer to test and find IDR mutant that fails to phase separate however retains nuclear localization and catalytic activity can be examined in future studies.

For all cellular experiments, it is not mentioned whether endogenous TIP60 was removed and absent in the cell lines used in this study. It's important to clarify this point because the localization and function of mutant TIP60 are affected by WT TIP60 (Fig 5).

Response: Endogenous TIP60 was present in in cellulo experiments, however as suggested by reviewer 1 we will perform some of the in cellulo experiments under endogenous TIP60 knockdown condition to validate our findings.

It is troubling that H4 peptide is used for in vitro HAT assay since TIP60 has much higher activity on nucleosomes and its preferred substrates include H2A.

Response: The purpose of using H4 peptide in the HAT assay is to determine the impact of mutations of TIP60’s catalytic activity. As H4 is one of the major histone substrate for TIP60, we believe it satisfy the objective of experiments.

Reviewer 3

This study presents results arguing that the mammalian acetyltransferase Tip60/KAT5 auto-acetylates itself on one specific lysine residue before the MYST domain, which in turn favors not only nuclear localization but also condensate formation on chromatin through LLPS. The authors further argue that this modification is responsible for the bulk of Tip60 autoacetylation and acetyltransferase activity towards histone H4. Finally, they suggest that it is required for association with txn factors and in vivo function in gene regulation and DNA damage response.

These are very wide and important claims and, while some results are interesting and intriguing, there is not really close to enough work performed/data presented to support them. In addition, some results are redundant between them, lack consistency in the mutants analyzed, and show contradiction between them. The most important shortcoming of the study is the fact that every single experiment in cells was done in over-expressed conditions, from transiently transfected cells. It is well known that these conditions can lead to non-specific mass effects, cellular localization not reflecting native conditions, and disruption of native interactome. On that topic, it is quite striking that the authors completely ignore the fact that Tip60 is exclusively found as part of a stable large multi-subunit complex in vivo, with more than 15 different proteins. Thus, arguing for a single residue acetylation regulating condensate formation and most Tip60 functions while ignoring native conditions (and the fact that Tip60 cannot function outside its native complex) does not allow me to support this study.

Response: We appreciate the reviewer’s point here, but it is important to note that the main purpose to use overexpression system in the study is to analyse the effect of different generated point/deletion mutations on TIP60. We have overexpressed proteins with different tags (GFP or RFP) or without tags (Figure 3C, Figure 5) to confirm the behaviour of protein which remains unperturbed due to presence of tags. To validate we have also examined localization of endogenous TIP60 protein which also depict similar localization behaviour as overexpressed protein. We would like to draw attention that there are several reports in literature where similar kind of overexpression system are used to determine functions of TIP60 and its mutants. Also nuclear foci pattern observed for TIP60 in our studies is also reported by several other groups.

Sun, Y., et. al. (2005) A role for the Tip60 histone acetyltransferase in the acetylation and activation of ATM. Proc Natl Acad Sci U S A, 102(37):13182-7.

Kim, C.-H. et al. (2015) ‘The chromodomain-containing histone acetyltransferase TIP60 acts as a code reader, recognizing the epigenetic codes for initiating transcription’, Bioscience, Biotechnology, and Biochemistry, 79(4), pp. 532–538.

Wee, C. L. et al. (2014) ‘Nuclear Arc Interacts with the Histone Acetyltransferase Tip60 to Modify H4K12 Acetylation(1,2,3).’, eNeuro, 1(1). doi: 10.1523/ENEURO.0019-14.2014.

However, as a caution and suggested by other reviewers also we will perform some of these overexpression experiments in absence of endogenous TIP60 by using 3’ UTR specific siRNA/shRNA.

We thank the reviewer for his comment on muti-subunit complex proteins and we would like to expand our study by determining the interaction of some of the complex subunits with TIP60 ((Wild-type) that forms nuclear condensates), TIP60 ((HAT mutant) that enters the nucleus but do not form condensates) and TIP60 ((K187R) that do not enter the nucleus and do not form condensates). We will include the result of these experiments in the revised manuscript.

• It is known that over-expression after transient transfection can lead to non-specific acetylation of lysines on the proteins, likely in part to protect from proteasome-mediated degradation. It is not clear whether the Kac sites targeted in the experiments are based on published/public data. In that sense, it is surprising that the K327R mutant does not behave like a HAT-dead mutant (which is what exactly?) or the K187R mutant as this site needs to be auto-acetylated to free the catalytic pocket, so essential for acetyltransferase activity like in all MYST-family HATs. In addition, the effect of K187R on the total acetyl-lysine signal of Tip60 is very surprising as this site does not seem to be a dominant one in public databases.

Response: We have chosen autoacetylation sites based on previously published studies where LC-MS/MS and in vitro acetylation assays were used to identified autoacetylation sites in TIP60 which includes K187. We have already mentioned about it in the manuscript and have quoted the references (1. Yang, C., et al (2012). Function of the active site lysine autoacetylation in Tip60 catalysis. PloS one 7, e32886. 10.1371/journal.pone.0032886. 2. Yi, J., et al (2014). Regulation of histone acetyltransferase TIP60 function by histone deacetylase 3. The Journal of biological chemistry 289, 33878–33886. 10.1074/jbc.M114.575266.). We would like to emphasize that both these studies have identified K187 as autoacetylation site in TIP60. Since TIP60 HAT mutant (with significantly reduced catalytic activity) can also enter nucleus, it is not surprising that K327 could also enter the nucleus.

• As the physiological relevance of the results is not clear, the mutants need to be analyzed at the native level of expression to study real functional effects on transcription and localization (ChIP/IF). It is not clear the claim that Tip60 forms nuclear foci/punctate signals at physiological levels is based on what. This is certainly debated because in part of the poor choice of antibodies available for IF analysis. In that sense, it is not clear which Ab is used in the Westerns. Endogenous Tip60 is known to be expressed in multiple isoforms from splice variants, the most dominant one being isoform 2 (PLIP) which lacks a big part (aa96-147) of the so-called IDR domain presented in the study. Does this major isoform behave the same?

Response: TIP60 antibody used in the study is from Santa Cruz (Cat. No.- sc-166323). This antibody is widely used for TIP60 detection by several methods and has been cited in numerous publications. Cat. No. will be mentioned in the manuscript. Regarding isoforms, three isoforms are known for TIP60 among which isoform 2 is majorly expressed and used in our study. Isoform and 1 and 2 have same length of IDR (150 amino acids) while isoform 3 has IDR of 97 amino acids. Interestingly, the K187 is present in all the isoforms (already mentioned in the manuscript) and missing region (96-147 amino acid) in isoform 3 has less propensity for disordered region (marked in blue circle). This clearly shows that all the isoforms of TIP60 has the tendency to phase separate.

Author response image 1.

• It is extremely strange to show that the K187R mutant fails to get in the nuclei by cell imaging but remains chromatin-bound by fractionation... If K187 is auto-acetylated and required to enter the nucleus, why would a HAT-dead mutant not behave the same?

Response: We would like to draw attention that both HAT mutant and K187R mutant are not completely catalytically dead. As our data shows both these mutants have catalytic activity although at significantly decreased levels. We believe that K187 acetylation is critical for TIP60 to enter the nucleus and once TIP60 shuttles inside the nucleus autoacetylation of other sites is required for efficient chromatin binding of TIP60. In fractionation assay, nuclear membrane is dissolved while preparing the soluble fraction so there is no hindrance for K187R mutant in accessing the chromatin. While in the case of HAT mutant, it can acetylate the K187 site and thus is able to enter the nucleus however this residual catalytic activity is either not able to autoacetylate other residues required for its efficient chromatin binding or to counter activities of HDAC’s deacetylating the TIP60.

• If K187 acetylation is key to Tip60 function, it would be most logical (and classical) to test a K187Q acetyl-mimic substitution. In that sense, what happens with the R188Q mutant? That all goes back to the fact that this cluster of basic residues looks quite like an NLS.

Response: As suggested we will generate acetylation mimicking mutant for K187 site and examine it. Result will be added in the revised manuscript.

• The effect of the mutant on the TIP60 complex itself needs to be analyzed, e.g. for associated subunits like p400, ING3, TRRAP, Brd8...

Response: As suggested we will examine the effect of mutations on TIP60 complex

Author Response:

Reviewer #1:

Summary:

This research study utilizes a realistic motoneuron model to explore the potential to trace back the appropriate levels of excitation, inhibition, and neuromodulation in the firing patterns of motoneurons observed in in-vitro and in-vivo experiments in mammals. The research employs high-performance computing power to achieve its objectives. The work introduces a new framework that enhances understanding of the neural inputs to motoneuron pools, thereby opening up new avenues for hypothesis testing research.

Strengths: The significance of the study holds relevance for all neuroscientists. Motoneurons are a unique class of neurons with known distribution of outputs for a wide range of voluntary and involuntary motor commands, and their physiological function is precisely understood. More importantly, they can be recorded in-vivo using minimally invasive methods, and they are directly impacted by many neurodegenerative diseases at the spinal cord level. The computational framework developed in this research offers the potential to reverse engineer the synaptic input distribution when assessing motor unit activity in humans, which holds particular importance. Overall, the strength of the findings focuses on providing a novel framework for studying and understanding the inputs that govern motoneuron behavior, with broad applications in neuroscience and potential implications for understanding neurodegenerative diseases. It highlights the significance of the study for various research domains, making it valuable to the scientific community.

Weaknesses: The exact levels of inhibition, excitation, and neuromodulatory inputs to neural networks are unknown. Therefore the work is based on fine-tuned measures that are indirectly based on experimental results. However, obtaining such physiological information is challenging and currently impossible. From a computational perspective it is a challenge that in theory can be solved. Thus, although we have no ground-truth evidence, this framework can provide compelling evidence for all hypothesis testing research and potentially solve this physiological problem with the use of computers.

We agree with the reviewer. This work was intended to determine the feasibility of reverse engineering motor unit firing patterns, using neuron models with a high degree realism. Given the results support this feasibility, our model and technique will therefore serve to construct new hypotheses as well as testing them.

Reviewer #2:

The study presents an extensive computational approach to identify the motor neuron input from the characteristics of single motor neuron discharge patterns during a ramp up/down contraction. This reverse engineering approach is relevant due to limitations in our ability to estimate this input experimentally. Using well-established models of single motor neurons, a (very) large number of simulations were performed that allowed identification of this relation. In this way, the results enable researchers to measure motor neuron behavior and from those results determine the underlying neural input scheme. Overall, the results are very convincing and represent an important step forward in understanding the neural strategies for controlling movement.

Nevertheless, I would suggest that the authors consider the following recommendations to strengthen the message further. First, I believe that the relation between individual motor neuron behavioral characteristics (delta F, brace height etc.) and the motor neuron input properties can be illustrated more clearly. Although this is explained in the text, I believe that this is not optimally supported by figures. Figure 6 to some extent shows this, but figures 8 and 9 as well as Table 1 shows primarily the goodness of fit rather than the actual fit.

We agree with the reviewer that showing the relationship between the motor neuron behavioral characteristics (delta F, brace height etc.) and the motor neuron input properties would be a great addition to the manuscript. Because the regression models have multiple dimensions (7 inputs and 3 outputs) it is difficult to show the relationship in a static image. We thought it best to show the goodness of fit even though it is more abstract and less intuitive. We added a supplemental diagram to Figure 8 to show the structure of the reverse engineered model that was fit (see Figure 8D).

Author response image 1: Figure 8. Residual plots showing the goodness of fit of the different predicted values: (A) Inhibition, (B) Neuromodulation and (C) excitatory Weight Ratio. The summary plots are for the models showing highest 𝑅2 results in Table 1. The predicted values are calculated using the features extracted from the firing rates (see Figure 7, section Machine learning inference of motor pool characteristics and Regression using motoneuron outputs to predict input organization). Diagram (D) shows the multidimensionality of the RE models (see Model fits) which have 7 feature inputs (see Feature Extraction) predicting 3 outputs (Inhibition, Neuromodulation and Weight Ratio).

Second, I would have expected the discussion to have addressed specifically the question of which of the two primary schemes (push-pull, balanced) is the most prevalent. This is the main research question of the study, but it is to some degree left unanswered. Now that the authors have identified the relation between the characteristics of motor neuron behaviors (which has been reported in many previous studies), why not exploit this finding by summarizing the results of previous studies (at least a few representative ones) and discuss the most likely underlying input scheme? Is there a consistent trend towards one of the schemes, or are both strategies commonly used?

We agree with the reviewer that our discussion should have addressed which of the two primary schemes – push-pull or balanced – is the most prevalent. At first glance, the upper right of Figure 6 looks the most realistic when compared to real data. We thus would expect that the push-pull scheme to dominate for the given task. We added a brief section (Push-Pull vs Balance Motor Command) in the discussion to address the reviewer’s comments. This section is not exhaustive but frames the debate using relevant literature. We are also now preparing to deploy these techniques on real data.

In addition, it seems striking to me that highly non-linear excitation profiles are necessary to obtain a linear CST ramp in many model configurations. Although somewhat speculative, one may expect that an approximately linear relation is desired for robust and intuitive motor control. It seems to me that humans generally have a good ability to accurately grade the magnitude of the motor output, which implies that either a non-linear relation has been learnt (complex task), or that the central nervous system can generally rely on a somewhat linear relation between the neural drive to the muscle and the output (simpler task).

We agree with the reviewer, and we were surprised by these results. Our motoneuron pool is equipped with persistent inward currents (PICs) which are nonlinear. Therefore, for the motoneuron to produce a linear output the central nervous system would have to incorporate these nonlinearities into its commands.

Following this reasoning, it could be interesting to report also for which input scheme, the excitation profile is most linear. I understand that this is not the primary aim of the study, but it may be an interesting way to elaborate on the finding that in many cases non-linear excitation profiles were needed to produce the linear ramp.

This is a very interesting point. The most realistic firing patterns – with respect to human data – are found in the parameter regions in the upper right in Figure 6, which in fact produce the most nonlinear input (see push-pull pattern in Figure 4C). However, in future studies we hope to separate the total motor command illustrated here into descending and feedback commands. This may result in a more linear descending drive.

Author response:

Public Reviews:

Reviewer #1 (Public Review):

(1) It will be interesting to monitor the levels of another MIM insertase namely, OXA1. This will help to understand whether some of the observed changes in levels of OXPHOS subunits are related to alterations in the amounts of this insertase.

OXA1 was not detected in the untargeted mass spectrometry analysis, most likely due to the fact that it is a polytopic membrane protein, spanning the membrane five times (1,2). Consequently, we measured OXA1 levels with immunoblotting, comparing patient fibroblast cells to the HC. No significant change in OXA1 steady state levels was observed. 

See the results below. These results will be added and discussed in the revised manuscript.

Author response image 1.

(2) Figure 3: How do the authors explain that although TIMM17 and TIMM23 were found to be significantly reduced by Western analysis they were not detected as such by the Mass Spec. method?

The untargeted mass spectrometry in the current study failed to detect the presence of TIMM17 for both, patient fibroblasts and mice neurons, while TIMM23 was detected only for mice neurons and a decrease was observed for this protein but was not significant. This is most likely due to the fact that TIMM17 and TIMM23 are both polytopic membrane proteins, spanning the membrane four times, which makes it difficult to extract them in quantities suitable for MS detection (2,3).

(3) How do the authors explain the higher levels of some proteins in the TIMM50 mutated cells?

The levels of fully functional TIM23 complex are deceased in patients' fibroblasts. Therefore, the mechanism by which the steady state level of some TIM23 substrate proteins is increased, can only be explained relying on events that occur outside the mitochondria. This could include increase in transcription, translation or post translation modifications, all of which may increase their steady state level albite the decrease in the steady state level of the import complex.

(4) Can the authors elaborate on why mutated cells are impaired in their ability to switch their energetic emphasis to glycolysis when needed?

Cellular regulation of the metabolic switch to glycolysis occurs via two known pathways: 1) Activation of AMP-activated protein kinase (AMPK) by increased levels of AMP/ADP (4). 2) Inhibition of pyruvate dehydrogenase (PDH) complexes by pyruvate dehydrogenase kinases (PDK) (5). Therefore, changes in the steady state levels of any of these regulators could push the cells towards anaerobic energy production, when needed. In our model systems, we did not observe changes in any of the AMPK, PDH or PDK subunits that were detected in our untargeted mass spectrometry analysis (see volcano plots below, no PDK subunits were detected in patient fibroblasts). Although this doesn’t directly explain why the cells have an impaired ability to switch their energetic emphasis, it does possibly explain why the switch did not occur de facto.

Author response image 2.

Reviewer #2 (Public Review):

(1) The authors claim in the abstract, the introduction, and the discussion that TIMM50 and the TIM23 translocase might not be relevant for mitochondrial protein import in mammals. This is misleading and certainly wrong!!!

Indeed, it was not in our intention to claim that the TIM23 complex might not be relevant. We have now rewritten the relevant parts to convey the correct message:

Abstract – 

Line 25 - “Strikingly, TIMM50 deficiency had no impact on the steady state levels of most of its putative substrates, suggesting that even low levels of a functional TIM23 complex are sufficient to maintain the majority of complex-dependent mitochondrial proteome.”

Introduction – 

Line 87 - Surprisingly, functional and physiological analysis points to the possibility that low levels of TIM23 complex core subunits (TIMM50, TIMM17 and TIMM23) are sufficient for maintaining steady-state levels of most presequence-containing proteins. However, the reduced TIM23CORE component levels do affect some critical mitochondrial properties and neuronal activity.

Discussion – 

Line 339 – “…surprising, as normal TIM23 complex levels are suggested to be indispensable for the translocation of presequence-containing mitochondrial proteins…”

Line 344 – “…it is possible that unlike what occurs in yeast, normal levels of mammalian TIMM50 and TIM23 complex are mainly essential for maintaining the steady state levels of intricate complexes/assemblies.”

Line 396 – “In summary, our results suggest that even low levels of TIMM50 and TIM23CORE components suffice in maintaining the majority of mitochondrial matrix and inner membrane proteome. Nevertheless, reductions in TIMM50 levels led to a decrease of many OXPHOS and MRP complex subunits, which indicates that normal TIMM50 levels might be mainly essential for maintaining the steady state levels and assembly of intricate complex proteins.”

(1) Homberg B, Rehling P, Cruz-Zaragoza LD. The multifaceted mitochondrial OXA insertase. Trends Cell Biol. 2023;33(9):765–72. 

(2) Carroll J, Altman MC, Fearnley IM, Walker JE. Identification of membrane proteins by tandem mass spectrometry of protein ions. Proc Natl Acad Sci U S A.

2007;104(36):14330–5. 

(3) Dekker PJT, Keil P, Rassow J, Maarse AC, Pfanner N, Meijer M. Identification of MIM23, a putative component of the protein import machinery of the mitochondrial inner membrane. FEBS Lett. 1993;330(1):66–70. 

(4) Trefts E, Shaw RJ. AMPK: restoring metabolic homeostasis over space and time. Mol Cell [Internet]. 2021;81(18):3677–90. Available from:

https://doi.org/10.1016/j.molcel.2021.08.015

(5) Zhang S, Hulver MW, McMillan RP, Cline MA, Gilbert ER. The pivotal role of pyruvate dehydrogenase kinases in metabolic flexibility. Nutr Metab. 2014;11(1):1–9.

Author response:

Public Reviews:

Reviewer #1 (Public review):

We thank the reviewer for his valuable input and careful assessment, which have significantly improved the clarity and rigor of our manuscript.

Summary:

Mazer & Yovel 2025 dissect the inverse problem of how echolocators in groups manage to navigate their surroundings despite intense jamming using computational simulations.

The authors show that despite the 'noisy' sensory environments that echolocating groups present, agents can still access some amount of echo-related information and use it to navigate their local environment. It is known that echolocating bats have strong small and large-scale spatial memory that plays an important role for individuals. The results from this paper also point to the potential importance of an even lower-level, short-term role of memory in the form of echo 'integration' across multiple calls, despite the unpredictability of echo detection in groups. The paper generates a useful basis to think about the mechanisms in echolocating groups for experimental investigations too.

Strengths:

(1) The paper builds on biologically well-motivated and parametrised 2D acoustics and sensory simulation setup to investigate the various key parameters of interest

(2) The 'null-model' of echolocators not being able to tell apart objects & conspecifics while echolocating still shows agents successfully emerge from groups - even though the probability of emergence drops severely in comparison to cognitively more 'capable' agents. This is nonetheless an important result showing the direction-of-arrival of a sound itself is the 'minimum' set of ingredients needed for echolocators navigating their environment.

(3) The results generate an important basis in unraveling how agents may navigate in sensorially noisy environments with a lot of irrelevant and very few relevant cues.

(4) The 2D simulation framework is simple and computationally tractable enough to perform multiple runs to investigate many variables - while also remaining true to the aim of the investigation.

Weaknesses:

There are a few places in the paper that can be misunderstood or don't provide complete details. Here is a selection:

(1) Line 61: '... studies have focused on movement algorithms while overlooking the sensory challenges involved' : This statement does not match the recent state of the literature. While the previous models may have had the assumption that all neighbours can be detected, there are models that specifically study the role of limited interaction arising from a potential inability to track all neighbours due to occlusion, and the effect of responding to only one/few neighbours at a time e.g. Bode et al. 2011 R. Soc. Interface, Rosenthal et al. 2015 PNAS, Jhawar et al. 2020 Nature Physics.

We appreciate the reviewer's comment and the relevant references. We have revised the manuscript accordingly to clarify the distinction between studies that incorporate limited interactions and those that explicitly analyze sensory constraints and interference. We have refined our statement to acknowledge these contributions while maintaining our focus on sensory challenges beyond limited neighbor detection, such as signal degradation, occlusion effects, and multimodal sensory integration (see lines 61-64):

While collective movement has been extensively studied in various species, including insect swarming, fish schooling, and bird murmuration (Pitcher, Partridge and Wardle, 1976; Partridge, 1982; Strandburg-Peshkin et al., 2013; Pearce et al., 2014; Rosenthal, Twomey, Hartnett, Wu, Couzin, et al., 2015; Bastien and Romanczuk, 2020; Davidson et al., 2021; Aidan, Bleichman and Ayali, 2024), as well as in swarm robotics agents performing tasks such as coordinated navigation and maze-solving (Faria Dias et al., 2021; Youssefi and Rouhani, 2021; Cheraghi, Shahzad and Graffi, 2022), most studies have focused on movement algorithms , often assuming full detection of neighbors (Parrish and Edelstein-Keshet, 1999; Couzin et al., 2002, 2005; Sumpter et al., 2008; Nagy et al., 2010; Bialek et al., 2012; Gautrais et al., 2012; Attanasi et al., 2014). Some models have incorporated limited interaction rules where individuals respond to one or a few neighbors due to sensory constraints (Bode, Franks and Wood, 2011; Jhawar et al., 2020). However, fewer studies explicitly examine how sensory interference, occlusion, and noise shape decision-making in collective systems (Rosenthal et al., 2015).

(2) The word 'interference' is used loosely places (Line 89: '...took all interference signals...', Line 319: 'spatial interference') - this is confusing as it is not clear whether the authors refer to interference in the physics/acoustics sense, or broadly speaking as a synonym for reflections and/or jamming.

To improve clarity, we have revised the manuscript to distinguish between different types of interference:

· Acoustic interference (jamming): Overlapping calls that completely obscure echo detection, preventing bats from perceiving necessary environmental cues.

· Acoustic interference (masking): Partial reduction in signal clarity due to competing calls.

· Spatial interference: Physical obstruction by conspecifics affecting movement and navigation.

We have updated the manuscript to use these terms consistently and explicitly define them in relevant sections (see lines 87-94 and 329-330). This distinction ensures that the reader can differentiate between interference as an acoustic phenomenon and its broader implications in navigation.

(3) The paper discusses original results without reference to how they were obtained or what was done. The lack of detail here must be considered while interpreting the Discussion e.g. Line 302 ('our model suggests...increasing the call-rate..' - no clear mention of how/where call-rate was varied) & Line 323 '..no benefit beyond a certain level..' - also no clear mention of how/where call-level was manipulated in the simulations.

All tested parameters, including call rate dynamics and call intensity variations, are detailed in the Methods section and Tables 1 and 2. Specifically:

· Call Rate Variation: The Inter-Pulse Interval (IPI) was modeled based on documented echolocation behavior, decreasing from 100 msec during the search phase to 35 msec (~28 calls per second) at the end of the approach phase, and to 5 msec (200 calls per second) during the final buzz (see Table 2). This natural variation in call rate was not manually manipulated in the model but emerged from the simulated bat behavior.

· Call Intensity Variation: The tested call intensity levels (100, 110, 120, 130 dB SPL) are presented in Table 1 under the “Call Level” parameter. The effect of increasing call intensity was analyzed in relation to exit probability, jamming probability, and collision rate. This is now explicitly referenced in the Discussion.

We have revised the manuscript to explicitly reference these aspects in the Results and Discussion sections.

Reviewer #2 (Public review):

We are grateful for the reviewer’s insightful feedback, which has helped us clarify key aspects of our research and strengthen our conclusions.

This manuscript describes a detailed model of bats flying together through a fixed geometry. The model considers elements that are faithful to both bat biosonar production and reception and the acoustics governing how sound moves in the air and interacts with obstacles. The model also incorporates behavioral patterns observed in bats, like one-dimensional feature following and temporal integration of cognitive maps. From a simulation study of the model and comparison of the results with the literature, the authors gain insight into how often bats may experience destructive interference of their acoustic signals and those of their peers, and how much such interference may actually negatively affect the groups' ability to navigate effectively. The authors use generalized linear models to test the significance of the effects they observe.

In terms of its strengths, the work relies on a thoughtful and detailed model that faithfully incorporates salient features, such as acoustic elements like the filter for a biological receiver and temporal aggregation as a kind of memory in the system. At the same time, the authors' abstract features are complicating without being expected to give additional insights, as can be seen in the choice of a two-dimensional rather than three-dimensional system. I thought that the level of abstraction in the model was perfect, enough to demonstrate their results without needless details. The results are compelling and interesting, and the authors do a great job discussing them in the context of the biological literature.

The most notable weakness I found in this work was that some aspects of the model were not entirely clear to me.

For example, the directionality of the bat's sonar call in relation to its velocity. Are these the same?

For simplicity, in our model, the head is aligned with the body, therefore the direction of the echolocation beam is the same as the direction of the flight.

Moreover, call directionality (directivity) is not directly influenced by velocity. Instead, directionality is estimated using the piston model, as described in the Methods section. The directionality is based on the emission frequency and is thus primarily linked to the behavioral phases of the bat, with frequency shifts occurring as the bat transitions from search to approach to buzz phases. During the approach phase, the bat emits calls with higher frequencies, resulting in increased directionality. This is supported by the literature (Jakobsen and Surlykke, 2010; Jakobsen, Brinkløv and Surlykke, 2013). This phase is also associated with a natural reduction in flight speed, which is a well-documented behavioral adaptation in echolocating bats (Jakobsen et al., 2024).

To clarify this in the manuscript, we have updated the text to explicitly state that directionality follows phase-dependent frequency changes rather than being a direct function of velocity, see lines 460-465.

If so, what is the difference between phi_target and phi_tx in the model equations?

represents the angle between the bat and the reflected object (target).

the angle [rad], between the masking bat and target (from the transmitter’s perspective)

refers to the angle between the transmitting conspecific and the receiving focal bat, from the transmitter’s point of view.

represents the angle between the receiving bat and the transmitting bat, from the receiver’s point of view.

These definitions have been explicitly stated in the revised manuscript to prevent any ambiguity (lines 467-468). Additionally, a Supplementary figure demonstrating the geometrical relations has been added to the manuscript.

Author response image 1.

What is a bat's response to colliding with a conspecific (rather than a wall)?

In nature, minor collisions between bats are common and typically do not result in significant disruptions to flight (Boerma et al., 2019; Roy et al., 2019; Goldstein et al., 2024).Given this, our model does not explicitly simulate the physical impact of a collision event. Instead, during the collision event the bat keeps decreasing its velocity and changing its flight direction until the distance between bats is above the threshold (0.4 m). We assume that the primary cost of such interactions arises from the effort required to avoid collisions, rather than from the collision itself. This assumption aligns with observations of bat behavior in dense flight environments, where individuals prioritize collision avoidance rather than modeling post-collision dynamics.

From the statistical side, it was not clear if replicate simulations were performed. If they were, which I believe is the right way due to stochasticity in the model, how many replicates were used, and are the standard errors referred to throughout the paper between individuals in the same simulation or between independent simulations, or both?

The number of repetitions for each scenario is detailed in Table 1, but we included it in a more prominent location in the text for clarity. Specifically, we now state (Lines 274-275):

"The number of repetitions for each scenario was as follows: 1 bat: 240; 2 bats: 120; 5 bats: 48; 10 bats: 24; 20 bats: 12; 40 bats: 12; 100 bats: 6."

Regarding the reported standard errors, they are calculated across all individuals within each scenario, without distinguishing between different simulation trials.

We clarified in the revised text (Lines 534-535 in Statistical Analysis)

Overall, I found these weaknesses to be superficial and easily remedied by the authors. The authors presented well-reasoned arguments that were supported by their results, and which were used to demonstrate how call interference impacts the collective's roost exit as measured by several variables. As the authors highlight, I think this work is valuable to individuals interested in bat biology and behavior, as well as to applications in engineered multi-agent systems like robotic swarms.

Reviewer #3 (Public review):

We sincerely appreciate the reviewer’s thoughtful comments and the time invested in evaluating our work, which have greatly contributed to refining our study.

We would like to note that in general, our model often simplifies some of the bats’ abilities, under the assumption that if the simulated bats manage to perform this difficult task with simpler mechanisms, real better adapted bats will probably perform even better. This thought strategy will be repeated in several of the answers below.

Summary:

The authors describe a model to mimic bat echolocation behavior and flight under high-density conditions and conclude that the problem of acoustic jamming is less severe than previously thought, conflating the success of their simulations (as described in the manuscript) with hard evidence for what real bats are actually doing. The authors base their model on two species of bats that fly at "high densities" (defined by the authors as colony sizes from tens to tens of thousands of individuals and densities of up to 33.3 bats/m2), Pipistrellus kuhli and Rhinopoma microphyllum. This work fits into the broader discussion of bat sensorimotor strategies during collective flight, and simulations are important to try to understand bat behavior, especially given a lack of empirical data. However, I have major concerns about the assumptions of the parameters used for the simulation, which significantly impact both the results of the simulation and the conclusions that can be made from the data. These details are elaborated upon below, along with key recommendations the authors should consider to guide the refinement of the model.

Strengths:

This paper carries out a simulation of bat behavior in dense swarms as a way to explain how jamming does not pose a problem in dense groups. Simulations are important when we lack empirical data. The simulation aims to model two different species with different echolocation signals, which is very important when trying to model echolocation behavior. The analyses are fairly systematic in testing all ranges of parameters used and discussing the differential results.

Weaknesses:

The justification for how the different foraging phase call types were chosen for different object detection distances in the simulation is unclear. Do these distances match those recorded from empirical studies, and if so, are they identical for both species used in the simulation?

The distances at which bats transition between echolocation phases are identical for both species in our model (see Table 2). These distances are based on well-documented empirical studies of bat hunting and obstacle avoidance behavior (Griffin, Webster and Michael, 1958; Simmons and Kick, 1983; Schnitzler et al., 1987; Kalko, 1995; Hiryu et al., 2008; Vanderelst and Peremans, 2018). These references provide extensive evidence that insectivorous bats systematically adjust their echolocation calls in response to object proximity, following the characteristic phases of search, approach, and buzz.

To improve clarity, we have updated the text to explicitly state that the phase transition distances are empirically grounded and apply equally to both modeled species (lines 430-447).

What reasoning do the authors have for a bat using the same call characteristics to detect a cave wall as they would for detecting a small insect?

In echolocating bats, call parameters are primarily shaped by the target distance and echo strength. Accordingly, there is little difference in call structure between prey capture and obstacles-related maneuvers, aside from intensity adjustments based on target strength (Hagino et al., 2007; Hiryu et al., 2008; Surlykke, Ghose and Moss, 2009; Kothari et al., 2014). In our study, due to the dense cave environment, the bats are found to operate in the approach phase nearly all the time, which is consistent with natural cave emergence, where they are navigating through a cluttered environment rather than engaging in open-space search. For one of the species (Rhinopoma M.), we also have empirical recordings of individuals flying under similar conditions (Goldstein et al., 2024). Our model was designed to remain as simple as possible while relying on conservative assumptions that may underestimate bat performance. If, in reality, bats fine-tune their echolocation calls even earlier or more precisely during navigation than assumed, our model would still conservatively reflect their actual capabilities.

We actually used logarithmically frequency modulated (FM) chirps, generated using the MATLAB built-in function chirp(t, f0, t1, f1, 'logarithmic'). This method aligns with the nonlinear FM characteristics of Pipistrellus kuhlii (PK) and Rhinopoma microphyllum (RM) and provides a realistic approximation of their echolocation signals. We acknowledge that this was not sufficiently emphasized in the original text, and we have now explicitly highlighted this in the revised version to ensure clarity (sell Lines 447-449 in Methods).

The two species modeled have different calls. In particular, the bandwidth varies by a factor of 10, meaning the species' sonars will have different spatial resolutions. Range resolution is about 10x better for PK compared to RM, but the authors appear to use the same thresholds for "correct detection" for both, which doesn't seem appropriate.

The detection process in our model is based on Saillant’s method using a filter bank, as detailed in the paper (Saillant et al., 1993; Neretti et al., 2003; Sanderson et al., 2003). This approach inherently incorporates the advantages of a wider bandwidth, meaning that the differences in range resolution between the species are already accounted for within the signal-processing framework. Thus, there is no need to explicitly adjust the model parameters for bandwidth variations, as these effects emerge from the applied method.

Also, the authors did not mention incorporating/correcting for/exploiting Doppler, which leads me to assume they did not model it.

The reviewer is correct. To maintain model simplicity, we did not incorporate the Doppler effect or its impact on echolocation. The exclusion of Doppler effects was based on the assumption that while Doppler shifts can influence frequency perception, their impact on jamming and overall navigation performance is minor within the modelled context.

The maximal Doppler shifts expected for the bats in this scenario are of ~ 1kHz. These shifts would be applied variably across signals due to the semi-random relative velocities between bats, leading to a mixed effect on frequency changes. This variability would likely result in an overall reduction in jamming rather than exacerbating it, aligning with our previous statement that our model may overestimate the severity of acoustic interference. Such Doppler shifts would result in errors of 2-4 cm in localization (i.e., 200-400 micro-seconds) (Boonman, Parsons and Jones, 2003). 

We have now explicitly highlighted this in the revised version (see Lines 468-470).

The success of the simulation may very well be due to variation in the calls of the bats, which ironically enough demonstrates the importance of a jamming avoidance response in dense flight. This explains why the performance of the simulation falls when bats are not able to distinguish their own echoes from other signals. For example, in Figure C2, there are calls that are labeled as conspecific calls and have markedly shorter durations and wider bandwidths than others. These three phases for call types used by the authors may be responsible for some (or most) of the performance of the model since the correlation between different call types is unlikely to exceed the detection threshold. But it turns out this variation in and of itself is what a jamming avoidance response may consist of. So, in essence, the authors are incorporating a jamming avoidance response into their simulation.

We fully agree that the natural variations in call design between the phases contribute significantly to interference reduction (see our discussion in a previous paper in Mazar & Yovel, 2020). However, we emphasize that this cannot be classified as a Jamming Avoidance Response (JAR). In our model, bats respond only to the physical presence of objects and not to the acoustic environment or interference itself. There is no active or adaptive adjustment of call design to minimize jamming beyond the natural phase-dependent variations in call structure. Therefore, while variation in call types does inherently reduce interference, this effect emerges passively from the modeled behavior rather than as an intentional strategy to avoid jamming.

The authors claim that integration over multiple pings (though I was not able to determine the specifics of this integration algorithm) reduces the masking problem. Indeed, it should: if you have two chances at detection, you've effectively increased your SNR by 3dB.

The reviewer is correct. Indeed, integration over multiple calls improves signal-to-noise ratio (SNR), effectively increasing it by approximately 3 dB per doubling of observations. The specifics of the integration algorithm are detailed in the Methods section, where we describe how sensory information is aggregated across multiple time steps to enhance detection reliability.

They also claim - although it is almost an afterthought - that integration dramatically reduces the degradation caused by false echoes. This also makes sense: from one ping to the next, the bat's own echo delays will correlate extremely well with the bat's flight path. Echo delays due to conspecifics will jump around kind of randomly. However, the main concern is regarding the time interval and number of pings of the integration, especially in the context of the bat's flight speed. The authors say that a 1s integration interval (5-10 pings) dramatically reduces jamming probability and echo confusion. This number of pings isn't very high, and it occurs over a time interval during which the bat has moved 5-10m. This distance is large compared to the 0.4m distance-to-obstacle that triggers an evasive maneuver from the bat, so integration should produce a latency in navigation that significantly hinders the ability to avoid obstacles. Can the authors provide statistics that describe this latency, and discussion about why it doesn't seem to be a problem?

As described in the Methods section, the bat’s collision avoidance response does not solely rely on the integration process. Instead, the model incorporates real-time echoes from the last calls, which are used independently of the integration process for immediate obstacle avoidance maneuvers. This ensures that bats can react to nearby obstacles without being hindered by the integration latency. The slower integration on the other hand is used for clustering, outlier removal and estimation wall directions to support the pathfinding process, as illustrated in Supplementary Figure 1.

Additionally, our model assumes that bats store the physical positions of echoes in an allocentric coordinate system (x-y). The integration occurs after transforming these detections from a local relative reference frame to a global spatial representation. This allows for stable environmental mapping while maintaining responsiveness to immediate changes in the bat’s surroundings.

See lines 518-523 in the revied version.

The authors are using a 2D simulation, but this very much simplifies the challenge of a 3D navigation task, and there is an explanation as to why this is appropriate. Bat densities and bat behavior are discussed per unit area when realistically it should be per unit volume. In fact, the authors reference studies to justify the densities used in the simulation, but these studies were done in a 3D world. If the authors have justification for why it is realistic to model a 3D world in a 2D simulation, I encourage them to provide references justifying this approach.

We acknowledge that this is a simplification; however, from an echolocation perspective, a 2D framework represents a worst-case scenario in terms of bat densities and maneuverability:

· Higher Effective Density: A 2D model forces all bats into a single plane rather than distributing them through a 3D volume, increasing the likelihood of overlap in calls and echoes and making jamming more severe. As described in the text: the average distance to the nearest bat in our simulation is 0.27m (with 100 bats), whereas reported distances in very dense colonies are 0.5m, as observed in Myotis grisescens and Tadarida brasiliensis (Fujioka et al., 2021; Sabol and Hudson, 1995; Betke et al., 2008; Gillam et al, 2010)

· Reduced Maneuverability: In 3D space, bats can use vertical movement to avoid obstacles and conspecifics. A 2D constraint eliminates this degree of freedom, increasing collision risk and limiting escape options.

Thus, our 2D model provides a conservative difficult test case, ensuring that our findings are valid under conditions where jamming and collision risks are maximized. Additionally, the 2D framework is computationally efficient, allowing us to perform multiple simulation runs to explore a broad parameter space and systematically test the impact of different variables.

To address the reviewer’s concern, we have clarified this justification in the revised text and will provide supporting references where applicable: (see Methods lines 407-412)

The focus on "masking" (which appears to be just in-band noise), especially relative to the problem of misassigned echoes, is concerning. If the bat calls are all the same waveform (downsweep linear FM of some duration, I assume - it's not clear from the text), false echoes would be a major problem. Masking, as the authors define it, just reduces SNR. This reduction is something like sqrt(N), where N is the number of conspecifics whose echoes are audible to the bat, so this allows the detection threshold to be set lower, increasing the probability that a bat's echo will exceed a detection threshold. False echoes present a very different problem. They do not reduce SNR per se, but rather they cause spurious threshold excursions (N of them!) that the bat cannot help but interpret as obstacle detection. I would argue that in dense groups the mis-assignment problem is much more important than the SNR problem.

There is substantial literature supporting the assumption that bats can recognize their own echoes and distinguish them from conspecific signals (Schnitzler and Bioscience, 2001‏; Kazial, Burnett and Masters, 2001; Burnett and Masters, 2002; Kazial, Kenny and Burnett, 2008; Chili, Xian and Moss, 2009; Yovel et al., 2009; Beetz and Hechavarría, 2022). However, we acknowledge that false echoes may present a major challenge in dense groups. To address this, we explicitly tested the impact of the self-echo identification assumption in our study see Results Figure 4: The impact of confusion on performance, and lines 345-355 in the Discussion.

Furthermore, we examined a full confusion scenario, where all reflected echoes from conspecifics were misinterpreted as obstacle reflections (i.e., 100% confusion). Our results show that this significantly degrades navigation performance, supporting the argument that echo misassignment is a critical issue. However, we also explored a simple mitigation strategy based on temporal integration with outlier rejection, which provided some improvement in performance. This suggests that real bats may possess additional mechanisms to enhance self-echo identification and reduce false detections. See lines XX in the manuscript for further discussion.

The criteria set for flight behavior (lines 393-406) are not justified with any empirical evidence of the flight behavior of wild bats in collective flight. How did the authors determine the avoidance distances? Also, what is the justification for the time limit of 15 seconds to emerge from the opening? Instead of an exit probability, why not instead use a time criterion, similar to "How long does it take X% of bats to exit?"

While we acknowledge that wild bats may employ more complex behaviors for collision avoidance, we chose to implement a simplified decision-making rule in our model to maintain computational tractability.

The avoidance distances (1.5 m from walls and 0.4 m from other bats) were selected as internal parameters to ensure coherent flight trajectories while maintaining a reasonable collision rate. These distances provide a balance between maneuverability and stability, preventing erratic flight patterns while still enabling effective obstacle avoidance. In the revised paper, we have added supplementary figures illustrating the effect of model parameters on performance, specifically focusing on the avoidance distance.

The 15-second exit limit was determined as described in the text (Lines 403-404): “A 15-second window was chosen because it is approximately twice the average exit time for 40 bats and allows for a second corrective maneuver if needed.” In other words, it allowed each bat to circle the ‘cave’ twice to exit even in the most crowded environment. This threshold was set to keep simulation time reasonable while allowing sufficient time for most bats to exit successfully.

We acknowledge that the alternative approach suggested by the reviewer—measuring the time taken for a certain percentage of bats to exit—is also valid. However, in our model, some outlier bats fail to exit and continue flying for many minutes, Such simulations would lead to excessive simulation times making it difficult to generate repetitions and not teaching us much – they usually resulted from the bat slightly missing the opening (see video S1. Our chosen approach ensures practical runtime constraints while still capturing relevant performance metrics.

What is the empirical justification for the 1-10 calls used for integration?

The "average exit time for 40 bats" is also confusing and not well explained. Was this determined empirically? From the simulation? If the latter, what are the conditions? Does it include masking, no masking, or which species?

Previous studies have demonstrated that bats integrate acoustic information received sequentially over several echolocation calls (2-15), effectively constructing an auditory scene in complex environments (Ulanovsky and Moss, 2008; Chili, Xian and Moss, 2009; Moss and Surlykke, 2010; Yovel and Ulanovsky, 2017; Salles, Diebold and Moss, 2020). Additionally, bats are known to produce echolocation sound groups when spatiotemporal localization demands are high (Kothari et al., 2014). Studies have documented call sequences ranging from 2 to 15 grouped calls (Moss et al., 2010), and it has been hypothesized that grouping facilitates echo segregation.

We did not use a single integration window - we tested integration sizes between 1 and 10 calls and presented the results in Figure 3A. This range was chosen based on prior empirical findings and to explore how different levels of temporal aggregation impact navigation performance. Indeed, the results showed that the performance levels between 5-10 calls integration window (Figure 3A)

Regarding the average exit time for 40 bats, this value was determined from our simulations, where it represents the mean time for successful exits under standard conditions with masking.

We have revised the text to clarify these details see, lines 466.

References:

Aidan, Y., Bleichman, I. and Ayali, A. (2024) ‘Pausing to swarm: locust intermittent motion is instrumental for swarming-related visual processing’, Biology letters, 20(2), p. 20230468. Available at: https://doi.org/10.1098/rsbl.2023.0468.

Attanasi, A. et al. (2014) ‘Collective Behaviour without Collective Order in Wild Swarms of Midges’. Edited by T. Vicsek, 10(7). Available at: https://doi.org/10.1371/journal.pcbi.1003697.

Bastien, R. and Romanczuk, P. (2020) ‘A model of collective behavior based purely on vision’, Science Advances, 6(6). Available at: https://doi.org/10.1126/sciadv.aay0792.

Beetz, M.J. and Hechavarría, J.C. (2022) ‘Neural Processing of Naturalistic Echolocation Signals in Bats’, Frontiers in Neural Circuits, 16, p. 899370. Available at: https://doi.org/10.3389/FNCIR.2022.899370/BIBTEX.

Betke, M. et al. (2008) ‘Thermal Imaging Reveals Significantly Smaller Brazilian Free-Tailed Bat Colonies Than Previously Estimated’, Journal of Mammalogy, 89(1), pp. 18–24. Available at: https://doi.org/10.1644/07-MAMM-A-011.1.

Bialek, W. et al. (2012) ‘Statistical mechanics for natural flocks of birds’, Proceedings of the National Academy of Sciences, 109(13), pp. 4786–4791. Available at: https://doi.org/10.1073/PNAS.1118633109.

Bode, N.W.F., Franks, D.W. and Wood, A.J. (2011) ‘Limited interactions in flocks: Relating model simulations to empirical data’, Journal of the Royal Society Interface, 8(55), pp. 301–304. Available at: https://doi.org/10.1098/RSIF.2010.0397.

Boerma, D.B. et al. (2019) ‘Wings as inertial appendages: How bats recover from aerial stumbles’, Journal of Experimental Biology, 222(20). Available at: https://doi.org/10.1242/JEB.204255/VIDEO-3.

Boonman, A.M., Parsons, S. and Jones, G. (2003) ‘The influence of flight speed on the ranging performance of bats using frequency modulated echolocation pulses’, The Journal of the Acoustical Society of America, 113(1), p. 617. Available at: https://doi.org/10.1121/1.1528175.

Burnett, S.C. and Masters, W.M. (2002) ‘Identifying Bats Using Computerized Analysis and Artificial Neural Networks’, North American Symposium on Bat Research, 9.

Cheraghi, A.R., Shahzad, S. and Graffi, K. (2022) ‘Past, Present, and Future of Swarm Robotics’, in Lecture Notes in Networks and Systems. Available at: https://doi.org/10.1007/978-3-030-82199-9_13.

Chili, C., Xian, W. and Moss, C.F. (2009) ‘Adaptive echolocation behavior in bats for the analysis of auditory scenes’, Journal of Experimental Biology, 212(9), pp. 1392–1404. Available at: https://doi.org/10.1242/jeb.027045.

Couzin, I.D. et al. (2002) ‘Collective Memory and Spatial Sorting in Animal Groups’, Journal of Theoretical Biology, 218(1), pp. 1–11. Available at: https://doi.org/10.1006/jtbi.2002.3065.

Couzin, I.D. et al. (2005) ‘Effective leadership and decision-making in animal groups on the move’, Nature, 433(7025), pp. 513–516. Available at: https://doi.org/10.1038/nature03236.

Davidson, J.D. et al. (2021) ‘Collective detection based on visual information in animal groups’, Journal of the Royal Society, 18(180), p. 2021.02.18.431380. Available at: https://doi.org/10.1098/rsif.2021.0142.

Faria Dias, P.G. et al. (2021) ‘Swarm robotics: A perspective on the latest reviewed concepts and applications’, Sensors. Available at: https://doi.org/10.3390/s21062062.

Fujioka, E. et al. (2021) ‘Three-Dimensional Trajectory Construction and Observation of Group Behavior of Wild Bats During Cave Emergence’, Journal of Robotics and Mechatronics, 33(3), pp. 556–563. Available at: https://doi.org/10.20965/jrm.2021.p0556.

Gautrais, J. et al. (2012) ‘Deciphering Interactions in Moving Animal Groups’, PLOS Computational Biology, 8(9), p. e1002678. Available at: https://doi.org/10.1371/JOURNAL.PCBI.1002678.

Gillam, E.H. et al. (2010) ‘Echolocation behavior of Brazilian free-tailed bats during dense emergence flights’, Journal of Mammalogy, 91(4), pp. 967–975. Available at: https://doi.org/10.1644/09-MAMM-A-302.1.

Goldstein, A. et al. (2024) ‘Collective Sensing – On-Board Recordings Reveal How Bats Maneuver Under Severe 4 Acoustic Interference’, Under Review, pp. 1–25.

Griffin, D.R., Webster, F.A. and Michael, C.R. (1958) ‘THE ECHOLOCATION OF FLYING INSECTS BY BATS ANIMAL BEHAVIOUR , Viii , 3-4’.

Hagino, T. et al. (2007) ‘Adaptive SONAR sounds by echolocating bats’, International Symposium on Underwater Technology, UT 2007 - International Workshop on Scientific Use of Submarine Cables and Related Technologies 2007, pp. 647–651. Available at: https://doi.org/10.1109/UT.2007.370829.

Hiryu, S. et al. (2008) ‘Adaptive echolocation sounds of insectivorous bats, Pipistrellus abramus, during foraging flights in the field’, The Journal of the Acoustical Society of America, 124(2), pp. EL51–EL56. Available at: https://doi.org/10.1121/1.2947629.

Jakobsen, L. et al. (2024) ‘Velocity as an overlooked driver in the echolocation behavior of aerial hawking vespertilionid bats’. Available at: https://doi.org/10.1016/j.cub.2024.12.042.

Jakobsen, L., Brinkløv, S. and Surlykke, A. (2013) ‘Intensity and directionality of bat echolocation signals’, Frontiers in Physiology, 4 APR(April), pp. 1–9. Available at: https://doi.org/10.3389/fphys.2013.00089.

Jakobsen, L. and Surlykke, A. (2010) ‘Vespertilionid bats control the width of their biosonar sound beam dynamically during prey pursuit’, 107(31). Available at: https://doi.org/10.1073/pnas.1006630107.

Jhawar, J. et al. (2020) ‘Noise-induced schooling of fish’, Nature Physics 2020 16:4, 16(4), pp. 488–493. Available at: https://doi.org/10.1038/s41567-020-0787-y.

Kalko, E.K. V. (1995) ‘Insect pursuit, prey capture and echolocation in pipistrelle bats (Microchirptera)’, Animal Behaviour, 50(4), pp. 861–880.

Kazial, K.A., Burnett, S.C. and Masters, W.M. (2001) ‘ Individual and Group Variation in Echolocation Calls of Big Brown Bats, Eptesicus Fuscus (Chiroptera: Vespertilionidae) ’, Journal of Mammalogy, 82(2), pp. 339–351. Available at: https://doi.org/10.1644/1545-1542(2001)082<0339:iagvie>2.0.co;2.

Kazial, K.A., Kenny, T.L. and Burnett, S.C. (2008) ‘Little brown bats (Myotis lucifugus) recognize individual identity of conspecifics using sonar calls’, Ethology, 114(5), pp. 469–478. Available at: https://doi.org/10.1111/j.1439-0310.2008.01483.x.

Kothari, N.B. et al. (2014) ‘Timing matters: Sonar call groups facilitate target localization in bats’, Frontiers in Physiology, 5 MAY. Available at: https://doi.org/10.3389/fphys.2014.00168.

Moss, C.F. and Surlykke, A. (2010) ‘Probing the natural scene by echolocation in bats’, Frontiers in Behavioral Neuroscience. Available at: https://doi.org/10.3389/fnbeh.2010.00033.

Nagy, M. et al. (2010) ‘Hierarchical group dynamics in pigeon flocks’, Nature 2010 464:7290, 464(7290), pp. 890–893. Available at: https://doi.org/10.1038/nature08891.

Neretti, N. et al. (2003) ‘Time-frequency model for echo-delay resolution in wideband biosonar’, The Journal of the Acoustical Society of America, 113(4), pp. 2137–2145. Available at: https://doi.org/10.1121/1.1554693.

Parrish, J.K. and Edelstein-Keshet, L. (1999) ‘Complexity, Pattern, and Evolutionary Trade-Offs in Animal Aggregation’, Science, 284(5411), pp. 99–101. Available at: https://doi.org/10.1126/SCIENCE.284.5411.99.

Partridge, B.L. (1982) ‘The Structure and Function of Fish Schools’, 246(6), pp. 114–123. Available at: https://doi.org/10.2307/24966618.

Pearce, D.J.G. et al. (2014) ‘Role of projection in the control of bird flocks’, Proceedings of the National Academy of Sciences of the United States of America, 111(29), pp. 10422–10426. Available at: https://doi.org/10.1073/pnas.1402202111.

Pitcher, T.J., Partridge, B.L. and Wardle, C.S. (1976) ‘A blind fish can school’, Science, 194(4268), pp. 963–965. Available at: https://doi.org/10.1126/science.982056.

Rosenthal, S.B., Twomey, C.R., Hartnett, A.T., Wu, H.S., Couzin, I.D., et al. (2015) ‘Revealing the hidden networks of interaction in mobile animal groups allows prediction of complex behavioral contagion’, Proceedings of the National Academy of Sciences of the United States of America, 112(15), pp. 4690–4695. Available at: https://doi.org/10.1073/pnas.1420068112.

Rosenthal, S.B., Twomey, C.R., Hartnett, A.T., Wu, H.S. and Couzin, I.D. (2015) ‘Revealing the hidden networks of interaction in mobile animal groups allows prediction of complex behavioral contagion’, Proceedings of the National Academy of Sciences of the United States of America, 112(15), pp. 4690–4695. Available at: https://doi.org/10.1073/PNAS.1420068112/-/DCSUPPLEMENTAL/PNAS.1420068112.SAPP.PDF.

Roy, S. et al. (2019) ‘Extracting interactions between flying bat pairs using model-free methods’, Entropy, 21(1). Available at: https://doi.org/10.3390/e21010042.

Sabol, B.M. and Hudson, M.K. (1995) ‘Technique using thermal infrared-imaging for estimating populations of gray bats’, Journal of Mammalogy, 76(4). Available at: https://doi.org/10.2307/1382618.

Saillant, P.A. et al. (1993) ‘A computational model of echo processing and acoustic imaging in frequency- modulated echolocating bats: The spectrogram correlation and transformation receiver’, The Journal of the Acoustical Society of America, 94(5). Available at: https://doi.org/10.1121/1.407353.

Salles, A., Diebold, C.A. and Moss, C.F. (2020) ‘Echolocating bats accumulate information from acoustic snapshots to predict auditory object motion’, Proceedings of the National Academy of Sciences of the United States of America, 117(46), pp. 29229–29238. Available at: https://doi.org/10.1073/PNAS.2011719117/SUPPL_FILE/PNAS.2011719117.SAPP.PDF.

Sanderson, M.I. et al. (2003) ‘Evaluation of an auditory model for echo delay accuracy in wideband biosonar’, The Journal of the Acoustical Society of America, 114(3), pp. 1648–1659. Available at: https://doi.org/10.1121/1.1598195.

Schnitzler, H., Bioscience, E.K.- and 2001‏, undefined (no date) ‘Echolocation by insect-eating bats: we define four distinct functional groups of bats and find differences in signal structure that correlate with the typical echolocation ‏’, academic.oup.com‏HU Schnitzler, EKV Kalko‏Bioscience, 2001‏•academic.oup.com‏ [Preprint]. Available at: https://academic.oup.com/bioscience/article-abstract/51/7/557/268230 (Accessed: 17 March 2025).

Schnitzler, H.-U. et al. (1987) ‘The echolocation and hunting behavior of the bat,Pipistrellus kuhli’, Journal of Comparative Physiology A, 161(2), pp. 267–274. Available at: https://doi.org/10.1007/BF00615246.

Simmons, J.A. and Kick, S.A. (1983) ‘Interception of Flying Insects by Bats’, Neuroethology and Behavioral Physiology, pp. 267–279. Available at: https://doi.org/10.1007/978-3-642-69271-0_20.

Strandburg-Peshkin, A. et al. (2013) ‘Visual sensory networks and effective information transfer in animal groups’, Current Biology. Cell Press. Available at: https://doi.org/10.1016/j.cub.2013.07.059.

Sumpter, D.J.T. et al. (2008) ‘Consensus Decision Making by Fish’, Current Biology, 18(22), pp. 1773–1777. Available at: https://doi.org/10.1016/J.CUB.2008.09.064.

Surlykke, A., Ghose, K. and Moss, C.F. (2009) ‘Acoustic scanning of natural scenes by echolocation in the big brown bat, Eptesicus fuscus’, Journal of Experimental Biology, 212(7), pp. 1011–1020. Available at: https://doi.org/10.1242/JEB.024620.

Theriault, D.H. et al. (no date) ‘Reconstruction and analysis of 3D trajectories of Brazilian free-tailed bats in flight‏’, cs-web.bu.edu‏ [Preprint]. Available at: https://cs-web.bu.edu/faculty/betke/papers/2010-027-3d-bat-trajectories.pdf (Accessed: 4 May 2023).

Ulanovsky, N. and Moss, C.F. (2008) ‘What the bat’s voice tells the bat’s brain’, Proceedings of the National Academy of Sciences of the United States of America, 105(25), pp. 8491–8498. Available at: https://doi.org/10.1073/pnas.0703550105.

Vanderelst, D. and Peremans, H. (2018) ‘Modeling bat prey capture in echolocating bats : The feasibility of reactive pursuit’, Journal of theoretical biology, 456, pp. 305–314.

Youssefi, K.A.R. and Rouhani, M. (2021) ‘Swarm intelligence based robotic search in unknown maze-like environments’, Expert Systems with Applications, 178. Available at: https://doi.org/10.1016/j.eswa.2021.114907.

Yovel, Y. et al. (2009) ‘The voice of bats: How greater mouse-eared bats recognize individuals based on their echolocation calls’, PLoS Computational Biology, 5(6). Available at: https://doi.org/10.1371/journal.pcbi.1000400.

Yovel, Y. and Ulanovsky, N. (2017) ‘Bat Navigation’, The Curated Reference Collection in Neuroscience and Biobehavioral Psychology, pp. 333–345. Available at: https://doi.org/10.1016/B978-0-12-809324-5.21031-6.

Author response:

We thank the reviewers for their thorough evaluation and constructive feedback on our manuscript.

We think that their valuable suggestions will strengthen the manuscript and help us clarify several important points.

All reviewers acknowledged the importance of our theoretical results and network classification in making pattern formation analysis a more tractable problem. At the same time, they have also raised a number of important concerns that we shall carefully consider.

A. A major clarification that the reviewers found important concerns the definition of non-trivial pattern transformations and its generalization to higher dimensions. In this regard, the reviewers’ comments are:

Reviewer #1:

(on non-trivial pattern transformations):

(3) All modelling is confined to one spatial dimension, and the very definition of a "non-trivial" transformation is framed in terms of peak positions along a line, which clearly must be reformulated for higher dimensions. It's well-known that diffusions in 1, 2, and 3 dimensions are also dramatically different, so the relevance of the three-class taxonomy to real multicellular tissues remains unclear, or at least should be explained in more detail. Reviewer #2 (on non-trivial pattern transformations):

(5) The definition of non-trivial pattern formation is provided only in the Supplementary Information, despite its central importance for interpreting the main results. It would significantly improve clarity if this definition were included and explained in the main text. Additionally, it remains unclear how the definition is consistently applied across the different initial conditions. In particular, the authors should clarify how slope-based measures are determined for both the random noise and sharp peak/step function initial states. Furthermore, the authors do not specify how the sign function is evaluated at zero. If the standard mathematical definition sgn(0)=0 is used, then even a simple widening of a peak could fulfill the criterion for nontrivial pattern transformation.

We agree with Reviewer #2 that including a more detailed definition of non-trivial pattern transformation in the main text would enhance the clarity of the paper. The one-dimensional (1D) definition currently provided in the Supplementary Information was chosen because all computations presented therein involve exclusively one-dimensional patterns. However, we acknowledge that this definition, as it was, did not have a totally unambiguous generalization  to higher dimensions. Therefore, in a revised version of the manuscript, we will incorporate an expanded definition applicable to higher-dimensional cases.

This general definition of a non-trivial pattern transformation should make no reference to the sign of spatial derivatives of either the initial or resulting patterns. Specifically, a pattern transformation is considered non-trivial if it satisfies the following criteria:

- It is heterogeneous: The resulting pattern is heterogeneous in space.

- It is rearranging: The arrangement of critical points (i.e. peaks, valleys and saddle points in a gene product concentration) along the domain in the resulting pattern of a gene product is different to the arrangement of critical points in its initial pattern. This includes the emergence of new critical points, the disappearance of existing ones, or the spatial displacement of critical points from one location to another.

- It is non-replicating: The spatial arrangement of critical points in the pattern of one gene product must differ from that of any other upstream gene product.

Nonetheless, our two initial patterns are spatially discontinuous functions: in homogeneous initial patterns, the white noise is discontinuous by definition; and for the spike and spike+homogeneous initial patterns, we use sharp spikes defined by the rectangular function, which is discontinuous at the spike boundaries. Therefore, the aforementioned definition should be supplemented with the following two ad hoc assumptions:

- Homogeneous initial patterns do not comprise any critical point. White noise in this type of initial patterns represents small thermodynamic fluctuations around the steady state and, for the purpose of pattern transformation, this is equivalent to a constant concentration along the domain.

- Spike and spike+homogeneous initial patterns each contain a single critical point located at the center of the spike. The sharp spikes, modeled using the rectangular function, serve as a theoretical idealization to facilitate mathematical analysis. Once diffusion begins to act, these sharp boundaries are smoothed into differentiable gradients, maintaining a unique critical point at the center of the initial spike, which is the most relevant information for pattern transformation.

Finally, it is worth recalling that our gene network classification is fundamentally based on an analysis of the dispersion relation associated with the gene network, and the construction of this dispersion relation is independent of the spatial dimensionality of the domain (i.e. it does not require assuming any specific number of dimensions). The fact that the description of this dispersion relation was in the SI may have been non-ideal for the understandability of the article and will, consequently, be moved to the main text in an upcoming version of the article. Thus, the gene networks that can lead to pattern transformation are the same in 1D, 2D or 3D. As for the resulting patterns, the broad description we provide also applies to any number of dimensions; these would be periodic, non periodic as in the amplified noise patterns or non periodic as in the hierarchic networks. For the latter notice that, except for boundary effects that we later discuss, the spike initial condition is radially symmetric and thus, the patterns resulting from it will also be radially symmetric. We will make this point more explicit in a revised version of the article, especially since, as suggested, this important portion of the Supplementary Information will be incorporated into the main text.

Reviewer 2 suggests that with our definition of non-trivial pattern transformation, the simple widening of a concentration peak would constitute a non-trivial pattern transformation. This is not the case, as already shown in the figures as a example, since in a widening there is no change in the position of the critical point. A different situation applies if a wide and completely flat concentration peak (i.e. a plateau) forms. As we will explain in the coming version this is not possible because of requirement R5.

We think that this clarification of the definition of non-trivial pattern transformation will also help clarify the next point (B below) since it would make it clearer that this article does not intend to explain which specific resulting pattern would arise from any given gene network.

B. The main concern among these relates to the validity of our linearization of the model equations and the extension of the results obtained for the linear system to the fully nonlinear system. In this regard, the reviewers’ comments are:

Reviewer #1:

(on linearization):

(2) A central step in the model formulation is the linearisation of the reaction term around a homogeneous steady state; higher-order kinetics, including ubiquitous bimolecular sinks such as A + B → AB, are simply collapsed into the Jacobian without any stated amplitude bound on the perturbations. Because the manuscript never analyses how far this assumption can be relaxed, the robustness of the three-class taxonomy under realistic nonlinear reactions or large spike amplitudes remains uncertain.

Reviewer #2:

(on linearization):

(2) Most of the proofs presented in the Supplementary Information rely on linearized versions of the governing equations, and it remains unclear how these results extend to the fully nonlinear system. We are concerned that the generality of the conclusions drawn from the linear analysis may be overstated in the main text. For example, in Section S3, the authors introduce the concept of dynamic equivalence of transitive chains (Proposition S3.1) and intracellular transitive M-branching (Proposition S3.2), which pertains to the system's steady-state behavior. However, the proof is based solely on the linearized equations, without additional justification for why the result should hold in the presence of nonlinearities. Moreover, the linearized system is used to analyze the response to a "spike initial pattern of arbitrary height C" (SI Chapter S5.1), yet it is not clear how conclusions derived from the linear regime can be valid for large perturbations, where nonlinear effects are expected to play a significant role. We encourage the authors to clarify the assumptions under which the linearized analysis remains valid and to discuss the potential limitations of applying these results to the nonlinear regime.

In this article, we address two main questions: first, which gene network topologies can give rise to non-trivial pattern transformations; and second, which broad types of resulting patterns can these gene network topologies give rise to resulting pattern. Thus, we are not intending to explain which exact resulting patterns would arise from any given gene network (i.e. a gene network topology with specific functions and interaction strengths or weights), a question for which non-linearities do indeed matter.

For most known gene regulatory networks, available empirical information is typically limited to the nature of gene product regulations -indicating whether they act as activators or inhibitors- while details about the specific functional form of these regulations are rare. For instance, given two gene products, i and j, the network may indicate that i acts as an activator of j, implying that the concentration of j increases with that of i. However, this increase could follow a variety of functional forms: it may be quadratic (e.g., ), cubic (e.g., ), or any other function f j(gi). As we explain in the description of our model, we restrict our study to functions with a monotonicity constraint: higher concentrations of i lead to increased production of j (i.e., ).  In other words, a given gene interaction is always inhibitory or activatory, it does not change of sign. This monotonicity constraint corresponds to requirement (R5) in our main text. This requirement it is based on the biologically plausible idea that the complexity of gene regulation in development stems more from the topology of gene networks than from the complexity of the regulation by which a gene product may regulate another (i.e. we use simple monotonic functions).

Question 1: A critical part to understand question 1 is in the dispersion relation that was explained in SI. From the reviewers’ comments it is clear that having this crucial part in the main text of an upcoming version of the article would improve understandability, specially for question 1.

In brief, any pattern transformation requires the initial pattern to change. The trigger of such change is a change in the concentration of some gene product, either conceptualized as a noise fluctuation (in the homogeneous initial pattern) or a regulated change in a specific point (in the spike initial pattern). Mathematically, both can be conceptualized as perturbations and, for pattern transformation to be possible, such perturbation should grow so that the initial pattern becomes unstable and can change to another resulting pattern.

If the perturbation is small, one can use the standard linear perturbation analysis in S6.2 of our Supplementary Information. In other words, the linear analysis is enough to ascertain if a small perturbation would grow or not. A gene network in which this will not happen would be unable to lead to pattern transformation, whichever the nonlinear part of f(g). In that sense, the linear approximation provides a necessary condition that any gene network needs to fulfill to lead to pattern transformation.

However, the linear analysis would not ascertain whether a specific gene network will actually lead to pattern transformation (i.e., the condition is not sufficient). This, as well as the shape of the specific resulting pattern, may actually depend on the non-linear parts too. As we discuss, based on the dispersion relation, and other complementing arguments along the article, we can also get some insights on the possible patterns from the linear approximation alone (question 2). This arguments hold thanks to the imposition of requirements (R1-R5) on function f(g), which prevent strange behaviors stemming from the nonlinear part of the equation.

The amplitude bound of perturbations mentioned by Reviewer #1 is addressed by requirements (R2) and (R4). Although the solution to the linear system predicts unbounded growth of unstable eigenmodes, the assume functions f(g) on which the nonlinear terms  eventually halt this growth, thereby ensuring the boundedness of solutions as imposed by (R4). This assumption on the nonlinear part is literally requirement R2 on f(g) in the main text.

The transitive chains and branchings in section S3 of the Supplementary Information mentioned by the Reviewer #2 are topological properties of gene networks and therefore they influence only the linear part of the reaction-diffusion equations. This is why the proofs in that section are based on the linearized equations. We agree that clarifying this point in the text, as suggested by the reviewer, would improve the reader’s understanding of the section.

Regarding Reviewer #2’s concerns about large perturbations, we acknowledge that the phrasing using “arbitrary height” may be confusing. For the homogeneous initial conditions these perturbations are assumed to be small because they are actually molecular noise (otherwise the initial condition could not be considered homogenous in the classical sense of developmental biology models). In the spike initial conditions in hierarchic networks the perturbation is not necessarily small. For the analysis provided in the SI we indeed assume that the perturbations are small enough for the linear approximation to be possible. Notice, however, that since these networks require an intracellular self-activating loop upstream of the first extracellular signal, the effective perturbation would rapidly grow to a value determined by such loop.

In general the height of the initial spike does not affect the fact that hierarchic networks can lead to non-trivial pattern transformation. By definition these networks require the secretion of an extracellular signal from the cells in the spike (otherwise no change in gene product concentrations can occur over space). By definition this signal is not produced by any other cells and, thus, its concentration is governed by diffusion from the spike and its production in the cells in the spike. Thus, whichever the initial height of the spike and whichever the non-linearities in f(g), the signal’s concentration would decrease with the distance from the spike. As explained in the main text, this would lead to non-trivial pattern transformations if other general conditions are met. In general, the height of the initial perturbation can affect which specific pattern transformation would arise from a specific gene network but not which gene network topologies can lead to pattern transformation. This will be more clearly stated in an upcoming version of the article. C. In the following, we respond to the remaining concerns raised by the reviewers:

Reviewer #1:

(1) The Results section is difficult to follow. Key logical steps and network configurations are described shortly in prose, which constantly require the reader to address either SI or other parts of the text (see numerous links on the requirements R1-R5 listed at the beginning of the paper) to gain minimal understanding. As a result, a scientifically literate but non-specialist reader may struggle to grasp the argument with a reasonable time invested.

We acknowledge that the current version of the main text may not be as clear as we intended. Initially, we believed that placing the more technical mathematical passages in the Supplementary Information would make the main text more accessible to readers. However, we agree with the reviewer that including some of these computations in the main text could improve clarity. We also believe that adding a summary table outlining all the model’s requirements would further contribute to that goal.

Reviewer #2:

(1) We have serious concerns regarding the validity of the simulation results presented in the manuscript. Rather than simulating the full nonlinear system described by Equation (1), the authors base their results on a truncated expansion (Equation S.8.2) that captures only the time evolution of small deviations around a spatially homogeneous steady state. However, it remains unclear how this reduced system is derived from the full equations specifically, which terms are retained or neglected and why- and how the expansion of the nonlinear function can be steady-state independent, as claimed. Additionally, in simulations involving the spike plus homogeneous initial condition, it is not evident -or, where equations are provided, it is not correct- that the assumed global homogeneous background actually corresponds to a steady state of the full dynamics. We elaborate on these concerns in the following:

We believe there has been a misunderstanding regarding the presentation of the model equations (S8.2) used throughout our simulations. Accordingly, we agree that this relevant section of the Supplementary Information should be rewritten in a revised version of the manuscript to clarify this issue. Below, we address all the concerns raised by the reviewer.

Equation (S8.2) represents the full nonlinear system described in Equation (1). While we recognize that the model may oversimplify real biological processes, its purpose is to illustrate our general statements about pattern formation rather than to capture any specific or detailed mechanism. In this context, model (S8.2) offers three key advantages for our goals: it allows rapid manipulation of gene network topology simply by modifying the matrix J, making it ideal for illustrating pattern formation across different network classes; it accommodates gene networks of arbitrary size -unlike other models, such as the classical Gierer-Meinhardt model, which are limited to two-element Turing or noise-amplifying networks-; and, due to the simplicity of its nonlinear terms, this model involves relatively few free parameters, facilitating the fine-tuning needed to identify parameter regions where non-trivial pattern transformations occur.

Indeed, we find that the ability of model (S8.2) to illustrate our results despite having such simple nonlinear terms -bearing in mind that at least some nonlinearity is always necessary for selforganization- strongly supports the claim that the capacity of a gene network to produce pattern transformations is fully determined by the linear part of Equation (1). In this sense, nonlinear terms primarily influence the precise parameter values at which these transformations occur and contribute to shaping specific features of the resulting patterns.

Model (S8.2) has been successfully employed in pattern formation studies elsewhere in the literature; accordingly, we provide relevant bibliographic references to support its widespread use.

We believe the misunderstanding arises from our explanation of the biological interpretation of the model. As noted in the accompanying bibliography, the model is based on a general reactiondiffusion mechanism assuming the existence of a steady state. However, this conceptual reactiondiffusion framework is not the same as our Equation (1); rather, it was introduced by the original proponents of the model in the seminal paper cited in our text. In this context, Equation (S8.2) describes small concentration perturbations around that steady state, where the variables represent deviations in concentration relative to the general steady state.

The aforementioned general steady state corresponds to the trivial equilibrium point g≡0 in equations (S8.2). Consequently, all our simulations based on model (S8.2) start from this steady state, to which we add white noise to generate homogeneous initial patterns or a sharp spike for the two types of spike initial patterns.

It is also worth noting that Equations (S8.2) represent a non-dimensional model.

It is assumed that the homogeneous steady states are given by g_i=0 and g_i=c_i, where 1/c_i = \mu_i or \hat{\mu}_i, independently of the specific network structure. However, the basis for this assumption is unclear, especially since some of the functions do not satisfy this condition -for example, f5 as defined below Eq. S8.10.5. Moreover, if g_i=c_i does not correspond to a true steady state, then the time evolution of deviations from this state is not correctly described by Eq. S8.2, as the zeroth-order terms do not vanish in that case.

From the explanations above, it is important to distinguish two scales in the process: the scale of small perturbations, where equations (S8.2) apply; and the global scale, where the conceptual general reaction-diffusion system operates. Since the specific form of this general system does not affect equations (S8.2), we assume that it follows any of the models cited in the text, which yield a non-zero steady state at .

In this sense, Equation (S8.2) represent a small concentration deviation of such global system and g(t ,x) is a relative concentration where g≡0 represents the steady-state at are concentrations above , and g<0 are concentrations below .

As previously mentioned, simulations are performed using Equations (S8.2) on the basis of the equilibrium point g≡0. The result of these simulations is then superimposed on the non-zero steady state and presented in the figures along the article.

Using the full model instead of the simplified Equations (S8.2) may result in slightly different resulting patterns, but it does not affect the gene network’s ability to produce pattern transformations, nor does it alter the main structural properties of the patterns—for example, the periodic nature of patterns generated by Turing networks.

Additionally, the equations used contain only linear terms and a cubic degradation term for each species g_i, while neglecting all quadratic terms and cubic terms involving cross-species interactions (i≠j). An explanation for this selective truncation is not provided, and without knowledge of the full equation (f), it is impossible to assess whether this expansion is mathematically justified. If, as suggested in the Supplementary Information, the linear and cubic terms are derived from f, then at the very least, the Jacobian matrix should depend on the background steady-state concentration. However, the equations for the small deviation around a steady state (including the Jacobian matrix) used in the simulations appear to be independent of the particular steady state concentration.

The Jacobian of Equation (S8.2) is independent of g because g represents a small perturbation around a steady state of a general reaction-diffusion system. Consequently, the matrix J corresponds to the Jacobian of the general system evaluated at that steady state. Evaluating the Jacobian of equations (S8.2) at the equilibrium point g≡0 -which represents the general steady state- recovers the matrix J.

This is why we believe that the differences observed between the spike-only initial condition and the spike superimposed on a homogeneous background are not due to the initial conditions themselves, but rather result from a modified reaction scheme introduced through a questionable cutoff.

"In simulations with spike initial patterns, the reference value g≡0 represents an actual concentration of 0 and therefore, we must add to (S8.2) a Heaviside function Φ acting of f (i.e., Φ(f(g))=f(g) if f(g)>0 , Φ(f(g))=0 if f(g){less than or equal to}0 ) to prevent the existence of negative concentrations for any gene product (i.e., g_i<0 for some i )." (SI chapter S8).

This cutoff alters the dynamics (no inhibition) and introduces a different reaction scheme between the two simulations. The need for this correction may itself reflect either a problem in the original equations (which should fulfill the necessary conditions and prevent negative concentrations (R4 in main text)) or the inappropriateness of using an expanded approximation which assumes independence on the steady state concentration. It is already questionable if the linearized equations with a cubic degradation term are valid for the spike initial conditions (with different background concentration values), as the amplitude of this perturbation seems rather large.

For homogeneous and spike+homogeneous initial conditions, we interpret equations (S8.2) as small perturbations around a non-zero steady state of a general reaction-diffusion system. For spike-only initial conditions, that steady state is zero. As we mention before, g≡0 will then represent such steady-state of zero concentration, g>0 are positive concentrations of the general system, and g<0 would represent unfeasible negative concentrations of the general system. Therefore, the use of a cutoff function to handle such initial conditions is justified. Moreover, this cutoff function is the same as the one employed in the reference general system cited in our paper.

We acknowledge that the cutoff influences the simulations and accounts for the differences observed between spike and spike+homogeneous initial conditions. However, this distinction reflects what occurs in real biological systems, which is precisely why we differentiate these two types of initial states. For instance, the emergence of a periodic pattern in a noise-amplifying network depends critically on the formation of regions with concentrations below the steady state near the initial spike. Such regions can form in spike-plus-homogeneous initial patterns but not in spike-only initial patterns, where concentrations below the steady state would correspond to biologically unfeasible negative values.

Lastly, we note that under the current simulation scheme, it is not possible to meaningfully assess criteria RH2a and RH2b, as they rely on nonlinear interactions that are absent from the implemented dynamics.

It is explicitly stated in the relevant subsections of Section S7 in the Supplementary Information that, for the simulations involving RH2a and RH2b, the function f(g) in equation (S8.2) is modified by adding an ad hoc quadratic term to enable the assessment of these criteria.

(3) Several statements in the main text are presented without accompanying proof or sufficient explanation, which makes it difficult to assess their validity. In some cases, the lack of justification raises serious doubts about whether the claims are generally true. Examples are:

"For the purpose of clarity we will explain our results as if these cells have a simple arrangement in space (e.g., a 1D line or a 2D square lattice) but, as we will discuss, our results shall apply with the same logic to any distribution of cells in space." (Main text l.145-l.148).

We believe that the confusion in this statement arises from the ambiguous use of the phrase “our results”. We will revise the text to provide a more precise description. Specifically, by “our results,” we refer to the conclusion that it is possible to determine whether a gene network leads to nontrivial pattern transformations based solely on its topology. This conclusion is independent of the dimensionality of space, as none of our arguments rely on assumptions specific to spatial dimensions. While one-dimensional examples are used for clarity and illustration, the underlying reasoning applies generally. In an improved version of the article, we will clarify this point explicitly and move relevant arguments from the Supplementary Information into the main text.

Critically, our classification of gene networks is ultimately based on an argument concerning the dispersion relation associated with the network, and the construction of this dispersion relation is independent of the spatial dimensionality of the domain. In this sense, the networks identified in the text as capable of producing pattern transformations will be able to generate non-trivial pattern transformations in any spatial domain and in any number of dimensions. While the specific parameter values that permit such transformations may vary depending on the geometry and dimensionality of the domain, the existence of at least one such parameter set remains unaffected.

The geometry of the domain can influence the specific form of the resulting patterns, but it does not alter the broader class of patterns (e.g., periodic patterns, peaks emerging around a spike, etc.) that a given gene network topology can produce. One such geometric influence, commonly observed in simulations, involves boundary effects. For example, structures such as peaks or rings forming near the boundaries may appear higher, broader, or spatially shifted compared to those arising in the central regions of the domain. However, we think a pattern consisting of a periodic train of peaks where only those near the boundary are slightly different can still be classified as a periodic pattern.

"For any non-trivial pattern transformation (as long as it is symmetric around the initial spike), there exists an H gene network capable of producing it from a spike initial pattern." (Main text l.366f).

A justification for this statement is provided shortly after the claim, although we acknowledge that the current explanation is somewhat cumbersome and would benefit from a clearer presentation in a revised version of the main text.

A more detailed justification is provided in the Supplementary Information, based on three key ideas. First, any pattern (provided it is symmetric with respect to the initial spike) can be described as an arrangement of peaks with varying heights and spatial positions along a one-dimensional domain. Second, there exists a simple gene network—the diamond network—that, through parameter tuning, can produce two peaks of arbitrary height and symmetric position relative to the initial spike. Third, by placing multiple diamond networks positively upstream of a common gene product, that gene product can express peaks at each location where the upstream diamond networks induce them. Under mild additional conditions, this mechanism allows the formation of essentially any symmetric pattern. These mild conditions, along with a detailed analysis of the diamond network’s ability to generate peaks with controllable height and position, are discussed in the Supplementary Information.

"In 2D there are no peaks but concentric rings of high gene product concentration centered around the spike, while in 3D there are concentric spherical shells." (Main text l. 447ff).

This result pertains specifically to pattern transformations arising from spike initial patterns. As defined in the text, spike initial patterns are radially symmetric. Since diffusion preserves radial symmetry, pattern transformations from spike initial patterns in two or three dimensions reduce to effectively one-dimensional transformations along each radial direction. In this framework, each pair of concentration peaks symmetric with respect to the spike in one dimension corresponds to a ring surrounding the spike in two dimensions, and each ring in two dimensions becomes a hollow spherical shell around the spike in three dimensions.

We agree that including a brief section in the Supplementary Information to clarify these subtleties would be helpful for readers to better understand the generalization of certain patterns to higher dimensions.

(4) The study identifies one-signal networks and examines how combinations of these structures can give rise to minimal pattern-forming subnetworks. However, the analysis of the combinations of these minimal pattern-forming subnetworks remains relatively brief, and the manuscript does not explore how the results might change if the subnetworks were combined in upstream and downstream configurations. In our view, it is not evident that all possible gene regulatory networks can be fully characterized by these categories, nor that the resulting patterns can be reliably predicted. Rather, the approach appears more suited to identifying which known subnetworks are present within a larger network, without necessarily capturing the full dynamics of more complex configurations.

We acknowledge that our explanation regarding the combination of sub-networks was relatively brief, and we intend to address this in a revised version. Our argument that combining sub-networks does not produce qualitatively new types of pattern transformations -beyond those already described- is based on the dispersion relation. Although this relation was only detailed in the Supplementary Information, it is central to our argument and will therefore be moved to the main text. Below, we provide an outline of this argument:

Our study identifies two distinct behaviors of the principal branch of the dispersion relation at large wavenumbers. Based on this, gene networks capable of pattern formation can be classified into two categories: networks of the first kind, where the real part of the principal branch diverges to infinity as the wavenumber increases; and networks of the second kind, where the real part of the principal branch converges to a positive finite value for large wavenumbers. Naturally this argument applies to any gene network irrespectively of which, or how many, sub-networks are used to built it.

Any gene regulatory network capable of pattern formation falls into one of these two categories. We identified that networks of the first kind contain at least one Turing sub-network, whereas networks of the second kind include either an H sub-network or a noise-amplifying sub-network. In this way, the primary objective of our study -namely, achieving a topological classification of gene regulatory networks capable of pattern formation- is fulfilled. It is important to note that while the dispersion relation provides broad information about the possible resulting patterns a gene network topology can produce (e.g., periodic versus noisy), it does not specify the exact patterns that emerge for each particular set of parameter values.

Finally, regarding the shape of the resulting patterns, Figure S10 in the Supplementary Information exemplifies the notion that the behavior of combined networks can be understood as a combination of the individual behaviors of each constituent sub-network (note that the contribution of each type of sub-network in the resulting pattern is readily distinguishable). Consequently, we focus our detailed analysis on the patterning properties of the fundamental classes.

(6) The manuscript lacks a clear and detailed explanation of the underlying model and its assumptions. In particular, it is not well-defined what constitutes a "cell" in the context of the model, nor is it justified why spatial features of cells -such as their size or boundaries- can be neglected. Furthermore, the concept of the extracellular space in the one-dimensional model remains ambiguous, making it unclear which gene products are assumed to diffuse.

The size of cells is ignored in our model because we assume that they are small enough with respect to the total size of the domain that the space continuous reaction-diffusion equation (equation (1) in the main text) holds. Conceptually, one could understand cells in our model each of the pieces in an even partition of the domain into small subdomains surrounding each position x. This is anyway the standard procedure in most models of pattern formation by reaction-diffusion in embryonic development.

For extracellular signals, we assume that g(t ,x) corresponds to the concentration of the signal in the extracellular space surrounding the cell located at position x. The extracellular space is any fluid medium for which Fick Laws apply and, therfore, the Fickian diffusion term in equation (1) is valid.

For intracellular gene products, we assume that g(t ,x) corresponds to the concentration of such gene product within the cell at position x (if the gene product in hand is a transcription factor, for example), or on its surface (if it is a membrane-bound receptor). When collapsed in the continuous equations there is not such difference between being strictly within the cell or on its boundary. The only important fact is that these gene products cannot diffuse.

Regarding cell boundaries, let us consider an extracellular signal s that regulates a transcriptor factor i within cells (in our model, i is an intracellular gene product). Such regulation shall be mediated by a membrane-bound receptor, which corresponds to intracellular gene product j. In terms of the gene regulatory network this is s→ j→i. Cell boundary effects mentioned by the reviewer should be encapsulated in the specific functional form of the regulation function f(g), but they have no effect in the actual topology of the network. Consequently, they are out of the scope of this study: as we mentioned before, considering different non-linear terms for f(g) will affect the parameter range for which a gene network is capable of producing non-trivial pattern transformations, but not their overall ability to produce non-trivial pattern transformations (i.e., the existence of at least one choice of model parameters for which such transformations take place).

Finally, we would like to once again express our sincere gratitude to all reviewers for their insightful and constructive feedback. We are confident that the thorough peer review process will significantly enhance both the clarity and depth of our work. We greatly value the detailed comments provided and will carefully incorporate them in the preparation of a revised manuscript, which we intend to submit in the coming months.

Author Response

Public Reviews:

Reviewer #1 (Public Review):

Summary:

Given knowledge of the amino acid sequence and of some version of the 3D structure of two monomers that are expected to form a complex, the authors investigate whether it is possible to accurately predict which residues will be in contact in the 3D structure of the expected complex. To this effect, they train a deep learning model that takes as inputs the geometric structures of the individual monomers, per-residue features (PSSMs) extracted from MSAs for each monomer, and rich representations of the amino acid sequences computed with the pre-trained protein language models ESM-1b, MSA Transformer, and ESM-IF. Predicting inter-protein contacts in complexes is an important problem. Multimer variants of AlphaFold, such as AlphaFold-Multimer, are the current state of the art for full protein complex structure prediction, and if the three-dimensional structure of a complex can be accurately predicted then the inter-protein contacts can also be accurately determined. By contrast, the method presented here seeks state-of-the-art performance among models that have been trained end-to-end for inter-protein contact prediction.

Strengths:

The paper is carefully written and the method is very well detailed. The model works both for homodimers and heterodimers. The ablation studies convincingly demonstrate that the chosen model architecture is appropriate for the task. Various comparisons suggest that PLMGraph-Inter performs substantially better, given the same input than DeepHomo, GLINTER, CDPred, DeepHomo2, and DRN-1D2D_Inter. As a byproduct of the analysis, a potentially useful heuristic criterion for acceptable contact prediction quality is found by the authors: namely, to have at least 50% precision in the prediction of the top 50 contacts.

We thank the reviewer for recognizing the strengths of our work!

Weaknesses:

My biggest issue with this work is the evaluations made using bound monomer structures as inputs, coming from the very complexes to be predicted. Conformational changes in protein-protein association are the key element of the binding mechanism and are challenging to predict. While the GLINTER paper (Xie & Xu, 2022) is guilty of the same sin, the authors of CDPred (Guo et al., 2022) correctly only report test results obtained using predicted unbound tertiary structures as inputs to their model. Test results using experimental monomer structures in bound states can hide important limitations in the model, and thus say very little about the realistic use cases in which only the unbound structures (experimental or predicted) are available. I therefore strongly suggest reducing the importance given to the results obtained using bound structures and emphasizing instead those obtained using predicted monomer structures as inputs.

We thank the reviewer for the suggestion! We evaluated PLMGraph-Inter with the predicted monomers and analyzed the result in details (see the “Impact of the monomeric structure quality on contact prediction” section and Figure 3). To mimic the real cases, we even deliberately reduced the performance of AF2 by using reduced MSAs (see the 2nd paragraph in the ““Impact of the monomeric structure quality on contact prediction” section). We leave some of the results in the supplementary of the current manuscript (Table S2). We will move these results to the main text to emphasize the performance of PLMGraph-Inter with the predicted monomers in the revision.

In particular, the most relevant comparison with AlphaFold-Multimer (AFM) is given in Figure S2, not Figure 6. Unfortunately, it substantially shrinks the proportion of structures for which AFM fails while PLMGraph-Inter performs decently. Still, it would be interesting to investigate why this occurs. One possibility would be that the predicted monomer structures are of bad quality there, and PLMGraph-Inter may be able to rely on a signal from its language model features instead. Finally, AFM multimer confidence values ("iptm + ptm") should be provided, especially in the cases in which AFM struggles.

We thank the reviewer for the suggestion! Yes! The performance of PLMGraph-Inter drops when the predicted monomers are used in the prediction. However, it is difficult to say which is a fairer comparison, Figure 6 or Figure S2, since AFM also searched monomer templates (see the third paragraph in 7. Supplementary Information : 7.1 Data in the AlphaFold-Multimer preprint: https://www.biorxiv.org/content/10.1101/2021.10.04.463034v2.full) in the prediction. When we checked our AFM runs, we found that 99% of the targets in our study (including all the targets in the four datasets: HomoPDB, HeteroPDB, DHTest and DB5.5) employed at least 20 templates in their predictions, and 87.8% of the targets employed the native templates. We will provide the AFM confidence values of the AFM predictions in the revision.

Besides, in cases where any experimental structures - bound or unbound - are available and given to PLMGraph-Inter as inputs, they should also be provided to AlphaFold-Multimer (AFM) as templates. Withholding these from AFM only makes the comparison artificially unfair. Hence, a new test should be run using AFM templates, and a new version of Figure 6 should be produced. Additionally, AFM's mean precision, at least for top-50 contact prediction, should be reported so it can be compared with PLMGraph-Inter's.

We thank the reviewers for the suggestion! We would like to notify that AFM also searched monomer templates (see the third paragraph in 7. Supplementary Information : 7.1 Data in the AlphaFold-Multimer preprint: https://www.biorxiv.org/content/10.1101/2021.10.04.463034v2.full) in the prediction. When we checked our AFM runs, we found that 99% of the targets in our study (including all the targets in the four datasets: HomoPDB, HeteroPDB, DHTest and DB5.5) employed at least 20 templates in their predictions, and 87.8% of the targets employed the native template.

It's a shame that many of the structures used in the comparison with AFM are actually in the AFM v2 training set. If there are any outside the AFM v2 training set and, ideally, not sequence- or structure-homologous to anything in the AFM v2 training set, they should be discussed and reported on separately. In addition, why not test on structures from the "Benchmark 2" or "Recent-PDB-Multimers" datasets used in the AFM paper?

We thank the reviewer for the suggestion! The biggest challenge to objectively evaluate AFM is that as far as we known, AFM does not release the PDB ids of its training set and the “Recent-PDB-Multimers” dataset. “Benchmark 2” only includes 17 heterodimer proteins, and the number can be further decreased after removing targets redundant to our training set. We think it is difficult to draw conclusions from such a small number of targets. In the revision, we will analyze the performance of AFM on targets released after the date cutoff of the AFM training set, but with which we cannot totally remove the redundancy between the training and the test sets of AFM.

It is also worth noting that the AFM v2 weights have now been outdated for a while, and better v3 weights now exist, with a training cutoff of 2021-09-30.

We thank the reviewer for reminding the new version of AFM. The only difference between AFM V3 and V2 is the cutoff date of the training set. Our test set would have more overlaps with the training set of AFM V3, which is one reason that we think AFM V2 is more appropriate to be used in the comparison.

Another weakness in the evaluation framework: because PLMGraph-Inter uses structural inputs, it is not sufficient to make its test set non-redundant in sequence to its training set. It must also be non-redundant in structure. The Benchmark 2 dataset mentioned above is an example of a test set constructed by removing structures with homologous templates in the AF2 training set. Something similar should be done here.

We agree with the reviewer that testing whether the model can keep its performance on targets with no templates (i.e. non-redundant in structure) is important. We will perform the analysis in the revision.

Finally, the performance of DRN-1D2D for top-50 precision reported in Table 1 suggests to me that, in an ablation study, language model features alone would yield better performance than geometric features alone. So, I am puzzled why model "a" in the ablation is a "geometry-only" model and not a "LM-only" one.

Using the protein geometric graph to integrate multiple protein language models is the main idea of PLMGraph-Inter. Comparing with our previous work (DRN-1D2D_Inter), we consider the building of the geometric graph as one major contribution of this work. To emphasize the efficacy of this geometric graph, we chose to use the “geometry-only” model as the base model. We will further clarity this in the revision.

Reviewer #2 (Public Review):

This work introduces PLMGraph-Inter, a new deep-learning approach for predicting inter-protein contacts, which is crucial for understanding protein-protein interactions. Despite advancements in this field, especially driven by AlphaFold, prediction accuracy and efficiency in terms of computational cost) still remains an area for improvement. PLMGraph-Inter utilizes invariant geometric graphs to integrate the features from multiple protein language models into the structural information of each subunit. When compared against other inter-protein contact prediction methods, PLMGraph-Inter shows better performance which indicates that utilizing both sequence embeddings and structural embeddings is important to achieve high-accuracy predictions with relatively smaller computational costs for the model training.

The conclusions of this paper are mostly well supported by data, but test examples should be revisited with a more strict sequence identity cutoff to avoid any potential information leakage from the training data. The main figures should be improved to make them easier to understand.

We thank the reviewer for recognizing the significance of our work! We will revise the manuscript carefully to address the reviewer’s concerns.

1. The sequence identity cutoff to remove redundancies between training and test set was set to 40%, which is a bit high to remove test examples having homology to training examples. For example, CDPred uses a sequence identity cutoff of 30% to strictly remove redundancies between training and test set examples. To make their results more solid, the authors should have curated test examples with lower sequence identity cutoffs, or have provided the performance changes against sequence identities to the closest training examples.

We thank the reviewer for the valuable suggestion! Using different thresholds to reduce the redundancy between the test set and the training set is a very good suggestion, and we will perform the analysis in the revision. In the current version of the manuscript, the 40% sequence identity is used as the cutoff for many previous studies used this cutoff (e.g. the Recent-PDB-Multimers used in AlphaFold-Multimer (see: 7.8 Datasets in the AlphaFold-Multimer paper); the work of DSCRIPT: https://www.cell.com/action/showPdf?pii=S2405-4712%2821%2900333-1 (see: the PPI dataset paragraph in the METHODS DETAILS section of the STAR METHODS)). One reason for using the relatively higher threshold for PPI studies is that PPIs are generally not as conserved as protein monomers.

We performed a preliminary analysis using different thresholds to remove redundancy when preparing this provisional response letter:

Author response table 1.

Table1. The performance of PLMGraph-Inter on the HomoPDB and HeteroPDB test sets using native structures(AlphaFold2 predicted structures).

Method:

To remove redundancy, we clustered 11096 sequences from the training set and test sets (HomoPDB, HeteroPDB) using MMSeq2 with different sequence identity threshold (40%, 30%, 20%, 10%) (the lowest cutoff for CD-HIT is 40%, so we switched to MMSeq2). Each sequence is then uniquely labeled by the cluster (e.g. cluster 0, cluster 1, …) to which it belongs, from which each PPI can be marked with a pair of clusters (e.g. cluster 0-cluster 1). The PPIs belonging to the same cluster pair (note: cluster n - cluster m and cluster n-cluster m were considered as the same pair) were considered as redundant. For each PPI in the test set, if the pair cluster it belongs to contains the PPI belonging to the training set, we remove that PPI from the test set.

We will perform more detailed analyses in the revised manuscript.

1. Figures with head-to-head comparison scatter plots are hard to understand as scatter plots because too many different methods are abstracted into a single plot with multiple colors. It would be better to provide individual head-to-head scatter plots as supplementary figures, not in the main figure.

We thank the reviewer for the suggestion! We will include the individual head-to-head scatter plots as supplementary figures in the revision.

3) The authors claim that PLMGraph-Inter is complementary to AlphaFold-multimer as it shows better precision for the cases where AlphaFold-multimer fails. To strengthen the point, the qualities of predicted complex structures via protein-protein docking with predicted contacts as restraints should have been compared to those of AlphaFold-multimer structures.

We thank the reviewer for the suggestion! We will add this comparison in the revision.

4) It would be interesting to further analyze whether there is a difference in prediction performance depending on the depth of multiple sequence alignment or the type of complex (antigen-antibody, enzyme-substrates, single species PPI, multiple species PPI, etc).

We thank the reviewer for the suggestion! We will perform such analysis in the revision.

Author response:

eLife Assessment 

This valuable study investigates how the neural representation of individual finger movements changes during the early period of sequence learning. By combining a new method for extracting features from human magnetoencephalography data and decoding analyses, the authors provide incomplete evidence of an early, swift change in the brain regions correlated with sequence learning, including a set of previously unreported frontal cortical regions. The addition of more control analyses to rule out that head movement artefacts influence the findings, and to further explain the proposal of offline contextualization during short rest periods as the basis for improvement performance would strengthen the manuscript. 

We appreciate the Editorial assessment on our paper’s strengths and novelty.  We have implemented additional control analyses to show that neither task-related eye movements nor increasing overlap of finger movements during learning account for our findings, which are that contextualized neural representations in a network of bilateral frontoparietal brain regions actively contribute to skill learning.  Importantly, we carried out additional analyses showing that contextualization develops predominantly during rest intervals.

Public Reviews:

We thank the Reviewers for their comments and suggestions, prompting new analyses and additions that strengthened our report.

Reviewer #1 (Public review): 

Summary: 

This study addresses the issue of rapid skill learning and whether individual sequence elements (here: finger presses) are differentially represented in human MEG data. The authors use a decoding approach to classify individual finger elements and accomplish an accuracy of around 94%. A relevant finding is that the neural representations of individual finger elements dynamically change over the course of learning. This would be highly relevant for any attempts to develop better brain machine interfaces - one now can decode individual elements within a sequence with high precision, but these representations are not static but develop over the course of learning. 

Strengths: The work follows a large body of work from the same group on the behavioural and neural foundations of sequence learning. The behavioural task is well established and neatly designed to allow for tracking learning and how individual sequence elements contribute. The inclusion of short offline rest periods between learning epochs has been influential because it has revealed that a lot, if not most of the gains in behaviour (ie speed of finger movements) occur in these so-called micro-offline rest periods. The authors use a range of new decoding techniques, and exhaustively interrogate their data in different ways, using different decoding approaches. Regardless of the approach, impressively high decoding accuracies are observed, but when using a hybrid approach that combines the MEG data in different ways, the authors observe decoding accuracies of individual sequence elements from the MEG data of up to 94%. 

We have previously showed that neural replay of MEG activity representing the practiced skill correlated with micro-offline gains during rest intervals of early learning, 1 consistent with the recent report that hippocampal ripples during these offline periods predict human motor sequence learning2.  However, decoding accuracy in our earlier work1 needed improvement.  Here, we reported a strategy to improve decoding accuracy that could benefit future studies of neural replay or BCI using MEG.

Weaknesses: 

There are a few concerns which the authors may well be able to resolve. These are not weaknesses as such, but factors that would be helpful to address as these concern potential contributions to the results that one would like to rule out. Regarding the decoding results shown in Figure 2 etc, a concern is that within individual frequency bands, the highest accuracy seems to be within frequencies that match the rate of keypresses. This is a general concern when relating movement to brain activity, so is not specific to decoding as done here. As far as reported, there was no specific restraint to the arm or shoulder, and even then it is conceivable that small head movements would correlate highly with the vigor of individual finger movements. This concern is supported by the highest contribution in decoding accuracy being in middle frontal regions - midline structures that would be specifically sensitive to movement artefacts and don't seem to come to mind as key structures for very simple sequential keypress tasks such as this - and the overall pattern is remarkably symmetrical (despite being a unimanual finger task) and spatially broad. This issue may well be matching the time course of learning, as the vigor and speed of finger presses will also influence the degree to which the arm/shoulder and head move. This is not to say that useful information is contained within either of the frequencies or broadband data. But it raises the question of whether a lot is dominated by movement "artefacts" and one may get a more specific answer if removing any such contributions. 

Reviewer #1 expresses concern that the combination of the low-frequency narrow-band decoder results, and the bilateral middle frontal regions displaying the highest average intra-parcel decoding performance across subjects is suggestive that the decoding results could be driven by head movement or other artefacts.

Head movement artefacts are highly unlikely to contribute meaningfully to our results for the following reasons. First, in addition to ICA denoising, all “recordings were visually inspected and marked to denoise segments containing other large amplitude artifacts due to movements” (see Methods). Second, the response pad was positioned in a manner that minimized wrist, arm or more proximal body movements during the task. Third, while head position was not monitored online for this study, the head was restrained using an inflatable air bladder, and head position was assessed at the beginning and at the end of each recording. Head movement did not exceed 5mm between the beginning and end of each scan for all participants included in the study. Fourth, we agree that despite the steps taken above, it is possible that minor head movements could still contribute to some remaining variance in the MEG data in our study. The Reviewer states a concern that “it is conceivable that small head movements would correlate highly with the vigor of individual finger movements”. However, in order for any such correlations to meaningfully impact decoding performance, such head movements would need to: (A) be consistent and pervasive throughout the recording (which might not be the case if the head movements were related to movement vigor and vigor changed over time); and (B) systematically vary between different finger movements, and also between the same finger movement performed at different sequence locations (see 5-class decoding performance in Figure 4B). The possibility of any head movement artefacts meeting all these conditions is extremely unlikely.

Given the task design, a much more likely confound in our estimation would be the contribution of eye movement artefacts to the decoder performance (an issue appropriately raised by Reviewer #3 in the comments below). Remember from Figure 1A in the manuscript that an asterisk marks the current position in the sequence and is updated at each keypress. Since participants make very few performance errors, the position of the asterisk on the display is highly correlated with the keypress being made in the sequence. Thus, it is possible that if participants are attending to the visual feedback provided on the display, they may move their eyes in a way that is systematically related to the task.  Since we did record eye movements simultaneously with the MEG recordings (EyeLink 1000 Plus; Fs = 600 Hz), we were able to perform a control analysis to address this question. For each keypress event during trials in which no errors occurred (which is the same time-point that the asterisk position is updated), we extracted three features related to eye movements: 1) the gaze position at the time of asterisk position update (or keyDown event), 2) the gaze position 150ms later, and 3) the peak velocity of the eye movement between the two positions. We then constructed a classifier from these features with the aim of predicting the location of the asterisk (ordinal positions 1-5) on the display. As shown in the confusion matrix below (Author response image 1), the classifier failed to perform above chance levels (Overall cross-validated accuracy = 0.21817):

Author response image 1.

Confusion matrix showing that three eye movement features fail to predict asterisk position on the task display above chance levels (Fold 1 test accuracy = 0.21718; Fold 2 test accuracy = 0.22023; Fold 3 test accuracy = 0.21859; Fold 4 test accuracy = 0.22113; Fold 5 test accuracy = 0.21373; Overall cross-validated accuracy = 0.2181). Since the ordinal position of the asterisk on the display is highly correlated with the ordinal position of individual keypresses in the sequence, this analysis provides strong evidence that keypress decoding performance from MEG features is not explained by systematic relationships between finger movement behavior and eye movements (i.e. – behavioral artefacts).

In fact, inspection of the eye position data revealed that a majority of participants on most trials displayed random walk gaze patterns around a center fixation point, indicating that participants did not attend to the asterisk position on the display. This is consistent with intrinsic generation of the action sequence, and congruent with the fact that the display does not provide explicit feedback related to performance. A similar real-world example would be manually inputting a long password into a secure online application. In this case, one intrinsically generates the sequence from memory and receives similar feedback about the password sequence position (also provided as asterisks), which is typically ignored by the user. The minimal participant engagement with the visual task display observed in this study highlights another important point – that the behavior in explicit sequence learning motor tasks is highly generative in nature rather than reactive to stimulus cues as in the serial reaction time task (SRTT).  This is a crucial difference that must be carefully considered when designing investigations and comparing findings across studies.

We observed that initial keypress decoding accuracy was predominantly driven by contralateral primary sensorimotor cortex in the initial practice trials before transitioning to bilateral frontoparietal regions by trials 11 or 12 as performance gains plateaued.  The contribution of contralateral primary sensorimotor areas to early skill learning has been extensively reported in humans and non-human animals. 1,3-5  Similarly, the increased involvement of bilateral frontal and parietal regions to decoding during early skill learning in the non-dominant hand is well known.  Enhanced bilateral activation in both frontal and parietal cortex during skill learning has been extensively reported6-11, and appears to be even more prominent during early fine motor skill learning in the non-dominant hand12,13.  The frontal regions identified in these studies are known to play crucial roles in executive control14, motor planning15, and working memory6,8,16-18 processes, while the same parietal regions are known to integrate multimodal sensory feedback and support visuomotor transformations6,8,16-18, in addition to working memory19. Thus, it is not surprising that these regions increasingly contribute to decoding as subjects internalize the sequential task.  We now include a statement reflecting these considerations in the revised Discussion.

A somewhat related point is this: when combining voxel and parcel space, a concern is whether a degree of circularity may have contributed to the improved accuracy of the combined data, because it seems to use the same MEG signals twice - the voxels most contributing are also those contributing most to a parcel being identified as relevant, as parcels reflect the average of voxels within a boundary. In this context, I struggled to understand the explanation given, ie that the improved accuracy of the hybrid model may be due to "lower spatially resolved whole-brain and higher spatially resolved regional activity patterns".

We strongly disagree with the Reviewer’s assertion that the construction of the hybrid-space decoder is circular. To clarify, the base feature set for the hybrid-space decoder constructed for all participants includes whole-brain spatial patterns of MEG source activity averaged within parcels. As stated in the manuscript, these 148 inter-parcel features reflect “lower spatially resolved whole-brain activity patterns” or global brain dynamics. We then independently test how well spatial patterns of MEG source activity for all voxels distributed within individual parcels can decode keypress actions. Again, the testing of these intra-parcel spatial patterns, intended to capture “higher spatially resolved regional brain activity patterns”, is completely independent from one another and independent from the weighting of individual inter-parcel features. These intra-parcel features could, for example, provide additional information about muscle activation patterns or the task environment. These approximately 1150 intra-parcel voxels (on average, within the total number varying between subjects) are then combined with the 148 inter-parcel features to construct the final hybrid-space decoder. In fact, this varied spatial filter approach shares some similarities to the construction of convolutional neural networks (CNNs) used to perform object recognition in image classification applications. One could also view this hybrid-space decoding approach as a spatial analogue to common time-frequency based analyses such as theta-gamma phase amplitude coupling (PAC), which combine information from two or more narrow-band spectral features derived from the same time-series data.

We directly tested this hypothesis – that spatially overlapping intra- and inter-parcel features portray different information – by constructing an alternative hybrid-space decoder (HybridAlt) that excluded average inter-parcel features which spatially overlapped with intra-parcel voxel features, and comparing the performance to the decoder used in the manuscript (HybridOrig). The prediction was that if the overlapping parcel contained similar information to the more spatially resolved voxel patterns, then removing the parcel features (n=8) from the decoding analysis should not impact performance. In fact, despite making up less than 1% of the overall input feature space, removing those parcels resulted in a significant drop in overall performance greater than 2% (78.15% ± SD 7.03% for HybridOrig vs. 75.49% ± SD 7.17% for HybridAlt; Wilcoxon signed rank test, z = 3.7410, p = 1.8326e-04) (Author response image 2).

Author response image 2.

Comparison of decoding performances with two different hybrid approaches. HybridAlt: Intra-parcel voxel-space features of top ranked parcels and inter-parcel features of remaining parcels. HybridOrig:  Voxel-space features of top ranked parcels and whole-brain parcel-space features (i.e. – the version used in the manuscript). Dots represent decoding accuracy for individual subjects. Dashed lines indicate the trend in performance change across participants. Note, that HybridOrig (the approach used in our manuscript) significantly outperforms the HybridAlt approach, indicating that the excluded parcel features provide unique information compared to the spatially overlapping intra-parcel voxel patterns.

Firstly, there will be a relatively high degree of spatial contiguity among voxels because of the nature of the signal measured, i.e. nearby individual voxels are unlikely to be independent. Secondly, the voxel data gives a somewhat misleading sense of precision; the inversion can be set up to give an estimate for each voxel, but there will not just be dependence among adjacent voxels, but also substantial variation in the sensitivity and confidence with which activity can be projected to different parts of the brain. Midline and deeper structures come to mind, where the inversion will be more problematic than for regions along the dorsal convexity of the brain, and a concern is that in those midline structures, the highest decoding accuracy is seen. 

We definitely agree with the Reviewer that some inter-parcel features representing neighboring (or spatially contiguous) voxels are likely to be correlated. This has been well documented in the MEG literature20,21 and is a particularly important confound to address in functional or effective connectivity analyses (not performed in the present study). In the present analysis, any correlation between adjacent voxels presents a multi-collinearity problem, which effectively reduces the dimensionality of the input feature space. However, as long as there are multiple groups of correlated voxels within each parcel (i.e. - the effective dimensionality is still greater than 1), the intra-parcel spatial patterns could still meaningfully contribute to the decoder performance. Two specific results support this assertion.

First, we obtained higher decoding accuracy with voxel-space features [74.51% (± SD 7.34%)] compared to parcel space features [68.77% (± SD 7.6%)] (Figure 3B), indicating individual voxels carry more information in decoding the keypresses than the averaged voxel-space features or parcel-space features.  Second, Individual voxels within a parcel showed varying feature importance scores in decoding keypresses (Author response image 3). This finding supports the Reviewer’s assertion that neighboring voxels express similar information, but also shows that the correlated voxels form mini subclusters that are much smaller spatially than the parcel they reside in.

Author response image 3.

Feature importance score of individual voxels in decoding keypresses: MRMR was used to rank the individual voxel space features in decoding keypresses and the min-max normalized MRMR score was mapped to a structural brain surface. Note that individual voxels within a parcel showed different contribution to decoding.

Some of these concerns could be addressed by recording head movement (with enough precision) to regress out these contributions. The authors state that head movement was monitored with 3 fiducials, and their time courses ought to provide a way to deal with this issue. The ICA procedure may not have sufficiently dealt with removing movement-related problems, but one could eg relate individual components that were identified to the keypresses as another means for checking. An alternative could be to focus on frequency ranges above the movement frequencies. The accuracy for those still seems impressive and may provide a slightly more biologically plausible assessment. 

We have already addressed the issue of movement related artefacts in the first response above. With respect to a focus on frequency ranges above movement frequencies, the Reviewer states the “accuracy for those still seems impressive and may provide a slightly more biologically plausible assessment”. First, it is important to note that cortical delta-band oscillations measured with local field potentials (LFPs) in macaques is known to contain important information related to end-effector kinematics22,23 muscle activation patterns24 and temporal sequencing25 during skilled reaching and grasping actions. Thus, there is a substantial body of evidence that low-frequency neural oscillatory activity in this range contains important information about the skill learning behavior investigated in the present study. Second, our own data shows (which the Reviewer also points out) that significant information related to the skill learning behavior is also present in higher frequency bands (see Figure 2A and Figure 3—figure supplement 1). As we pointed out in our earlier response to questions about the hybrid space decoder architecture (see above), it is likely that different, yet complimentary, information is encoded across different temporal frequencies (just as it is encoded across different spatial frequencies). Again, this interpretation is supported by our data as the highest performing classifiers in all cases (when holding all parameters constant) were always constructed from broadband input MEG data (Figure 2A and Figure 3—figure supplement 1).  

One question concerns the interpretation of the results shown in Figure 4. They imply that during the course of learning, entirely different brain networks underpin the behaviour. Not only that, but they also include regions that would seem rather unexpected to be key nodes for learning and expressing relatively simple finger sequences, such as here. What then is the biological plausibility of these results? The authors seem to circumnavigate this issue by moving into a distance metric that captures the (neural network) changes over the course of learning, but the discussion seems detached from which regions are actually involved; or they offer a rather broad discussion of the anatomical regions identified here, eg in the context of LFOs, where they merely refer to "frontoparietal regions". 

The Reviewer notes the shift in brain networks driving keypress decoding performance between trials 1, 11 and 36 as shown in Figure 4A. The Reviewer questions whether these substantial shifts in brain network states underpinning the skill are biologically plausible, as well as the likelihood that bilateral superior and middle frontal and parietal cortex are important nodes within these networks.

First, previous fMRI work in humans performing a similar sequence learning task showed that flexibility in brain network composition (i.e. – changes in brain region members displaying coordinated activity) is up-regulated in novel learning environments and explains differences in learning rates across individuals26.  This work supports our interpretation of the present study data, that brain networks engaged in sequential motor skills rapidly reconfigure during early learning.

Second, frontoparietal network activity is known to support motor memory encoding during early learning27,28. For example, reactivation events in the posterior parietal29 and medial prefrontal30,31 cortex (MPFC) have been temporally linked to hippocampal replay, and are posited to support memory consolidation across several memory domains32, including motor sequence learning1,33,34.  Further, synchronized interactions between MPFC and hippocampus are more prominent during early learning as opposed to later stages27,35,36, perhaps reflecting “redistribution of hippocampal memories to MPFC” 27.  MPFC contributes to very early memory formation by learning association between contexts, locations, events and adaptive responses during rapid learning37. Consistently, coupling between hippocampus and MPFC has been shown during, and importantly immediately following (rest) initial memory encoding38,39.  Importantly, MPFC activity during initial memory encoding predicts subsequent recall40. Thus, the spatial map required to encode a motor sequence memory may be “built under the supervision of the prefrontal cortex” 28, also engaged in the development of an abstract representation of the sequence41.  In more abstract terms, the prefrontal, premotor and parietal cortices support novice performance “by deploying attentional and control processes” 42-44 required during early learning42-44. The dorsolateral prefrontal cortex DLPFC specifically is thought to engage in goal selection and sequence monitoring during early skill practice45, all consistent with the schema model of declarative memory in which prefrontal cortices play an important role in encoding46,47.  Thus, several prefrontal and frontoparietal regions contributing to long term learning 48 are also engaged in early stages of encoding. Altogether, there is strong biological support for the involvement of bilateral prefrontal and frontoparietal regions to decoding during early skill learning.  We now address this issue in the revised manuscript.

If I understand correctly, the offline neural representation analysis is in essence the comparison of the last keypress vs the first keypress of the next sequence. In that sense, the activity during offline rest periods is actually not considered. This makes the nomenclature somewhat confusing. While it matches the behavioural analysis, having only key presses one can't do it in any other way, but here the authors actually do have recordings of brain activity during offline rest. So at the very least calling it offline neural representation is misleading to this reviewer because what is compared is activity during the last and during the next keypress, not activity during offline periods. But it also seems a missed opportunity - the authors argue that most of the relevant learning occurs during offline rest periods, yet there is no attempt to actually test whether activity during this period can be useful for the questions at hand here. 

We agree with the Reviewer that our previous “offline neural representation” nomenclature could be misinterpreted. In the revised manuscript we refer to this difference as the “offline neural representational change”. Please, note that our previous work did link offline neural activity (i.e. – 16-22 Hz beta power and neural replay density during inter-practice rest periods) to observed micro-offline gains49.

Reviewer #2 (Public review): 

Summary 

Dash et al. asked whether and how the neural representation of individual finger movements is "contextualized" within a trained sequence during the very early period of sequential skill learning by using decoding of MEG signal. Specifically, they assessed whether/how the same finger presses (pressing index finger) embedded in the different ordinal positions of a practiced sequence (4-1-3-2-4; here, the numbers 1 through 4 correspond to the little through the index fingers of the non-dominant left hand) change their representation (MEG feature). They did this by computing either the decoding accuracy of the index finger at the ordinal positions 1 vs. 5 (index_OP1 vs index_OP5) or pattern distance between index_OP1 vs. index_OP5 at each training trial and found that both the decoding accuracy and the pattern distance progressively increase over the course of learning trials. More interestingly, they also computed the pattern distance for index_OP5 for the last execution of a practice trial vs. index_OP1 for the first execution in the next practice trial (i.e., across the rest period). This "off-line" distance was significantly larger than the "on-line" distance, which was computed within practice trials and predicted micro-offline skill gain. Based on these results, the authors conclude that the differentiation of representation for the identical movement embedded in different positions of a sequential skill ("contextualization") primarily occurs during early skill learning, especially during rest, consistent with the recent theory of the "micro-offline learning" proposed by the authors' group. I think this is an important and timely topic for the field of motor learning and beyond. <br /> Strengths 

The specific strengths of the current work are as follows. First, the use of temporally rich neural information (MEG signal) has a large advantage over previous studies testing sequential representations using fMRI. This allowed the authors to examine the earliest period (= the first few minutes of training) of skill learning with finer temporal resolution. Second, through the optimization of MEG feature extraction, the current study achieved extremely high decoding accuracy (approx. 94%) compared to previous works. As claimed by the authors, this is one of the strengths of the paper (but see my comments). Third, although some potential refinement might be needed, comparing "online" and "offline" pattern distance is a neat idea. 

Weaknesses 

Along with the strengths I raised above, the paper has some weaknesses. First, the pursuit of high decoding accuracy, especially the choice of time points and window length (i.e., 200 msec window starting from 0 msec from key press onset), casts a shadow on the interpretation of the main result. Currently, it is unclear whether the decoding results simply reflect behavioral change or true underlying neural change. As shown in the behavioral data, the key press speed reached 3~4 presses per second already at around the end of the early learning period (11th trial), which means inter-press intervals become as short as 250-330 msec. Thus, in almost more than 60% of training period data, the time window for MEG feature extraction (200 msec) spans around 60% of the inter-press intervals. Considering that the preparation/cueing of subsequent presses starts ahead of the actual press (e.g., Kornysheva et al., 2019) and/or potential online planning (e.g., Ariani and Diedrichsen, 2019), the decoder likely has captured these future press information as well as the signal related to the current key press, independent of the formation of genuine sequential representation (e.g., "contextualization" of individual press). This may also explain the gradual increase in decoding accuracy or pattern distance between index_OP1 vs. index_OP5 (Figure 4C and 5A), which co-occurred with performance improvement, as shorter inter-press intervals are more favorable for the dissociating the two index finger presses followed by different finger presses. The compromised decoding accuracies for the control sequences can be explained in similar logic. Therefore, more careful consideration and elaborated discussion seem necessary when trying to both achieve high-performance decoding and assess early skill learning, as it can impact all the subsequent analyses.

The Reviewer raises the possibility that (given the windowing parameters used in the present study) an increase in “contextualization” with learning could simply reflect faster typing speeds as opposed to an actual change in the underlying neural representation. The issue can essentially be framed as a mixing problem. As correct sequences are generated at higher and higher speeds over training, MEG activity patterns related to the planning, execution, evaluation and memory of individual keypresses overlap more in time. Thus, increased overlap between the “4” and “1” keypresses (at the start of the sequence) and “2” and “4” keypresses (at the end of the sequence) could artefactually increase contextualization distances even if the underlying neural representations for the individual keypresses remain unchanged (assuming this mixing of representations is used by the classifier to differentially tag each index finger press). If this were the case, it follows that such mixing effects reflecting the ordinal sequence structure would also be observable in the distribution of decoder misclassifications. For example, “4” keypresses would be more likely to be misclassified as “1” or “2” keypresses (or vice versa) than as “3” keypresses. The confusion matrices presented in Figures 3C and 4B and Figure 3—figure supplement 3A in the previously submitted manuscript do not show this trend in the distribution of misclassifications across the four fingers.

Moreover, if the representation distance is largely driven by this mixing effect, it’s also possible that the increased overlap between consecutive index finger keypresses during the 4-4 transition marking the end of one sequence and the beginning of the next one could actually mask contextualization-related changes to the underlying neural representations and make them harder to detect. In this case, a decoder tasked with separating individual index finger keypresses into two distinct classes based upon sequence position might show decreased performance with learning as adjacent keypresses overlapped in time with each other to an increasing extent. However, Figure 4C in our previously submitted manuscript does not support this possibility, as the 2-class hybrid classifier displays improved classification performance over early practice trials despite greater temporal overlap.

We also conducted a new multivariate regression analysis to directly assess whether the neural representation distance score could be predicted by the 4-1, 2-4 and 4-4 keypress transition times observed for each complete correct sequence (both predictor and response variables were z-score normalized within-subject). The results of this analysis affirmed that the possible alternative explanation put forward by the Reviewer is not supported by our data (Adjusted R2 = 0.00431; F = 5.62). We now include this new negative control analysis result in the revised manuscript.

Overall, we do strongly agree with the Reviewer that the naturalistic, self-paced, generative task employed in the present study results in overlapping brain processes related to planning, execution, evaluation and memory of the action sequence. We also agree that there are several tradeoffs to consider in the construction of the classifiers depending on the study aim. Given our aim of optimizing keypress decoder accuracy in the present study, the set of trade-offs resulted in representations reflecting more the latter three processes, and less so the planning component. Whether separate decoders can be constructed to tease apart the representations or networks supporting these overlapping processes is an important future direction of research in this area. For example, work presently underway in our lab constrains the selection of windowing parameters in a manner that allows individual classifiers to be temporally linked to specific planning, execution, evaluation or memory-related processes to discern which brain networks are involved and how they adaptively reorganize with learning. Results from the present study (Figure 4—figure supplement 2) showing hybrid-space decoder prediction accuracies exceeding 74% for temporal windows spanning as little as 25ms and located up to 100ms prior to the keyDown event strongly support the feasibility of such an approach.

Related to the above point, testing only one particular sequence (4-1-3-2-4), aside from the control ones, limits the generalizability of the finding. This also may have contributed to the extremely high decoding accuracy reported in the current study. 

The Reviewer raises a question about the generalizability of the decoder accuracy reported in our study. Fortunately, a comparison between decoder performances on Day 1 and Day 2 datasets does provide some insight into this issue. As the Reviewer points out, the classifiers in this study were trained and tested on keypresses performed while practicing a specific sequence (4-1-3-2-4). The study was designed this way as to avoid the impact of interference effects on learning dynamics. The cross-validated performance of classifiers on MEG data collected within the same session was 90.47% overall accuracy (4-class; Figure 3C). We then tested classifier performance on data collected during a separate MEG session conducted approximately 24 hours later (Day 2; see Figure 3—supplement 3). We observed a reduction in overall accuracy rate to 87.11% when tested on MEG data recorded while participants performed the same learned sequence, and 79.44% when they performed several previously unpracticed sequences. Both changes in accuracy are important with regards to the generalizability of our findings. First, 87.11% performance accuracy for the trained sequence data on Day 2 (a reduction of only 3.36%) indicates that the hybrid-space decoder performance is robust over multiple MEG sessions, and thus, robust to variations in SNR across the MEG sensor array caused by small differences in head position between scans.  This indicates a substantial advantage over sensor-space decoding approaches. Furthermore, when tested on data from unpracticed sequences, overall performance dropped an additional 7.67%. This difference reflects the performance bias of the classifier for the trained sequence, possibly caused by high-order sequence structure being incorporated into the feature weights. In the future, it will be important to understand in more detail how random or repeated keypress sequence training data impacts overall decoder performance and generalization. We strongly agree with the Reviewer that the issue of generalizability is extremely important and have added a new paragraph to the Discussion in the revised manuscript highlighting the strengths and weaknesses of our study with respect to this issue.

In terms of clinical BCI, one of the potential relevance of the study, as claimed by the authors, it is not clear that the specific time window chosen in the current study (up to 200 msec since key press onset) is really useful. In most cases, clinical BCI would target neural signals with no overt movement execution due to patients' inability to move (e.g., Hochberg et al., 2012). Given the time window, the surprisingly high performance of the current decoder may result from sensory feedback and/or planning of subsequent movement, which may not always be available in the clinical BCI context. Of course, the decoding accuracy is still much higher than chance even when using signal before the key press (as shown in Figure 4 Supplement 2), but it is not immediately clear to me that the authors relate their high decoding accuracy based on post-movement signal to clinical BCI settings.

The Reviewer questions the relevance of the specific window parameters used in the present study for clinical BCI applications, particularly for paretic patients who are unable to produce finger movements or for whom afferent sensory feedback is no longer intact. We strongly agree with the Reviewer that any intended clinical application must carefully consider these specific input feature constraints dictated by the clinical cohort, and in turn impose appropriate and complimentary constraints on classifier parameters that may differ from the ones used in the present study.  We now highlight this issue in the Discussion of the revised manuscript and relate our present findings to published clinical BCI work within this context.

One of the important and fascinating claims of the current study is that the "contextualization" of individual finger movements in a trained sequence specifically occurs during short rest periods in very early skill learning, echoing the recent theory of micro-offline learning proposed by the authors' group. Here, I think two points need to be clarified. First, the concept of "contextualization" is kept somewhat blurry throughout the text. It is only at the later part of the Discussion (around line #330 on page 13) that some potential mechanism for the "contextualization" is provided as "what-and-where" binding. Still, it is unclear what "contextualization" actually is in the current data, as the MEG signal analyzed is extracted from 0-200 msec after the keypress. If one thinks something is contextualizing an action, that contextualization should come earlier than the action itself. 

The Reviewer requests that we: 1) more clearly define our use of the term “contextualization” and 2) provide the rationale for assessing it over a 200ms window aligned to the keyDown event. This choice of window parameters means that the MEG activity used in our analysis was coincident with, rather than preceding, the actual keypresses.  We define contextualization as the differentiation of representation for the identical movement embedded in different positions of a sequential skill. That is, representations of individual action elements progressively incorporate information about their relationship to the overall sequence structure as the skill is learned. We agree with the Reviewer that this can be appropriately interpreted as “what-and-where” binding. We now incorporate this definition in the Introduction of the revised manuscript as requested.

The window parameters for optimizing accurate decoding individual finger movements were determined using a grid search of the parameter space (a sliding window of variable width between 25-350 ms with 25 ms increments variably aligned from 0 to +100ms with 10ms increments relative to the keyDown event). This approach generated 140 different temporal windows for each keypress for each participant, with the final parameter selection determined through comparison of the resulting performance between each decoder.  Importantly, the decision to optimize for decoding accuracy placed an emphasis on keypress representations characterized by the most consistent and robust features shared across subjects, which in turn maximize statistical power in detecting common learning-related changes. In this case, the optimal window encompassed a 200ms epoch aligned to the keyDown event (t0 = 0 ms).  We then asked if the representations (i.e. – spatial patterns of combined parcel- and voxel-space activity) of the same digit at two different sequence positions changed with practice within this optimal decoding window.  Of course, our findings do not rule out the possibility that contextualization can also be found before or even after this time window, as we did not directly address this issue in the present study.  Ongoing work in our lab, as pointed out above, is investigating contextualization within different time windows tailored specifically for assessing sequence skill action planning, execution, evaluation and memory processes.

The second point is that the result provided by the authors is not yet convincing enough to support the claim that "contextualization" occurs during rest. In the original analysis, the authors presented the statistical significance regarding the correlation between the "offline" pattern differentiation and micro-offline skill gain (Figure 5. Supplement 1), as well as the larger "offline" distance than "online" distance (Figure 5B). However, this analysis looks like regressing two variables (monotonically) increasing as a function of the trial. Although some information in this analysis, such as what the independent/dependent variables were or how individual subjects were treated, was missing in the Methods, getting a statistically significant slope seems unsurprising in such a situation. Also, curiously, the same quantitative evidence was not provided for its "online" counterpart, and the authors only briefly mentioned in the text that there was no significant correlation between them. It may be true looking at the data in Figure 5A as the online representation distance looks less monotonically changing, but the classification accuracy presented in Figure 4C, which should reflect similar representational distance, shows a more monotonic increase up to the 11th trial. Further, the ways the "online" and "offline" representation distance was estimated seem to make them not directly comparable. While the "online" distance was computed using all the correct press data within each 10 sec of execution, the "offline" distance is basically computed by only two presses (i.e., the last index_OP5 vs. the first index_OP1 separated by 10 sec of rest). Theoretically, the distance between the neural activity patterns for temporally closer events tends to be closer than that between the patterns for temporally far-apart events. It would be fairer to use the distance between the first index_OP1 vs. the last index_OP5 within an execution period for "online" distance, as well. 

The Reviewer suggests that the current data is not convincing enough to show that contextualization occurs during rest and raises two important concerns: 1) the relationship between online contextualization and micro-online gains is not shown, and 2) the online distance was calculated differently from its offline counterpart (i.e. - instead of calculating the distance between last IndexOP5 and first IndexOP1 from a single trial, the distance was calculated for each sequence within a trial and then averaged).

We addressed the first concern by performing individual subject correlations between 1) contextualization changes during rest intervals and micro-offline gains; 2) contextualization changes during practice trials and micro-online gains, and 3) contextualization changes during practice trials and micro-offline gains (Author response image 4). We then statistically compared the resulting correlation coefficient distributions and found that within-subject correlations for contextualization changes during rest intervals and micro-offline gains were significantly higher than online contextualization and micro-online gains (t = 3.2827, p = 0.0015) and online contextualization and micro-offline gains (t = 3.7021, p = 5.3013e-04). These results are consistent with our interpretation that micro-offline gains are supported by contextualization changes during the inter-practice rest period.

Author response image 4.

Distribution of individual subject correlation coefficients between contextualization changes occurring during practice or rest with  micro-online and micro-offline performance gains. Note that, the correlation distributions were significantly higher for the relationship between contextualization changes during rest and micro-offline gains than for contextualization changes during practice and either micro-online or offline gain.

With respect to the second concern highlighted above, we agree with the Reviewer that one limitation of the analysis comparing online versus offline changes in contextualization as presented in the reviewed manuscript, is that it does not eliminate the possibility that any differences could simply be explained by the passage of time (which is smaller for the online analysis compared to the offline analysis). The Reviewer suggests an approach that addresses this issue, which we have now carried out.   When quantifying online changes in contextualization from the first IndexOP1 the last IndexOP5 keypress in the same trial we observed no learning-related trend (Author response image 5, right panel). Importantly, offline distances were significantly larger than online distances regardless of the measurement approach and neither predicted online learning (Author response image 6).

Author response image 5.

Trial by trial trend of offline (left panel) and online (middle and right panels) changes in contextualization. Offline changes in contextualization were assessed by calculating the distance between neural representations for the last IndexOP5 keypress in the previous trial and the first IndexOP1 keypress in the present trial. Two different approaches were used to characterize online contextualization changes. The analysis included in the reviewed manuscript (middle panel) calculated the distance between IndexOP1 and IndexOP5 for each correct sequence, which was then averaged across the trial. This approach is limited by the lack of control for the passage of time when making online versus offline comparisons. Thus, the second approach controlled for the passage of time by calculating distance between the representations associated with the first IndexOP1 keypress and the last IndexOP5 keypress within the same trial. Note that while the first approach showed an increase online contextualization trend with practice, the second approach did not.

Author response image 6.

Relationship between online contextualization and online learning is shown for both within-sequence (left; note that this is the online contextualization measure used in the reviewd manuscript) and across-sequence (right) distance calculation. There was no significant relationship between online learning and online contextualization regardless of the measurement approach.

A related concern regarding the control analysis, where individual values for max speed and the degree of online contextualization were compared (Figure 5 Supplement 3), is whether the individual difference is meaningful. If I understood correctly, the optimization of the decoding process (temporal window, feature inclusion/reduction, decoder, etc.) was performed for individual participants, and the same feature extraction was also employed for the analysis of representation distance (i.e., contextualization). If this is the case, the distances are individually differently calculated and they may need to be normalized relative to some stable reference (e.g., 1 vs. 4 or average distance within the control sequence presses) before comparison across the individuals. 

The Reviewer makes a good point here. We have now implemented the suggested normalization procedure in the analysis provided in the revised manuscript.

Reviewer #3 (Public review): 

Summary: 

One goal of this paper is to introduce a new approach for highly accurate decoding of finger movements from human magnetoencephalography data via dimension reduction of a "multi-scale, hybrid" feature space. Following this decoding approach, the authors aim to show that early skill learning involves "contextualization" of the neural coding of individual movements, relative to their position in a sequence of consecutive movements. Furthermore, they aim to show that this "contextualization" develops primarily during short rest periods interspersed with skill training and correlates with a performance metric which the authors interpret as an indicator of offline learning. <br /> Strengths: 

A clear strength of the paper is the innovative decoding approach, which achieves impressive decoding accuracies via dimension reduction of a "multi-scale, hybrid space". This hybrid-space approach follows the neurobiologically plausible idea of the concurrent distribution of neural coding across local circuits as well as large-scale networks. A further strength of the study is the large number of tested dimension reduction techniques and classifiers (though the manuscript reveals little about the comparison of the latter). 

We appreciate the Reviewer’s comments regarding the paper’s strengths.

A simple control analysis based on shuffled class labels could lend further support to this complex decoding approach. As a control analysis that completely rules out any source of overfitting, the authors could test the decoder after shuffling class labels. Following such shuffling, decoding accuracies should drop to chance level for all decoding approaches, including the optimized decoder. This would also provide an estimate of actual chance-level performance (which is informative over and beyond the theoretical chance level). Furthermore, currently, the manuscript does not explain the huge drop in decoding accuracies for the voxel-space decoding (Figure 3B). Finally, the authors' approach to cortical parcellation raises questions regarding the information carried by varying dipole orientations within a parcel (which currently seems to be ignored?) and the implementation of the mean-flipping method (given that there are two dimensions - space and time - what do the authors refer to when they talk about the sign of the "average source", line 477?). 

The Reviewer recommends that we: 1) conduct an additional control analysis on classifier performance using shuffled class labels, 2) provide a more detailed explanation regarding the drop in decoding accuracies for the voxel-space decoding following LDA dimensionality reduction (see Fig 3B), and 3) provide additional details on how problems related to dipole solution orientations were addressed in the present study.  

In relation to the first point, we have now implemented a random shuffling approach as a control for the classification analyses. The results of this analysis indicated that the chance level accuracy was 22.12% (± SD 9.1%) for individual keypress decoding (4-class classification), and 18.41% (± SD 7.4%) for individual sequence item decoding (5-class classification), irrespective of the input feature set or the type of decoder used. Thus, the decoding accuracy observed with the final model was substantially higher than these chance levels.  

Second, please note that the dimensionality of the voxel-space feature set is very high (i.e. – 15684). LDA attempts to map the input features onto a much smaller dimensional space (number of classes-1; e.g. –  3 dimensions, for 4-class keypress decoding). Given the very high dimension of the voxel-space input features in this case, the resulting mapping exhibits reduced accuracy. Despite this general consideration, please refer to Figure 3—figure supplement 3, where we observe improvement in voxel-space decoder performance when utilizing alternative dimensionality reduction techniques.

The decoders constructed in the present study assess the average spatial patterns across time (as defined by the windowing procedure) in the input feature space.  We now provide additional details in the Methods of the revised manuscript pertaining to the parcellation procedure and how the sign ambiguity problem was addressed in our analysis.

Weaknesses: 

A clear weakness of the paper lies in the authors' conclusions regarding "contextualization". Several potential confounds, described below, question the neurobiological implications proposed by the authors and provide a simpler explanation of the results. Furthermore, the paper follows the assumption that short breaks result in offline skill learning, while recent evidence, described below, casts doubt on this assumption. 

We thank the Reviewer for giving us the opportunity to address these issues in detail (see below).

The authors interpret the ordinal position information captured by their decoding approach as a reflection of neural coding dedicated to the local context of a movement (Figure 4). One way to dissociate ordinal position information from information about the moving effectors is to train a classifier on one sequence and test the classifier on other sequences that require the same movements, but in different positions50. In the present study, however, participants trained to repeat a single sequence (4-1-3-2-4). As a result, ordinal position information is potentially confounded by the fixed finger transitions around each of the two critical positions (first and fifth press). Across consecutive correct sequences, the first keypress in a given sequence was always preceded by a movement of the index finger (=last movement of the preceding sequence), and followed by a little finger movement. The last keypress, on the other hand, was always preceded by a ring finger movement, and followed by an index finger movement (=first movement of the next sequence). Figure 4 - Supplement 2 shows that finger identity can be decoded with high accuracy (>70%) across a large time window around the time of the key press, up to at least +/-100 ms (and likely beyond, given that decoding accuracy is still high at the boundaries of the window depicted in that figure). This time window approaches the keypress transition times in this study. Given that distinct finger transitions characterized the first and fifth keypress, the classifier could thus rely on persistent (or "lingering") information from the preceding finger movement, and/or "preparatory" information about the subsequent finger movement, in order to dissociate the first and fifth keypress. Currently, the manuscript provides no evidence that the context information captured by the decoding approach is more than a by-product of temporally extended, and therefore overlapping, but independent neural representations of consecutive keypresses that are executed in close temporal proximity - rather than a neural representation dedicated to context. 

Such temporal overlap of consecutive, independent finger representations may also account for the dynamics of "ordinal coding"/"contextualization", i.e., the increase in 2-class decoding accuracy, across Day 1 (Figure 4C). As learning progresses, both tapping speed and the consistency of keypress transition times increase (Figure 1), i.e., consecutive keypresses are closer in time, and more consistently so. As a result, information related to a given keypress is increasingly overlapping in time with information related to the preceding and subsequent keypresses. The authors seem to argue that their regression analysis in Figure 5 - Figure Supplement 3 speaks against any influence of tapping speed on "ordinal coding" (even though that argument is not made explicitly in the manuscript). However, Figure 5 - Figure Supplement 3 shows inter-individual differences in a between-subject analysis (across trials, as in panel A, or separately for each trial, as in panel B), and, therefore, says little about the within-subject dynamics of "ordinal coding" across the experiment. A regression of trial-by-trial "ordinal coding" on trial-by-trial tapping speed (either within-subject or at a group-level, after averaging across subjects) could address this issue. Given the highly similar dynamics of "ordinal coding" on the one hand (Figure 4C), and tapping speed on the other hand (Figure 1B), I would expect a strong relationship between the two in the suggested within-subject (or group-level) regression. Furthermore, learning should increase the number of (consecutively) correct sequences, and, thus, the consistency of finger transitions. Therefore, the increase in 2-class decoding accuracy may simply reflect an increasing overlap in time of increasingly consistent information from consecutive keypresses, which allows the classifier to dissociate the first and fifth keypress more reliably as learning progresses, simply based on the characteristic finger transitions associated with each. In other words, given that the physical context of a given keypress changes as learning progresses - keypresses move closer together in time and are more consistently correct - it seems problematic to conclude that the mental representation of that context changes. To draw that conclusion, the physical context should remain stable (or any changes to the physical context should be controlled for). 

The issues raised by Reviewer #3 here are similar to two issues raised by Reviewer #2 above and agree they must both be carefully considered in any evaluation of our findings.

As both Reviewers pointed out, the classifiers in this study were trained and tested on keypresses performed while practicing a specific sequence (4-1-3-2-4). The study was designed this way as to avoid the impact of interference effects on learning dynamics. The cross-validated performance of classifiers on MEG data collected within the same session was 90.47% overall accuracy (4-class; Figure 3C). We then tested classifier performance on data collected during a separate MEG session conducted approximately 24 hours later (Day 2; see Figure 3—supplement 3). We observed a reduction in overall accuracy rate to 87.11% when tested on MEG data recorded while participants performed the same learned sequence, and 79.44% when they performed several previously unpracticed sequences. This classification performance difference of 7.67% when tested on the Day 2 data could reflect the performance bias of the classifier for the trained sequence, possibly caused by mixed information from temporally close keypresses being incorporated into the feature weights.

Along these same lines, both Reviewers also raise the possibility that an increase in “ordinal coding/contextualization” with learning could simply reflect an increase in this mixing effect caused by faster typing speeds as opposed to an actual change in the underlying neural representation. The basic idea is that as correct sequences are generated at higher and higher speeds over training, MEG activity patterns related to the planning, execution, evaluation and memory of individual keypresses overlap more in time. Thus, increased overlap between the “4” and “1” keypresses (at the start of the sequence) and “2” and “4” keypresses (at the end of the sequence) could artefactually increase contextualization distances even if the underlying neural representations for the individual keypresses remain unchanged (assuming this mixing of representations is used by the classifier to differentially tag each index finger press). If this were the case, it follows that such mixing effects reflecting the ordinal sequence structure would also be observable in the distribution of decoder misclassifications. For example, “4” keypresses would be more likely to be misclassified as “1” or “2” keypresses (or vice versa) than as “3” keypresses. The confusion matrices presented in Figures 3C and 4B and Figure 3—figure supplement 3A in the previously submitted manuscript do not show this trend in the distribution of misclassifications across the four fingers.

Following this logic, it’s also possible that if the ordinal coding is largely driven by this mixing effect, the increased overlap between consecutive index finger keypresses during the 4-4 transition marking the end of one sequence and the beginning of the next one could actually mask contextualization-related changes to the underlying neural representations and make them harder to detect. In this case, a decoder tasked with separating individual index finger keypresses into two distinct classes based upon sequence position might show decreased performance with learning as adjacent keypresses overlapped in time with each other to an increasing extent. However, Figure 4C in our previously submitted manuscript does not support this possibility, as the 2-class hybrid classifier displays improved classification performance over early practice trials despite greater temporal overlap.

As noted in the above replay to Reviewer #2, we also conducted a new multivariate regression analysis to directly assess whether the neural representation distance score could be predicted by the 4-1, 2-4 and 4-4 keypress transition times observed for each complete correct sequence (both predictor and response variables were z-score normalized within-subject). The results of this analysis affirmed that the possible alternative explanation put forward by the Reviewer is not supported by our data (Adjusted R2 = 0.00431; F = 5.62). We now include this new negative control analysis result in the revised manuscript.

Finally, the Reviewer hints that one way to address this issue would be to compare MEG responses before and after learning for sequences typed at a fixed speed. However, given that the speed-accuracy trade-off should improve with learning, a comparison between unlearned and learned skill states would dictate that the skill be evaluated at a very low fixed speed. Essentially, such a design presents the problem that the post-training test is evaluating the representation in the unlearned behavioral state that is not representative of the acquired skill. Thus, this approach would not address our experimental question: “do neural representations of the same action performed at different locations within a skill sequence contextually differentiate or remain stable as learning evolves”.

A similar difference in physical context may explain why neural representation distances ("differentiation") differ between rest and practice (Figure 5). The authors define "offline differentiation" by comparing the hybrid space features of the last index finger movement of a trial (ordinal position 5) and the first index finger movement of the next trial (ordinal position 1). However, the latter is not only the first movement in the sequence but also the very first movement in that trial (at least in trials that started with a correct sequence), i.e., not preceded by any recent movement. In contrast, the last index finger of the last correct sequence in the preceding trial includes the characteristic finger transition from the fourth to the fifth movement. Thus, there is more overlapping information arising from the consistent, neighbouring keypresses for the last index finger movement, compared to the first index finger movement of the next trial. A strong difference (larger neural representation distance) between these two movements is, therefore, not surprising, given the task design, and this difference is also expected to increase with learning, given the increase in tapping speed, and the consequent stronger overlap in representations for consecutive keypresses. Furthermore, initiating a new sequence involves pre-planning, while ongoing practice relies on online planning (Ariani et al., eNeuro 2021), i.e., two mental operations that are dissociable at the level of neural representation (Ariani et al., bioRxiv 2023). 

The Reviewer argues that the comparison of last finger movement of a trial and the first in the next trial are performed in different circumstances and contexts. This is an important point and one we tend to agree with. For this task, the first sequence in a practice trial (which is pre-planned offline) is performed in a somewhat different context from the sequence iterations that follow, which involve temporally overlapping planning, execution and evaluation processes.  The Reviewer is particularly concerned about a difference in the temporal mixing effect issue raised above between the first and last keypresses performed in a trial. However, in contrast to the Reviewers stated argument above, findings from Korneysheva et. al (2019) showed that neural representations of individual actions are competitively queued during the pre-planning period in a manner that reflects the ordinal structure of the learned sequence.  Thus, mixing effects are likely still present for the first keypress in a trial. Also note that we now present new control analyses in multiple responses above confirming that hypothetical mixing effects between adjacent keypresses do not explain our reported contextualization finding. A statement addressing these possibilities raised by the Reviewer has been added to the Discussion in the revised manuscript.

In relation to pre-planning, ongoing MEG work in our lab is investigating contextualization within different time windows tailored specifically for assessing how sequence skill action planning evolves with learning.

Given these differences in the physical context and associated mental processes, it is not surprising that "offline differentiation", as defined here, is more pronounced than "online differentiation". For the latter, the authors compared movements that were better matched regarding the presence of consistent preceding and subsequent keypresses (online differentiation was defined as the mean difference between all first vs. last index finger movements during practice).  It is unclear why the authors did not follow a similar definition for "online differentiation" as for "micro-online gains" (and, indeed, a definition that is more consistent with their definition of "offline differentiation"), i.e., the difference between the first index finger movement of the first correct sequence during practice, and the last index finger of the last correct sequence. While these two movements are, again, not matched for the presence of neighbouring keypresses (see the argument above), this mismatch would at least be the same across "offline differentiation" and "online differentiation", so they would be more comparable. 

This is the same point made earlier by Reviewer #2, and we agree with this assessment. As stated in the response to Reviewer #2 above, we have now carried out quantification of online contextualization using this approach and included it in the revised manuscript. We thank the Reviewer for this suggestion.

A further complication in interpreting the results regarding "contextualization" stems from the visual feedback that participants received during the task. Each keypress generated an asterisk shown above the string on the screen, irrespective of whether the keypress was correct or incorrect. As a result, incorrect (e.g., additional, or missing) keypresses could shift the phase of the visual feedback string (of asterisks) relative to the ordinal position of the current movement in the sequence (e.g., the fifth movement in the sequence could coincide with the presentation of any asterisk in the string, from the first to the fifth). Given that more incorrect keypresses are expected at the start of the experiment, compared to later stages, the consistency in visual feedback position, relative to the ordinal position of the movement in the sequence, increased across the experiment. A better differentiation between the first and the fifth movement with learning could, therefore, simply reflect better decoding of the more consistent visual feedback, based either on the feedback-induced brain response, or feedback-induced eye movements (the study did not include eye tracking). It is not clear why the authors introduced this complicated visual feedback in their task, besides consistency with their previous studies.

We strongly agree with the Reviewer that eye movements related to task engagement are important to rule out as a potential driver of the decoding accuracy or contextualization effect. We address this issue above in response to a question raised by Reviewer #1 about the impact of movement related artefacts in general on our findings.

First, the assumption the Reviewer makes here about the distribution of errors in this task is incorrect. On average across subjects, 2.32% ± 1.48% (mean ± SD) of all keypresses performed were errors, which were evenly distributed across the four possible keypress responses. While errors increased progressively over practice trials, they did so in proportion to the increase in correct keypresses, so that the overall ratio of correct-to-incorrect keypresses remained stable over the training session. Thus, the Reviewer’s assumptions that there is a higher relative frequency of errors in early trials, and a resulting systematic trend phase shift differences between the visual display updates (i.e. – a change in asterisk position above the displayed sequence) and the keypress performed is not substantiated by the data. To the contrary, the asterisk position on the display and the keypress being executed remained highly correlated over the entire training session. We now include a statement about the frequency and distribution of errors in the revised manuscript.

Given this high correlation, we firmly agree with the Reviewer that the issue of eye movement-related artefacts is still an important one to address. Fortunately, we did collect eye movement data during the MEG recordings so were able to investigate this. As detailed in the response to Reviewer #1 above, we found that gaze positions and eye-movement velocity time-locked to visual display updates (i.e. – a change in asterisk position above the displayed sequence) did not reflect the asterisk location above chance levels (Overall cross-validated accuracy = 0.21817; see Author response image 1). Furthermore, an inspection of the eye position data revealed that a majority of participants on most trials displayed random walk gaze patterns around a center fixation point, indicating that participants did not attend to the asterisk position on the display. This is consistent with intrinsic generation of the action sequence, and congruent with the fact that the display does not provide explicit feedback related to performance. As pointed out above, a similar real-world example would be manually inputting a long password into a secure online application. In this case, one intrinsically generates the sequence from memory and receives similar feedback about the password sequence position (also provided as asterisks), which is typically ignored by the user. Notably, the minimal participant engagement with the visual task display observed in this study highlights an important difference between behavior observed during explicit sequence learning motor tasks (which is highly generative in nature) with reactive responses to stimulus cues in a serial reaction time task (SRTT).  This is a crucial difference that must be carefully considered when comparing findings across studies. All elements pertaining to this new control analysis are now included in the revised manuscript.

The authors report a significant correlation between "offline differentiation" and cumulative micro-offline gains. However, it would be more informative to correlate trial-by-trial changes in each of the two variables. This would address the question of whether there is a trial-by-trial relation between the degree of "contextualization" and the amount of micro-offline gains - are performance changes (micro-offline gains) less pronounced across rest periods for which the change in "contextualization" is relatively low? Furthermore, is the relationship between micro-offline gains and "offline differentiation" significantly stronger than the relationship between micro-offline gains and "online differentiation"? 

In response to a similar issue raised above by Reviewer #2, we now include new analyses comparing correlation magnitudes between (1) “online differention” vs micro-online gains, (2) “online differention” vs micro-offline gains and (3) “offline differentiation” and micro-offline gains (see Author response images 4, 5 and 6 above). These new analyses and results have been added to the revised manuscript. Once again, we thank both Reviewers for this suggestion.

The authors follow the assumption that micro-offline gains reflect offline learning.

This statement is incorrect. The original Bonstrup et al (2019) 49 paper clearly states that micro-offline gains must be carefully interpreted based upon the behavioral context within which they are observed, and lays out the conditions under which one can have confidence that micro-offline gains reflect offline learning.  In fact, the excellent meta-analysis of Pan & Rickard (2015) 51, which re-interprets the benefits of sleep in overnight skill consolidation from a “reactive inhibition” perspective, was a crucial resource in the experimental design of our initial study49, as well as in all our subsequent work. Pan & Rickard stated:

“Empirically, reactive inhibition refers to performance worsening that can accumulate during a period of continuous training (Hull, 1943). It tends to dissipate, at least in part, when brief breaks are inserted between blocks of training. If there are multiple performance-break cycles over a training session, as in the motor sequence literature, performance can exhibit a scalloped effect, worsening during each uninterrupted performance block but improving across blocks52,53. Rickard, Cai, Rieth, Jones, and Ard (2008) and Brawn, Fenn, Nusbaum, and Margoliash (2010) 52,53 demonstrated highly robust scalloped reactive inhibition effects using the commonly employed 30 s–30 s performance break cycle, as shown for Rickard et al.’s (2008) massed practice sleep group in Figure 2. The scalloped effect is evident for that group after the first few 30 s blocks of each session. The absence of the scalloped effect during the first few blocks of training in the massed group suggests that rapid learning during that period masks any reactive inhibition effect.”

Crucially, Pan & Rickard51 made several concrete recommendations for reducing the impact of the reactive inhibition confound on offline learning studies. One of these recommendations was to reduce practice times to 10s (most prior sequence learning studies up until that point had employed 30s long practice trials). They stated:

“The traditional design involving 30 s-30 s performance break cycles should be abandoned given the evidence that it results in a reactive inhibition confound, and alternative designs with reduced performance duration per block used instead 51. One promising possibility is to switch to 10 s performance durations for each performance-break cycle Instead 51. That design appears sufficient to eliminate at least the majority of the reactive inhibition effect 52,53.”

We mindfully incorporated recommendations from Pan and Rickard51  into our own study designs including 1) utilizing 10s practice trials and 2) constraining our analysis of micro-offline gains to early learning trials (where performance monotonically increases and 95% of overall performance gains occur), which are prior to the emergence of the “scalloped” performance dynamics that are strongly linked to reactive inhibition effects. 

However, there is no direct evidence in the literature that micro-offline gains really result from offline learning, i.e., an improvement in skill level.

We strongly disagree with the Reviewer’s assertion that “there is no direct evidence in the literature that micro-offline gains really result from offline learning, i.e., an improvement in skill level.”  The initial Bönstrup et al. (2019) 49 report was followed up by a large online crowd-sourcing study (Bönstrup et al., 2020) 54. This second (and much larger) study provided several additional important findings supporting our interpretation of micro-offline gains in cases where the important behavioral conditions clarified above were met (see Author response image 7 below for further details on these conditions).

Author response image 7.

Micro-offline gains observed in learning and non-learning contexts are attributed to different underlying causes. (A) Micro-offline and online changes relative to overall trial-by-trial learning. This figure is based on data from Bönstrup et al. (2019) 49. During early learning, micro-offline gains (red bars) closely track trial-by-trial performance gains (green line with open circle markers), with minimal contribution from micro-online gains (blue bars). The stated conclusion in Bönstrup et al. (2019) is that micro-offline gains only during this Early Learning stage reflect rapid memory consolidation (see also 54). After early learning, about practice trial 11, skill plateaus. This plateau skill period is characterized by a striking emergence of coupled (and relatively stable) micro-online drops and micro-offline increases. Bönstrup et al. (2019) as well as others in the literature 55-57, argue that micro-offline gains during the plateau period likely reflect recovery from inhibitory performance factors such as reactive inhibition or fatigue, and thus must be excluded from analyses relating micro-offline gains to skill learning.  The Non-repeating groups in Experiments 3 and 4 from Das et al. (2024) suffer from a lack of consideration of these known confounds.

Evidence documented in that paper54 showed that micro-offline gains during early skill learning were: 1) replicable and generalized to subjects learning the task in their daily living environment (n=389); 2) equivalent when significantly shortening practice period duration, thus confirming that they are not a result of recovery from performance fatigue (n=118);  3) reduced (along with learning rates) by retroactive interference applied immediately after each practice period relative to interference applied after passage of time (n=373), indicating stabilization of the motor memory at a microscale of several seconds consistent with rapid consolidation; and 4) not modified by random termination of the practice periods, ruling out a contribution of predictive motor slowing (N = 71) 54.  Altogether, our findings were strongly consistent with the interpretation that micro-offline gains reflect memory consolidation supporting early skill learning. This is precisely the portion of the learning curve Pan and Rickard51 refer to when they state “…rapid learning during that period masks any reactive inhibition effect”.

This interpretation is further supported by brain imaging evidence linking known memory-related networks and consolidation mechanisms to micro-offline gains. First, we reported that the density of fast hippocampo-neocortical skill memory replay events increases approximately three-fold during early learning inter-practice rest periods with the density explaining differences in the magnitude of micro-offline gains across subjects1. Second, Jacobacci et al. (2020) independently reproduced our original behavioral findings and reported BOLD fMRI changes in the hippocampus and precuneus (regions also identified in our MEG study1) linked to micro-offline gains during early skill learning. 33 These functional changes were coupled with rapid alterations in brain microstructure in the order of minutes, suggesting that the same network that operates during rest periods of early learning undergoes structural plasticity over several minutes following practice58. Third, even more recently, Chen et al. (2024) provided direct evidence from intracranial EEG in humans linking sharp-wave ripple events (which are known markers for neural replay59) in the hippocampus (80-120 Hz in humans) with micro-offline gains during early skill learning. The authors report that the strong increase in ripple rates tracked learning behavior, both across blocks and across participants. The authors conclude that hippocampal ripples during resting offline periods contribute to motor sequence learning. 2

Thus, there is actually now substantial evidence in the literature directly supporting the assertion “that micro-offline gains really result from offline learning”.  On the contrary, according to Gupta & Rickard (2024) “…the mechanism underlying RI [reactive inhibition] is not well established” after over 80 years of investigation60, possibly due to the fact that “reactive inhibition” is a categorical description of behavioral effects that likely result from several heterogenous processes with very different underlying mechanisms.

On the contrary, recent evidence questions this interpretation (Gupta & Rickard, npj Sci Learn 2022; Gupta & Rickard, Sci Rep 2024; Das et al., bioRxiv 2024). Instead, there is evidence that micro-offline gains are transient performance benefits that emerge when participants train with breaks, compared to participants who train without breaks, however, these benefits vanish within seconds after training if both groups of participants perform under comparable conditions (Das et al., bioRxiv 2024). 

It is important to point out that the recent work of Gupta & Rickard (2022,2024) 55 does not present any data that directly opposes our finding that early skill learning49 is expressed as micro-offline gains during rest breaks. These studies are essentially an extension of the Rickard et al (2008) paper that employed a massed (30s practice followed by 30s breaks) vs spaced (10s practice followed by 10s breaks) to assess if recovery from reactive inhibition effects could account for performance gains measured after several minutes or hours. Gupta & Rickard (2022) added two additional groups (30s practice/10s break and 10s practice/10s break as used in the work from our group). The primary aim of the study was to assess whether it was more likely that changes in performance when retested 5 minutes after skill training (consisting of 12 practice trials for the massed groups and 36 practice trials for the spaced groups) had ended reflected memory consolidation effects or recovery from reactive inhibition effects. The Gupta & Rickard (2024) follow-up paper employed a similar design with the primary difference being that participants performed a fixed number of sequences on each trial as opposed to trials lasting a fixed duration. This was done to facilitate the fitting of a quantitative statistical model to the data.  To reiterate, neither study included any analysis of micro-online or micro-offline gains and did not include any comparison focused on skill gains during early learning. Instead, Gupta & Rickard (2022), reported evidence for reactive inhibition effects for all groups over much longer training periods. Again, we reported the same finding for trials following the early learning period in our original Bönstrup et al. (2019) paper49 (Author response image 7). Also, please note that we reported in this paper that cumulative micro-offline gains over early learning did not correlate with overnight offline consolidation measured 24 hours later49 (see the Results section and further elaboration in the Discussion). Thus, while the composition of our data is supportive of a short-term memory consolidation process operating over several seconds during early learning, it likely differs from those involved over longer training times and offline periods, as assessed by Gupta & Rickard (2022).

In the recent preprint from Das et al (2024) 61,  the authors make the strong claim that “micro-offline gains during early learning do not reflect offline learning” which is not supported by their own data.   The authors hypothesize that if “micro-offline gains represent offline learning, participants should reach higher skill levels when training with breaks, compared to training without breaks”.  The study utilizes a spaced vs. massed practice group between-subjects design inspired by the reactive inhibition work from Rickard and others to test this hypothesis. Crucially, the design incorporates only a small fraction of the training used in other investigations to evaluate early skill learning1,33,49,54,57,58,62.  A direct comparison between the practice schedule designs for the spaced and massed groups in Das et al., and the training schedule all participants experienced in the original Bönstrup et al. (2019) paper highlights this issue as well as several others (Author response image 8):

Author response image 8.

(A) Comparison of Das et al. Spaced & Massed group training session designs, and the training session design from the original Bönstrup et al. (2019) 49 paper. Similar to the approach taken by Das et al., all practice is visualized as 10-second practice trials with a variable number (either 0, 1 or 30) of 10-second-long inter-practice rest intervals to allow for direct comparisons between designs. The two key takeaways from this comparison are that (1) the intervention differences (i.e. – practice schedules) between the Massed and Spaced groups from the Das et al. report are extremely small (less than 12% of the overall session schedule) and (2) the overall amount of practice is much less than compared to the design from the original Bönstrup report 49  (which has been utilized in several subsequent studies). (B) Group-level learning curve data from Bönstrup et al. (2019) 49 is used to estimate the performance range accounted for by the equivalent periods covering Test 1, Training 1 and Test 2 from Das et al (2024). Note that the intervention in the Das et al. study is limited to a period covering less than 50% of the overall learning range.

First, participants in the original Bönstrup et al. study 49 experienced 157.14% more practice time and 46.97% less inter-practice rest time than the Spaced group in the Das et al. study (Author response image 8).  Thus, the overall amount of practice and rest differ substantially between studies, with much more limited training occurring for participants in Das et al.  

Second, and perhaps most importantly, the actual intervention (i.e. – the difference in practice schedule between the Spaced and Massed groups) employed by Das et al. covers a very small fraction of the overall training session. Identical practice schedule segments for both the Spaced & Massed groups are indicated by the red shaded area in Author response image 8. Please note that these identical segments cover 94.84% of the Massed group training schedule and 88.01% of the Spaced group training schedule (since it has 60 seconds of additional rest). This means that the actual interventions cover less than 5% (for Massed) and 12% (for Spaced) of the total training session, which minimizes any chance of observing a difference between groups.

Also note that the very beginning of the practice schedule (during which Figure R9 shows substantial learning is known to occur) is labeled in the Das et al. study as Test 1.  Test 1 encompasses the first 20 seconds of practice (alternatively viewed as the first two 10-second-long practice trials with no inter-practice rest). This is immediately followed by the Training 1 intervention, which is composed of only three 10-second-long practice trials (with 10-second inter-practice rest for the Spaced group and no inter-practice rest for the Massed group). Author response image 8 also shows that since there is no inter-practice rest after the third Training practice trial for the Spaced group, this third trial (for both Training 1 and 2) is actually a part of an identical practice schedule segment shared by both groups (Massed and Spaced), reducing the magnitude of the intervention even further.

Moreover, we know from the original Bönstrup et al. (2019) paper49 that 46.57% of all overall group-level performance gains occurred between trials 2 and 5 for that study. Thus, Das et al. are limiting their designed intervention to a period covering less than half of the early learning range discussed in the literature, which again, minimizes any chance of observing an effect.

This issue is amplified even further at Training 2 since skill learning prior to the long 5-minute break is retained, further constraining the performance range over these three trials. A related issue pertains to the trials labeled as Test 1 (trials 1-2) and Test 2 (trials 6-7) by Das et al. Again, we know from the original Bönstrup et al. paper 49 that 18.06% and 14.43% (32.49% total) of all overall group-level performance gains occurred during trials corresponding to Das et al Test 1 and Test 2, respectively. In other words, Das et al averaged skill performance over 20 seconds of practice at two time-points where dramatic skill improvements occur. Pan & Rickard (1995) previously showed that such averaging is known to inject artefacts into analyses of performance gains.

Furthermore, the structure of the Test in Das et. al study appears to have an interference effect on the Spaced group performance after the training intervention.  This makes sense if you consider that the Spaced group is required to now perform the task in a Massed practice environment (i.e., two 10-second-long practice trials merged into one long trial), further blurring the true intervention effects. This effect is observable in Figure 1C,E of their pre-print. Specifically, while the Massed group continues to show an increase in performance during test relative to the last 10 seconds of practice during training, the Spaced group displays a marked decrease. This decrease is in stark contrast to the monotonic increases observed for both groups at all other time-points.

Interestingly, when statistical comparisons between the groups are made at the time-points when the intervention is present (as opposed to after it has been removed) then the stated hypothesis, “If micro-offline gains represent offline learning, participants should reach higher skill levels when training with breaks, compared to training without breaks”, is confirmed.

The data presented by Gupta and Rickard (2022, 2024) and Das et al. (2024) is in many ways more confirmatory of the constraints employed by our group and others with respect to experimental design, analysis and interpretation of study findings, rather than contradictory. Still, it does highlight a limitation of the current micro-online/offline framework, which was originally only intended to be applied to early skill learning over spaced practice schedules when reactive inhibition effects are minimized49. Extrapolation of this current framework to post-plateau performance periods, longer timespans, or non-learning situations (e.g. – the Non-repeating groups from Experiments 3 & 4 in Das et al. (2024)), when reactive inhibition plays a more substantive role, is not warranted. Ultimately, it will be important to develop new paradigms allowing one to independently estimate the different coincident or antagonistic features (e.g. - memory consolidation, planning, working memory and reactive inhibition) contributing to micro-online and micro-offline gains during and after early skill learning within a unifying framework.

References

(1) Buch, E. R., Claudino, L., Quentin, R., Bonstrup, M. & Cohen, L. G. Consolidation of human skill linked to waking hippocampo-neocortical replay. Cell Rep 35, 109193 (2021). https://doi.org:10.1016/j.celrep.2021.109193

(2) Chen, P.-C., Stritzelberger, J., Walther, K., Hamer, H. & Staresina, B. P. Hippocampal ripples during offline periods predict human motor sequence learning. bioRxiv, 2024.2010.2006.614680 (2024). https://doi.org:10.1101/2024.10.06.614680

(3) Classen, J., Liepert, J., Wise, S. P., Hallett, M. & Cohen, L. G. Rapid plasticity of human cortical movement representation induced by practice. J Neurophysiol 79, 1117-1123 (1998).

(4) Karni, A. et al. Functional MRI evidence for adult motor cortex plasticity during motor skill learning. Nature 377, 155-158 (1995). https://doi.org:10.1038/377155a0

(5) Kleim, J. A., Barbay, S. & Nudo, R. J. Functional reorganization of the rat motor cortex following motor skill learning. J Neurophysiol 80, 3321-3325 (1998).

(6) Shadmehr, R. & Holcomb, H. H. Neural correlates of motor memory consolidation. Science 277, 821-824 (1997).

(7) Doyon, J. et al. Experience-dependent changes in cerebellar contributions to motor sequence learning. Proc Natl Acad Sci U S A 99, 1017-1022 (2002).

(8) Toni, I., Ramnani, N., Josephs, O., Ashburner, J. & Passingham, R. E. Learning arbitrary visuomotor associations: temporal dynamic of brain activity. Neuroimage 14, 1048-1057 (2001).

(9) Grafton, S. T. et al. Functional anatomy of human procedural learning determined with regional cerebral blood flow and PET. J Neurosci 12, 2542-2548 (1992).

(10) Kennerley, S. W., Sakai, K. & Rushworth, M. F. Organization of action sequences and the role of the pre-SMA. J Neurophysiol 91, 978-993 (2004). https://doi.org:10.1152/jn.00651.2003 00651.2003 [pii]

(11) Hardwick, R. M., Rottschy, C., Miall, R. C. & Eickhoff, S. B. A quantitative meta-analysis and review of motor learning in the human brain. Neuroimage 67, 283-297 (2013). https://doi.org:10.1016/j.neuroimage.2012.11.020

(12) Sawamura, D. et al. Acquisition of chopstick-operation skills with the non-dominant hand and concomitant changes in brain activity. Sci Rep 9, 20397 (2019). https://doi.org:10.1038/s41598-019-56956-0

(13) Lee, S. H., Jin, S. H. & An, J. The difference in cortical activation pattern for complex motor skills: A functional near- infrared spectroscopy study. Sci Rep 9, 14066 (2019). https://doi.org:10.1038/s41598-019-50644-9

(14) Battaglia-Mayer, A. & Caminiti, R. Corticocortical Systems Underlying High-Order Motor Control. J Neurosci 39, 4404-4421 (2019). https://doi.org:10.1523/JNEUROSCI.2094-18.2019

(15) Toni, I., Thoenissen, D. & Zilles, K. Movement preparation and motor intention. Neuroimage 14, S110-117 (2001). https://doi.org:10.1006/nimg.2001.0841

(16) Wolpert, D. M., Goodbody, S. J. & Husain, M. Maintaining internal representations: the role of the human superior parietal lobe. Nat Neurosci 1, 529-533 (1998). https://doi.org:10.1038/2245

(17) Andersen, R. A. & Buneo, C. A. Intentional maps in posterior parietal cortex. Annu Rev Neurosci 25, 189-220 (2002). https://doi.org:10.1146/annurev.neuro.25.112701.142922 112701.142922 [pii]

(18) Buneo, C. A. & Andersen, R. A. The posterior parietal cortex: sensorimotor interface for the planning and online control of visually guided movements. Neuropsychologia 44, 2594-2606 (2006). https://doi.org:S0028-3932(05)00333-7 [pii] 10.1016/j.neuropsychologia.2005.10.011

(19) Grover, S., Wen, W., Viswanathan, V., Gill, C. T. & Reinhart, R. M. G. Long-lasting, dissociable improvements in working memory and long-term memory in older adults with repetitive neuromodulation. Nat Neurosci 25, 1237-1246 (2022). https://doi.org:10.1038/s41593-022-01132-3

(20) Colclough, G. L. et al. How reliable are MEG resting-state connectivity metrics? Neuroimage 138, 284-293 (2016). https://doi.org:10.1016/j.neuroimage.2016.05.070

(21) Colclough, G. L., Brookes, M. J., Smith, S. M. & Woolrich, M. W. A symmetric multivariate leakage correction for MEG connectomes. NeuroImage 117, 439-448 (2015). https://doi.org:10.1016/j.neuroimage.2015.03.071

(22) Mollazadeh, M. et al. Spatiotemporal variation of multiple neurophysiological signals in the primary motor cortex during dexterous reach-to-grasp movements. J Neurosci 31, 15531-15543 (2011). https://doi.org:10.1523/JNEUROSCI.2999-11.2011

(23) Bansal, A. K., Vargas-Irwin, C. E., Truccolo, W. & Donoghue, J. P. Relationships among low-frequency local field potentials, spiking activity, and three-dimensional reach and grasp kinematics in primary motor and ventral premotor cortices. J Neurophysiol 105, 1603-1619 (2011). https://doi.org:10.1152/jn.00532.2010

(24) Flint, R. D., Ethier, C., Oby, E. R., Miller, L. E. & Slutzky, M. W. Local field potentials allow accurate decoding of muscle activity. J Neurophysiol 108, 18-24 (2012). https://doi.org:10.1152/jn.00832.2011

(25) Churchland, M. M. et al. Neural population dynamics during reaching. Nature 487, 51-56 (2012). https://doi.org:10.1038/nature11129

(26) Bassett, D. S. et al. Dynamic reconfiguration of human brain networks during learning. Proc Natl Acad Sci U S A 108, 7641-7646 (2011). https://doi.org:10.1073/pnas.1018985108

(27) Albouy, G., King, B. R., Maquet, P. & Doyon, J. Hippocampus and striatum: dynamics and interaction during acquisition and sleep-related motor sequence memory consolidation. Hippocampus 23, 985-1004 (2013). https://doi.org:10.1002/hipo.22183

(28) Albouy, G. et al. Neural correlates of performance variability during motor sequence acquisition. Neuroimage 60, 324-331 (2012). https://doi.org:10.1016/j.neuroimage.2011.12.049

(29) Qin, Y. L., McNaughton, B. L., Skaggs, W. E. & Barnes, C. A. Memory reprocessing in corticocortical and hippocampocortical neuronal ensembles. Philos Trans R Soc Lond B Biol Sci 352, 1525-1533 (1997). https://doi.org:10.1098/rstb.1997.0139

(30) Euston, D. R., Tatsuno, M. & McNaughton, B. L. Fast-forward playback of recent memory sequences in prefrontal cortex during sleep. Science 318, 1147-1150 (2007). https://doi.org:10.1126/science.1148979

(31) Molle, M. & Born, J. Hippocampus whispering in deep sleep to prefrontal cortex--for good memories? Neuron 61, 496-498 (2009). https://doi.org:S0896-6273(09)00122-6 [pii] 10.1016/j.neuron.2009.02.002

(32) Frankland, P. W. & Bontempi, B. The organization of recent and remote memories. Nat Rev Neurosci 6, 119-130 (2005). https://doi.org:10.1038/nrn1607

(33) Jacobacci, F. et al. Rapid hippocampal plasticity supports motor sequence learning. Proc Natl Acad Sci U S A 117, 23898-23903 (2020). https://doi.org:10.1073/pnas.2009576117

(34) Albouy, G. et al. Maintaining vs. enhancing motor sequence memories: respective roles of striatal and hippocampal systems. Neuroimage 108, 423-434 (2015). https://doi.org:10.1016/j.neuroimage.2014.12.049

(35) Gais, S. et al. Sleep transforms the cerebral trace of declarative memories. Proc Natl Acad Sci U S A 104, 18778-18783 (2007). https://doi.org:0705454104 [pii] 10.1073/pnas.0705454104

(36) Sterpenich, V. et al. Sleep promotes the neural reorganization of remote emotional memory. J Neurosci 29, 5143-5152 (2009). https://doi.org:10.1523/JNEUROSCI.0561-09.2009

(37) Euston, D. R., Gruber, A. J. & McNaughton, B. L. The role of medial prefrontal cortex in memory and decision making. Neuron 76, 1057-1070 (2012). https://doi.org:10.1016/j.neuron.2012.12.002

(38) van Kesteren, M. T., Fernandez, G., Norris, D. G. & Hermans, E. J. Persistent schema-dependent hippocampal-neocortical connectivity during memory encoding and postencoding rest in humans. Proc Natl Acad Sci U S A 107, 7550-7555 (2010). https://doi.org:10.1073/pnas.0914892107

(39) van Kesteren, M. T., Ruiter, D. J., Fernandez, G. & Henson, R. N. How schema and novelty augment memory formation. Trends Neurosci 35, 211-219 (2012). https://doi.org:10.1016/j.tins.2012.02.001

(40) Wagner, A. D. et al. Building memories: remembering and forgetting of verbal experiences as predicted by brain activity. Science (New York, N.Y.) 281, 1188-1191 (1998).

(41) Ashe, J., Lungu, O. V., Basford, A. T. & Lu, X. Cortical control of motor sequences. Curr Opin Neurobiol 16, 213-221 (2006).

(42) Hikosaka, O., Nakamura, K., Sakai, K. & Nakahara, H. Central mechanisms of motor skill learning. Curr Opin Neurobiol 12, 217-222 (2002).

(43) Penhune, V. B. & Steele, C. J. Parallel contributions of cerebellar, striatal and M1 mechanisms to motor sequence learning. Behav. Brain Res. 226, 579-591 (2012). https://doi.org:10.1016/j.bbr.2011.09.044

(44) Doyon, J. et al. Contributions of the basal ganglia and functionally related brain structures to motor learning. Behavioural brain research 199, 61-75 (2009). https://doi.org:10.1016/j.bbr.2008.11.012

(45) Schendan, H. E., Searl, M. M., Melrose, R. J. & Stern, C. E. An FMRI study of the role of the medial temporal lobe in implicit and explicit sequence learning. Neuron 37, 1013-1025 (2003). https://doi.org:10.1016/s0896-6273(03)00123-5

(46) Morris, R. G. M. Elements of a neurobiological theory of hippocampal function: the role of synaptic plasticity, synaptic tagging and schemas. The European journal of neuroscience 23, 2829-2846 (2006). https://doi.org:10.1111/j.1460-9568.2006.04888.x

(47) Tse, D. et al. Schemas and memory consolidation. Science 316, 76-82 (2007). https://doi.org:10.1126/science.1135935

(48) Berlot, E., Popp, N. J. & Diedrichsen, J. A critical re-evaluation of fMRI signatures of motor sequence learning. Elife 9 (2020). https://doi.org:10.7554/eLife.55241

(49) Bonstrup, M. et al. A Rapid Form of Offline Consolidation in Skill Learning. Curr Biol 29, 1346-1351 e1344 (2019). https://doi.org:10.1016/j.cub.2019.02.049

(50) Kornysheva, K. et al. Neural Competitive Queuing of Ordinal Structure Underlies Skilled Sequential Action. Neuron 101, 1166-1180 e1163 (2019). https://doi.org:10.1016/j.neuron.2019.01.018

(51) Pan, S. C. & Rickard, T. C. Sleep and motor learning: Is there room for consolidation? Psychol Bull 141, 812-834 (2015). https://doi.org:10.1037/bul0000009

(52) Rickard, T. C., Cai, D. J., Rieth, C. A., Jones, J. & Ard, M. C. Sleep does not enhance motor sequence learning. J Exp Psychol Learn Mem Cogn 34, 834-842 (2008). https://doi.org:10.1037/0278-7393.34.4.834

53) Brawn, T. P., Fenn, K. M., Nusbaum, H. C. & Margoliash, D. Consolidating the effects of waking and sleep on motor-sequence learning. J Neurosci 30, 13977-13982 (2010). https://doi.org:10.1523/JNEUROSCI.3295-10.2010

(54) Bonstrup, M., Iturrate, I., Hebart, M. N., Censor, N. & Cohen, L. G. Mechanisms of offline motor learning at a microscale of seconds in large-scale crowdsourced data. NPJ Sci Learn 5, 7 (2020). https://doi.org:10.1038/s41539-020-0066-9

(55) Gupta, M. W. & Rickard, T. C. Dissipation of reactive inhibition is sufficient to explain post-rest improvements in motor sequence learning. NPJ Sci Learn 7, 25 (2022). https://doi.org:10.1038/s41539-022-00140-z

(56) Jacobacci, F. et al. Rapid hippocampal plasticity supports motor sequence learning. Proceedings of the National Academy of Sciences 117, 23898-23903 (2020).

(57) Brooks, E., Wallis, S., Hendrikse, J. & Coxon, J. Micro-consolidation occurs when learning an implicit motor sequence, but is not influenced by HIIT exercise. NPJ Sci Learn 9, 23 (2024). https://doi.org:10.1038/s41539-024-00238-6

(58) Deleglise, A. et al. Human motor sequence learning drives transient changes in network topology and hippocampal connectivity early during memory consolidation. Cereb Cortex 33, 6120-6131 (2023). https://doi.org:10.1093/cercor/bhac489

(59) Buzsaki, G. Hippocampal sharp wave-ripple: A cognitive biomarker for episodic memory and planning. Hippocampus 25, 1073-1188 (2015). https://doi.org:10.1002/hipo.22488

(60) Gupta, M. W. & Rickard, T. C. Comparison of online, offline, and hybrid hypotheses of motor sequence learning using a quantitative model that incorporate reactive inhibition. Sci Rep 14, 4661 (2024). https://doi.org:10.1038/s41598-024-52726-9

(61) Das, A., Karagiorgis, A., Diedrichsen, J., Stenner, M.-P. & Azanon, E. “Micro-offline gains” convey no benefit for motor skill learning. bioRxiv, 2024.2007.2011.602795 (2024). https://doi.org:10.1101/2024.07.11.602795

(62) Mylonas, D. et al. Maintenance of Procedural Motor Memory across Brief Rest Periods Requires the Hippocampus. J Neurosci 44 (2024). https://doi.org:10.1523/JNEUROSCI.1839-23.2024

Author Response

Reviewer #1 (Public Review):

Summary:

By examining the prevalence of interactions with ancient amino acids of coenzymes in ancient versus recent folds, the authors noticed an increased interaction propensity for ancient interactions. They infer from this that coenzymes might have played an important role in prebiotic proteins.

Strengths:

(1) The analysis, which is very straightforward, is technically correct. However, the conclusions might not be as strong as presented.

(2) This paper presents an excellent summary of contemporary thought on what might have constituted prebiotic proteins and their properties.

(3) The paper is clearly written.

We are grateful for the kind comments of the reviewer on our manuscript. However, we would like to clarify a possible misunderstanding in the summary of our study. Specifically, analysis of "ancient versus recent folds" was not really reported in our results. Our analysis concerned "coenzyme age" rather than the "protein folds age" and was focused mainly on interaction with early vs. late amino acids in protein sequence. While structural propensities of the coenzyme binding sites were also analyzed, no distinction on the level of ancient vs. recent folds was assumed and this was only commented on in the discussion, based on previous work of others.

Weaknesses:

(1) The conclusions might not be as strong as presented. First of all, while ancient amino acids interact less frequently in late with a given coenzyme, maybe this just reflects the fact that proteins that evolved later might be using residues that have a more favorable binding free energy.

We would like to point out that there was no distinction to proteins that evolved early or late in our dataset of coenzyme-binding proteins. The aim of our analysis was purely to observe trends in the age of amino acids vs. age of coenzymes. While no direct inference can be made from this about early life as all the proteins are from extant life (as highlighted in the discussion of our work), our goal was to look for intrinsic propensities of early vs. late amino acids in binding to the different coenzyme entities. Indeed, very early interactions would be smeared by the eons of evolutionary history (perhaps also towards more favourable binding free energy, as pointed out also by the reviewer). Nevertheless, significant trends have been recorded across the PDB dataset, pointing to different propensities and mechanistic properties of the binding events. Rather than to a specific evolutionary past, our data therefore point to a “capacity” of the early amino acids to bind certain coenzymes and we believe that this is the major (and standing) conclusion of our work, along with the properties of such interactions. In our revised version, we will carefully go through all the conclusions and make sure that this message stands out but we are confident that the following concluding sentences copied from the abstract and the discussion of our manuscript fully comply with our data:

“These results imply the plausibility of a coenzyme-peptide functional collaboration preceding the establishment of the Central Dogma and full protein alphabet evolution”

“While no direct inferences about distant evolutionary past can be drawn from the analysis of extant proteins, the principles guiding these interactions can imply their potential prebiotic feasibility and significance.”

“This implies that late amino acids would not be necessarily needed for the sovereignty of coenzyme-peptide interplay.”

We would also like to add that proteins that evolved later might not always have higher free energy of binding. Musil et al., 2021 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8294521/) showed in their study on the example of haloalkane dehalogenase Dha A that the ancestral sequence reconstruction is a powerful tool for designing more stable, but also more active proteins. Ancestral sequence reconstruction relies on finding ancient states of protein families to suggest mutations that will lead to more stable proteins than are currently existing proteins. Their study did not explore the ligand-protein interactions specifically, but showed that ancient states often show more favourable properties than modern proteins.

(2) What about other small molecules that existed in the probiotic soup? Do they also prefer such ancient amino acids? If so, this might reflect the interaction propensity of specific amino acids rather than the inferred important role of coenzymes.

We appreciate the comment of the reviewer towards other small molecules, which we assume points mainly towards metal ions (i.e. inorganic cofactors). We completely agree with the reviewer that such interactions are of utmost importance to the origins of life. Intentionally, they were not part of our study, as these have already been studied previously by others (e.g. Bromberg et al., 2022; and reviewed in Frenkel-Pinter et al., 2020) and also us (Fried et al., 2022). For example, it is noteworthy that prebiotically relevant metal binding sites (e.g. of Mg2+) exhibit enrichment in early amino acids such as Asp and Glu while more recent metal (e.g. Cu and Zn) site in the late amino acids His and Cys (Fried et al., 2022). At the same time, comparable analyses of amino acid - coenzyme trends were not available.

Nevertheless, involvement of metal ions in the coenzyme binding sites was also studied here and pointed to their bigger involvement with the Ancient coenzymes. In the revised version of the manuscript, we will be happy to enlarge the discussion of the studies concerning inorganic cofactors.

(3) Perhaps the conclusions just reflect the types of active sites that evolved first and nothing more.

We partly agree on this point with the reviewer but not on the fact why it is listed as the weakness of our study and on the “nothing more” notion. Understanding what the properties of the earliest binding sites is key to merging the gap between prebiotic chemistry and biochemistry. The potential of peptides preceding ribosomal synthesis (and the full alphabet evolution) along with prebiotically plausible coenzymes addresses exactly this gap, which is currently not understood.

Reviewer #2 (Public Review):

I enjoyed reading this paper and appreciate the careful analysis performed by the investigators examining whether 'ancient' cofactors are preferentially bound by the first-available amino acids, and whether later 'LUCA' cofactors are bound by the late-arriving amino acids. I've always found this question fascinating as there is a contradiction in inorganic metal-protein complexes (not what is focused on here). Metal coordination of Fe, Ni heavily relies on softer ligands like His and Cys - which are by most models latecomer amino acids. There are no traces of thiols or imidazoles in meteorites - although work by Dvorkin has indicated that could very well be due to acid degradation during extraction. Chris Dupont (PNAS 2005) showed that metal speciation in the early earth (such as proposed by Anbar and prior RJP Williams) matched the purported order of fold emergence.

As such, cofactor-protein interactions as a driving force for evolution has always made sense to me and I admittedly read this paper biased in its favor. But to make sure, I started to play around with the data that the authors kindly and importantly shared in the supplementary files. Here's what I found:

Point 1: The correlation between abundance of amino acids and protein age is dominated by glycine. There is a small, but visible difference in old vs new amino acid fractional abundance between Ancient and LUCA proteins (Figure 3, Supplementary Table 3). However, the bias is not evenly distributed among the amino acids - which Figure 4A shows but is hard to digest as presented. So instead I used the spreadsheet in Supplement 3 to calculate the fractional difference FDaa = F(old aa)-F(new aa). As expected from Figure 3, the mean FD for Ancient is greater than the mean FD for LUCA. But when you look at the same table for each amino acid FDcofactor = F(ancient cofactor) - F(LUCA cofactor), you now see that the bias is not evenly distributed between older and newer amino acids at all. In fact, most of the difference can be explained by glycine (FDcofactor = 3.8) and the rest by also including tryptophan (FDcofactor = -3.8). If you remove these two amino acids from the analysis, the trend seen in Figure 3 all but disappears.

Troubling - so you might argue that Gly is the oldest of the old and Trp is the newest of the new so the argument still stands. Unfortunately, Gly is a lot of things - flexible, small, polar - so what is the real correlation, age, or chemistry? This leads to point 2.

We truly acknowledge the effort that the reviewer made in the revision of the data and for the thoughtful, deeper analysis. We agree that this deserves further discussion of our data. As invited by the reviewer, we indeed repeated the analysis on the whole dataset. First, we would like to point out that the reviewer was most probably referring to the Supplementary Fig. 2 (and not 3, which concerns protein folds). While the difference between Ancient and LUCA coenzyme binding is indeed most pronounced for Gly and Trp, we failed to confirm that the trend disappears if those two amino acids are removed from the analysis (additional FDcofactors of 3.2 and -3.2 are observed for the early and late amino acids, resp.), as seen in Table I below. The main additional contributors to this effect are Asp (FD of 2.1) and Ser (FD of 1.8) from the early amino acids and Arg (FD of -2.6) and Cys (FD of -1.7) of the late amino acids. Hence, while we agree with the reviewer that Gly and Trp (the oldest and the youngest) contribute to this effect the most, we disagree that the trend reduces to these two amino acids.

In addition, the most recent coenzyme temporality (the Post-LUCA) was neglected in the reviewer’s analysis. The difference between F (old) and F (new) is even more pronounced in PostLUCA than in LUCA, vs. Ancient (Table II) and depends much less on Trp. Meanwhile, Asp, Ser, Leu, Phe, and Arg dominate the observed phenomenon (Table I). This further supports our lack of agreement with the reviewer’s point. Nevertheless, we remain grateful for this discussion and we will happily include this additional analysis in the Supplementary Material of our revised manuscript.

Author response table 1.

Amino acid fractional difference of all coenzymes at residue level

Author response table 2.

Amino acid fractional difference of all coenzymes

Point 2 - The correlation is dominated by phosphate.

In the ancient cofactor list, all but 4 comprise at least one phosphate (SAM, tetrahydrofolic acid, biopterin, and heme). Except for SAM, the rest have very low Gly abundance. The overall high Gly abundance in the ancient enzymes is due to the chemical property of glycine that can occupy the right-hand side of the Ramachandran plot. This allows it to make the alternating alphaleftalpharight conformation of the P-loop forming Milner-White's anionic nest. If you remove phosphate binding folds from the analysis the trend in Figure 3 vanishes.

Likewise, Trp is an important functional residue for binding quinones and tuning its redox potential. The LUCA cofactor set is dominated by quinone and derivatives, which likely drives up the new amino acid score for this class of cofactors.

Once again, we are thankful to the reviewer for raising this point. The role of Gly in the anionic nests proposed by Milner-White and Russel, as well as the Trp role in quinone binding are important points that we would be happy to highlight more in the discussion of the revised manuscript.<br /> Nevertheless, we disagree that the trends reduce only to the phosphate-containing coenzymes and importantly, that “the trend in Figure 3 vanishes” upon their removal. Table III and IV (below) show the data for coenzymes excluding those with phosphate moiety and the trend in Fig. 3 remains, albeit less pronounced.

Author response table 3.

Amino acid fractional difference of non-phosphate containing coenzymes

Author response table 4.

Amino acid fractional difference of non-phosphate containing coenzymes at residue level

In summary, while I still believe the premise that cofactors drove the shape of peptides and the folds that came from them - and that Rossmann folds are ancient phosphate-binding proteins, this analysis does not really bring anything new to these ideas that have already been stated by Tawfik/Longo, Milner-White/Russell, and many others.

I did this analysis ad hoc on a slice of the data the authors provided and could easily have missed something and I encourage the authors to check my work. If it holds up it should be noted that negative results can often be as informative as strong positive ones. I think the signal here is too weak to see in the noise using the current approach.

We are grateful to the reviewer for encouraging further look at our data. While we hope that the analysis on the whole dataset (listed in Tables I - IV) will change the reviewer’s standpoint on our work, we would still like to comment on the questioned novelty of our results. In fact, the extraordinary works by Tawfik/Longo and Milner-While/Russel (which were cited in our manuscript multiple times) presented one of the motivations for this study. We take the opportunity to copy the part of our discussion that specifically highlights the relevance of their studies, and points out the contribution of our work with respect to theirs.

“While all the coenzymes bind preferentially to protein residue sidechains, more backbone interactions appear in the ancient coenzyme class when compared to others. This supports an earlier hypothesis that functions of the earliest peptides (possibly of variable compositions and lengths) would be performed with the assistance of the main chain atoms rather than their sidechains (Milner-White and Russel 2011). Longo et al., recently analyzed binding sites of different phosphate-containing ligands which were arguably of high relevance during earliest stages of life, connecting all of today’s core metabolism (Longo et al., 2020 (b)). They observed that unlike the evolutionary younger binding motifs (which rely on sidechain binding), the most ancient lineages indeed bind to phosphate moieties predominantly via the protein backbone. Our analysis assigns this phenomenon primarily to interactions via early amino acids that (as mentioned above) are generally enriched in the binding interface of the ancient coenzymes. This implies that late amino acids would not be necessarily needed for the sovereignty of coenzymepeptide interplay.”

Unlike any other previous work, our study involves all the major coenzymes (not just the phosphate-containing ones) and is based on their evolutionary age, as well as age of amino acids. It is the first PDB-wide systematic evolutionary analysis of coenzyme-amino acid binding. Besides confirming some earlier theoretical assertions (such as role of backbone interactions in early peptide-coenzyme evolution) and observations (such as occurrence of the ancient phosphatecontaining coenzymes in the oldest protein folds), it uncovers substantial novel knowledge. For example, (i) enrichment of early amino acids in the binding of ancient coenzymes, vs. enrichment of late amino acids in the binding of LUCA and Post-LUCA coenzymes, (ii) the trends in secondary structure content of the binding sites of coenzyme of different temporalities, (iii) increased involvement of metal ions in the ancient coenzyme binding events, and (iv) the capacity of only early amino acids to bind ancient coenzymes. In our humble opinion, all of these points bring important contributions in the peptide-coenzyme knowledge gap which has been discussed in a number of previous studies.

Author response:

eLife assessment

This potentially useful study involves neuro-imaging and electrophysiology in a small cohort of congenital cataract patients after sight recovery and age-matched control participants with normal sight. It aims to characterize the effects of early visual deprivation on excitatory and inhibitory balance in the visual cortex. While the findings are taken to suggest the existence of persistent alterations in Glx/GABA ratio and aperiodic EEG signals, the evidence supporting these claims is incomplete. Specifically, small sample sizes, lack of a specific control cohort, and other methodological limitations will likely restrict the usefulness of the work, with relevance limited to scientists working in this particular subfield.

As pointed out in the public reviews, there are only very few human models which allow for assessing the role of early experience on neural circuit development. While the prevalent research in permanent congenital blindness reveals the response and adaptation of the developing brain to an atypical situation (blindness), research in sight restoration addresses the question of whether and how atypical development can be remediated if typical experience (vision) is restored. The literature on the role of visual experience in the development of E/I balance in humans, assessed via Magnetic Resonance Spectroscopy (MRS), has been limited to a few studies on congenital permanent blindness. Thus, we assessed sight recovery individuals with a history of congenital blindness, as limited evidence from other researchers indicated that the visual cortex E/I ratio might differ compared to normally sighted controls.

Individuals with total bilateral congenital cataracts who remained untreated until later in life are extremely rare, particularly if only carefully diagnosed patients are included in a study sample. A sample size of 10 patients is, at the very least, typical of past studies in this population, even for exclusively behavioral assessments. In the present study, in addition to behavioral assessment as an indirect measure of sensitive periods, we investigated participants with two neuroimaging methods (Magnetic Resonance Spectroscopy and electroencephalography) to directly assess the neural correlates of sensitive periods in humans. The electroencephalography data allowed us to link the results of our small sample to findings documented in large cohorts of both, sight recovery individuals and permanently congenitally blind individuals. As pointed out in a recent editorial recommending an “exploration-then-estimation procedure,” (“Consideration of Sample Size in Neuroscience Studies,” 2020), exploratory studies like ours provide crucial direction and specific hypotheses for future work.

We included an age-matched sighted control group recruited from the same community, measured in the same scanner and laboratory, to assess whether early experience is necessary for a typical excitatory/inhibitory (E/I) ratio to emerge in adulthood. The present findings indicate that this is indeed the case. Based on these results, a possible question to answer in future work, with individuals who had developmental cataracts, is whether later visual deprivation causes similar effects. Note that even if visual deprivation at a later stage in life caused similar effects, the current results would not be invalidated; by contrast, they are essential to understand future work on late (permanent or transient) blindness.

Thus, we think that the present manuscript has far reaching implications for our understanding of the conditions under which E/I balance, a crucial characteristic of brain functioning, emerges in humans.

Finally, our manuscript is one of the first few studies which relates MRS neurotransmitter concentrations to parameters of EEG aperiodic activity. Since present research has been using aperiodic activity as a correlate of the E/I ratio, and partially of higher cognitive functions, we think that our manuscript additionally contributes to a better understanding of what might be measured with aperiodic neurophysiological activity.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

In this human neuroimaging and electrophysiology study, the authors aimed to characterize the effects of a period of visual deprivation in the sensitive period on excitatory and inhibitory balance in the visual cortex. They attempted to do so by comparing neurochemistry conditions ('eyes open', 'eyes closed') and resting state, and visually evoked EEG activity between ten congenital cataract patients with recovered sight (CC), and ten age-matched control participants (SC) with normal sight.

First, they used magnetic resonance spectroscopy to measure in vivo neurochemistry from two locations, the primary location of interest in the visual cortex, and a control location in the frontal cortex. Such voxels are used to provide a control for the spatial specificity of any effects because the single-voxel MRS method provides a single sampling location. Using MR-visible proxies of excitatory and inhibitory neurotransmission, Glx and GABA+ respectively, the authors report no group effects in GABA+ or Glx, no difference in the functional conditions 'eyes closed' and 'eyes open'. They found an effect of the group in the ratio of Glx/GABA+ and no similar effect in the control voxel location. They then performed multiple exploratory correlations between MRS measures and visual acuity, and reported a weak positive correlation between the 'eyes open' condition and visual acuity in CC participants.

The same participants then took part in an EEG experiment. The authors selected only two electrodes placed in the visual cortex for analysis and reported a group difference in an EEG index of neural activity, the aperiodic intercept, as well as the aperiodic slope, considered a proxy for cortical inhibition. They report an exploratory correlation between the aperiodic intercept and Glx in one out of three EEG conditions.

The authors report the difference in E/I ratio, and interpret the lower E/I ratio as representing an adaptation to visual deprivation, which would have initially caused a higher E/I ratio. Although intriguing, the strength of evidence in support of this view is not strong. Amongst the limitations are the low sample size, a critical control cohort that could provide evidence for a higher E/I ratio in CC patients without recovered sight for example, and lower data quality in the control voxel.

Strengths of study:

How sensitive period experience shapes the developing brain is an enduring and important question in neuroscience. This question has been particularly difficult to investigate in humans. The authors recruited a small number of sight-recovered participants with bilateral congenital cataracts to investigate the effect of sensitive period deprivation on the balance of excitation and inhibition in the visual brain using measures of brain chemistry and brain electrophysiology. The research is novel, and the paper was interesting and well-written.

Limitations:

(1.1) Low sample size. Ten for CC and ten for SC, and a further two SC participants were rejected due to a lack of frontal control voxel data. The sample size limits the statistical power of the dataset and increases the likelihood of effect inflation.

Applying strict criteria, we only included individuals who were born with no patterned vision in the CC group. The population of individuals who have remained untreated past infancy is small in India, despite a higher prevalence of childhood cataract than Germany. Indeed, from the original 11 CC and 11 SC participants tested, one participant each from the CC and SC group had to be rejected, as their data had been corrupted, resulting in 10 participants in each group.

It was a challenge to recruit participants from this rare group with no history of neurological diagnosis/intake of neuromodulatory medications, who were able and willing to undergo both MRS and EEG. For this study, data collection took more than 1.5 years.

We took care of the validity of our results with two measures; first, assessed not just MRS, but additionally, EEG measures of E/I ratio. The latter allowed us to link results to a larger population of CC individuals, that is, we replicated the results of a larger group of 38 individuals (Ossandón et al., 2023) in our sub-group.

Second, we included a control voxel. As predicted, all group effects were restricted to the occipital voxel.

(1.2) Lack of specific control cohort. The control cohort has normal vision. The control cohort is not specific enough to distinguish between people with sight loss due to different causes and patients with congenital cataracts with co-morbidities. Further data from more specific populations, such as patients whose cataracts have not been removed, with developmental cataracts, or congenitally blind participants, would greatly improve the interpretability of the main finding. The lack of a more specific control cohort is a major caveat that limits a conclusive interpretation of the results.

The existing work on visual deprivation and neurochemical changes, as assessed with MRS, has been limited to permanent congenital blindness. In fact, most of the studies on permanent blindness included only congenitally blind or early blind humans (Coullon et al., 2015; Weaver et al., 2013), or, in separate studies, only late-blind individuals (Bernabeu et al., 2009). Thus, accordingly, we started with the most “extreme” visual deprivation model, sight recovery after congenital blindness. If we had not observed any group difference compared to normally sighted controls, investigating other groups might have been trivial. Based on our results, subsequent studies in late blind individuals, and then individuals with developmental cataracts, can be planned with clear hypotheses.

(1.3) MRS data quality differences. Data quality in the control voxel appears worse than in the visual cortex voxel. The frontal cortex MRS spectrum shows far broader linewidth than the visual cortex (Supplementary Figures). Compared to the visual voxel, the frontal cortex voxel has less defined Glx and GABA+ peaks; lower GABA+ and Glx concentrations, lower NAA SNR values; lower NAA concentrations. If the data quality is a lot worse in the FC, then small effects may not be detectable.

Worse data quality in the frontal than the visual cortex has been repeatedly observed in the MRS literature, attributable to magnetic field distortions (Juchem & Graaf, 2017) resulting from the proximity of the region to the sinuses (recent example: (Rideaux et al., 2022)). Nevertheless, we chose the frontal control region rather than a parietal voxel, given the potential  neurochemical changes in multisensory regions of the parietal cortex due to blindness. Such reorganization would be less likely in frontal areas associated with higher cognitive functions. Further, prior MRS studies of the visual cortex have used the frontal cortex as a control region as well (Pitchaimuthu et al., 2017; Rideaux et al., 2022).

In the present study, we checked that the frontal cortex datasets for Glx and GABA+ concentrations were of sufficient quality: the fit error was below 8.31% in both groups (Supplementary Material S3). For reference, Mikkelsen et al. reported a mean GABA+ fit error of 6.24 +/- 1.95% from a posterior cingulate cortex voxel across 8 GE scanners, using the Gannet pipeline. No absolute cutoffs have been proposed for fit errors. However, MRS studies in special populations (I/E ratio assessed in narcolepsy (Gao et al., 2024), GABA concentration assessed in Autism Spectrum Disorder (Maier et al., 2022)) have used frontal cortex data with a fit error of <10% to identify differences between cohorts (Gao et al., 2024; Pitchaimuthu et al., 2017). Based on the literature, MRS data from the frontal voxel of the present study would have been of sufficient quality to uncover group differences.

In the revised manuscript, we will add the recently published MRS quality assessment form to the supplementary materials. Additionally, we would like to allude to our apriori prediction of group differences for the visual cortex, but not for the frontal cortex voxel.

(1.4) Because of the direction of the difference in E/I, the authors interpret their findings as representing signatures of sight improvement after surgery without further evidence, either within the study or from the literature. However, the literature suggests that plasticity and visual deprivation drive the E/I index up rather than down. Decreasing GABA+ is thought to facilitate experience-dependent remodelling. What evidence is there that cortical inhibition increases in response to a visual cortex that is over-sensitised due to congenital cataracts? Without further experimental or literature support this interpretation remains very speculative.

Indeed, higher inhibition was not predicted, which we attempt to reconcile in our discussion section. We base our discussion mainly on the non-human animal literature, which has shown evidence of homeostatic changes after prolonged visual deprivation in the adult brain (Barnes et al., 2015). It is also interesting to note that after monocular deprivation in adult humans, resting GABA+ levels decreased in the visual cortex (Lunghi et al., 2015). Assuming that after delayed sight restoration, adult neuroplasticity mechanisms must be employed, these studies would predict a “balancing” of the increased excitatory drive following sight restoration by a commensurate increase in inhibition (Keck et al., 2017). Additionally, the EEG results of the present study allowed for speculation regarding the underlying neural mechanisms of an altered E/I ratio. The aperiodic EEG activity suggested higher spontaneous spiking (increased intercept) and increased inhibition (steeper aperiodic slope between 1-20 Hz) in CC vs SC individuals (Ossandón et al., 2023).

In the revised manuscript, we will more clearly indicate that these speculations are based primarily on non-human animal work, due to the lack of human studies on the subject.

(1.5) Heterogeneity in the patient group. Congenital cataract (CC) patients experienced a variety of duration of visual impairment and were of different ages. They presented with co-morbidities (absorbed lens, strabismus, nystagmus). Strabismus has been associated with abnormalities in GABAergic inhibition in the visual cortex. The possible interactions with residual vision and confounds of co-morbidities are not experimentally controlled for in the correlations, and not discussed.

The goal of the present study was to assess whether we would observe changes in E/I ratio after restoring vision at all. We would not have included patients without nystagmus in the CC group of the present study, since it would have been unlikely that they experienced congenital patterned visual deprivation. Amongst diagnosticians, nystagmus or strabismus might not be considered genuine “comorbidities” that emerge in people with congenital cataracts. Rather, these are consequences of congenital visual deprivation, which we employed as diagnostic criteria. Similarly, absorbed lenses are clear signs that cataracts were congenital. As in other models of experience dependent brain development (e.g. the extant literature on congenital permanent blindness, including anophthalmic individuals (Coullon et al., 2015; Weaver et al., 2013), some uncertainty remains regarding whether the (remaining, in our case) abnormalities of the eye, or the blindness they caused, are the factors driving neural changes. In case of people with reversed congenital cataracts, at least the retina is considered to be intact, as they would otherwise not receive cataract removal surgery.

However, we consider it unlikely that strabismus caused the group differences, because the present study shows group differences in the Glx/GABA+ ratio at rest, regardless of eye opening or eye closure, for which strabismus would have caused distinct effects. By contrast, the link between GABA concentration and, for example, interocular suppression in strabismus, have so far been documented during visual stimulation (Mukerji et al., 2022; Sengpiel et al., 2006), and differed in direction depending on the amblyopic vs. non-amblyopic eye. Further, one MRS study did not find group differences in GABA concentration between the visual cortices of 16 amblyopic individuals and sighted controls (Mukerji et al., 2022), supporting that the differences in Glx/GABA+ concentration which we observed were driven by congenital deprivation, and not amblyopia-associated visual acuity or eye movement differences.  

In the revised manuscript, we will discuss the inclusion criteria in more detail, and the aforementioned reasons why our data remains interpretable.

(1.6) Multiple exploratory correlations were performed to relate MRS measures to visual acuity (shown in Supplementary Materials), and only specific ones were shown in the main document. The authors describe the analysis as exploratory in the 'Methods' section. Furthermore, the correlation between visual acuity and E/I metric is weak, and not corrected for multiple comparisons. The results should be presented as preliminary, as no strong conclusions can be made from them. They can provide a hypothesis to test in a future study.

In the revised manuscript, we will clearly indicate that the exploratory correlation analyses are reported to put forth hypotheses for future studies.

(1.7) P.16 Given the correlation of the aperiodic intercept with age ("Age negatively correlated with the aperiodic intercept across CC and SC individuals, that is, a flattening of the intercept was observed with age"), age needs to be controlled for in the correlation between neurochemistry and the aperiodic intercept. Glx has also been shown to negatively correlate with age.

The correlation between chronological age and aperiodic intercept was observed across groups, but the correlation between Glx and the intercept of the aperiodic EEG activity was seen only in the CC group, even though the SC group was matched for age. Thus, such a correlation was very unlikely to  be predominantly driven by an effect of chronological age.

In the revised manuscript, we will add the linear regressions with age as a covariate included below, for the relationship between aperiodic intercept and Glx concentration in the CC group. 

a. A linear regression was conducted within the CC group to predict the intercept during visual stimulation, based on age and visual cortex Glx concentration. The results of the regression analysis indicated that the model explained a significant proportion of the variance in the aperiodic intercept, 𝑅2\=0.82_, t_(2,7)=16.1_, 𝑝=0.0024._ Note that the coefficient for age was not significant, 𝛽=0.007, t(7)=0.82, 𝑝=0.439. The regression coefficients and their respective statistics are presented in Author response table 1.

Author response table 1.

Regression Analysis Summary for Predicting Aperiodic Intercept (Visual Stimulation) in the CC group

b. A linear regression was conducted to predict the intercept during eye opening at rest, based on age and visual cortex Glx concentration. The results of the regression analysis indicated that the model explained a significant proportion of the variance in the aperiodic intercept, 𝑅2\=0.842_, t_(2,7)=18.6,  𝑝=0.00159_._ Note that the coefficient for age was not significant, 𝛽=−0.005, t(7)=−0.90, 𝑝=0.400. The regression coefficients and their respective statistics are presented in Author response table 2.

Author response table 2.

Regression Analysis Summary for Predicting Aperiodic Intercept (Eyes Open) in the CC group

c. Given that the Glx coefficient is significant in both models and age does not significantly predict either outcome, it can be concluded that Glx independently predicts the intercept of the aperiodic intercept.

(1.8) Multiple exploratory correlations were performed to relate MRS to EEG measures (shown in Supplementary Materials), and only specific ones were shown in the main document. Given the multiple measures from the MRS, the correlations with the EEG measures were exploratory, as stated in the text, p.16, and in Figure 4. Yet the introduction said that there was a prior hypothesis "We further hypothesized that neurotransmitter changes would relate to changes in the slope and intercept of the EEG aperiodic activity in the same subjects." It would be great if the text could be revised for consistency and the analysis described as exploratory.

In the revised manuscript, we will improve the phrasing. We consider the correlation analyses as exploratory due to our sample size and the absence of prior work. However, we did hypothesize that both MRS and EEG markers would concurrently be altered in CC vs SC individuals.

(1.9) The analysis for the EEG needs to take more advantage of the available data. As far as I understand, only two electrodes were used, yet far more were available as seen in their previous study (Ossandon et al., 2023). The spatial specificity is not established. The authors could use the frontal cortex electrode (FP1, FP2) signals as a control for spatial specificity in the group effects, or even better, all available electrodes and correct for multiple comparisons. Furthermore, they could use the aperiodic intercept vs Glx in SC to evaluate the specificity of the correlation to CC.

The aperiodic intercept and slope did not differ between CC and SC individuals for Fp1 and Fp2, suggesting the spatial specificity of the results. In the revised manuscript, we will add this analysis to the supplementary material.

Author response image 1.

Aperiodic intercept (top) and slope (bottom) for congenital cataract-reversal (CC, red) and age-matched normally sighted control (SC, blue) individuals. Distributions of these parameters are displayed as violin plots for three conditions; at rest with eyes closed (EC), at rest with eyes open (EO) and during visual stimulation (LU). Aperiodic parameters were calculated across electrodes Fp1 and Fp2. Solid black lines indicate mean values, dotted black lines indicate median values. Coloured lines connect values of individual participants across conditions.

Further, Glx concentration in the visual cortex did not correlate with the aperiodic intercept in the SC group (Figure 4), suggesting that this relationship was indeed specific to the CC group.

The data from all electrodes has been analyzed and published in other studies as well (Pant et al., 2023; Ossandón et al., 2023).

Reviewer #2 (Public Review):

Summary:

The manuscript reports non-invasive measures of activity and neurochemical profiles of the visual cortex in congenitally blind patients who recovered vision through the surgical removal of bilateral dense cataracts. The declared aim of the study is to find out how restoring visual function after several months or years of complete blindness impacts the balance between excitation and inhibition in the visual cortex.

Strengths:

The findings are undoubtedly useful for the community, as they contribute towards characterising the many ways this special population differs from normally sighted individuals. The combination of MRS and EEG measures is a promising strategy to estimate a fundamental physiological parameter - the balance between excitation and inhibition in the visual cortex, which animal studies show to be heavily dependent upon early visual experience. Thus, the reported results pave the way for further studies, which may use a similar approach to evaluate more patients and control groups.

Weaknesses:

(2.1) The main issue is the lack of an appropriate comparison group or condition to delineate the effect of sight recovery (as opposed to the effect of congenital blindness). Few previous studies suggested an increased excitation/Inhibition ratio in the visual cortex of congenitally blind patients; the present study reports a decreased E/I ratio instead. The authors claim that this implies a change of E/I ratio following sight recovery. However, supporting this claim would require showing a shift of E/I after vs. before the sight-recovery surgery, or at least it would require comparing patients who did and did not undergo the sight-recovery surgery (as common in the field).

Longitudinal studies would indeed be the best way to test the hypothesis that the lower E/I ratio in the CC group observed by the present study is a consequence of sight restoration. However, longitudinal studies involving neuroimaging are an effortful challenge, particularly in research conducted outside of major developed countries and dedicated neuroimaging research facilities. Crucially, however, had CC and SC individuals, as well as permanently congenitally blind vs SC individuals (Coullon et al., 2015; Weaver et al., 2013), not differed on any neurochemical markers, such a longitudinal study might have been trivial. Thus, in order to justify and better tailor longitudinal studies, cross-sectional studies are an initial step.

(2.2) MR Spectroscopy shows a reduced GLX/GABA ratio in patients vs. sighted controls; however, this finding remains rather isolated, not corroborated by other observations. The difference between patients and controls only emerges for the GLX/GABA ratio, but there is no accompanying difference in either the GLX or the GABA concentrations. There is an attempt to relate the MRS data with acuity measurements and electrophysiological indices, but the explorative correlational analyses do not help to build a coherent picture. A bland correlation between GLX/GABA and visual impairment is reported, but this is specific to the patients' group (N=10) and would not hold across groups (the correlation is positive, predicting the lowest GLX/GABA ratio values for the sighted controls - the opposite of what is found). There is also a strong correlation between GLX concentrations and the EEG power at the lowest temporal frequencies. Although this relation is intriguing, it only holds for a very specific combination of parameters (of the many tested): only with eyes open, only in the patient group.

We interpret these findings differently, that is, in the context of experiments from non-human animals and the larger MRS literature.

Homeostatic control of E/I balance assumes that the ratio of excitation (reflected here by Glx) and inhibition (reflected here by GABA+) is regulated. Like prior work (Gao et al., 2024, 2024; Narayan et al., 2022; Perica et al., 2022; Steel et al., 2020; Takado et al., 2022; Takei et al., 2016), we assumed that the ratio of Glx/GABA+ is indicative of E/I balance rather than solely the individual neurotransmitter levels. One of the motivations for assessing the ratio vs the absolute concentration is that as per the underlying E/I balance hypothesis, a change in excitation would cause a concomitant change in inhibition, and vice versa, which has been shown in non-human animal work (Fang et al., 2021; Haider et al., 2006; Tao & Poo, 2005) and modeling research (Vreeswijk & Sompolinsky, 1996; Wu et al., 2022). Importantly, our interpretation of the lower E/I ratio is not just from the Glx/GABA+ ratio, but additionally, based on the steeper EEG aperiodic slope (1-20 Hz).  

As in the discussion section and response 1.4, we did not expect to see a lower Glx/GABA+ ratio in CC individuals. We discuss the possible reasons for the direction of the correlation with visual acuity and aperiodic offset during passive visual stimulation, and offer interpretations and (testable) hypotheses.

We interpret the direction of the  Glx/GABA+ correlation with visual acuity to imply that patients with highest (compensatory) balancing of the consequences of congenital blindness (hyperexcitation), in light of visual stimulation, are those who recover best. Note, the sighted control group was selected based on their “normal” vision. Thus, clinical visual acuity measures are not expected to sufficiently vary, nor have the resolution to show strong correlations with neurophysiological measures. By contrast, the CC group comprised patients highly varying in visual outcomes, and thus were ideal to investigate such correlations.

This holds for the correlation between Glx and the aperiodic intercept, as well. Previous work has suggested that the intercept of the aperiodic activity is associated with broadband spiking activity in neural circuits (Manning et al., 2009). Thus, an atypical increase of spiking activity during visual stimulation, as indirectly suggested by “old” non-human primate work on visual deprivation (Hyvärinen et al., 1981) might drive a correlation not observed in healthy populations.

In the revised manuscript, we will more clearly indicate in the discussion that these are possible post-hoc interpretations. We argue that given the lack of such studies in humans, it is all the more important that extant data be presented completely, even if the direction of the effects are not as expected.

(2.3) For these reasons, the reported findings do not allow us to draw firm conclusions on the relation between EEG parameters and E/I ratio or on the impact of early (vs. late) visual experience on the excitation/inhibition ratio of the human visual cortex.

Indeed, the correlations we have tested between the E/I ratio and EEG parameters were exploratory, and have been reported as such. The goal of our study was not to compare the effects of early vs. late visual experience. The goal was to study whether early visual experience is necessary for a typical E/I ratio in visual neural circuits. We provided clear evidence in favor of this hypothesis. Thus, the present results suggest the necessity of investigating the effects of late visual deprivation. In fact, such research is missing in permanent blindness as well.

Reviewer #3 (Public Review):

This manuscript examines the impact of congenital visual deprivation on the excitatory/inhibitory (E/I) ratio in the visual cortex using Magnetic Resonance Spectroscopy (MRS) and electroencephalography (EEG) in individuals whose sight was restored. Ten individuals with reversed congenital cataracts were compared to age-matched, normally sighted controls, assessing the cortical E/I balance and its interrelationship to visual acuity. The study reveals that the Glx/GABA ratio in the visual cortex and the intercept and aperiodic signal are significantly altered in those with a history of early visual deprivation, suggesting persistent neurophysiological changes despite visual restoration.

My expertise is in EEG (particularly in the decomposition of periodic and aperiodic activity) and statistical methods. I have several major concerns in terms of methodological and statistical approaches along with the (over)interpretation of the results. These major concerns are detailed below.

(3.1) Variability in visual deprivation:

- The document states a large variability in the duration of visual deprivation (probably also the age at restoration), with significant implications for the sensitivity period's impact on visual circuit development. The variability and its potential effects on the outcomes need thorough exploration and discussion.

We work with a rare, unique patient population, which makes it difficult to systematically assess the effects of different visual histories while maintaining stringent inclusion criteria such as complete patterned visual deprivation at birth. Regardless, we considered the large variance in age at surgery and time since surgery as supportive of our interpretation: group differences were found despite the large variance in duration of visual deprivation. Moreover, the existing variance was used to explore possible associations between behavior and neural measures, as well as neurochemical and EEG measures.

In the revised manuscript, we will detail the advantages and disadvantages of our CC sample, with respect to duration of congenital visual deprivation.

(3.2) Sample size:

- The small sample size is a major concern as it may not provide sufficient power to detect subtle effects and/or overestimate significant effects, which then tend not to generalize to new data. One of the biggest drivers of the replication crisis in neuroscience.

We address the small sample size in our discussion, and make clear that small sample sizes were due to the nature of investigations in special populations. It is worth noting that our EEG results fully align  with those of a larger sample of CC individuals (Ossandón et al., 2023), providing us confidence about their validity and reproducibility. Moreover, our MRS results and correlations of those with EEG parameters were spatially specific to occipital cortex measures, as predicted.

The main problem with the correlation analyses between MRS and EEG measures is that the sample size is simply too small to conduct such an analysis. Moreover, it is unclear from the methods section that this analysis was only conducted in the patient group (which the reviewer assumed from the plots), and not explained why this was done only in the patient group. I would highly recommend removing these correlation analyses.

We marked the correlation analyses as exploratory; note that we do not base most of our discussion on the results of these analyses. As indicated by Reviewer 1, reporting them allows for deriving more precise hypothesis for future studies. It has to be noted that we investigate an extremely rare population, tested outside of major developed economies and dedicated neuroimaging research facilities. In addition to being a rare patient group, these individuals come from poor communities. Therefore, we consider it justified to report these correlations as exploratory, providing direction for future research.

(3.3) Statistical concerns:

- The statistical analyses, particularly the correlations drawn from a small sample, may not provide reliable estimates (see https://www.sciencedirect.com/science/article/pii/S0092656613000858, which clearly describes this problem).

It would undoubtedly be better to have a larger sample size. We nonetheless think it is of value to the research community to publish this dataset, since 10 multimodal data sets from a carefully diagnosed, rare population, representing a human model for the effects of early experience on brain development, are quite a lot.  Sample sizes in prior neuroimaging studies in transient blindness have most often ranged from n = 1 to n = 10. They nevertheless provided valuable direction for future research, and integration of results across multiple studies provides scientific insights.  

Identifying possible group differences was the goal of our study, with the correlations being an exploratory analysis, which we have clearly indicated in the methods, results and discussion.

- Statistical analyses for the MRS: The authors should consider some additional permutation statistics, which are more suitable for small sample sizes. The current statistical model (2x2) design ANOVA is not ideal for such small sample sizes. Moreover, it is unclear why the condition (EO & EC) was chosen as a predictor and not the brain region (visual & frontal) or neurochemicals. Finally, the authors did not provide any information on the alpha level nor any information on correction for multiple comparisons (in the methods section). Finally, even if the groups are matched w.r.t. age, the time between surgery and measurement, the duration of visual deprivation, (and sex?), these should be included as covariates as it has been shown that these are highly related to the measurements of interest (especially for the EEG measurements) and the age range of the current study is large.

In our ANOVA models, the neurochemicals were the outcome variables, and the conditions were chosen as predictors based on prior work suggesting that Glx/GABA+ might vary with eye closure (Kurcyus et al., 2018). The study was designed based on a hypothesis of group differences localized to the occipital cortex, due to visual deprivation. The frontal cortex voxel was chosen to indicate whether these differences were spatially specific. Therefore, we conducted separate ANOVAs based on this study design.

In the revised manuscript, we will add permutation analyses for our outcomes, as well as multiple regression models investigating whether the variance in visual history might have driven these results. Note that in the supplementary materials (S6, S7), we have reported the correlations between visual history metrics and MRS/EEG outcomes.

The alpha level used for the ANOVA models specified in the methods section was 0.05. The alpha level for the exploratory analyses reported in the main manuscript was 0.008, after correcting for (6) multiple comparisons using the Bonferroni correction, also specified in the methods. Note that the p-values following correction are expressed as multiplied by 6, due to most readers assuming an alpha level of 0.05 (see response regarding large p-values).

We used a control group matched for age and sex. Moreover, the controls were recruited and tested in the same institutes, using the same setup. We feel that we followed the gold standards for recruiting a healthy control group for a patient group.

- EEG statistical analyses: The same critique as for the MRS statistical analyses applies to the EEG analysis. In addition: was the 2x3 ANOVA conducted for EO and EC independently? This seems to be inconsistent with the approach in the MRS analyses, in which the authors chose EO & EC as predictors in their 2x2 ANOVA.

The 2x3 ANOVA was not conducted independently for the eyes open/eyes closed condition, the ANOVA conducted on the EEG metrics was 2x3 because it had group (CC, SC) and condition (eyes open (EO), eyes closed (EC) and visual stimulation (LU)) as predictors.

- Figure 4: The authors report a p-value of >0.999 with a correlation coefficient of -0.42 with a sample size of 10 subjects. This can't be correct (it should be around: p = 0.22). All statistical analyses should be checked.

As specified in the methods and figure legend, the reported p values in Figure 4 have been corrected using the Bonferroni correction, and therefore multiplied by the number of comparisons, leading to the seemingly large values.

Additionally, to check all statistical analyses, we put the manuscript through an independent Statistics Check (Nuijten & Polanin, 2020) (https://michelenuijten.shinyapps.io/statcheck-web/) and will upload the consistency report with the revised supplementary material.

- Figure 2c. Eyes closed condition: The highest score of the *Glx/GABA ratio seems to be ~3.6. In subplot 2a, there seem to be 3 subjects that show a Glx/GABA ratio score > 3.6. How can this be explained? There is also a discrepancy for the eyes-closed condition.

The three subjects that show the Glx/GABA+ ratio > 3.6 in subplot 2a are in the SC group, whereas the correlations plotted in figure 2c are only for the CC group, where the highest score is indeed ~3.6.

(3.4) Interpretation of aperiodic signal:

- Several recent papers demonstrated that the aperiodic signal measured in EEG or ECoG is related to various important aspects such as age, skull thickness, electrode impedance, as well as cognition. Thus, currently, very little is known about the underlying effects which influence the aperiodic intercept and slope. The entire interpretation of the aperiodic slope as a proxy for E/I is based on a computational model and simulation (as described in the Gao et al. paper).

Apart from the modeling work from Gao et al., multiple papers which have also been cited which used ECoG, EEG and MEG and showed concomitant changes in aperiodic activity with pharmacological manipulation of the E/I ratio (Colombo et al., 2019; Molina et al., 2020; Muthukumaraswamy & Liley, 2018). Further, several prior studies have interpreted changes in the aperiodic slope as reflective of changes in the E/I ratio, including studies of developmental groups (Favaro et al., 2023; Hill et al., 2022; McSweeney et al., 2023; Schaworonkow & Voytek, 2021) as well as patient groups (Molina et al., 2020; Ostlund et al., 2021).

In the revised manuscript, we will cite those studies not already included in the introduction.

- Especially the aperiodic intercept is a very sensitive measure to many influences (e.g. skull thickness, electrode impedance...). As crucial results (correlation aperiodic intercept and MRS measures) are facing this problem, this needs to be reevaluated. It is safer to make statements on the aperiodic slope than intercept. In theory, some of the potentially confounding measures are available to the authors (e.g. skull thickness can be computed from T1w images; electrode impedances are usually acquired alongside the EEG data) and could be therefore controlled.

All electrophysiological measures indeed depend on parameters such as skull thickness and electrode impedance. As in the extant literature using neurophysiological measures to compare brain function between patient and control groups, we used a control group matched in age/ sex, recruited in the same region, tested with the same devices, and analyzed with the same analysis pipeline. For example, impedance was kept below 10 kOhm for all subjects. There is no evidence available suggesting that congenital cataracts are associated with changes in skull thickness that would cause the observed pattern of group results. Moreover, we cannot think of how any of the exploratory correlations between neurophysiological measures and MRS measures could be accounted for by a difference e.g. in skull thickness.

- The authors wrote: "Higher frequencies (such as 20-40 Hz) have been predominantly associated with local circuit activity and feedforward signaling (Bastos et al., 2018; Van Kerkoerle et al., 2014); the increased 20-40 Hz slope may therefore signal increased spontaneous spiking activity in local networks. We speculate that the steeper slope of the aperiodic activity for the lower frequency range (1-20 Hz) in CC individuals reflects the concomitant increase in inhibition." The authors confuse the interpretation of periodic and aperiodic signals. This section refers to the interpretation of the periodic signal (higher frequencies). This interpretation cannot simply be translated to the aperiodic signal (slope).

Prior work has not always separated the aperiodic and periodic components, making it unclear what might have driven these effects in our data. The interpretation of the higher frequency range was intended to contrast with the interpretations of lower frequency range, in order to speculate as to why the two aperiodic fits might go in differing directions. We will clarify our interpretation in the revised manuscript. Note that Ossandon et al. reported highly similar results (group differences for CC individuals and for permanently congenitally blind humans) for the aperiodic activity between 20-40 Hz and oscillatory activity in the gamma range. We will allude to these findings in the revised manuscript.

- The authors further wrote: We used the slope of the aperiodic (1/f) component of the EEG spectrum as an estimate of E/I ratio (Gao et al., 2017; Medel et al., 2020; Muthukumaraswamy & Liley, 2018). This is a highly speculative interpretation with very little empirical evidence. These papers were conducted with ECoG data (mostly in animals) and mostly under anesthesia. Thus, these studies only allow an indirect interpretation by what the 1/f slope in EEG measurements is actually influenced.

Note that Muthukumaraswamy et al. (2018) used different types of pharmacological manipulations and analyzed periodic and aperiodic MEG activity in addition to monkey ECoG (Medel et al., 2020) (now published as (Medel et al., 2023)) compared EEG activity in addition to ECoG data after propofol administration. The interpretation of our results are in line with a number of recent studies in developing (Hill et al., 2022; Schaworonkow & Voytek, 2021) and special populations using EEG. As mentioned above, several prior studies have used the slope of the 1/f component/aperiodic activity as an indirect measure of the E/I ratio (Favaro et al., 2023; Hill et al., 2022; McSweeney et al., 2023; Molina et al., 2020; Ostlund et al., 2021; Schaworonkow & Voytek, 2021), including studies using scalp-recorded EEG. We will make more clear in the introduction of the revised manuscript that this metric is indirect.

While a full understanding of aperiodic activity needs to be provided, some convergent ideas have emerged . We think that our results contribute to this enterprise, since our study is, to the best of our knowledge, the first which assessed MRS measured neurotransmitter levels and EEG aperiodic activity.

(3.5) Problems with EEG preprocessing and analysis:

- It seems that the authors did not identify bad channels nor address the line noise issue (even a problem if a low pass filter of below-the-line noise was applied).

As pointed out in the methods and Figure 1, we only analyzed data from two channels, O1 and O2, neither of which were rejected for any participant. Channel rejection was performed for the larger dataset, published elsewhere (Ossandón et al., 2023; Pant et al., 2023).

In both published works, we did not consider frequency ranges above 40 Hz to avoid any possible contamination with line noise. Here, we focused on activity between 0 and 20 Hz, definitely excluding line noise contaminations. The low pass filter (FIR, 1-45 Hz) guaranteed that any spill-over effects of line noise would be restricted to frequencies just below the upper cutoff frequency.

Additionally, a prior version of the analysis used the cleanline.m function to remove line noise before filtering, and the group differences remained stable. We will report this analysis in the supplementary version of the revised manuscript. Further, both groups were measured in the same lab, making line noise as an account for the observed group effects highly unlikely. Finally, any of the exploratory MRS-EEG correlations would be hard to explain if the EEG parameters would be contaminated with line noise.

- What was the percentage of segments that needed to be rejected due to the 120μV criteria? This should be reported specifically for EO & EC and controls and patients.

The mean percentage of 1 second segments rejected for each resting state condition is below. Mean percentage of 6.25 long segments rejected in each group for the visual stimulation condition are also included, and will be added to the revised manuscript:

Author response table 3.

- The authors downsampled the data to 60Hz to "to match the stimulation rate". What is the intention of this? Because the subsequent spectral analyses are conflated by this choice (see Nyquist theorem).

This data were collected as part of a study designed to evoke alpha activity with visual white-noise, which ranged in luminance with equal power at all frequencies from 1-60 Hz, restricted by the refresh rate of the monitor on which stimuli were presented (Pant et al., 2023). This paradigm and method was developed by VanRullen and colleagues (Schwenk et al., 2020; Vanrullen & MacDonald, 2012), wherein the analysis requires the same sampling rate between the presented frequencies and the EEG data. The downsampling function used here automatically applies an anti-aliasing filter (EEGLAB 2019) .

- "Subsequently, baseline removal was conducted by subtracting the mean activity across the length of an epoch from every data point." The actual baseline time segment should be specified.

The time segment was the length of the epoch, that is, 1 second for the resting state conditions and 6.25 seconds for the visual stimulation conditions. This will be explicitly stated in the revised manuscript.

- "We excluded the alpha range (8-14 Hz) for this fit to avoid biasing the results due to documented differences in alpha activity between CC and SC individuals (Bottari et al., 2016; Ossandón et al., 2023; Pant et al., 2023)." This does not really make sense, as the FOOOF algorithm first fits the 1/f slope, for which the alpha activity is not relevant.

We did not use the FOOOF algorithm/toolbox in this manuscript. As stated in the methods, we used a 1/f fit to the 1-20 Hz spectrum in the log-log space, and subtracted this fit from the original spectrum to obtain the corrected spectrum. Given the pronounced difference in alpha power between groups (Bottari et al., 2016; Ossandón et al., 2023; Pant et al., 2023), we were concerned it might drive differences in the exponent values.  Our analysis pipeline had been adapted from previous publications of our group and other labs (Ossandón et al., 2023; Voytek et al., 2015; Waschke et al., 2017).

We have conducted the analysis with and without the exclusion of the alpha range, as well as using the FOOOF toolbox both in the 1-20 Hz and 20-40 Hz ranges (Ossandón et al., 2023); The findings of a steeper slope in the 1-20 Hz range as well as lower alpha power in CC vs SC individuals remained stable. In Ossandón et al., the comparison between the piecewise fits and FOOOF fits led the authors to use the former as it outperformed the FOOOF algorithm for their data.

- The model fits of the 1/f fitting for EO, EC, and both participant groups should be reported.

In Figure 3 of the manuscript, we depicted the mean spectra and 1/f fits for each group. We will add the fit quality metrics and show individual subjects’ fits in the revised manuscript.

(3.6) Validity of GABA measurements and results:

- According the a newer study by the authors of the Gannet toolbox (https://analyticalsciencejournals.onlinelibrary.wiley.com/doi/abs/10.1002/nbm.5076), the reliability and reproducibility of the gamma-aminobutyric acid (GABA) measurement can vary significantly depending on acquisition and modeling parameter. Thus, did the author address these challenges?

We took care of data quality while acquiring MRS data by ensuring appropriate voxel placement and linewidth prior to scanning. Acquisition as well as modeling parameters were constant for both groups, so they cannot have driven group differences.

The linked article compares the reproducibility of GABA measurement using Osprey, which was released in 2020 and uses linear combination modeling to fit the peak as opposed to Gannet’s simple peak fitting (Hupfeld et al., 2024). The study finds better test-retest reliability for Osprey compared to Gannet’s method.

As the present work was conceptualized in 2018, we used Gannet 3.0, which was the state-of-the-art edited spectral analysis toolbox at the time, and still is widely used. In the revised manuscript, we will include a supplementary section reanalyzing the main findings with Osprey.

- Furthermore, the authors wrote: "We confirmed the within-subject stability of metabolite quantification by testing a subset of the sighted controls (n=6) 2-4 weeks apart. Looking at the supplementary Figure 5 (which would be rather plotted as ICC or Blant-Altman plots), the within-subject stability compared to between-subject variability seems not to be great. Furthermore, I don't think such a small sample size qualifies for a rigorous assessment of stability.

Indeed, we did not intend to provide a rigorous assessment of within-subject stability. Rather, we aimed to confirm that data quality/concentration ratios did not systematically differ between the same subjects tested longitudinally; driven, for example, by scanner heating or time of day. As with the phantom testing, we attempted to give readers an idea of the quality of the data, as they were collected from a primarily clinical rather than a research site.

In the revised manuscript we will remove the statement regarding stability, and add the Blant-Altman plot.

- "Why might an enhanced inhibitory drive, as indicated by the lower Glx/GABA ratio" Is this interpretation really warranted, as the results of the group differences in the Glx/GABA ratio seem to be rather driven by a decreased Glx concentration in CC rather than an increased GABA (see Figure 2).

We used the Glx/GABA+ ratio as a measure, rather than individual Glx or GABA+ concentration, which did not significantly differ between groups. As detailed in Response 2.2, we think this metric aligns better with an underlying E/I balance hypothesis and has been used in many previous studies (Gao et al., 2024; Liu et al., 2015; Narayan et al., 2022; Perica et al., 2022).

Our interpretation of an enhanced inhibitory drive additionally comes from the combination of aperiodic EEG (1-20 Hz) and MRS measures, which, when considered together, are consistent with a decreased E/I ratio.

In the revised manuscript, we will rephrase this sentence accordingly. 

- Glx concentration predicted the aperiodic intercept in CC individuals' visual cortices during ambient and flickering visual stimulation. Why specifically investigate the Glx concentration, when the paper is about E/I ratio?

As stated in the methods, we exploratorily assessed the relationship between all MRS parameters (Glx, GABA+ and Glx/GABA+ ratio) with the aperiodic parameters (slope, offset), and corrected for multiple comparisons accordingly. We think this is a worthwhile analysis considering the rarity of the dataset/population (see 1.2, 1.6, 2.1 and reviewer 1’s comments about future hypotheses). We only report the Glx – aperiodic intercept correlation in the main manuscript as it survived correction for multiple comparisons.

(3.7) Interpretation of the correlation between MRS measurements and EEG aperiodic signal:

- The authors wrote: "The intercept of the aperiodic activity was highly correlated with the Glx concentration during rest with eyes open and during flickering stimulation (also see Supplementary Material S11). Based on the assumption that the aperiodic intercept reflects broadband firing (Manning et al., 2009; Winawer et al., 2013), this suggests that the Glx concentration might be related to broadband firing in CC individuals during active and passive visual stimulation." These results should not be interpreted (or with very caution) for several reasons (see also problem with influences on aperiodic intercept and small sample size). This is a result of the exploratory analyses of correlating every EEG parameter with every MRS parameter. This requires well-powered replication before any interpretation can be provided. Furthermore and importantly: why should this be specifically only in CC patients, but not in the SC control group?

We indicate clearly in all parts of the manuscript that these correlations are presented as exploratory. Further, we interpret the Glx-aperiodic offset correlation, and none of the others, as it survived the Bonferroni correction for multiple comparisons. We offer a hypothesis in the discussion section as to why such a correlation might exist in the CC but not the SC group (see response 2.2), and do not speculate further.

(3.8) Language and presentation:

- The manuscript requires language improvements and correction of numerous typos. Over-simplifications and unclear statements are present, which could mislead or confuse readers (see also interpretation of aperiodic signal).

In the revision, we will check that speculations are clearly marked and typos are removed.

- The authors state that "Together, the present results provide strong evidence for experience-dependent development of the E/I ratio in the human visual cortex, with consequences for behavior." The results of the study do not provide any strong evidence, because of the small sample size and exploratory analyses approach and not accounting for possible confounding factors.

We disagree with this statement and allude to convergent evidence of both MRS and neurophysiological measures. The latter link to corresponding results observed in a larger sample of CC individuals (Ossandón et al., 2023).

- "Our results imply a change in neurotransmitter concentrations as a consequence of *restoring* vision following congenital blindness." This is a speculative statement to infer a causal relationship on cross-sectional data.

As mentioned under 2.1, we conducted a cross-sectional study which might justify future longitudinal work. In order to advance science, new testable hypotheses were put forward at the end of a manuscript.

In the revised manuscript we will add “might imply” to better indicate the hypothetical character of this idea.

- In the limitation section, the authors wrote: "The sample size of the present study is relatively high for the rare population , but undoubtedly, overall, rather small." This sentence should be rewritten, as the study is plein underpowered. The further justification "We nevertheless think that our results are valid. Our findings neurochemically (Glx and GABA+ concentration), and anatomically (visual cortex) specific. The MRS parameters varied with parameters of the aperiodic EEG activity and visual acuity. The group differences for the EEG assessments corresponded to those of a larger sample of CC individuals (n=38) (Ossandón et al., 2023), and effects of chronological age were as expected from the literature." These statements do not provide any validation or justification of small samples. Furthermore, the current data set is a subset of an earlier published paper by the same authors "The EEG data sets reported here were part of data published earlier (Ossandón et al., 2023; Pant et al., 2023)." Thus, the statement "The group differences for the EEG assessments corresponded to those of a larger sample of CC individuals (n=38) " is a circular argument and should be avoided.

Our intention was not to justify having a small sample, but to justify why we think the results might be valid as they align with/replicate existing literature.

In the revised manuscript, we will add a figure showing that the EEG results of the 10 subjects considered here correspond to those of the 28 other subjects of Ossandon et al. We will adapt the text accordingly, clearly stating that the pattern of EEG results of the ten subjects reported here replicate those of the 28 additional subjects of Ossandon et al. (2023).

References

Barnes, S. J., Sammons, R. P., Jacobsen, R. I., Mackie, J., Keller, G. B., & Keck, T. (2015). Subnetwork-specific homeostatic plasticity in mouse visual cortex in vivo. Neuron, 86(5), 1290–1303. https://doi.org/10.1016/J.NEURON.2015.05.010

Bernabeu, A., Alfaro, A., García, M., & Fernández, E. (2009). Proton magnetic resonance spectroscopy (1H-MRS) reveals the presence of elevated myo-inositol in the occipital cortex of blind subjects. NeuroImage, 47(4), 1172–1176. https://doi.org/10.1016/j.neuroimage.2009.04.080

Bottari, D., Troje, N. F., Ley, P., Hense, M., Kekunnaya, R., & Röder, B. (2016). Sight restoration after congenital blindness does not reinstate alpha oscillatory activity in humans. Scientific Reports. https://doi.org/10.1038/srep24683

Colombo, M. A., Napolitani, M., Boly, M., Gosseries, O., Casarotto, S., Rosanova, M., Brichant, J. F., Boveroux, P., Rex, S., Laureys, S., Massimini, M., Chieregato, A., & Sarasso, S. (2019). The spectral exponent of the resting EEG indexes the presence of consciousness during unresponsiveness induced by propofol, xenon, and ketamine. NeuroImage, 189(September 2018), 631–644. https://doi.org/10.1016/j.neuroimage.2019.01.024

Consideration of Sample Size in Neuroscience Studies. (2020). Journal of Neuroscience, 40(21), 4076–4077. https://doi.org/10.1523/JNEUROSCI.0866-20.2020

Coullon, G. S. L., Emir, U. E., Fine, I., Watkins, K. E., & Bridge, H. (2015). Neurochemical changes in the pericalcarine cortex in congenital blindness attributable to bilateral anophthalmia. Journal of Neurophysiology. https://doi.org/10.1152/jn.00567.2015

Fang, Q., Li, Y. T., Peng, B., Li, Z., Zhang, L. I., & Tao, H. W. (2021). Balanced enhancements of synaptic excitation and inhibition underlie developmental maturation of receptive fields in the mouse visual cortex. Journal of Neuroscience, 41(49), 10065–10079. https://doi.org/10.1523/JNEUROSCI.0442-21.2021

Favaro, J., Colombo, M. A., Mikulan, E., Sartori, S., Nosadini, M., Pelizza, M. F., Rosanova, M., Sarasso, S., Massimini, M., & Toldo, I. (2023). The maturation of aperiodic EEG activity across development reveals a progressive differentiation of wakefulness from sleep. NeuroImage, 277. https://doi.org/10.1016/J.NEUROIMAGE.2023.120264

Gao, Y., Liu, Y., Zhao, S., Liu, Y., Zhang, C., Hui, S., Mikkelsen, M., Edden, R. A. E., Meng, X., Yu, B., & Xiao, L. (2024). MRS study on the correlation between frontal GABA+/Glx ratio and abnormal cognitive function in medication-naive patients with narcolepsy. Sleep Medicine, 119, 1–8. https://doi.org/10.1016/j.sleep.2024.04.004

Haider, B., Duque, A., Hasenstaub, A. R., & McCormick, D. A. (2006). Neocortical network activity in vivo is generated through a dynamic balance of excitation and inhibition. Journal of Neuroscience. https://doi.org/10.1523/JNEUROSCI.5297-05.2006

Hill, A. T., Clark, G. M., Bigelow, F. J., Lum, J. A. G., & Enticott, P. G. (2022). Periodic and aperiodic neural activity displays age-dependent changes across early-to-middle childhood. Developmental Cognitive Neuroscience, 54, 101076. https://doi.org/10.1016/J.DCN.2022.101076

Hupfeld, K. E., Zöllner, H. J., Hui, S. C. N., Song, Y., Murali-Manohar, S., Yedavalli, V., Oeltzschner, G., Prisciandaro, J. J., & Edden, R. A. E. (2024). Impact of acquisition and modeling parameters on the test–retest reproducibility of edited GABA+. NMR in Biomedicine, 37(4), e5076. https://doi.org/10.1002/nbm.5076

Hyvärinen, J., Carlson, S., & Hyvärinen, L. (1981). Early visual deprivation alters modality of neuronal responses in area 19 of monkey cortex. Neuroscience Letters, 26(3), 239–243. https://doi.org/10.1016/0304-3940(81)90139-7

Juchem, C., & Graaf, R. A. de. (2017). B0 magnetic field homogeneity and shimming for in vivo magnetic resonance spectroscopy. Analytical Biochemistry, 529, 17–29. https://doi.org/10.1016/j.ab.2016.06.003

Keck, T., Hübener, M., & Bonhoeffer, T. (2017). Interactions between synaptic homeostatic mechanisms: An attempt to reconcile BCM theory, synaptic scaling, and changing excitation/inhibition balance. Current Opinion in Neurobiology, 43, 87–93. https://doi.org/10.1016/J.CONB.2017.02.003

Kurcyus, K., Annac, E., Hanning, N. M., Harris, A. D., Oeltzschner, G., Edden, R., & Riedl, V. (2018). Opposite Dynamics of GABA and Glutamate Levels in the Occipital Cortex during Visual Processing. Journal of Neuroscience, 38(46), 9967–9976. https://doi.org/10.1523/JNEUROSCI.1214-18.2018

Liu, B., Wang, G., Gao, D., Gao, F., Zhao, B., Qiao, M., Yang, H., Yu, Y., Ren, F., Yang, P., Chen, W., & Rae, C. D. (2015). Alterations of GABA and glutamate-glutamine levels in premenstrual dysphoric disorder: A 3T proton magnetic resonance spectroscopy study. Psychiatry Research - Neuroimaging, 231(1), 64–70. https://doi.org/10.1016/J.PSCYCHRESNS.2014.10.020

Lunghi, C., Berchicci, M., Morrone, M. C., & Russo, F. D. (2015). Short‐term monocular deprivation alters early components of visual evoked potentials. The Journal of Physiology, 593(19), 4361. https://doi.org/10.1113/JP270950

Maier, S., Düppers, A. L., Runge, K., Dacko, M., Lange, T., Fangmeier, T., Riedel, A., Ebert, D., Endres, D., Domschke, K., Perlov, E., Nickel, K., & Tebartz van Elst, L. (2022). Increased prefrontal GABA concentrations in adults with autism spectrum disorders. Autism Research, 15(7), 1222–1236. https://doi.org/10.1002/aur.2740

Manning, J. R., Jacobs, J., Fried, I., & Kahana, M. J. (2009). Broadband shifts in local field potential power spectra are correlated with single-neuron spiking in humans. The Journal of Neuroscience : The Official Journal of the Society for Neuroscience, 29(43), 13613–13620. https://doi.org/10.1523/JNEUROSCI.2041-09.2009

McSweeney, M., Morales, S., Valadez, E. A., Buzzell, G. A., Yoder, L., Fifer, W. P., Pini, N., Shuffrey, L. C., Elliott, A. J., Isler, J. R., & Fox, N. A. (2023). Age-related trends in aperiodic EEG activity and alpha oscillations during early- to middle-childhood. NeuroImage, 269, 119925. https://doi.org/10.1016/j.neuroimage.2023.119925

Medel, V., Irani, M., Crossley, N., Ossandón, T., & Boncompte, G. (2023). Complexity and 1/f slope jointly reflect brain states. Scientific Reports, 13(1), 21700. https://doi.org/10.1038/s41598-023-47316-0

Medel, V., Irani, M., Ossandón, T., & Boncompte, G. (2020). Complexity and 1/f slope jointly reflect cortical states across different E/I balances. bioRxiv, 2020.09.15.298497. https://doi.org/10.1101/2020.09.15.298497

Molina, J. L., Voytek, B., Thomas, M. L., Joshi, Y. B., Bhakta, S. G., Talledo, J. A., Swerdlow, N. R., & Light, G. A. (2020). Memantine Effects on Electroencephalographic Measures of Putative Excitatory/Inhibitory Balance in Schizophrenia. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging, 5(6), 562–568. https://doi.org/10.1016/j.bpsc.2020.02.004

Mukerji, A., Byrne, K. N., Yang, E., Levi, D. M., & Silver, M. A. (2022). Visual cortical γ−aminobutyric acid and perceptual suppression in amblyopia. Frontiers in Human Neuroscience, 16. https://doi.org/10.3389/fnhum.2022.949395

Muthukumaraswamy, S. D., & Liley, D. T. (2018). 1/F electrophysiological spectra in resting and drug-induced states can be explained by the dynamics of multiple oscillatory relaxation processes. NeuroImage, 179(November 2017), 582–595. https://doi.org/10.1016/j.neuroimage.2018.06.068

Narayan, G. A., Hill, K. R., Wengler, K., He, X., Wang, J., Yang, J., Parsey, R. V., & DeLorenzo, C. (2022). Does the change in glutamate to GABA ratio correlate with change in depression severity? A randomized, double-blind clinical trial. Molecular Psychiatry, 27(9), 3833—3841. https://doi.org/10.1038/s41380-022-01730-4

Nuijten, M. B., & Polanin, J. R. (2020). “statcheck”: Automatically detect statistical reporting inconsistencies to increase reproducibility of meta-analyses. Research Synthesis Methods, 11(5), 574–579. https://doi.org/10.1002/jrsm.1408

Ossandón, J. P., Stange, L., Gudi-Mindermann, H., Rimmele, J. M., Sourav, S., Bottari, D., Kekunnaya, R., & Röder, B. (2023). The development of oscillatory and aperiodic resting state activity is linked to a sensitive period in humans. NeuroImage, 275, 120171. https://doi.org/10.1016/J.NEUROIMAGE.2023.120171

Ostlund, B. D., Alperin, B. R., Drew, T., & Karalunas, S. L. (2021). Behavioral and cognitive correlates of the aperiodic (1/f-like) exponent of the EEG power spectrum in adolescents with and without ADHD. Developmental Cognitive Neuroscience, 48, 100931. https://doi.org/10.1016/j.dcn.2021.100931

Pant, R., Ossandón, J., Stange, L., Shareef, I., Kekunnaya, R., & Röder, B. (2023). Stimulus-evoked and resting-state alpha oscillations show a linked dependence on patterned visual experience for development. NeuroImage: Clinical, 103375. https://doi.org/10.1016/J.NICL.2023.103375

Perica, M. I., Calabro, F. J., Larsen, B., Foran, W., Yushmanov, V. E., Hetherington, H., Tervo-Clemmens, B., Moon, C.-H., & Luna, B. (2022). Development of frontal GABA and glutamate supports excitation/inhibition balance from adolescence into adulthood. Progress in Neurobiology, 219, 102370. https://doi.org/10.1016/j.pneurobio.2022.102370

Pitchaimuthu, K., Wu, Q. Z., Carter, O., Nguyen, B. N., Ahn, S., Egan, G. F., & McKendrick, A. M. (2017). Occipital GABA levels in older adults and their relationship to visual perceptual suppression. Scientific Reports, 7(1). https://doi.org/10.1038/S41598-017-14577-5

Rideaux, R., Ehrhardt, S. E., Wards, Y., Filmer, H. L., Jin, J., Deelchand, D. K., Marjańska, M., Mattingley, J. B., & Dux, P. E. (2022). On the relationship between GABA+ and glutamate across the brain. NeuroImage, 257, 119273. https://doi.org/10.1016/J.NEUROIMAGE.2022.119273

Schaworonkow, N., & Voytek, B. (2021). Longitudinal changes in aperiodic and periodic activity in electrophysiological recordings in the first seven months of life. Developmental Cognitive Neuroscience, 47. https://doi.org/10.1016/j.dcn.2020.100895

Schwenk, J. C. B., VanRullen, R., & Bremmer, F. (2020). Dynamics of Visual Perceptual Echoes Following Short-Term Visual Deprivation. Cerebral Cortex Communications, 1(1). https://doi.org/10.1093/TEXCOM/TGAA012

Sengpiel, F., Jirmann, K.-U., Vorobyov, V., & Eysel, U. T. (2006). Strabismic Suppression Is Mediated by Inhibitory Interactions in the Primary Visual Cortex. Cerebral Cortex, 16(12), 1750–1758. https://doi.org/10.1093/cercor/bhj110

Steel, A., Mikkelsen, M., Edden, R. A. E., & Robertson, C. E. (2020). Regional balance between glutamate+glutamine and GABA+ in the resting human brain. NeuroImage, 220. https://doi.org/10.1016/J.NEUROIMAGE.2020.117112

Takado, Y., Takuwa, H., Sampei, K., Urushihata, T., Takahashi, M., Shimojo, M., Uchida, S., Nitta, N., Shibata, S., Nagashima, K., Ochi, Y., Ono, M., Maeda, J., Tomita, Y., Sahara, N., Near, J., Aoki, I., Shibata, K., & Higuchi, M. (2022). MRS-measured glutamate versus GABA reflects excitatory versus inhibitory neural activities in awake mice. Journal of Cerebral Blood Flow & Metabolism, 42(1), 197. https://doi.org/10.1177/0271678X211045449

Takei, Y., Fujihara, K., Tagawa, M., Hironaga, N., Near, J., Kasagi, M., Takahashi, Y., Motegi, T., Suzuki, Y., Aoyama, Y., Sakurai, N., Yamaguchi, M., Tobimatsu, S., Ujita, K., Tsushima, Y., Narita, K., & Fukuda, M. (2016). The inhibition/excitation ratio related to task-induced oscillatory modulations during a working memory task: A multtimodal-imaging study using MEG and MRS. NeuroImage, 128, 302–315. https://doi.org/10.1016/J.NEUROIMAGE.2015.12.057

Tao, H. W., & Poo, M. M. (2005). Activity-dependent matching of excitatory and inhibitory inputs during refinement of visual receptive fields. Neuron, 45(6), 829–836. https://doi.org/10.1016/J.NEURON.2005.01.046

Vanrullen, R., & MacDonald, J. S. P. (2012). Perceptual echoes at 10 Hz in the human brain. Current Biology. https://doi.org/10.1016/j.cub.2012.03.050

Voytek, B., Kramer, M. A., Case, J., Lepage, K. Q., Tempesta, Z. R., Knight, R. T., & Gazzaley, A. (2015). Age-related changes in 1/f neural electrophysiological noise. Journal of Neuroscience, 35(38). https://doi.org/10.1523/JNEUROSCI.2332-14.2015

Vreeswijk, C. V., & Sompolinsky, H. (1996). Chaos in neuronal networks with balanced excitatory and inhibitory activity. Science, 274(5293), 1724–1726. https://doi.org/10.1126/SCIENCE.274.5293.1724

Waschke, L., Wöstmann, M., & Obleser, J. (2017). States and traits of neural irregularity in the age-varying human brain. Scientific Reports 2017 7:1, 7(1), 1–12. https://doi.org/10.1038/s41598-017-17766-4

Weaver, K. E., Richards, T. L., Saenz, M., Petropoulos, H., & Fine, I. (2013). Neurochemical changes within human early blind occipital cortex. Neuroscience. https://doi.org/10.1016/j.neuroscience.2013.08.004

Wu, Y. K., Miehl, C., & Gjorgjieva, J. (2022). Regulation of circuit organization and function through inhibitory synaptic plasticity. Trends in Neurosciences, 45(12), 884–898. https://doi.org/10.1016/J.TINS.2022.10.006

Author response:

Reviewer #1 (Public review):

(1) Legionella effectors are often activated by binding to eukaryote-specific host factors, including actin. The authors should test the following: a) whether Lfat1 can fatty acylate small G-proteins in vitro; b) whether this activity is dependent on actin binding; and c) whether expression of the Y240A mutant in mammalian cells affects the fatty acylation of Rac3 (Figure 6B), or other small G-proteins.

We were not able to express and purify the full-length recombinant Lfat1 to perform fatty acylation of small GTPases in vitro. However, in cellulo overexpression of the Y240A mutant still retained ability to fatty acylate Rac3 and another small GTPase RheB (see Author response image 1 below). We postulate that under infection conditions, actin-binding might be required to fatty acylate certain GTPases due to the small amount of effector proteins that secreted into the host cell.

Author response image 1.

(2) It should be demonstrated that lysine residues on small G-proteins are indeed targeted by Lfat1. Ideally, the functional consequences of these modifications should also be investigated. For example, does fatty acylation of G-proteins affect GTPase activity or binding to downstream effectors?

We have mutated K178 on RheB and showed that this mutation abolished its fatty acylation by Lfat1 (see Author response image 2 below). We were not able to test if fatty acylation by Lfat1 affect downstream effector binding.

Author response image 2.

(3) Line 138: Can the authors clarify whether the Lfat1 ABD induces bundling of F-actin filaments or promotes actin oligomerization? Does the Lfat1 ABD form multimers that bring multiple filaments together? If Lfat1 induces actin oligomerization, this effect should be experimentally tested and reported. Additionally, the impact of Lfat1 binding on actin filament stability should be assessed. This is particularly important given the proposed use of the ABD as an actin probe.

The ABD domain does not form oligomer as evidenced by gel filtration profile of the ABD domain. However, we do see F-actin bundling in our in vitro -F-actin polymerization experiment when both actin and ABD are in high concentration (data not shown). Under low concentration of ABD, there is not aggregation/bundling effect of F-actin.

(4) Line 180: I think it's too premature to refer to the interaction as having "high specificity and affinity." We really don't know what else it's binding to.

We have revised the text and reworded the sentence by removing "high specificity and affinity."

(5) The authors should reconsider the color scheme used in the structural figures, particularly in Figures 2D and S4.

Not sure the comments on the color scheme of the structure figures.

(6) In Figure 3E, the WT curve fits the data poorly, possibly because the actin concentration exceeds the Kd of the interaction. It might fit better to a quadratic.

We have performed quadratic fitting and replaced Figure 3E.

(7) The authors propose that the individual helices of the Lfat1 ABD could be expressed on separate proteins and used to target multi-component biological complexes to F-actin by genetically fusing each component to a split alpha-helix. This is an intriguing idea, but it should be tested as a proof of concept to support its feasibility and potential utility.

It is a good suggestion. We plan to thoroughly test the feasibility of this idea as one of our future directions.

(7) The plot in Figure S2D appears cropped on the X-axis or was generated from a ~2× binned map rather than the deposited one (pixel size ~0.83 Å, plot suggests ~1.6 Å). The reported pixel size is inconsistent between the Methods and Table 1-please clarify whether 0.83 Å refers to super-resolution.

Yes, 0.83 Å is super-resolution. We have updated in the cryoEM table

Reviewer #2 (Public review):

Weaknesses:

(1) The authors should use biochemical reactions to analyze the KFAT of Llfat1 on one or two small GTPases shown to be modified by this effector in cellulo. Such reactions may allow them to determine the role of actin binding in its biochemical activity. This notion is particularly relevant in light of recent studies that actin is a co-factor for the activity of LnaB and Ceg14 (PMID: 39009586; PMID: 38776962; PMID: 40394005). In addition, the study should be discussed in the context of these recent findings on the role of actin in the activity of L. pneumophila effectors.

We have new data showed that Actin binding does not affect Lfat1 enzymatic activity. (see figure; response to Reviewer #1). We have added this new data as Figure S7 to the paper. Accordingly, we also revised the discussion by adding the following paragraph.

“The discovery of Lfat1 as an F-actin–binding lysine fatty acyl transferase raised the intriguing question of whether its enzymatic activity depends on F-actin binding. Recent studies have shown that other Legionella effectors, such as LnaB and Ceg14, use actin as a co-factor to regulate their activities. For instance, LnaB binds monomeric G-actin to enhance its phosphoryl-AMPylase activity toward phosphorylated residues, resulting in unique ADPylation modifications in host proteins (Fu et al, 2024; Wang et al, 2024). Similarly, Ceg14 is activated by host actin to convert ATP and dATP into adenosine and deoxyadenosine monophosphate, thereby modulating ATP levels in L. pneumophila–infected cells (He et al, 2025). However, this does not appear to be the case for Lfat1. We found that Lfat1 mutants defective in F-actin binding retained the ability to modify host small GTPases when expressed in cells (Figure S7). These findings suggest that, rather than serving as a co-factor, F-actin may serve to localize Lfat1 via its actin-binding domain (ABD), thereby confining its activity to regions enriched in F-actin and enabling spatial specificity in the modification of host targets.”

(2) The development of the ABD domain of Llfat1 as an F-actin domain is a nice extension of the biochemical and structural experiments. The authors need to compare the new probe to those currently commonly used ones, such as Lifeact, in labeling of the actin cytoskeleton structure.

We fully agree with the reviewer’s insightful suggestion. However, a direct comparison of the Lfat1 ABD domain with commonly used actin probes such as Lifeact, as well as evaluation of the split α-helix probe (as suggested by Reviewer #1), would require extensive and technically demanding experiments. These are important directions that we plan to pursue in future studies.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

This study reveals that TRPV1 signaling plays a key role in tympanic membrane (TM) healing by promoting macrophage recruitment and angiogenesis. Using a mouse TM perforation model, researchers found that blood-derived macrophages accumulated near the wound, driving angiogenesis and repair. TRPV1-expressing nerve fibers triggered neuroinflammatory responses, facilitating macrophage recruitment. Genetic Trpv1 mutation reduced macrophage infiltration, angiogenesis, and delayed healing. These findings suggest that targeting TRPV1 or stimulating sensory nerve fibers could enhance TM repair, improve blood flow, and prevent infections. This offers new therapeutic strategies for TM perforations and otitis media in clinical settings. This is an excellent and high-quality study that provides valuable insights into the mechanisms underlying TM wound healing.

Strengths:

The work is particularly important for elucidating the cellular and molecular processes involved in TM repair. However, there are several concerns about the current version.

We sincerely thank Reviewer #1 for their time and effort in evaluating and improving our study. Below, we are pleased to address the Reviewer's concerns point by point.

Weaknesses:

Major concerns

(1) The method of administration will be a critical factor when considering potential therapeutic strategies to promote TM healing. It would be beneficial if the authors could discuss possible delivery methods, such as topical application, transtympanic injection, or systemic administration, and their respective advantages and limitations for targeting TRPV1 signaling. For example, Dr. Kanemaru and his colleagues have proposed the use of Trafermin and Spongel to regenerate the eardrum.

We are grateful to the reviewer for raising this important point. While the present study primarily focuses on the mechanistic role of TRPV1 in TM repair, we agree that the mode of therapeutic delivery will be pivotal in translating these findings into clinical practice. In response, we will expand the discussion to explore possible delivery methods—such as topical application, transtympanic injection, and systemic routes—along with their respective benefits and challenges. We will also cite the work by Dr. Kanemaru and colleagues as an example of how local delivery systems may facilitate TM regeneration.

(2) The authors appear to have used surface imaging techniques to observe the TM. However, the TM consists of three distinct layers: the epithelial layer, the fibrous middle layer, and the inner mucosal layer. The authors should clarify whether the proposed mechanism involving TRPV1-mediated macrophage recruitment and angiogenesis is limited to the epithelial layer or if it extends to the deeper layers of the TM.

We apologize for any confusion caused by our previous description. In our study, we utilized Z-stack confocal imaging to capture the full thickness of the TM, as illustrated in Author response image 1 (reconstructed from the acquired Z-sections). This imaging technique allowed us to encompass all three layers of the TM entirely. Each sample was imaged using a 10X objective on an Olympus fluorescence microscope. Given the conical shape and size of the TM, we imaged it in four quadrants, acquiring approximately 30 optical sections (with a 3 µm step) per region. Each acquired images were projected and exported using FV10ASW 4.2 Viewer, then stitched together using Photoshop. The resulting Z-stack projections enabled us to visualize the distribution of macrophages, angiogenesis, and the localization of nerve fibers throughout the TM. We will include this detailed methodology in our revision to clarify any potential confusion.

Author response image 1.

Representative confocal images showing one quadrant of the TM collected from collected from CSR1F^EGFP bone marrow transplanted mouse at day 7 post-perforation. (A-B) 3D-rendered views from different angles reveal the close spatial relationship between CSF1R^EGFP cells (green) and blood vessels (red) within the TM. (C) Cross-sectional view highlights the depth-wise distribution of CSF1R^EGFP cells (green) and blood vessels (red) across the layered TM architecture. All images were processed using Imaris Viewer x64 (version 10.2.0).

Minor concerns

In Figure 8, the schematic illustration presents a coronal section of the TM. However, based on the data provided in the manuscript, it is unclear whether the authors directly obtained coronal images in their study. To enhance the clarity and impact of the schematic, it would be helpful to include representative images of coronal sections showing macrophage infiltration, angiogenesis, and nerve fiber distribution in the TM.

As noted above, we utilized Z-stack confocal imaging to capture the full thickness of the TM, enabling us to visualize structures across all three layers. This approach ensured that all layers were included in our analysis. Due to the thin and curved nature of the TM, traditional cross-sectional imaging often struggles to clearly depict the spatial relationships between macrophages, blood vessels, and nerve fibers, especially at low magnification as shown in Author response image 2. In response to the reviewer's suggestion, we will include representative coronal images in the revised manuscript to better illustrate the distribution of these structures at higher magnification.

Author response image 2.

Confocal images of eardrum cross-sections collected at day 1 (A), 3 (B), and 7 (C) post perforation to demonstrate the wound healing processes.

Reviewer #2 (Public review):

Summary:

This study examines the role of TRPV1 signaling in the recruitment of monocyte-derived macrophages and the promotion of angiogenesis during tympanic membrane (TM) wound healing. The authors use a combination of genetic mouse models, macrophage depletion, and transcriptomic approaches to suggest that neuronal TRPV1 activity contributes to macrophage-driven vascular responses necessary for tissue repair.

Strengths:

(1) The topic of neuroimmune interactions in tissue regeneration is of interest and underexplored in the context of the TM, which presents a unique model due to its anatomical features.

(2) The use of reporter mice and bone marrow chimeras allows for some dissection of immune cell origin.

(3) The authors incorporate transcriptomic data to contextualize inflammatory and angiogenic processes during wound healing.

We sincerely thank Reviewer #2 for their time and effort in improving our study and recognizing its strengths. Below, we are pleased to address the reviewer's concerns point by point.

Weaknesses:

(1) The primary claims of the manuscript are not convincingly supported by the evidence presented. Most of the data are correlative in nature, and no direct mechanistic experiments are included to establish causality between TRPV1 signaling and macrophage recruitment or function.

We appreciate Reviewer #2's perspective on the lack of molecular mechanisms linking TRPV1 signaling and macrophages. However, our data demonstrates that TRPV1 mutations significantly affect macrophage recruitment and angiogenesis. This initial study primarily focuses on the intriguing phenomenon of how sensory nerve fibers are involved in eardrum immunity and wound healing, an area that has not been clearly reported in the literature before. We believe that further research is necessary to explore this topic in greater depth.

(2) Functional validation of key molecular players (such as Tac1 or Spp1) is lacking, and their roles are inferred primarily from gene expression data rather than experimentally tested.

Although we have identified the TAC1 and SPP1 signals as potentially important for TM wound healing for the first time, we agree with the Reviewer's view regarding the lack of molecular mechanisms explored in this study. We have not yet tested the downstream signaling pathways, but we plan to investigate them in a series of future studies. As this is an early report, we will continue to explore these signals and their potential clinical applications based on our initial findings moving forward.

(3) The reuse of publicly available scRNA-seq data is not sufficiently integrated or extended to yield new biological insights, and it remains largely descriptive.

We appreciate Reviewer #2 for highlighting this point. Leveraging publicly available scRNA-seq databases and established analysis pipelines not only saves time and resources—my lab recently collected macrophages from the eardrums of postnatal P15 mice, with each trial requiring 20 eardrums from 10 animals to obtain a sufficient number of cells—but also allows researchers to build on previous work and focus on new biological questions without the need to repeat experiments. A prior study conducted by Dr. Tward and his team utilized scRNA-seq data to make initial discoveries related to eardrum wound healing, primarily focusing on epithelial cells rather than macrophages. We are building on their raw data to uncover new biological insights regarding macrophages, even though we have not yet tested the unidentified signals, which we believe will be valuable to our peers.

(4) The macrophage depletion model (CX3CR1CreER; iDTR) lacks specificity, and possible off-target or systemic effects are not addressed.

We agree with reviewer #2, although macrophage depletion model used in our study is a standard and well-used animal model (Shi, Hua et al. 2018), which has been used by many other laboratories, it is important to note that any macrophage depletion model may have potential issues. We will discuss this in our revision.

(5) Several interpretations of the data appear overstated, particularly regarding the necessity of TRPV1 for monocyte recruitment and wound healing.

We thank the reviewer for pointing this out. We will revise our manuscript where it is overstated accordingly.

(6) Overall, the study appears to apply known concepts - namely, TRPV1-mediated neurogenic inflammation and macrophage-driven angiogenesis - to a new anatomical site without providing new mechanistic insight or advancing the field substantially.

Although our study may not seem highly innovative at first glance, it reveals a previously unknown role of the TRPV1 pain signaling pathway in promoting eardrum healing for the first time. This healing process includes the recruitment of monocyte-derived macrophages and the formation of new blood vessels (angiogenesis). While this process has been documented in other organs, most research on macrophage-driven angiogenesis has been conducted using in vitro models, with very few studies demonstrating this process in vivo. Our findings could lead to new translational opportunities, especially considering that tympanic membrane perforation, along with damage-induced otitis media and conductive hearing loss, are common clinical issues affecting millions of people worldwide. Targeting TRPV1 signaling could enhance tympanic membrane immunity, improve blood circulation, promote the repair of damaged tympanic membranes, and ultimately prevent middle ear infections—an idea that has not been previously proposed.

Overall:

While the study addresses an interesting topic, the current version does not provide sufficiently strong or novel evidence to support its major conclusions. Additional mechanistic experiments and more rigorous validation would be necessary to substantiate the proposed model and clarify the relevance of the findings beyond this specific tissue context.

We greatly thank the two reviewers for their helpful critiques to improve our study. We especially thank the Section Editors for their insightful and constructive comments on this initial study.

References:

Shi, J., L. Hua, D. Harmer, P. Li and G. Ren (2018). "Cre Driver Mice Targeting Macrophages." Methods Mol Biol 1784: 263-275.

Author response:

Reviewer #1 (Public review):

Summary:

This article investigates the origin of movement slowdown in weightlessness by testing two possible hypotheses: the first is based on a strategic and conservative slowdown, presented as a scaling of the motion kinematics without altering its profile, while the second is based on the hypothesis of a misestimation of effective mass by the brain due to an alteration of gravity-dependent sensory inputs, which alters the kinematics following a controller parameterization error.

Strengths:

The article convincingly demonstrates that trajectories are affected in 0g conditions, as in previous work. It is interesting, and the results appear robust. However, I have two major reservations about the current version of the manuscript that prevent me from endorsing the conclusion in its current form.

Weaknesses:

(1) First, the hypothesis of a strategic and conservative slow down implicitly assumes a similar cost function, which cannot be guaranteed, tested, or verified. For example, previous work has suggested that changing the ratio between the state and control weight matrices produced an alteration in movement kinematics similar to that presented here, without changing the estimated mass parameter (Crevecoeur et al., 2010, J Neurophysiol, 104 (3), 1301-1313). Thus, the hypothesis of conservative slowing cannot be rejected. Such a strategy could vary with effective mass (thus showing a statistical effect), but the possibility that the data reflect a combination of both mechanisms (strategic slowing and mass misestimation) remains open.

We test whether changing the ratio between the state and control weight matrices can generate the observed effect. As shown in Author response image 1 and Author response image 2, the cost function change cannot produce a reduced peak velocity/acceleration and their timing advance simultaneously, but a mass estimation change can. In other words, using mass underestimation alone can explain the two key findings, amplitude reduction and timing advance. Yes, we cannot exclude the possibility of a change in cost function on top of the mass underestimation, but the principle of Occam’s Razor would support to adhering to a simple explanation, i.e., using body mass underestimation to explain the key findings. We will include our exploration on possible changes in cost function in the revision (in the Supplemental Materials).

Author response image 1.

Simulation using an altered cost function with α = 3.0. Panels A, B, and E show simulated position, velocity, and acceleration profiles, respectively, for the three movement directions. Solid lines correspond to pre- and post-exposure conditions, while dashed lines represent the in-flight condition. Panels C and D display the peak velocity and its timing across the three phases (Pre, In, Post), and Panels F and G show the corresponding peak acceleration and its timing. Note, varying the cost function, while leading to reduced peak velocity/acceleration, leads to an erroneous prediction of delayed timing of peak velocity/acceleration.

Author response image 2.

Simulation results using a cost function with α = 0.3. The format is the same as in Author response image 1. Note, this ten-fold decrease in α, while finally getting the timing of peak velocity/acceleration right (advanced or reduced), leads to an erroneous prediction of increased peak velocity/acceleration.

(2) The main strength of the article is the presence of directional effects expected under the hypothesis of mass estimation error. However, the article lacks a clear demonstration of such an effect: indeed, although there appears to be a significant effect of direction, I was not sure that this effect matched the model's predictions. A directional effect is not sufficient because the model makes clear quantitative predictions about how this effect should vary across directions. In the absence of a quantitative match between the model and the data, the authors' claims regarding the role of misestimating the effective mass remain unsupported.

Our paper does not aim to quantitatively reproduce human reaching movements in microgravity. We will make this more clearly in the revision.

(1) The model is a simplification of the actual situation. For example, the model simulates an ideal case of moving a point mass (effective mass) without friction and without considering Coriolis and centripetal torques, while the actual situation is that people move their finger across a touch screen. The two-link arm model assumes planar movements, but our participants move their hand on a table top without vertical support to constrain their movement in 2D.

(2) Our study merely uses well-established (though simplified) models to qualitatively predict the overall behavioral patterns if mass underestimation is at play. For this purpose, the results are well in line with models’ qualitative predictions: we indeed confirm that key kinematic features (peak velocity and acceleration) follow the same ranking order of movement direction conditions as predicted.

(3) Using model simulation to qualitatively predict human behavioral patterns is a common practice in motor control studies, prominent examples including the papers on optimal feedback control (Todorov, 2004 and 2005) and movement vigor (Shadmehr et al., 2016). In fact, our model was inspired by the model in the latter paper.

Citations:

Todorov, E. (2004). Optimality principles in sensorimotor control. Nature Neuroscience, 7(9), 907.

Todorov, E. (2005). Stochastic optimal control and estimation methods adapted to the noise characteristics of the sensorimotor system. Neural Computation, 17(5), 1084–1108.

Shadmehr, R., Huang, H. J., & Ahmed, A. A. (2016). A Representation of Effort in Decision-Making and Motor Control. Current Biology: CB, 26(14), 1929–1934.

In general, both the hypotheses of slowing motion (out of caution) and misestimating mass have been put forward in the past, and the added value of this article lies in demonstrating that the effect depended on direction. However, (1) a conservative strategy with a different cost function can also explain the data, and (2) the quantitative match between the directional effect and the model's predictions has not been established.

Specific points:

(1) I noted a lack of presentation of raw kinematic traces, which would be necessary to convince me that the directional effect was related to effective mass as stated.

We are happy to include exemplary speed and acceleration trajectories. One example subject’s detailed trajectories are shown below and will be included in the revision. The reduced and advanced velocity/acceleration peaks are visible in typical trials.

Author response image 3.

Hand speed profiles (upper panels), hand acceleration profiles (middle panels) and speed profiles of the primary submovements (lower panels) towards different directions from an example participant.

(2) The presentation and justification of the model require substantial improvement; the reason for their presence in the supplementary material is unclear, as there is space to present the modelling work in detail in the main text. Regarding the model, some choices require justification: for example, why did the authors ignore the nonlinear Coriolis and centripetal terms?

Response: In brief, our simulations show that Coriolis and centripetal forces, despite having some directional anisotropy, only have small effects on predicted kinematics (see our responses to Reviewer 2). We will move descriptions of the model into the main text with more justifications for using a simple model.

(3) The increase in the proportion of trials with subcomponents is interesting, but the explanatory power of this observation is limited, as the initial percentage was already quite high (from 60-70% during the initial study to 70-85% in flight). This suggests that the potential effect of effective mass only explains a small increase in a trend already present in the initial study. A more critical assessment of this result is warranted.

Response: Indeed, the percentage of submovements only increases slightly, but the more important change is that the IPI (the inter-peak interval between submovements) also increases at the same time. Moreover, it is the effect of IPI that significantly predicts the duration increase in our linear mixed model. We will highlight this fact in our revision to avoid confusion.

Reviewer #2 (Public review):

This study explores the underlying causes of the generalized movement slowness observed in astronauts in weightlessness compared to their performance on Earth. The authors argue that this movement slowness stems from an underestimation of mass rather than a deliberate reduction in speed for enhanced stability and safety.

Overall, this is a fascinating and well-written work. The kinematic analysis is thorough and comprehensive. The design of the study is solid, the collected dataset is rare, and the model tends to add confidence to the proposed conclusions. That being said, I have several comments that could be addressed to consolidate interpretations and improve clarity.

Main comments:

(1) Mass underestimation

a) While this interpretation is supported by data and analyses, it is not clear whether this gives a complete picture of the underlying phenomena. The two hypotheses (i.e., mass underestimation vs deliberate speed reduction) can only be distinguished in terms of velocity/acceleration patterns, which should display specific changes during the flight with a mass underestimation. The experimental data generally shows the expected changes but for the 45{degree sign} condition, no changes are observed during flight compared to the pre- and post-phases (Figure 4). In Figure 5E, only a change in the primary submovement peak velocity is observed for 45{degree sign}, but this finding relies on a more involved decomposition procedure. It suggests that there is something specific about 45{degree sign} (beyond its low effective mass). In such planar movements, 45{degree sign} often corresponds to a movement which is close to single-joint, whereas 90{degree sign} and 135{degree sign} involve multi-joint movements. If so, the increased proportion of submovements in 90{degree sign} and 135{degree sign} could indicate that participants had more difficulties in coordinating multi-joint movements during flight. Besides inertia, Coriolis and centripetal effects may be non-negligible in such fast planar reaching (Hollerbach & Flash, Biol Cyber, 1982) and, interestingly, they would also be affected by a mass underestimation (thus, this is not necessarily incompatible with the author's view; yet predicting the effects of a mass underestimation on Coriolis/centripetal torques would require a two-link arm model). Overall, I found the discrepancy between the 45{degree sign} direction and the other directions under-exploited in the current version of the article. In sum, could the corrective submovements be due to a misestimation of Coriolis/centripetal torques in the multi-joint dynamics (caused specifically -or not- by a mass underestimation)?

We agree that the effect of mass underestimation is less in the 45° direction than the other two directions, possibly related to its reliance on single-joint (elbow) as opposed to two-joints (elbow and shoulder) movements. Plus, movement correction using one joint is probably easier (as also suggested by another reviewer), this possibility will be further discussed in the revision. However, we find that our model simplification (excluding Coriolis and centripetal torques) does not affect our main conclusions at all. First, we performed a simple simulation and found that, under the current optimal hand trajectory, incorporating Coriolis and centripetal torques has only a limited impact on the resulting joint torques (see simulations in Author response image 4). One reason is that we used smaller movements than Hallerbach & Flash did. In addition, we applied an optimal feedback control model to a more realistic 2-joint arm configuration. Despite its simplicity, this model produced a speed profile consistent with our current predictions and made similar predictions regarding the effects of mass underestimation (Author response image 5). We will provide a more realistic 2-joint arm model muscle dynamics in the revision to improve the simulation further, but the message will be same: including or excluding Coriolis and centripetal torques will not affect the theoretical predictions about mass underestimation. Second, as the reviewer correctly pointed out, the mass (and its underestimation) also affects these two torque terms, thus its effect on kinematic measures is not affected much even with the full model.

Author response image 4.

Joint angles and joint torque of shoulder and elbow with simulated trajectories towards different directions. A. Shoulder (green) and elbow (blue) angles change with time for the 45° movement direction. B. Components of joint interaction torques at the shoulder. Solid line: net torque at the shoulder; dotted line: shoulder inertia torque; dashed line: shoulder Coriolis and centripetal torque. C. Same plot as B for the elbow joint. D–F. Coriolis and centripetal components in the full 360° workspace, beyond three movement directions (45°, 90°, and 135°). D. Net torque. E. Inertial torque. F. Combined Coriolis and centripetal torque. Note the polar plots of Coriolis/centripetal torques (F) have a scale that is two magnitudes smaller than that of inertial torque in our simulation. All torques were simulated with the optimal movement duration. Torques were squared and integrated over each trajectory.

Author response image 5.

Comparison between simulation results from the full model with the addition of Coriolis/centripetal torques (left) and the simplified model (right). The position profiles (top) and the corresponding speed profiles low) are shown. Solid lines are for normal mass estimation and dashed lines for mass underestimation in microgravity. The three colors represent three movement directions (dark red: 45°, red: 90°, yellow: 135°). The full model used a 2-link arm model without realistic muscle dynamics yet (will include in the formal revision) thus the speed profile is not smooth. Importantly, the full model also predict the same effect of mass underestimation, i.e., reduced peak velocity/acceleration and their timing advance.

b) Additionally, since the taikonauts are tested after 2 or 3 weeks in flight, one could also assume that neuromuscular deconditioning explains (at least in part) the general decrease in movement speed. Can the authors explain how to rule out this alternative interpretation? For instance, weaker muscles could account for slower movements within a classical time-effort trade-off (as more neural effort would be needed to generate a similar amount of muscle force, thereby suggesting a purposive slowing down of movement). Therefore, could the observed results (slowing down + more submovements) be explained by some neuromuscular deconditioning combined with a difficulty in coordinating multi-joint movements in weightlessness (due to a misestimation or Coriolis/centripetal torques) provide an alternative explanation for the results?

Response: Neuromuscular deconditioning is indeed a space or microgravity effect; thanks for bringing this up as we omitted the discussion of its possible contribution in the initial submission. However, muscle weakness is less for upper-limb muscles than for postural and lower-limb muscles (Tesch et al., 2005). The handgrip strength decreases 5% to 15% after several months (Moosavi et al., 2021); shoulder and elbow muscles atrophy, though not directly measured, was estimated to be minimal (Shen et al., 2017). The muscle weakness is unlikely to play a major role here since our reaching task involves small movements (~12cm) with joint torques of a magnitude of ~2N·m. Coriolis/centripetal torques does not affect the putative mass effect (as shown above simulations). The reviewer suggests that their poor coordination in microgravity might contribute to slowing down + more submovements. Poor coordination is an umbrella term for any motor control problems, and it can explain any microgravity effect. The feedforward control changes caused by mass underestimation can also be viewed as poor coordination. If we limit it as the coordination of the two joints or coordinating Coriolis/centripetal torques, we should expect to see some trajectory curvature changes in microgravity. However, we further analyzed our reaching trajectories and found no sign of curvature increase in our large collection of reaching movements. We probably have the largest dataset of reaching movements collected in microgravity thus far, given that we had 12 taikonauts and each of them performed about 480 to 840 reaching trials during their spaceflight. We believe the probability of Type II error is quite low here. We will include descriptive statistics of these new analyses in our revision.

Citation: Tesch, P. A., Berg, H. E., Bring, D., Evans, H. J., & LeBlanc, A. D. (2005). Effects of 17-day spaceflight on knee extensor muscle function and size. European journal of applied physiology, 93(4), 463-468.

Moosavi, D., Wolovsky, D., Depompeis, A., Uher, D., Lennington, D., Bodden, R., & Garber, C. E. (2021). The effects of spaceflight microgravity on the musculoskeletal system of humans and animals, with an emphasis on exercise as a countermeasure: A systematic scoping review. Physiological Research, 70(2), 119.

Shen, H., Lim, C., Schwartz, A. G., Andreev-Andrievskiy, A., Deymier, A. C., & Thomopoulos, S. (2017). Effects of spaceflight on the muscles of the murine shoulder. The FASEB Journal, 31(12), 5466.

(2) Modelling

a) The model description should be improved as it is currently a mix of discrete time and continuous time formulations. Moreover, an infinite-horizon cost function is used, but I thought the authors used a finite-horizon formulation with the prefixed duration provided by the movement utility maximization framework of Shadmehr et al. (Curr Biol, 2016). Furthermore, was the mass underestimation reflected both in the utility model and the optimal control model? If so, did the authors really compute the feedback control gain with the underestimated mass but simulate the system with the real mass? This is important because the mass appears both in the utility framework and in the LQ framework. Given the current interpretations, the feedforward command is assumed to be erroneous, and the feedback command would allow for motor corrections. Therefore, it could be clarified whether the feedback command also misestimates the mass or not, which may affect its efficiency. For instance, if both feedforward and feedback motor commands are based on wrong internal models (e.g., due to the mass underestimation), one may wonder how the astronauts would execute accurate goal-directed movements.

b) The model seems to be deterministic in its current form (no motor and sensory noise). Since the framework developed by Todorov (2005) is used, sensorimotor noise could have been readily considered. One could also assume that motor and sensory noise increase in microgravity, and the model could inform on how microgravity affects the number of submovements or endpoint variance due to sensorimotor noise changes, for instance.

c) Finally, how does the model distinguish the feedforward and feedback components of the motor command that are discussed in the paper, given that the model only yields a feedback control law? Does 'feedforward' refer to the motor plan here (i.e., the prefixed duration and arguably the precomputed feedback gain)?

We appreciate these very helpful suggestions about our model presentation. Indeed, our initial submission did not give detailed model descriptions in the main text, due to text limits for early submissions. We actually used a finite-horizon framework throughout, with a pre-specified duration derived from the utility model. In the revision, we will make that point clear, and we will also revise the Methods section to explicitly distinguish feedforward vs. feedback components, clarify the use of mass underestimation in both utility and control models, and update the equations accordingly.

(3) Brevity of movements and speed-accuracy trade-off

The tested movements are much faster (average duration approx. 350 ms) than similar self-paced movements that have been studied in other works (e.g., Wang et al., J Neurophysiology, 2016; Berret et al., PLOS Comp Biol, 2021, where movements can last about 900-1000 ms). This is consistent with the instructions to reach quickly and accurately, in line with a speed-accuracy trade-off. Was this instruction given to highlight the inertial effects related to the arm's anisotropy? One may however, wonder if the same results would hold for slower self-paced movements (are they also with reduced speed compared to Earth performance?). Moreover, a few other important questions might need to be addressed for completeness: how to ensure that astronauts did remember this instruction during the flight? (could the control group move faster because they better remembered the instruction?). Did the taikonauts perform the experiment on their own during the flight, or did one taikonaut assume the role of the experimenter?

Thanks for highlighting the brevity of movements in our experiment. Our intention in emphasizing fast movements is to rigorously test whether movement is indeed slowed down in microgravity. The observed prolonged movement duration clearly shows that microgravity affects people’s movement duration, even when they are pushed to move fast. The second reason for using fast movement is to highlight that feedforward control is affected in microgravity. Mass underestimation specifically affects feedforward control in the first place. Slow movement would inevitably have online corrections that might obscure the effect of mass underestimation. Note that movement slowing is not only observed in our speed-emphasized reaching task, but also in whole-arm pointing in other astronauts studies (Berger, 1997; Sangals, 1999), which have been quoted in our paper. We thus believe these findings are generalizable.

Regarding the consistency of instructions: all our experiments conducted in the Tiangong space station were monitored in real time by experimenters in the Control Center located in Beijing. The task instructions were presented on the initial display of the data acquisition application and ample reading time was allowed. In fact, all the pre-, in-, and post-flight test sessions were administered by the same group of experimenters with the same instruction. It is common that astronauts serve both as participants and experimenters at the same time. And, they were well trained for this type of role on the ground. Note that we had multiple pre-flight test sessions to familiarize them with the task. All these rigorous measures were in place to obtain high-quality data. We will include these experimental details and the rationales for emphasizing fast movements in the revision.

Citations:

Berger, M., Mescheriakov, S., Molokanova, E., Lechner-Steinleitner, S., Seguer, N., & Kozlovskaya, I. (1997). Pointing arm movements in short- and long-term spaceflights. Aviation, Space, and Environmental Medicine, 68(9), 781–787.

Sangals, J., Heuer, H., Manzey, D., & Lorenz, B. (1999). Changed visuomotor transformations during and after prolonged microgravity. Experimental Brain Research. Experimentelle Hirnforschung. Experimentation Cerebrale, 129(3), 378–390.

(4) No learning effect

This is a surprising effect, as mentioned by the authors. Other studies conducted in microgravity have indeed revealed an optimal adaptation of motor patterns in a few dozen trials (e.g., Gaveau et al., eLife, 2016). Perhaps the difference is again related to single-joint versus multi-joint movements. This should be better discussed given the impact of this claim. Typically, why would a "sensory bias of bodily property" persist in microgravity and be a "fundamental constraint of the sensorimotor system"?

We believe the differences between our study and Gaveau et al.’s study cannot be simply attributed to single-joint versus multi-joint movements. One of the most salient differences is that their adaptation is about incorporating microgravity in control for minimizing effort, while our adaptation is about rightfully perceiving body mass. We will elaborate on possible reasons for the lack of learning in the light of this previous study.

We can elaborate on “sensory bias” and “fundamental constraint of the sensorimotor system”. If an inertial change is perceived (like an extra weight attached to the forearm, as in previous motor adaptation studies), people can adapt their reaching in tens of trials. In this case, sensory cues are veridical as they correctly inform about the inertial perturbation. However, in microgravity, reduced gravitational pull and proprioceptive inputs constantly inform the controller that the body mass is less than its actual magnitude. In other words, sensory cues in space are misleading for estimating body mass. The resulting sensory bias prevents the sensorimotor system from correctly adapt. Our statement was too brief in the initial submission; we will expand it in the revision.

Reviewer #3 (Public review):

Summary:

The authors describe an interesting study of arm movements carried out in weightlessness after a prolonged exposure to the so-called microgravity conditions of orbital spaceflight. Subjects performed radial point-to-point motions of the fingertip on a touch pad. The authors note a reduction in movement speed in weightlessness, which they hypothesize could be due to either an overall strategy of lowering movement speed to better accommodate the instability of the body in weightlessness or an underestimation of body mass. They conclude for the latter, mainly based on two effects. One, slowing in weightlessness is greater for movement directions with higher effective mass at the end effector of the arm. Two, they present evidence for an increased number of corrective submovements in weightlessness. They contend that this provides conclusive evidence to accept the hypothesis of an underestimation of body mass.

Strengths:

In my opinion, the study provides a valuable contribution, the theoretical aspects are well presented through simulations, the statistical analyses are meticulous, the applicable literature is comprehensively considered and cited, and the manuscript is well written.

Weaknesses:

Nevertheless, I am of the opinion that the interpretation of the observations leaves room for other possible explanations of the observed phenomenon, thus weakening the strength of the arguments.

First, I would like to point out an apparent (at least to me) divergence between the predictions and the observed data. Figures 1 and S1 show that the difference between predicted values for the 3 movement directions is almost linear, with predictions for 90º midway between predictions for 45º and 135º. The effective mass at 90º appears to be much closer to that of 45º than to that of 135º (Figure S1A). But the data shown in Figure 2 and Figure 3 indicate that movements at 90º and 135º are grouped together in terms of reaction time, movement duration, and peak acceleration, while both differ significantly from those values for movements at 45º.

Furthermore, in Figure 4, the change in peak acceleration time and relative time to peak acceleration between 1g and 0g appears to be greater for 90º than for 135º, which appears to me to be at least superficially in contradiction with the predictions from Figure S1. If the effective mass is the key parameter, wouldn't one expect as much difference between 90º and 135º as between 90º and 45º? It is true that peak speed (Figure 3B) and peak speed time (Figure 4B) appear to follow the ordering according to effective mass, but is there a mathematical explanation as to why the ordering is respected for velocity but not acceleration? These inconsistencies weaken the author's conclusions and should be addressed.

Indeed, the model predicts an almost equal separation between 45° and 90° and between 90° and 135°, while the data indicate that the spacing between 45° and 90° is much smaller than between 90° and 135°. We do not regard the divergence as evidence undermining our main conclusion since 1) the model is a simplification of the actual situation. For example, the model simulates an ideal case of moving a point mass (effective mass) without friction and without considering Coriolis and centripetal torques. 2) Our study does not make quantitative predictions of all the key kinematic measures; that will require model fitting and parameter estimation; instead, our study uses well-established (though simplified) models to qualitatively predict the overall behavioral pattern we would observe. For this purpose, our results are well in line with our expectations: though we did not find equal spacing between direction conditions, we do confirm that the key kinematic properties (Figure 2 and Figure 3 as questioned) follow the same ranking order of directions as predicted.

We thank the reviewer for pointing out the apparent discrepancy between model simulation and observed data. We will elaborate on the reasons behind the discrepancy in the revision.

Then, to strengthen the conclusions, I feel that the following points would need to be addressed:

(1) The authors model the movement control through equations that derive the input control variable in terms of the force acting on the hand and treat the arm as a second-order low-pass filter (Equation 13). Underestimation of the mass in the computation of a feedforward command would lead to a lower-than-expected displacement to that command. But it is not clear if and how the authors account for a potential modification of the time constants of the 2nd order system. The CNS does not effectuate movements with pure torque generators. Muscles have elastic properties that depend on their tonic excitation level, reflex feedback, and other parameters. Indeed, Fisk et al.* showed variations of movement characteristics consistent with lower muscle tone, lower bandwidth, and lower damping ratio in 0g compared to 1g. Could the variations in the response to the initial feedforward command be explained by a misrepresentation of the limbs' damping and natural frequency, leading to greater uncertainty about the consequences of the initial command? This would still be an argument for unadapted feedforward control of the movement, leading to the need for more corrective movements. But it would not necessarily reflect an underestimation of body mass.

*Fisk, J. O. H. N., Lackner, J. R., & DiZio, P. A. U. L. (1993). Gravitoinertial force level influences arm movement control. Journal of neurophysiology, 69(2), 504-511.

We agree that muscle properties, tonic excitation level, proprioception-mediated reflexes all contribute to reaching control. Fisk et al. (1993) study indeed showed that arm movement kinematics change, possibly owing to lower muscle tone and/or damping. However, reduced muscle damping and reduced spindle activity are more likely to affect feedback-based movements. Like in Fisk et al.’s study, people performed continuous arm movements with eyes closed; thus their movements largely relied on proprioceptive control. Our major findings are about the feedforward control, i.e., the reduced and “advanced” peak velocity/acceleration in discrete and ballistic reaching movements. Note that the peak acceleration happens as early as approximately 90-100ms into the movements, clearly showing that feedforward control is affected -- a different effect from Fisk et al’s findings. It is unlikely that people “advanced” their peak velocity/acceleration because they feel the need for more later corrective movements. Thus, underestimation of body mass remains the most plausible explanation.

(2) The movements were measured by having the subjects slide their finger on the surface of a touch screen. In weightlessness, the implications of this contact are expected to be quite different than those on the ground. In weightlessness, the taikonauts would need to actively press downward to maintain contact with the screen, while on Earth, gravity will do the work. The tangential forces that resist movement due to friction might therefore be different in 0g. This could be particularly relevant given that the effect of friction would interact with the limb in a direction-dependent fashion, given the anisotropy of the equivalent mass at the fingertip evoked by the authors. Is there some way to discount or control for these potential effects?

We agree that friction might play a role here, but normal interaction with a touch screen typically involves friction between 0.1 and 0.5N (e.g., Ayyildiz et al., 2018). We believe that the directional variation is even smaller than 0.1N. It is very small compared to the force used to accelerate the arm for the reaching movement (10-15N). Thus, friction anisotropy is unlikely to explain our data.

Citation: Ayyildiz M, Scaraggi M, Sirin O, Basdogan C, Persson BNJ. Contact mechanics between the human finger and a touchscreen under electroadhesion. Proc Natl Acad Sci U S A. 2018 Dec 11;115(50):12668-12673.

(3) The carefully crafted modelling of the limb neglects, nevertheless, the potential instability of the base of the arm. While the taikonauts were able to use their left arm to stabilize their bodies, it is not clear to what extent active stabilization with the contralateral limb can reproduce the stability of the human body seated in a chair in Earth gravity. Unintended motion of the shoulder could account for a smaller-than-expected displacement of the hand in response to the initial feedforward command and/or greater propensity for errors (with a greater need for corrective submovements) in 0g. The direction of movement with respect to the anchoring point could lead to the dependence of the observed effects on movement direction. Could this be tested in some way, e.g., by testing subjects on the ground while standing on an unstable base of support or sitting on a swing, with the same requirement to stabilize the torso using the contralateral arm?

Body stabilization is always a challenge for human movement studies in space. We minimized its potential confounding effects by using left-hand grasping and foot straps for postural support throughout the experiment. We would argue shoulder stability is an unlikely explanation because unexpected shoulder instability should not affect the feedforward (early) part of the ballistic reaching movement: the reduced peak acceleration and its early peak were observed at about 90-100ms after movement initiation. This effect is too early to be explained by an expected stability issue.

The arguments for an underestimation of body mass would be strengthened if the authors could address these points in some way.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

In this study from Zhu and colleagues, a clear role for MED26 in mouse and human erythropoiesis is demonstrated that is also mapped to amino acids 88-480 of the human protein. The authors also show the unique expression of MED26 in later-stage erythropoiesis and propose transcriptional pausing and condensate formation mechanisms for MED26's role in promoting erythropoiesis. Despite the author's introductory claim that many questions regarding Pol II pausing in mammalian development remain unanswered, the importance of transcriptional pausing in erythropoiesis has actually already been demonstrated (Martell-Smart, et al. 2023, PMID: 37586368, which the authors notably did not cite in this manuscript). Here, the novelty and strength of this study is MED26 and its unique expression kinetics during erythroid development.

Strengths:

The widespread characterization of kinetics of mediator complex component expression throughout the erythropoietic timeline is excellent and shows the interesting divergence of MED26 expression pattern from many other mediator complex components. The genetic evidence in conditional knockout mice for erythropoiesis requiring MED26 is outstanding. These are completely new models from the investigators and are an impressive amount of work to have both EpoR-driven deletion and inducible deletion. The effect on red cell number is strong in both. The genetic over-expression experiments are also quite impressive, especially the investigators' structure-function mapping in primary cells. Overall the data is quite convincing regarding the genetic requirement for MED26. The authors should be commended for demonstrating this in multiple rigorous ways.

Thank you for your positive feedback.

Weaknesses:

(1) The authors state that MED26 was nominated for study based on RNA-seq analysis of a prior published dataset. They do not however display any of that RNA-seq analysis with regards to Mediator complex subunits. While they do a good job showing protein-level analysis during erythropoiesis for several subunits, the RNA-seq analysis would allow them to show the developmental expression dynamics of all subunit members.

Thank you for this helpful suggestion. While we did not originally nominate MED26 based on RNA-seq analysis, we have analyzed the transcript levels of Mediator complex subunits in our RNA-seq data across different stages of erythroid differentiation (Author response image 1). The results indicate that most Mediator subunits, including MED26, display decreased RNA expression over the course of differentiation, with the exception of MED25, as reported previously (Pope et al., Mol Cell Biol 2013. PMID: 23459945).

Notably, our study is based on initial observations at the protein level, where we found that, unlike most other Mediator subunits that are downregulated during erythropoiesis, MED26 remains relatively abundant. Protein expression levels more directly reflect the combined influences of transcription, translation and degradation processes within cells, and are likely more closely related to biological functions in this context. It is possible that post-transcriptional regulation (such as m6A-mediated improvement of translational efficiency) or post-translational modifications (like escape from ubiquitination) could contribute to the sustained levels of MED26 protein, and this will be an interesting direction for future investigation.

Author response image 1.

Relative RNA expression of Mediator complex subunits during erythropoiesis in human CD34+ erythroid cultures. Different differentiation stages from HSPCs to late erythroblasts were identified using CD71 and CD235a markers, progressing sequentially as CD71-CD235a-, CD71+CD235a-, CD71+CD235a+, and CD71-CD235a+. Expression levels were presented as TPM (transcripts per million).

(2) The authors use an EpoR Cre for red cell-specific MED26 deletion. However, other studies have now shown that the EpoR Cre can also lead to recombination in the macrophage lineage, which clouds some of the in vivo conclusions for erythroid specificity. That being said, the in vitro erythropoiesis experiments here are convincing that there is a major erythroid-intrinsic effect.

Thank you for this insightful comment. We recognize that EpoR-Cre can drive recombination in both erythroid and macrophage lineages (Zhang et al., Blood 2021, PMID: 34098576). However, EpoR-Cre remains the most widely used Cre for studying erythroid lineage effects in the hematopoietic community. Numerous studies have employed EpoR-Cre for erythroid-specific gene knockout models (Pang et al, Mol Cell Biol 2021, PMID: 22566683; Santana-Codina et al., Haematologica 2019, PMID: 30630985; Xu et al., Science 2013, PMID: 21998251.).

While a GYPA (CD235a)-Cre model with erythroid specificity has recently been developed (https://www.sciencedirect.com/science/article/pii/S0006497121029074), it has not yet been officially published. We look forward to utilizing the GYPA-Cre model for future studies. As you noted, our in vivo mouse model and primary human CD34+ erythroid differentiation system both demonstrate that MED26 is essential for erythropoiesis, suggesting that the regulatory effects of MED26 in our study are predominantly erythroid-intrinsic.

(3) Te donor chimerism assessment of mice transplanted with MED26 knockout cells is a bit troubling. First, there are no staining controls shown and the full gating strategy is not shown. Furthermore, the authors use the CD45.1/CD45.2 system to differentiate between donor and recipient cells in erythroblasts. However, CD45 is not expressed from the CD235a+ stage of erythropoiesis onwards, so it is unclear how the authors are detecting essentially zero CD45-negative cells in the erythroblast compartment. This is quite odd and raises questions about the results. That being said, the red cell indices in the mice are the much more convincing data.

Thank you for your careful and thorough feedback. We have now included negative staining controls (Author response image 2A, top). We agree that CD45 is typically not expressed in erythroid precursors in normal development. Prior studies have characterized BFU-E and CFU-E stages as c-Kit+CD45+Ter119−CD71low and c-Kit+CD45−Ter119−CD71high cells in fetal liver (Katiyar et al, Cells 2023, PMID: 37174702).

However, our observations indicate that erythroid surface markers differ during hematopoiesis reconstitution following bone marrow transplantation.  We found that nearly all nucleated erythroid progenitors/precursors (Ter119+Hoechst+) express CD45 after hematopoiesis reconstitution (Author response image 2A, bottom).

To validate our assay, we performed next-generation sequencing by first mixing mouse CD45.1 and CD45.2 total bone marrow cells at a 1:2 ratio. We then isolated nucleated erythroid progenitors/precursors (Ter119+Hoechst+) by FACS and sequenced the CD45 gene locus by targeted sequencing. The resulting CD45 allele distribution matched our initial mixing ratio, confirming the accuracy of our approach (Author response image 2B).

Moreover, a recent study supports that reconstituted erythroid progenitors can indeed be distinguished by CD45 expression following bone marrow transplantation (He et al., Nature Aging 2024, PMID: 38632351. Extended Data Fig. 8). 

In conclusion, our data indicate that newly formed erythroid progenitors/precursors post-transplant express CD45, enabling us to identify nucleated erythroid progenitors/precursors by Ter119+Hoechst+ and determine their origin using CD45.1 and CD45.2 markers.

Author response image 2.

Representative flow cytometry gating strategy of erythroid chimerism following mouse bone marrow transplantation. A. Gating strategy used in the erythroid chimerism assay. B. Targeted sequencing result of Ter119+Hoechst+ cells isolated by FACS. The cell sample was pre-mixed with 1/3 CD45.2 and 2/3 CD45.1 bone marrow cells. Ptprc is the gene locus for CD45.

(4) The authors make heavy use of defining "erythroid gene" sets and "non-erythroid gene" sets, but it is unclear what those lists of genes actually are. This makes it hard to assess any claims made about erythroid and non-erythroid genes.

Thank you for this helpful suggestion. We defined "erythroid genes" and "non-erythroid genes" based on RNA-seq data from Ludwig et al. (Cell Reports 2019. PMID: 31189107. Figure 2 and Table S1). Genes downregulated from stages k1 to k5 are classified as “non-erythroid genes,” while genes upregulated from stages k6 to k7 are classified as “erythroid genes.” We will add this description in the revised manuscript.

(5) Overall the data regarding condensate formation is difficult to interpret and is the weakest part of this paper. It is also unclear how studies of in vitro condensate formation or studies in 293T or K562 cells can truly relate to highly specialized erythroid biology. This does not detract from the major findings regarding genetic requirements of MED26 in erythropoiesis.

Thank you for the rigorous feedback. Assessing the condensate properties of MED26 protein in primary CD34+ erythroid cells or mouse models is indeed challenging. As is common in many condensate studies, we used in vitro assays and cellular assays in HEK293T and K562 cells to examine the biophysical properties (Figure S7), condensation formation capacity (Figure 5C and Figure S7C), key phase-separation regions of MED26 protein (Figure S6), and recruitment of pausing factors (Figure 6A-B) in live cells. We then conducted functional assays to demonstrate that the phase-separation region of MED26 can promote erythroid differentiation similarly to the full-length protein in the CD34+ system and K562 cells (Figure 5A). Specifically, overexpressing the MED26 phase-separation domain accelerates erythropoiesis in primary human erythroid culture, while deleting the Intrinsically Disordered Region (IDR) impairs MED26’s ability to form condensates and recruit PAF1 in K562 cells.

In summary, we used HEK293T cells to study the biochemical and biophysical properties of MED26, and the primary CD34+ differentiation system to examine its developmental roles. Our findings support the conclusion that MED26-associated condensate formation promotes erythropoiesis.

(6) For many figures, there are some panels where conclusions are drawn, but no statistical quantification of whether a difference is significant or not.

Thank you for your thorough feedback. We have checked all figures for statistical quantification and added the relevant statistical analysis methods to the corresponding figure legends (Figure 2L and Figure S4C) to clarify the significance of the observed differences. The updated information will be incorporated into the revised manuscript.

Reviewer #2 (Public review):

Summary:

The manuscript by Zhu et al describes a novel role for MED26, a subunit of the Mediator complex, in erythroid development. The authors have discovered that MED26 promotes transcriptional pausing of RNA Pol II, by recruiting pausing-related factors.

Strengths:

This is a well-executed study. The authors have employed a range of cutting-edge and appropriate techniques to generate their data, including: CUT&Tag to profile chromatin changes and mediator complex distribution; nuclear run-on sequencing (PRO-seq) to study Pol II dynamics; knockout mice to determine the phenotype of MED26 perturbation in vivo; an ex vivo erythroid differentiation system to perform additional, important, biochemical and perturbation experiments; immunoprecipitation mass spectrometry (IP-MS); and the "optoDroplet" assay to study phase-separation and molecular condensates.

This is a real highlight of the study. The authors have managed to generate a comprehensive picture by employing these multiple techniques. In doing so, they have also managed to provide greater molecular insight into the workings of the MEDIATOR complex, an important multi-protein complex that plays an important role in a range of biological contexts. The insights the authors have uncovered for different subunits in erythropoiesis will very likely have ramifications in many other settings, in both healthy biology and disease contexts.

Thank you for your thoughtful summary and encouraging feedback.

Weaknesses:

There are almost no discernible weaknesses in the techniques used, nor the interpretation of the data. The IP-MS data was generated in HEK293 cells when it could have been performed in the human CD34+ HSPC system that they employed to generate a number of the other data. This would have been a more natural setting and would have enabled a more like-for-like comparison with the other data.

Thank you for your positive feedback and insightful suggestions. We will perform validation of the immunoprecipitation results in CD34+ derived erythroid cells to further confirm our findings.

Reviewer #3 (Public review):

Summary:

The authors aim to explore whether other subunits besides MED1 exert specific functions during the process of terminal erythropoiesis with global gene repression, and finally they demonstrated that MED26-enriched condensates drive erythropoiesis through modulating transcription pausing.

Strengths:

Through both in vitro and in vivo models, the authors showed that while MED1 and MED26 co-occupy a plethora of genes important for cell survival and proliferation at the HSPC stage, MED26 preferentially marks erythroid genes and recruits pausing-related factors for cell fate specification. Gradually, MED26 becomes the dominant factor in shaping the composition of transcription condensates and transforms the chromatin towards a repressive yet permissive state, achieving global transcription repression in erythropoiesis.

Thank you for your positive summary and feedback.

Weaknesses:

In the in vitro model, the author only used CD34+ cell-derived erythropoiesis as the validation, which is relatively simple, and more in vitro erythropoiesis models need to be used to strengthen the conclusion.

Thank you for your thoughtful suggestions. We have shown that MED26 promotes erythropoiesis using the primary human CD34+ differentiation system (Figure 2 K-M and Figure S4) and have demonstrated its essential role in erythropoiesis through multiple mouse models (Figure 2A-G and Figure S1-3). Together, these in vitro and in vivo results support our conclusion that MED26 regulates erythropoiesis. However, we are open to further validating our findings with additional in vitro erythropoiesis models, such as iPSC or HUDEP erythroid differentiation systems.

Author Response

Reviewer #1 (Public Review):

[...] Genes expressed in the same direction in lowland individuals facing hypoxia (the plastic state) as what is found in the colonised state are defined as adaptative, while genes with the opposite expression pattern were labelled as maladaptive, using the assumption that the colonised state must represent the result of natural selection. Furthermore, genes could be classified as representing reversion plasticity when the expression pattern differed between the plasticity and colonised states and as reinforcement when they were in the same direction (for example more expressed in the plastic state and the colonised state than in the ancestral state). They found that more genes had a plastic expression pattern that was labelled as maladaptive than adaptive. Therefore, some of the genes have an expression pattern in accordance with what would be predicted based on the plasticity-first hypothesis, while others do not.

Thank you for a precise summary of our work. We appreciate the very encouraging comments recognizing the value of our work. We have addressed concerns from the reviewer in greater detail below.

Q1. As pointed out by the authors themselves, the fact that temperature was not included as a variable, which would make the experimental design much more complex, misses the opportunity to more accurately reflect the environmental conditions that the colonizer individuals face at high altitude. Also pointed out by the authors, the acclimation experiment in hypoxia lasted 4 weeks. It is possible that longer term effects would be identifiable in gene expression in the lowland individuals facing hypoxia on a longer time scale. Furthermore, a sample size of 3 or 4 individuals per group depending on the tissue for wild individuals may miss some of the natural variation present in these populations. Stating that they have a n=7 for the plastic stage and n= 14 for the ancestral and colonized stages refers to the total number of tissue samples and not the number of individuals, according to supplementary table 1.

We shared the same concerns as the reviewer. This is partly because it is quite challenging to bring wild birds into captivity to conduct the hypoxia acclimation experiments. We had to work hard to perform acclimation experiments by taking lowland sparrows in a hypoxic condition for a month. We indeed have recognized the similar set of limitations as the review pointed out and have discussed the limitations in the study, i.e., considering hypoxic condition alone, short time acclimation period, etc. Regarding sample sizes, we have collected cardiac muscle from nine individuals (three individuals for each stage) and flight muscle from 12 individuals (four individuals for each stage). We have clarified this in Supplementary Table 1.

Q2. Finally, I could not find a statement indicating that the lowland individuals placed in hypoxia (plastic stage) were from the same population as the lowland individuals for which transcriptomic data was already available, used as the "ancestral state" group (which themselves seem to come from 3 populations Qinghuangdao, Beijing, and Tianjin, according to supplementary table 2) nor if they were sampled in the same time of year (pre reproduction, during breeding, after, or if they were juveniles, proportion of males or females, etc). These two aspects could affect both gene expression (through neutral or adaptive genetic variation among lowland populations that can affect gene expression, or environmental effects other than hypoxia that differ in these populations' environments or because of their sexes or age). This could potentially also affect the FST analysis done by the authors, which they use to claim that strong selective pressure acted on the expression level of some of the genes in the colonised group.

The reviewer asked how individual tree sparrows used in the transcriptomic analyses were collected. The individuals used for the hypoxia acclimation experiment and represented the ancestral lowland population were collected from the same locality (Beijing) and at the same season (i.e., pre-breeding) of the year. They are all adults and weight approximately 18g. We have clarified this in the Supplementary Table S1 and Methods. We did not distinguish males from females (both sexes look similar) under the assumption that both sexes respond similarly to hypoxia acclimation in their cardiac and flight muscle gene expression.

The Supplementary Table 2 lists the individuals that were used for sequence analyses. These individuals were only used for sequence comparisons but not for the transcriptomic analyses. The population genetic structure analyzed in a previously published study showed that there is no clear genetic divergence within the lowland population (i.e., individuals collected from Beijing, Tianjing and Qinhuangdao) or the highland population (i.e., Gangcha and Qinghai Lake). In addition, there was no clear genetic divergence between the highland and lowland populations (Qu et al. 2020).

Author response image 1.

Population genetic structure of the Eurasian Tree Sparrow (Passer montanus). The genetic structure generated using FRAPPE. The colors in each column represent the contribution from each subcluster (Qu et al. 2020). Yellow, highland population; blue, lowland population.

Q4. Impact of the work There has been work showing that populations adapted to high altitude environments show changes in their hypoxia response that differs from the short-term acclimation response of lowland population of the same species. For example, in humans, see Erzurum et al. 2007 and Peng et al. 2017, where they show that the hypoxia response cascade, which starts with the gene HIF (Hypoxia-Inducible Factor) and includes the EPO gene, which codes for erythropoietin, which in turns activates the production of red blood cell, is LESS activated in high altitude individuals compared to the activation level in lowland individuals (which gives it its name). The present work adds to this body of knowledge showing that the short-term response to hypoxia and the long term one can affect different pathways and that acclimation/plasticity does not always predict what physiological traits will evolve in populations that colonize these environments over many generations and additional selection pressure (UV exposure, temperature, nutrient availability). Altogether, this work provides new information on the evolution of reaction norms of genes associated with the physiological response to one of the main environmental variables that affects almost all animals, oxygen availability. It also provides an interesting model system to study this type of question further in a natural population of homeotherms.

Erzurum, S. C., S. Ghosh, A. J. Janocha, W. Xu, S. Bauer, N. S. Bryan, J. Tejero et al. "Higher blood flow and circulating NO products offset high-altitude hypoxia among Tibetans." Proceedings of the National Academy of Sciences 104, no. 45 (2007): 17593-17598. Peng, Y., C. Cui, Y. He, Ouzhuluobu, H. Zhang, D. Yang, Q. Zhang, Bianbazhuoma, L. Yang, Y. He, et al. 2017. Down-regulation of EPAS1 transcription and genetic adaptation of Tibetans to high-altitude hypoxia. Molecular biology and evolution 34:818-830.

Thank you for highlighting the potential novelty of our work in light of the big field. We found it very interesting to discuss our results (from a bird species) together with similar findings from humans. In the revised version of manuscript, we have discussed short-term acclimation response and long-term adaptive evolution to a high-elevation environment, as well as how our work provides understanding of the relative roles of short-term plasticity and long-term adaptation. We appreciate the two important work pointed out by the reviewer and we have also cited them in the revised version of manuscript.

Reviewer #2 (Public Review):

This is a well-written paper using gene expression in tree sparrow as model traits to distinguish between genetic effects that either reinforce or reverse initial plastic response to environmental changes. Tree sparrow tissues (cardiac and flight muscle) collected in lowland populations subject to hypoxia treatment were profiled for gene expression and compared with previously collected data in 1) highland birds; 2) lowland birds under normal condition to test for differences in directions of changes between initial plastic response and subsequent colonized response. The question is an important and interesting one but I have several major concerns on experimental design and interpretations.

Thank you for a precise summary of our work and constructive comments to improve this study. We have addressed your concerns in greater detail below.

Q1. The datasets consist of two sources of data. The hypoxia treated birds collected from the current study and highland and lowland birds in their respective native environment from a previous study. This creates a complete confounding between the hypoxia treatment and experimental batches that it is impossible to draw any conclusions. The sample size is relatively small. Basically correlation among tens of thousands of genes was computed based on merely 12 or 9 samples.

We appreciate the critical comments from the reviewer. The reviewer raised the concerns about the batch effect from birds collected from the previous study and this study. There is an important detail we didn’t describe in the previous version. All tissues from hypoxia acclimated birds and highland and lowland birds have been collected at the same time (i.e., Qu et al. 2020). RNA library construction and sequencing of these samples were also conducted at the same time, although only the transcriptomic data of lowland and highland tree sparrows were included in Qu et al. (2020). The data from acclimated birds have not been published before.

In the revised version of manuscript, we also compared log-transformed transcript per million (TPM) across all genes and determined the most conserved genes (i.e., coefficient of variance ≤  0.3 and average TPM ≥ 1 for each sample) for the flight and cardiac muscles, respectively (Hao et al. 2023). We compared the median expression levels of these conserved genes and found no difference among the lowland, hypoxia-exposed lowland, and highland tree sparrows (Wilcoxon signed-rank test, P<0.05). As these results suggested little batch effect on the transcriptomic data, we used TPM values to calculate gene expression level and intensity. This methodological detail has been further clarified in the Methods and we also provided a new supplementary Figure (Figure S5) to show the comparative results.

Author response image 2.

The median expression levels of the conserved genes (i.e., coefficient of variance ≤ 0.3 and average TPM ≥ 1 for each sample) did not differ among the lowland, hypoxia-exposed lowland, and highland tree sparrows (Wilcoxon signed-rank test, P<0.05).

The reviewer also raised the issue of sample size. We certainly would have liked to have more individuals in the study, but this was not possible due to the logistical problem of keeping wild bird in a common garden experiment for a long time. We have acknowledged this in the manuscript. In order to mitigate this we have tested the hypothesis of plasticity following by genetic change using two different tissues (cardiac and flight muscles) and two different datasets (co-expressed gene-set and muscle-associated gene-set). As all these analyses show similar results, they indicate that the main conclusion drawn from this study is robust.

Q2. Genes are classified into two classes (reversion and reinforcement) based on arbitrarily chosen thresholds. More "reversion" genes are found and this was taken as evidence reversal is more prominent. However, a trivial explanation is that genes must be expressed within a certain range and those plastic changes simply have more space to reverse direction rather than having any biological reason to do so.

Thank you for the critical comments. There are two questions raised we should like to address them separately. The first concern centered on the issue of arbitrarily chosen thresholds. In our manuscript, we used a range of thresholds, i.e., 50%, 100%, 150% and 200% of change in the gene expression levels of the ancestral lowland tree sparrow to detect genes with reinforcement and reversion plasticity. By this design we wanted to explore the magnitudes of gene expression plasticity (i.e., Ho & Zhang 2018), and whether strength of selection (i.e., genetic variation) changes with the magnitude of gene expression plasticity (i.e., Campbell-Staton et al. 2021).

As the reviewer pointed out, we have now realized that this threshold selection is arbitrarily. We have thus implemented two other categorization schemes to test the robustness of the observation of unequal proportions of genes with reinforcement and reversion plasticity. Specifically, we used a parametric bootstrap procedure as described in Ho & Zhang (2019), which aimed to identify genes resulting from genuine differences rather than random sampling errors. Bootstrap results suggested that genes exhibiting reversing plasticity significantly outnumber those exhibiting reinforcing plasticity, suggesting that our inference of an excess of genes with reversion plasticity is robust to random sampling errors. We have added these analyses to the revised version of manuscript, and provided results in the Figure 2d and Figure 3d.

Author response image 3.

Figure 2a (left) and Figure 2b (right). Frequencies of genes with reinforcement and reversion plasticity (>50%) and their subsets that acquire strong support in the parametric bootstrap analyses (≥ 950/1000).

In addition, we adapted a bin scheme (i.e., 20%, 40% and 60% bin settings along the spectrum of the reinforcement/reversion plasticity). These analyses based on different categorization schemes revealed similar results, and suggested that our inference of an excess of genes with reversion plasticity is robust. We have provided these results in the Supplementary Figure S2 and S4.

Author response image 4.

(A) and Figure S4 (B). Frequencies of genes with reinforcement and reversion plasticity in the flight and cardiac muscle. (A) For genes identified by WGCNA, all comparisons show that there are more genes showing reversion plasticity than those showing reinforcement plasticity for both the flight and cardiac msucles. (B) For genes that associated with muscle phentoypes, all comparisons show that there are more genes showing reversion plasticity than those showing reinforcement plasticity for the flight muscle, while more than 50% of comparisons support an excess of genes with reversion plasticity for the cardiac muscle. Two-tailed binomial test, NS, non-significant; , P < 0.05; , P < 0.01; **, P < 0.001.

The second issue that the reviewer raised is that the plastic changes simply have more space to reverse direction rather than having any biological reason to do so. While a causal reason why there are more genes with expression levels being reversed than those with expression levels being reinforced at the late stages is still contentious, increasingly many studies show that genes expression plasticity at the early stage may be functionally maladapted to novel environment that the species have recently colonized (i.e., lizard, Campbell-Staton et al. 2021; Escherichia coli, yeast, guppies, chickens and babblers, Ho and Zhang 2018; Ho et al. 2020; Kuo et al. 2023). Our comparisons based on the two genesets that are associated with muscle phenotypes corroborated with these previous studies and showed that initial gene expression plasticity may be nonadaptive to the novel environments (i.e., Ghalambor et al. 2015; Ho & Zhang 2018; Ho et al. 2020; Kuo et al. 2023; Campbell-Staton et al. 2021).

Q3. The correlation between plastic change and evolved divergence is an artifact due to the definitions of adaptive versus maladaptive changes. For example, the definition of adaptive changes requires that plastic change and evolved divergence are in the same direction (Figure 3a), so the positive correlation was a result of this selection (Figure 3d).

The reviewer raised an issue that the correlation between plastic change and evolved divergence is an artifact because of the definition of adaptive versus maladaptive changes, for example, Figure 3d. We agree with the reviewer that the correlation analysis is circular because the definition of adaptive and maladaptive plasticity depends on the direction of plastic change matched or opposed that of the colonized tree sparrows. We have thus removed previous Figure 3d-e and related texts from the revised version of manuscript. Meanwhile, we have changed Figure 3a to further clarify the schematic framework.

eLife Assessment

This study presents a fundamental discovery of how cerebellar climbing fibers modulate plastic changes in the somatosensory cortex by identifying both the responsible cortical circuit and the anatomical pathways. The evidence supporting the conclusions is convincing and well supported by modern neuroscience methodologies. Overall, this work represents a significant contribution that will be of broad interest to neuroscientists, especially those studying the long-distance cerebellar influence on non-motor brain functions.

Reviewer #1 (Public review):

Summary:

Silbaugh, Koster, and Hansel investigated how the cerebellar climbing fiber (CF) signals influence neuronal activity and plasticity in mouse primary somatosensory (S1) cortex. They found that optogenetic activation of CFs in the cerebellum modulates responses of cortical neurons to whisker stimulation in a cell-type-specific manner and suppresses potentiation of layer 2/3 pyramidal neurons induced by repeated whisker stimulation. This suppression of plasticity by CF activation is mediated through modulation of VIP- and SST-positive interneurons. Using transsynaptic tracing and chemogenetic approaches, the authors identified a pathway from the cerebellum through the zona incerta and the thalamic posterior medial (POm) nucleus to the S1 cortex, which underlies this functional modulation.

Strengths:

This study employed a combination of modern neuroscientific techniques, including two-photon imaging, opto- and chemo-genetic approaches, and transsynaptic tracing. The experiments were thoroughly conducted, and the results were clearly and systematically described. The interplay between the cerebellum and other brain regions - and its functional implications - is one of the major topics in this field. This study provides solid evidence for an instructive role of the cerebellum in experience-dependent plasticity in the S1 cortex.

Weaknesses:

There may be some methodological limitations, and the physiological relevance of the CF-induced plasticity modulation in the S1 cortex remains unclear. In particular, it has not been elucidated how CF activity influences the firing patterns of downstream neurons along the pathway to the S1 cortex during stimulation.

(1) Optogenetic stimulation may have activated a large population of CFs synchronously, potentially leading to strong suppression followed by massive activation in numerous cerebellar nuclear (CN) neurons. Given that there is no quantitative estimation of the stimulated area or number of activated CFs, observed effects are difficult to interpret directly. The authors should at least provide the basic stimulation parameters (coordinates of stim location, power density, spot size, estimated number of Purkinje cells included, etc.).

(2) There are CF collaterals directly innervating CN (PMID:10982464). Therefore, antidromic spikes induced by optogenetic stimulation may directly activate CN neurons. On the other hand, a previous study reported that CN neurons exhibit only weak responses to CF collateral inputs (PMID: 27047344). The authors should discuss these possibilities and the potential influence of CF collaterals on the interpretation of the results.

(3) The rationale behind the plasticity induction protocol for RWS+CF (50 ms light pulses at 1 Hz during 5 min of RWS, with a 45 ms delay relative to the onset of whisker stimulation) is unclear.

a) The authors state that 1 Hz was chosen to match the spontaneous CF firing rate (line 107); however, they also introduced a delay to mimic the CF response to whisker stimulation (line 108). This is confusing, and requires further clarification, specifically, whether the protocol was designed to reproduce spontaneous or sensory-evoked CF activity.

b) Was the timing of delivering light pulses constant or random? Given the stochastic nature of CF firing, randomly timed light pulses with an average rate of 1Hz would be more physiologically relevant. At the very least, the authors should provide a clear explanation of how the stimulation timing was implemented.

(4) CF activation modulates inhibitory interneurons in the S1 cortex (Figure 2): responses of interneurons in S1 to whisker stimulation were enhanced upon CF coactivation (Figure 2C), and these neurons were predominantly SST- and PV-positive interneurons (Figure 2H, I). In contrast, VIP-positive neurons were suppressed only in the late time window of 650-850 ms (Figure 2G). If the authors' hypothesis-that the activity of VIP neurons regulates SST- and PV-neuron activity during RWS+CF-is correct, then the activity of SST- and PV-neurons should also be increased during this late time window. The authors should clarify whether such temporal dynamics were observed or could be inferred from their data.

(5) Transsynaptic tracing from CN nicely identified zona incerta (ZI) neurons and their axon terminals in both POm and S1 (Figure 6 and Figure S7).

a) Which part of the CN (medial, interposed, or lateral) is involved in this pathway is unclear.

b) Were the electrophysiological properties of these ZI neurons consistent with those of PV neurons?

c) There appears to be a considerable number of axons of these ZI neurons projecting to the S1 cortex (Figure S7C). Would it be possible to estimate the relative density of axons projecting to the POm versus those projecting to S1? In addition, the authors should discuss the potential functional role of this direct pathway from the ZI to the S1 cortex.

Reviewer #2 (Public review):

Summary:

The authors examined long-distance influence of climbing fiber (CF) signaling in the somatosensory cortex by manipulating whiskers through stimulation. Also, they examined CF signaling using two-photon imaging and mapped projections from the cerebellum to the somatosensory cortex using transsynaptic tracing. As a final manipulation, they used chemogenetics to perturb parvalbumin-positive neurons in the zona incerta and recorded from climbing fibers.

Strengths:

There are several strengths to this paper. The recordings were carefully performed, and AAVs used were selective and specific for the cell types and pathways being analyzed. In addition, the authors used multiple approaches that support climbing fiber pathways to distal regions of the brain. This work will impact the field and describes nice methods to target difficult-to-reach brain regions, such as the inferior olive.

Weaknesses:

There are some details in the methods that could be explained further. The discussion was very short and could connect the findings in a broader way.

Reviewer #3 (Public review):

Summary:

The authors developed an interesting novel paradigm to probe the effects of cerebellar climbing fiber activation on short-term adaptation of somatosensory neocortical activity during repetitive whisker stimulation. Normally, RWS potentiated whisker responses in pyramidal cells and weakly suppressed them in interneurons, lasting for at least 1h. Crusii Optogenetic climbing fiber activation during RWS reduced or inverted these adaptive changes. This effect was generally mimicked or blocked with chemogenetic SST or VIP activation/suppression as predicted based on their "sign" in the circuit.

Strengths:

The central finding about CF modulation of S1 response adaptation is interesting, important, and convincing, and provides a jumping-off point for the field to start to think carefully about cerebellar modulation of neocortical plasticity.

Weaknesses:

The SST and VIP results appeared slightly weaker statistically, but I do not personally think this detracts from the importance of the initial finding (if there are multiple underlying mechanisms, modulating one may reproduce only a fraction of the effect size). I found the suggestion that zona incerta may be responsible for the cerebellar effects on S1 to be a more speculative result (it is not so easy with existing technology to effectively modulate this type of polysynaptic pathway), but this may be an interesting topic for the authors to follow up on in more detail in the future.

eLife Assessment

This valuable manuscript presents findings supported by solid data to identify a surprising glia-exclusive function for betapix in vascular integrity and angiogenesis. The manuscript also describes the optimisation of a modified CRISPR-based Zwitch approach to generate conditional knockouts in zebrafish

Reviewer #1 (Public review):

The manuscript by Chiu et al describes the modification of the Zwitch strategy to efficiently generate conditional knockouts of zebrafish betapix. They leverage this system to identify a surprising glia-exclusive function of betapix in mediating vascular integrity and angiogenesis. Betapix has been previously associated with vascular integrity and angiogenesis in zebrafish, and betapix function in glia has also been proposed. However, this study identifies glial betapix in vascular stability and angiogenesis for the first time.

The study derives its strength from the modified CRISPR-based Zwitch approach to identify the specific role of glial betapix (and not neuronal, mural or endothelial). Using RNA-in situ hybridisation and analysis of scRNA-Seq data, they also identify delayed maturation of neurons and glia and implicate a reduction in stathmin levels in the glial knockouts in mediating vascular homeostasis and angiogenesis. The study also implicates a betapix-zfhx3/4-vegfa axis in mediating cerebral angiogenesis.

There is both technical (the generation of conditional KOs) and knowledge-related (the exclusive role of glial betapix in vascular stability/angiogenesis) novelty in this work that is going to benefit the community significantly.

However, the study has the following major weaknesses:

(1) The lack of glia-specific rescue of betapix in the global KOs/mutants prevents the study from making a compelling case for the unexpected glial-specific function in vascular development and stability.

(2) Given the known splice-isoform specific function of betapix in haemorrhaging (Liu et al, 2007), at least an expression profile of the isoforms in glia at the relevant timepoints would have further underscored betapix function.

(3) Direct evidence of the status of endothelial cell proliferation/survival deficits, if any, in the glial betapix KOs would have provided a key mechanistic handle. It becomes all the more relevant as Liu et al, 2012 have demonstrated reduced proliferation of endothelial cells in bbh fish and linked it to deficits in angiogenesis.

Author response:

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

Reviewer #1 (Public review):

The manuscript by Chiu et al describes the modification of the Zwitch strategy to efficiently generate conditional knockouts of zebrafish betapix. They leverage this system to identify a surprising glia-exclusive function of betapix in mediating vascular integrity and angiogenesis. Betapix has been previously associated with vascular integrity and angiogenesis in zebrafish, and betapix function in glia has also been proposed. However, this study identifies glial betapix in vascular stability and angiogenesis for the first time.

The study derives its strength from the modified CRISPR-based Zwitch approach to identify the specific role of glial betapix (and not neuronal, mural, or endothelial). Using RNA-in situ hybridization and analysis of scRNA-Seq data, they also identify delayed maturation of neurons and glia and implicate a reduction in stathmin levels in the glial knockouts in mediating vascular homeostasis and angiogenesis. The study also implicates a betapix-zfhx3/4-vegfa axis in mediating cerebral angiogenesis.

There is both technical (the generation of conditional KOs) and knowledge-related (the exclusive role of glial betapix in vascular stability/angiogenesis) novelty in this work that is going to benefit the community significantly.

While the text is well written, it often elides details of experiments and relies on implicit understanding on the part of the reader. Similarly, the figure legends are laconic and often fail to provide all the relevant details.

Thanks for this reviewer on his/her overall supports on our manuscript. We have now revised the manuscript text and figure legends making them to have all relevant details as much as we can. 

Specific comments:

(1) While the evidence from cKO's implicating glial betapix in vascular stability/angiogenesis is exciting, glia-specific rescue of betapix in the global KOs/mutants (like those performed for stathmin) would be necessary to make a water-tight case for glial betapix.

We fully agree with the reviewer that it would be ideal to examine glia-specific rescue of betaPix in its global KOs. At the same time, it is difficult to achieve optimal transient expression of betaPix by injecting plasmid clone of gfap:betaPix while it takes long time to establish stable transgenic line gfap:betaPix for rescuing mutant phenotypes. We would like to pursue this line of researches in the future.

(2) Splice variants of betapix have been shown to have differential roles in haemorrhaging (Liu, 2007). What are the major glial isoforms, and are there specific splice variants in the glial that contribute to the phenotypes described?

We agree that it would be important to address whether any specific splice variants in glia contribute to betaPix mutant phenotypes. Previous studies have shown that the isoform a of betaPix is ubiquitously expressed across various tissues, while isoforms b, c, and d are predominantly expressed in the nervous system. In mice, the expression level of isoform betaPix-d is essential for the neurite outgrowth and migration. In the nervous system, we have not assessed glial specific betaPix isoforms directly. Our current data cannot rule out whether specific isoform is involved in its function in glial responses. The Zwitch cassette of betaPix resides on intron 5, thus disrupting all transcripts when Cre is activated. However, we are fully aware of the potential of identifying glial betaPix isoform with direct downstream targets. Further studies to dissect their roles in cerebral vascular development and diseases are part of our future plans.

(3) Liu et al, 2012 demonstrated reduced proliferation of endothelial cells in bbh fish and linked it to deficits in angiogenesis. Are there proliferation/survival defects in endothelial cells in the glial KOs?

We thank the reviewer for highlighting endothelial cell phenotypes in betaPix mutants. We are aware of endothelial cells might directly link to the mutant defects in angiogenesis. We assessed and quantified endothelial migration by measuring the length of developing central arteries, but we did not examine endothelial cell proliferation/survival defects in glial KOs. In our scRNA-seq analysis, the proportion of endothelial cells reduced among betaPix deficiency, indicating that endothelial cell proliferation/survival might decrease in mutants. In this endothelial cell cluster, we found disrupted transcriptional landscape in a set of angiogenic associated genes (Figure 6M). While these analysis highlights altered angiogenic transcriptome profile in endothelial cells of betaPix knockouts, we acknowledge that our study does not directly address proliferation/survival phenotypes in endothelial cells, which warrants future investigations on the role of betaPix in regulating glia-endothelial cell interaction.  

Reviewer #2 (Public review):

Summary:

Using a genetic model of beta-pix conditional trap, the authors are able to regulate the spatio-temporal depletion of beta-pix, a gene with an established role in maintaining vascular integrity (shown elsewhere). This study provides strong in vivo evidence that glial beta-pix is essential to the development of the blood-brain barrier and maintaining vascular integrity. Using genetic and biochemical approaches, the authors show that PAK1 and Stathmins are in the same signaling axis as beta-pix, and act downstream to it, potentially regulating cytoskeletal remodeling and controlling glial migration. How exactly the glial-specific (beta-pix driven-) signaling influences angiogenesis or vascular integrity is not clear.

Strengths:

(1) Developing a conditional gene-trap genetic model which allows for tracking knockin reporter driven by endogenous promoter, plus allowing for knocking down genes. This genetic model enabled the authors to address the relevant scientific questions they were interested in, i.e., a) track expression of beta-pix gene, b) deletion of beta-pix gene in a cell-specific manner.

(2) The study reveals the glial-specific role of beta-pix, which was unknown earlier. This opens up avenues for further research. (For instance, how do such (multiple) cell-specific signaling converge onto endothelial cells which build the central artery and maintain the blood-brain barriers?)

We thank this reviewer for his/her overall supports on our work.

Weaknesses:

Major:

(1) The study clearly establishes a role of beta-pix in glial cells, which regulates the length of the central artery and keeps the hemorrhages under control. Nevertheless, it is not clear how this is accomplished.

(a) Is this phenotype (hemorrhage) a result of the direct interaction of glial cells and the adjacent endothelial cells? If direct, is the communication established through junctions or through secreted molecules?

Thanks for this critical question. We attempted to address this issue by performing live imaging using light-sheet confocal microscopy, but failed to achieve sub-cellular resolution. We don’t have data to address this critical issue that warrants future investigations. 

(b) The authors do not exclude the possibility that the effects observed on endothelial cells (quantified as length of central artery) could be secondary to the phenotype observed with deletion of glial beta-pix. For instance, can glial beta-pix regulate angiogenic factors secreted by peri-vascular cells, which consequently regulate the length of the central artery or vascular integrity?

Thank the reviewer for this critical point. While we found the major defects of endothelial cell migration quantified by the central artery length, could not rule out the participation of signals from other peri-vascular cells. We fully agree that it will be important to address the cell-type specific relationship by angiogenic factors. Of note, degradation of extracellular matrix and focal adhesion is critical for the hemorrhagic phenotypes of bbh mutants. In a previous published study in our group, we found that suppressing the globally induced MEK/ERK/MMP9 signaling in bbh mutants significantly decreases hemorrhages. Accordingly, we edited a paragraph in the Discussion section on pages 24-25. We plan to continue investigating whether the complex interactions in the perivascular space contribute to vascular integrity disruption, as well as the cross-talks among different cell types during vascular development in these mutants. We believe that our model of glial specific betaPix function will guide us to further study cellular interactions in the follow-up studies.

(c) The pictorial summary of the findings (Figure 7) does not include Zfhx or Vegfa. The data do not provide clarity on how these molecules contribute (directly or indirectly) to endothelial cell integrity. Vegfaa is expressed in the central artery, but the expression of the receptor in these endothelial cells is not shown. Similarly, all other experimental analyses for Zfhx and Vegfa expression were performed in glial cells. More experimental evidence is necessary to show the regulation of angiogenesis (of endothelial cells) by glial beta-pix. Is the Vegfaa receptor present on central arteries, and how does glial depletion of beta-pix affect its expression or response of central artery endothelial cells (both pertaining to angiogenesis and vascular integrity).

Thank this reviewer for pointing out this critical issue. We have now revised the pictorial summary including Zfhx or Vegfa information in Figure 7. The key receptors of VEGF-A ligand are VEGFR-1 and VEGFR-2. In zebrafish, expression of Vegfr-2, as known as kdrl, is well-documented at endothelial cells including the hindbrain central arteries. We fully agree that it would indeed be of great value to assess changes of kdrl expression pattern after betaPix deficiency in vivo. It warrants future investigations to address how the VEGFA-VEGFR2 signaling in endothelial cells is altered in betaPix mutants.

(2) Microtubule stabilization via glial beta-pix, claimed in Figure 5M, is unclear. Magnified images for h-betapix OE and h-stmn-1 glial cells are absent. Is this migration regulated by beta-pix through its GEF activity for Cdc42/Rac?

We have now revised Figure 5M to include magnified images for h-betaPIX and h-STMN1 overexpression groups. It has been shown that there is a positive feedback loop of microtubule regulation consisting of Rac1-Pak1-Stathmin at the cell edge (Zeitz and Kierfeld, 2014 Biophys J.). Previous studies have shown betaPix activates Rac1 through its GEF activity and also regulates the activity of Pak1 via direct binding. As reported by Kwon et al., betaPix-d isoform promotes neurite outgrowth via the PAK-dependent inactivation of Stathmin1. In this work, we did not assess binding activity of betaPix to Rac1 or Pak1. Nevertheless, our data on the rescue experiments via IPA-3 suggest that betaPix deficiency impaired migration through Pak1 signaling. 

(3) Hemorrhages are caused by compromised vascular integrity, which was not measured (either qualitatively or quantitatively) throughout the manuscript. The authors do measure the length of the central artery in several gene deletion models (2I, 3C. 5F/J, 6G/K), which is indicative of artery growth/ angiogenesis. How (if at all) defects in angiogenesis are an indication of hemorrhage should be explained or established. Do these angiogenic growth defects translate into junctional defects at later developmental time points? Formation and maintenance of endothelial cell junctions within the hemorrhaging arteries should be assessed in fish with deleted beta-pix from astrocytes.

We appreciate the reviewer’s point and agree that this is a key aspect we need to clarify. To address junctional defects in our model, we re-examined the scRNA-seq data and found mild downregulation of junction protein claudin-5a (cldn5a) levels in the transcriptome analysis of the endothelial cluster (Author response image 1). We agree in principle that single cell RNA sequencing findings should be validated by immunostaining. While we did not measure junctional defects directly in this work, we have previously reported comparable tight junction protein zonula occludens-1 (ZO1) expression between siblings and bbh mutants (Yang et al., 2017 Dis Model Mech). In zebrafish, functionally characterized blood brain barrier (BBB) is only identified after 3 dpf. The lack of mature BBB might be due to the immature status of barrier signature at this developmental stage. Hemorrhage phenotype occurred around 40 hpf, and hematomas would be almost completely absorbed at later stage since most mutants recover and survive to adulthood. Thus future studies are needed to address the junctional characteristics on the cellular and molecular level in later developmental stages of betaPix mutants.   

Author response image 1.

Violin plots showing cdh5, cldn5a, cldn5b and oclna expression levels in endothelial sub-cluster. ctrl, control siblings; ko, betaPix knockouts (CRISPR mutants); 1d or 2d, 1 or 2 days post fertilization.

(4) More information is required about the quality control steps for 10X sequencing (Figure 4, number of cells, reads, etc.). What steps were taken to validate the data quality? The EC groups, 1 and 2-days post-KO are not visible in 4C. One appreciates that the progenitor group is affected the most 2 days post-KO. But since the effects are expected to be on the endothelial cell group as well (which is shown in in vivo data), an extensive analysis should be done on the EC group (like markers for junctional integrity, angiogenesis, mesenchymal interaction, etc.). Are Stathmins limited to glial cells? Are there indicators for angiogenic responses in endothelial cells?

Thank the reviewer for these critical suggestions. The detailed statements about the quality control steps for 10X sequencing are now provided in the Materials and Methods section. We validate the data quality through multiple steps, including verification of the number of viable cells used in experiment, assessment of peak shapes and fragment sizes of scRNA-seq libraries, confirmation of sufficient cell counts and sequencing reads for data analyses, and implementation of stringent filtering steps to exclude low-quality cells. Stathmins expressions as shown in Violin plots in Figure 4E and stmn1a, stmn1b and stmn4l expressions in UMAP plots in Figure S6C. These expressions are not limited to glial cells but distributed more widely among zebrafish tissues. We would like to point out that despite the small amount, the endothelial cell clusters are presented in Figure 4C with color brown. The proportions of EC groups split by four sample are visualized in Figure S6B and shown significant reduction among betaPix knockouts at 2 dpf, which had similar trend as glial progenitors. In addition, gene ontology analysis identified a set of down-regulated angiogenic genes expression in endothelial cluster (Figure 6M). We realize our interpretation of endothelial cell phenotypes was not sufficiently clear in this work and have now added sentences to the manuscript text on pages 16-17. As noted above, future studies are needed to address how glial betaPix regulates endothelial cell and BBB function. 

Reviewing Editor Comments:

comments on your manuscript. Addressing comments 1-3 from Reviewer 1 and comment 1 and its subparts from Reviewer 2 (major weaknesses) will significantly improve the manuscript by reinforcing the cell autonomous requirement of betaPix and also gain mechanistic insights. In addition, extensive proofreading and editing of the text, as well as changes to the figure, figure legends, and the discussion as indicated by both reviewers, will improve the readability and clarity of this manuscript.

Thanks for Reviewing Editor on his/her supports on this manuscript. As noted above, we are trying to address the reviewers’ comments using the data we obtained in this work, as well as our plans for future investigations. We have now made extensive proofreading and editing of manuscript text and figure legends for improving the readability and clarity of this manuscript.

Reviewer #1 (Recommendations for the authors):

(1) The Discussion is written like an introduction with very little engagement with the data generated in the manuscript. The role of betapix-Pak-stathmin and betapix-zfhx3/4-vegfaa is barely discussed and contextualised vis-à-vis the current knowledge in the field.

We appreciate the reviewer’s critical comments regarding the Discussion section. We have now revised the manuscript text on pages 20-23 to address the role of betapix-Pak-stathmin and betapix-zfhx3/4-vegfaa axis with contributions from this work.

(2) Line 145: "light sheet microscopy" - explain that this was only for experiments involving fluorescence. Currently, it reads as if the data presented in Figures 1D and E are also obtained via light sheet microscopy. E.g., the paragraph starting on line 139 does not say what line was imaged (and what it labels) to reach the conclusions reached. This detail is not there even in the associated figure legend. Similarly, line 153 discusses radial glia, but there is no indication that these were labelled using Tg (GFAP:GFP) except in the figure annotation. There are various instances of such omissions throughout the text, and they should be remedied to indicate what each line is and what it labels, at least in the first instance.

Thank the reviewer for their thoughtful points. In this revised version, we have incorporated more statements of the objectives and methodologies in the text in pages 8-9. We hope that the revised manuscript can better present the data with clarifying methodologies and materials used in this work. 

(3) Figure 1E legend: What is the haemorrhage percentage? Is it the number of embryos per experiment showing hemorrhage? Indicate in the text. In the right panel, what is the number of embryos used? Please ensure all numbers (number of embryos, experiments, etc) used to plot any data in the set of figures in the entire manuscript are clearly indicated.

Thank the reviewer for the suggestion. In this revised version, we have incorporated more detailed statements in figures and figure legends in the manuscript to show the numbers of embryos used.

(4) The Discussion section suddenly introduces the blood-brain barrier and extensively discusses it. However, while cerebral haemorrhage can disrupt the BBB and exacerbate the effects of the haemorrhage, this manuscript does not suggest that a weakened BBB is the cause of haemorrhages in betapix mutants. More likely, betapix stabilises and maintains vascular integrity, and loss of this function causes haemorrhaging and subsequent disruption of the BBB. The glial function noted in this study is likely to be distinct from the glial function in BBB development and maintenance. The authors do not show any direct evidence for the latter. These should be shortened, and only relevant aspects facilitating contextualisation of data generated in this manuscript should be retained.

We have now revised the Discussion section to reduce the introduction of blood-brain barrier and add statements according to the suggestions from both reviewers. We hope that the revisions provide a more relevant and balanced discussion.

(5) Is the scratch assay in Figure 5 controlled for differences in cell proliferation among the different manipulations?

We plated the same numbers of cells and cultured them in the same condition. Before conducted scratch assay we replaced medium with serum-free culture medium to reduce the effect from cell proliferation among the different manipulation groups. 

(6) In the glioblastoma experiments involving betapix KD, does stathmin RNA/protein decrease? What about Ser 16 phosphorylation (as shown for neurons in Kwon et al, 2020)?

STMN1 RNA was down-regulated by betaPIX deficiency, which was rescued by betaPIX overexpression in glial cells (Author response image 2). These results are similar to those from in vivo analysis (Figure 5A, 5B and S7A). We agree with the reviewer that it would been ideal to examine Ser 16 phosphorylation of Stathmin in our models. However, we believe that our data have established Stathmins function downstream to betaPix.

Author response image 2.

qRT-PCR analysis showing that betaPIX over-expression (betaPix OE) rescued STMN1 expression in betaPIX siRNA knockdown (betaPix KD) in U251 cells. Data are presented in mean ± SEM; one-way ANOVA analysis with Dunnett's test, individual P values mentioned in the figure

(7) How was the rescue of betapix in glioblastoma cells with siRNA-mediated betapix knockdown performed? Is this by betapix-resistant cDNA? Further, no information about isoforms of betapix (both for siRNA-mediated KD and rescue) or stathmin is provided.

As similar to our Zwitch method that disrupting all betaPix transcripts in vivo, the knockdown of human betaPIX were designed to target conserved region of all transcripts in glioblastoma cell lines. And the rescue human betaPIX were obtained from the U251 cDNA library, ideally all isoforms enriched in the glioblastoma cell line would be isolated. The missing details are now provided in the Materials and Methods section, page 26. 

(8) It is unclear what the authors' thoughts are on the decrease in stathmin observed and the functional outcome of this decrease. The Discussion could benefit from this.

Thanks. We have now incorporated a new paragraph in the Discussion section at pages 21-22 addressing that down-regulated expression of Stathmins is associated with functional outcome of this decrease.

(9) Zfhx4 mRNA injection is performed on bbh and betapixKO (is this a global or glial KO?) and found to rescue haemorrhaging. While vegfaa mRNA increases, it is formally possible that the rescue is not due to the increase in vegfaa (or that vegfaa is sufficient). Injection of vegfaa mRNA could address this issue.

Zfhx4 mRNA injection was performed on bbh mutants and global betapix knockouts (crispr mutants). To avoid confusion, we have now included a sentence highlighting global knockout mutants used for this rescue experiment. For the second part, we acknowledge that this study cannot definitively prove the necessity of increased vegfaa levels in the rescue experiment. However, our data established Zhfx3/4 as novel downstream effectors to betaPix in cerebral vessel development. And these effects might partly be linked to angiogenic responses regulated by Zhfx3/4. In this revised version, we carefully proposed that Vegfaa signals act downstream of betaPix-Zfhx3/4 axis and highlighted the weakness of our manuscript on not fully investigating sufficiency of Vegfaa in the Discussion section at page 24. We intend to pursue more extensive analysis in our follow-up studies.

(10) A significant part of the manuscript looks at angiogenesis/vascularisation, however, the title of the paper only reflects vessel integrity (which can be distinct from angiogenesis).

Thanks. We have now changed the title to: Glial betaPix is essential for blood vessel development in the zebrafish brain

(11) Line 366: The BBB abbreviation is used without indicating the full form. Perhaps this can be introduced in the preceding sentence.

We have now edited the following sentence: “The maturation hallmark of central nervous system (CNS) vasculature is acquisition of blood brain barrier (BBB) properties, establishing a stable environment ...” in lines 386-387, Discussion section.

(12) Line 371: "rupture" and not "rapture".

We thank the reviewer for pointing out the spelling error, and have now made this correction. 

(13) Line 416: "is enriched" instead of "enriches"?

We have now edited as: “...end feet that is enriched with aquaporin-4 ...” in line 411, page 19. 

(14) The sentence in lines 121-123 should be simplified.

We have now revised this sentence as the following: “A previous work has shown that bubblehead (bbh^fn40a) mutant has a global reduction in betaPix transcripts, and bbh^m292 mutant has a hypomorphic mutation in betaPix, thus establishing that betaPix is responsible for bubblehead mutant phenotypes [10]”. 

(15) No mention in the text of what o-dianisine labels.

We have now edited the following sentence: “By using o-dianisidine staining to label hemoglobins, we found severe brain hemorrhages ...” in lines 131-133.

(16) Line 165: Sentence requires improvement. Perhaps "Vascularisation of the central arteries in the zebrafish hindbrain ...".

We have now edited this sentence as: “Vascularisation of the central arteries in the zebrafish hindbrain starts at 29 hpf.” in this revised version (line 176). 

(17) Line 184: Why is "hematopoiesis" mentioned? The genesis of blood cells is not tested anywhere in the manuscript.

Thanks. We have now edited this statement as: “IPA-3 treatment had no effect on heamorrhage induction in betaPix^ct/ct control siblings.” 

(18) Line 222-223: Improve "increasing trends". Perhaps "increased relative proportions". Clarify "progenitors" means neuronal and glial progenitors.

We have now edited this statement: “we found that most neuronal clusters increased relative proportions ...” in this revised version.

(19) Line 232-233: "arrow indicates" - perhaps "indicated by the arrow"? Also, the arrow indicating gfap needs to be mentioned in the Figure S6A legend. Cannot understand what is meant by "as of its enriched gfap".

We have now edited in the text as: “Figure S6A, indicated by the arrow”, and added “Box area and arrow highlighting gfap expressions.” in Figure S6 legend. To avoid confusion, we have revised "as of its enriched gfap" sentence as the following: “We next focused on the progenitor cluster owing to the enriched gfap expression and the significantly reduced numbers of cells in this cluster by betaPix deficiency.”

(20) Line 239 - 240: While the sentence says "... revealed three major categories:", well, more than 3 are mentioned subsequently.

To avoid possible confusion in the text, we have now removed the sub-category examples and presented the data as: “three major categories: epigenetic remodeling, microtubule organizations and neurotransmitter secretion/transportation (Figure 4D).” 

(21) Line 252: Stathmins negatively regulate microtubule stability. Why are they referred to as "microtubule polymerization genes stathmins"?

We are thankful to the reviewer for pointing out this error, and we have now made correction in the text as “microtubule-destabilizing protein Stathmins”.

(22) Line 262-265: The citation used to indicate concurrence with mouse data is disingenuous. That study did not show a reduction in stathmin levels upon betapix loss. Rather, it showed an increase in Ser16 phosphorylation on stathmin, which reduces stathmin's microtubule destabilising function. Please elaborate on the difference between the two studies.

We completely agree with the reviewer’s statement that in the cited article, increased Ser16 phosphorylation on stathmin reduces its microtubule destabilising function. While that study did not show a reduction in Stathmin levels, others have shown that transcriptionally downregulated Stathmins are associated with the impaired neuronal and glial development. We have now revised the Discussion section by adding a new paragraph to address the disrupted homeostasis of Stathmins in these previous studies and their possible association with our data. We hope that these changes we made can clarify this issue. 

(23) Line 310: While ZFHX3 levels are reduced in betapix mutants and KD in glioblastomas, were ZFHX3 and 4 up- or downregulated in the scRNA-Seq data?

Thanks for this critical point. Indeed, our results showed that ZFHX3 and 4 down-regulated in the glial progenitor cluster in the scRNA-Seq data (Figure S8A) in betaPix knockouts and the FACS-sorted glia cells (Figure S8B). 

(24) Line 317: "... betaPix acts upstream to Zfhx3/4-VEGFA signaling in regulating angiogenesis ...". While this is established later, the data at the time of this sentence does not warrant this claim.

We agree with the reviewer’s statement and restated this sentence in the following way: “Zfhx3/4 might act as downstream effector of betaPix.”

Reviewer #2 (Recommendations for the authors):

(1) The images shown in 2E/H, 3B, 6F/J can use a schematic that helps readers to understand what to expect or look for. Splitting up the channels may also help in visualizing the vasculature clearly.

Thank the reviewer for these suggestions. In this revised version, we have included schematic diagrams in the figures and incorporated more detailed statements in the legends.

(2) Many times, arrows are pointing to structures (2E/H, 3B), but are not explained clearly (neither in the text nor in the legends). In 3B, the arrow is pointing to a negative space.

(3) Legends are minimalistic and do not provide much information. The reader is left to interpret the data on their own.

We apologize for not explaining the figures in enough details. In this revised version, we have now incorporated more detailed statements in the figure legends and have adjusted arrows in all figures.

(4) The text needs heavy proofreading. For example:

(a) Line 208- the title does not seem appropriate since the following text does not discuss Stathmins at all, which comes later.

We agree with the reviewer’s statement and restated the title in the following way: “Single-cell transcriptome profiling reveals that gfap-positive progenitors were affected in betaPix knockouts.”

(b) There is no mention of Figure 7 throughout the text.

(c) Figure 7 does not include Zfhx or Vegfaa.

Thank the reviewer for pointing out these errors. We have now revised Figure 7 and incorporated it to corresponding paragraphs in the Discussion section. 

(5) The discussion seems incoherent in its current state.

We have now revised the Discussion section according to the suggestions from both reviewers. We hope these revisions adequately address your concerns.

(6) Please include some of the following points, if possible, in the discussion.

(a) How is GEF activity of Rac/Cdc42 expected to be affected in beta-pix KO fishes?

(b) What are the possible different ways the angiogenic pathways merge onto endothelial cells? Or do the authors imagine this process to be entirely driven by glial cells (directly)?

We would like to thank the reviewer for his/her invaluable suggestions. We have now revised the Discussion section and hope that these changes can provide better and more balanced discussion. Since we have no data directly related to GEF activity of Rac/Cdc42 that might be affected in betaPix mutants, as well as have very limited data showing how glial betaPix regulates cerebral endothelial cells and BBB function, we would like to have the Discussion focused on the CRISPR-induced KI and cKO technologies, glial betaPix function and brain hemorrhage, and the putative role of betaPix-Zfhx3/4-VEGF function in central artery development. 

References:

Daub, H., Gevaert, K., Vandekerckhove, J., Sobel, A., and Hall, A. (2001). Rac/Cdc42 and p65PAK regulate the microtubule-destabilizing protein stathmin through phosphorylation at serine 16. J Biol Chem 276, 1677-1680. 10.1074/jbc.C000635200.

Kim S, Park H, Kang J, Choi S, Sadra A, Huh SO. β-PIX-d, a Member of the ARHGEF7 Guanine Nucleotide Exchange Factor Family, Activates Rac1 and Induces Neuritogenesis in Primary Cortical Neurons. Exp Neurobiol. 2024;33(5):215-224. doi:10.5607/en24026

Kwon Y, Jeon YW, Kwon M, Cho Y, Park D, Shin JE. βPix-d promotes tubulin acetylation and neurite outgrowth through a PAK/Stathmin1 signaling pathway [published correction appears in PLoS One. 2020 May 13;15(5):e0233327. doi: 10.1371/journal.pone.0233327.]. PLoS One. 2020;15(4):e0230814. Published 2020 Apr 6. doi:10.1371/journal.pone.0230814

Kwon Y, Lee SJ, Shin YK, Choi JS, Park D, Shin JE. Loss of neuronal βPix isoforms impairs neuronal morphology in the hippocampus and causes behavioral defects. Anim Cells Syst (Seoul). 2025;29(1):57-71. Published 2025 Jan 8. doi:10.1080/19768354.2024.2448999

Wittmann, T., Bokoch, G.M., and Waterman-Storer, C.M. (2004). Regulation of microtubule destabilizing activity of Op18/stathmin downstream of Rac1. J Biol Chem 279, 6196-6203.10.1074/jbc.M307261200.

Zeitz, M., and Kierfeld, J. (2014). Feedback mechanism for microtubule length regulation by stathmin gradients. Biophys J 107, 2860-2871.10.1016/j.bpj.2014.10.056.

I completely agree because giving credit to sources shows honesty and respects for others' work. This helps prevent plagiarism and makes your writing more trustworthy.

you always want the reader to know where to get the actual information. also for MLA text citations.

your MLA citing should look have this evidence.

The video talks about how good design isn’t just about looks , it’s about solving problems and making things simple and easy to use. It also shows how testing and getting feedback help make the overall experience better, so people actually enjoy using the website or app.

hi

@brinds_

Why were the Huns successful in battle but unable to effectively capture and control cities?

This means that Aryabhata made major contributions to mathematics by creating ideas and systems, like zero and decimals, that are still used in math today.

This means that Constantine shifted the main center of the Roman Empire to a new city in the East, giving it a fresh identity and greater political importance.

This means that Constantine believed a divine sign from God, shown as a cross with words promising victory, encouraged him to fight under the Christian symbol.

This means that Paul helped Christianity grow quickly by making it easier for non Jewish people to join and by spreading the faith in major cities where his message reached more people.

It’s interesting how the Gothic War started right after Justinian’s big win over the Vandals in Africa! He saw the Ostrogoths in Italy as both illegitimate rulers and religious threats, so he set out to take Italy back under his control.

It’s amazing how Muhammad and the Muslims kept going even in exile! They continued trading in Medina, spreading their faith across Arabia, and after years of battles in the 620s, they finally returned to Mecca in 630 when their enemies realized they couldn’t stop them.

This outbreak sounds terrifying! It was probably a bubonic plague, not a pneumonic one, and it attacked people’s lymph systems, causing high fevers, black swollen buboes, and delirium. Half the people who caught it didn’t survive—it must have been absolutely devastating!

Counter definition of human economy in the XXth century.

Where do they pull these four? Others?

R0:

Reviewer #1: Manuscript as reviewed meets PLOS Global Public Health publication requirements, the author(s) clearly presented the study background, methods, results, discussions and conclusion. My comments and revision request are minor formatting and suggested input. No ethics concerns at this time. Reviewer #2: This is a well-written paper with clear methodology. From the perspective of data science applied to public health, this manuscript does a great job of clearly discussing and defining its methodology, which are all the current best practices. Correcting for class imbalance was a good choice, given the low prevalence of EC in the survey population. The use of SMOTE on the training set only ensured minimal data leakage, and is the current best practice. Using such a large variety of machine learning models creates a challenge in describing each model well enough within one manuscript, and the author did a good job of balancing that challenge.

I only have a few minor suggestions toc clarify the methodology of the manuscript:

Please specify upfront how many observations were used in training and testing, and specify how many positive EC outcomes were included in the testing set. With such a low prevalence of a positive outcome in a relatively small set of observations, it is worth mentioning that there are perhaps only 10-20 positive outcomes being predicted in the test set. In the absence of weighting, it may be that characteristics of those few positive outcomes in test set are biasing the predictors, and this is worth mentioning.

Please discuss how the initial 38 variables were selected from the survey. If there was an initial expert judgment on inclusion into the variable set for feature selection, that should be mentioned.

Cluster design was mentioned in the PMA survey. This indicates that the survey includes survey weights of some kind. Please discuss whether those weights were addressed in the machine learning methods, or defend why they were not included in the model design. Survey weights can be included in machine learning models to make the predictors more representative of the population of interest.

In the discussion, please discuss the impact of low precision, where there were many false positives compared to true positives. While it is mentioned, there are consequences (e.g., loss of trust) for low precision prediction models in public health, and this characteristic of the findings could be discussed more.

Consider including a SHAP dependance plot, because potential interactions are discussed (e.g., knowledge and ad exposure) without showing evidence. A SHAP Dependence plot could take care of this.

Consider explicitly discussing the limitation of cross-sectional survey data used for prediction, where proxies were used in place of quantitative evidence (e.g., exposure to ads to proxy perceptions).

Overall, great work, timely, and well constructed. Reviewer #3: SEE word document attached with clear table

Manuscript Number: PGPH-D-25-01837 Review report

This manuscript demonstrates a significant strength in its application of advanced machine learning and Explainable AI (XAI) to address the critical public health challenge of low emergency contraceptive (EC) use in Ethiopia. By rigorously testing multiple models and using SMOTE to handle severe class imbalance, it identifies key modifiable predictors like primarily EC awareness and media exposure rather than static socioeconomic factors. The use of SHAP values transforms complex model outputs into actionable insights, revealing that knowledge gaps are the primary barrier. This approach provides a powerful, data-driven blueprint for designing targeted interventions, such as tailored media campaigns and improved health counselling, to effectively increase EC uptake and reduce unintended pregnancies. However, the following points may need to be considered, so as to improve the quality of the paper.

Topic/ subtopic Issue Suggestions Title: Predicting Utilization of Emergency Contraceptive Usage in Ethiopia and Identifying Its Predictors Using Machine Learning Redundancy. "Utilization" and "Usage" mean the same thing. Predicting the Utilization of Emergency Contraception in Ethiopia and Identifying Its Predictors Using Machine Learning. Affiliation Inconsistent institution name. on page 1 says "College of Medicine Health Science" while first page of manuscript is "College of Health Science". Use consistent affiliation name Abstract "Traditional analyses have struggled to identify complex predictors." For flow, consider: Traditional statistical analyses have struggled to… Abstract "with SMOTE used to address class imbalance" – Grammar: This is a dependent clause. It should be connected to the previous sentence. ..., and the SMOTE was used to address class imbalance. Abstract "Findings highlight that knowledge gaps, not poverty or access, are key barriers to EC use." – Clarity: "access" is vague. Be more specific. ...not poverty or physical access barriers, are key. Introduction Page 3: "moderate’s" Change to moderates ("the way the education level moderate’s religion-based stigma"). Introduction "drives excessive maternal mortality rates of over 500 deaths per 100,000 live births, drives poverty cycles, constrains girls' and women's educational and economic opportunities, and overwhelms poor healthcare infrastructures." – The word "drives" is used twice in close succession. ...contributes to high maternal mortality rates of over 500 deaths per 100,000 live births, perpetuates cycles of poverty, constrains... Introduction "is a central preventive intervention" is a crucial preventive intervention Introduction "the use of EC remains embarrassingly low" "Embarrassingly" is subjective and informal. ...remains critically low. Introduction "tempts women to shun services" Word choice not good. ...pressures women to shun services. Introduction "woefully underserved" Informal. ...significantly underserved. Introduction "yield the predictive resolution necessary" "Resolution" unusual in this context. ...yield the predictive accuracy necessary Introduction "vastness tests for fairness" – Phrase is unclear and likely an error. Correct the phrase to clarity Methods Data Source & Inclusion Criteria: The criteria for selecting the 2,334 women from the larger PMA sample of 8,943 are not explicitly stated. Was it a complete case analysis? This needs clarification as it affects the generalizability of the findings. Clarify if sampling was done or it was a complete case study Methods "The dataset demonstrates low overall missing data prevalence" –"Prevalence" is for diseases outbreaks. The missing data were minimal overall; Methods "offering robust classifier building while preserving real performance measurement." ...facilitating the development of robust classifiers while preserving a realistic assessment of performance. Results "nailing 17 true positives" Informal word choice. ...correctly identifying 17 true positives... Results "It manages this recall strength at the expense of precision, though, which sits at approximately 11%." – "Sits at" is informal. It achieves this high recall at the expense of precision, which was approximately 11%. Results "The most influential positive feature was “heard_emergency”, indicating awareness of emergency services has the greatest influence..." add which . The most influential positive feature was “heard_emergency”, which indicates that awareness of emergency contraception has the greatest influence... Results "This resonates with core assumptions of health behavior theories like the Health Belief Model, which posit perceived knowledge as a harbinger of action." "Harbinger" misused. ...which posit knowledge as a prerequisite for action. Results Page 18: "radio-implemented" Change to radio-delivered or radio-based. Results "Even positive, this reflects continued systemic disincentives documented elsewhere" – Unclear Even not a correct word. Although positively associated, this factor reflects... Results "all the sources of blunting the effect of being in contact with the health system." Grammatically incorrect and unclear. ...all of which blunt the effect of health system contact. Results "One of the thoughtful discoveries of SHAP values was the sizeable negative impact" "Thoughtful" incorrect. A notable discovery from the SHAP analysis was. Results "Isolated use of SMOTE in the training set" – "Isolated" wrong word. Applying SMOTE exclusively to the training set Results "It shifted the ML model from being a prediction device to an analysis tool, not just deciding which features were significant, but the size and sign of their effects, and significantly, potential interactions" Not clear because of parallel verbs. It transformed the ML model from a prediction device into an analytical tool, revealing not only which features were significant but also the magnitude and direction of their effects, as well as potential interactions. Results "Simulation by counterfactual SHAP analysis suggests a hypothetical 30% increase in EC knowledge might boost utilization by approximately 12.7%, a valuable public health gain." The sentence needs clearer explanation. Counterfactual simulation using SHAP values (e.g., calculating the mean impact of increasing the "heard_emergency" feature value) suggested that a 30% increase in EC knowledge could potentially increase utilization by approximately 12.7%, representing a valuable public health gain. Results "Geographic ML modeling over the geographic data would also potentially be able to further optimize resource deployment" Repetition: "Geographic" used twice. Rewrite the sentence for clarity Results "the implied vulnerability evidenced by the 'forced pregnancy' variable (despite missing data concerns) underscore" Not clear as the subject-verb disagreement. .use the word..underscores. Methods Model Selection Justification: The list of eight algorithms is comprehensive, but justification for simpler models like Naive Bayes is weak. Justify the inclusion of Naïve Bayes. Is it possible because they were included as benchmarks. Methods Evaluation Metrics: AUC-ROC emphasized, but for imbalanced problems F1-Score or Precision-Recall AUC may be better. Also consider using F1-Score or Precision as the data is not balanced or Justify the use of AUC-ROC Methods Model Performance Presentation: Logistic Regression focus unclear since Gradient Boosting achieved higher AUC-ROC (0.85). Consider Gradient Boosting as it achieved AUC-ROC 0.85 OR Explain rationale (e.g., performance vs. interpretability). Results Confusion Matrix Analysis (Figure 3): Issue: The analysis states precision is "approximately 11%." Based on the described confusion matrix (TP=17, FP=138), precision is 17 / (17+138) = 11.0%. This is a critical weakness of the model that deserves more emphasis. It means ~89% of the people predicted to be EC users were actually non-users. This has huge implications for the cost and efficiency of any intervention based on this model Discuss this trade-off explicitly: "The model's high recall (85%) comes at the cost of low precision (11%), resulting in a high false positive rate. This suggests the model is well-suited as a screening tool where identifying most true cases is prioritized over resource efficiency, but would require secondary screening or low-cost interventions to target the large number of false positives." Discussion Addressing Limitations More Forcefully: Underreporting of EC likely major issue. Add: "A key limitation is the potential for significant underreporting of EC use due to social desirability bias and stigma..." Conclusion "myth-busting" Word choice is Informal. myth-dispelling Conclusion "stock guarantees of EC" Not clear Consider write as guaranteed EC stock availability Conclusion "This research provides an ethical and evidence-based blueprint to accelerate gains in reducing maternal mortality and advancing reproductive autonomy in Ethiopia and similar settings." – Awkward phrasing. .Conside rephrasing as ..blueprint to reduce maternal mortality and advance... Reviewer #4: This manuscript applies machine learning (ML) and explainable AI (XAI) methods to predict emergency contraceptive (EC) use among women in Ethiopia, using data from the 2023 PMA survey. The authors compare eight algorithms, address severe class imbalance with SMOTE, and use SHAP values to interpret predictors. They find that awareness of EC is the strongest predictor, followed by media exposure and health facility discussions, while demographic variables show limited predictive value.

However, the results as currently presented are unreliable. Major inconsistencies in reported performance metrics (e.g., contradictory precision values, implausible Naive Bayes results, inflated accuracy) call into question the validity of the analyses. In addition, the small number of EC users makes the modeling unstable, and subgroup analyses are not feasible with this dataset. These issues, combined with over-interpretation of SHAP as causal, limit both the methodological credibility and substantive contribution of the paper.

Contradictory precision results The performance metrics are inconsistent. Table 4 shows Logistic Regression with SMOTE achieving precision = 0.72 and recall = 0.85, yet the confusion matrix description reports precision at only ~11%. These cannot both be correct. This discrepancy raises questions about the accuracy of the reported results and must be clarified.

Inflated accuracy The reported accuracy of 0.95 for Logistic Regression with SMOTE appears implausibly high given the extreme class imbalance (4.4% EC use). Accuracy is not an informative measure in this context, and such values raise concerns about potential data leakage or overly optimistic validation. The authors should confirm that the outcome variable or proxy features were not inadvertently included in the predictors.

Over-interpretation of SHAP The SHAP analysis is framed in causal terms (e.g., a 30% increase in knowledge leading to a 12.7% increase in use). SHAP values describe associations within the model, not causal effects. The manuscript should temper these statements and present SHAP findings as indicators of relative predictive importance, not intervention outcomes.

Implausible Naive Bayes results Naive Bayes is reported as having accuracy of only 0.06 pre-SMOTE. Given that 95% of the sample did not use EC, even a trivial majority-class classifier would achieve ~95% accuracy. Such a result suggests an error in coding or reporting that must be checked.

Small minority class vs. model complexity Only 103 EC users were present in the dataset. Training and tuning eight algorithms with hyperparameter searches on such a small minority class risks overfitting and unstable results, even with SMOTE. This limitation should be acknowledged explicitly, with emphasis on the need for validation on independent samples.

Subgroup analysis claims The manuscript claims fairness testing across subgroups (rural/urban, religion, age), but no results are presented. With so few EC users, subgroup analyses would be underpowered and unreliable. It would be more appropriate to note this limitation rather than imply subgroup robustness.

Causality Issue The manuscript repeatedly interprets predictive associations as though they were causal effects. For example, SHAP values are used to suggest that increasing knowledge by 30% would increase EC use by 12.7%. Since the data are cross-sectional and observational, such statements are not justified. Machine learning models in this setting can identify predictive patterns, but they cannot establish causal relationships between predictors and outcomes. This overreach is particularly concerning because it could mislead policymakers or practitioners into believing the study provides evidence of causal effects. Reviewer #5: Summary This study investigates the underuse of emergency contraception in Ethiopia using a machine learning framework. Strengths include the application of multiple algorithms, careful handling of class imbalance, and the use of Explainable AI to interpret model outputs. The paper is generally well-structured, and the methodological workflow is presented clearly. At the same time, the results are presented in a way that overstates the model’s practical utility while giving insufficient attention to the precision–recall trade-off. The manuscript should be revised to consistently acknowledge the low precision across the abstract, results, and discussion, and to provide a clear justification for the relevance of a high-recall, low-precision model in this public health context. The limitation posed by the small number of positive cases in the validation set should also be explicitly discussed. Addressing these points is necessary to strengthen the scientific validity of the work. Specific comments 1. Title; It should be shortened to remove redundancy since Utilization and Usage mean the same thing 2. Abstract. I think something key was missed. The aurthors state a recall of 0.85 without mentioning the precision. I see that (Figure 3, page 20) show that the precision is approximately 11%. My understanding of this that for every 100 women the model flags as likely EC non-users who need intervention, 89 of them are false alarms. An abstract must present a balanced view of performance. 3. Methods (About the data): A sample size of 2,334 with a 4.4% prevalence means you only have ~103 positive cases (EC users). After an 80/20 train-test split, your test set contains only ~21 positive cases. This number is critically small and raises serious questions about the stability and generalizability of your reported performance metrics. A different random split could yield vastly different results. I suggest that such a major limitation is addressed upfront in the limitations section and acknowledged in the methods section. 4. Data balancing; I like the write up of this section 5. Evaluation Metrics; The text states the test set has 18.7% EC users, but the abstract and data balancing section state the overall prevalence is 4.4%. Please clarify this discrepancy. Is 18.7% a typo? Or did the stratified split result in a test set with a much higher prevalence than the overall dataset? This needs to be consistent. Could you also add the precision-recall plots, since you state that they were tracked. 6. Results: - In Table 4, the columns are F1 and Score. This seems like a typo. It should likely be a single column: F1 Score. Please correct. - Lastly, i think it would be good to acknowledge the weaknesses of SMOTE Reviewer #6: The title of the article is: Predicting Utilization of Emergency Contraceptive Usage in Ethiopia and Identifying Its Predictors Using Machine Learning. The author explains that traditional analyses have struggled to identify complex predictors and therefore they used machine learning (ML) and Explainable AI (XAI) to improve the prediction and interpretability of Emergency Contraceptive (EC) use. The paper can be published with the following corrections and some are extremely important. In particular methodological perspectives. Category Authors Contribution Comments Objectives The primary objectives are twofold:

one, to predict the likelihood of EC use with far greater accuracy than conventional regression techniques;

two, to identify the key modifiable socio-behavioural predictors e.g., self-efficacy, mass media exposure, provider perception, and women's autonomy through XAI methods like SHAP values to yield interpretability and actionable insights. First objective can be modified. Far greater is a vague statement. Measuring accuracy is an indicator of choosing between models but conventional regression techniques why has a problem in this study should focus on that.

Second objective seems motivation of the study. This objective should be written in clear sentence. Identify predictors to yield interpretability and actionable insights are subjective things. These objective seems ambiguous.

Methodological view Page 5: Methodologically, it represents a new contribution by rigorously testing the performance of eight alternative ML classifiers and developing an optimized analytical pipeline specifically designed to handle skewed healthcare datasets prevalent in rare outcomes like EC use

Theoretically, it applies the Socio-Ecological Model (SEM) framework to hierarchically analyze predictors at levels of individual (knowledge, attitudes), interpersonal (partner communication, family influence), community (stigma norms, access), and policy (health system factors) providing an integrated explanation for the interrelating influences on EC behavior. It is not methodological contribution.

Moreover, author mentioned theoretical contribution. However, it is just exploratory of the data.

Methodology In page 4: In contrast to conventional statistical approaches, ML algorithms, such as random forests, gradient boosting machines (e.g., XGBoost), and neural networks, can particularly identify complex, high-dimensional patterns within diverse data sets, properly manage missing data, and produce personalized risk predictions with improved accuracy Author mentioned several times about conventional statistical technique. However, in the report author directly reported the model performance of ML. My suggestion is to first run the analysis using traditional or conventional methods and then compare with ML techniques. This is very important. Outcome Variable Page 8: The outcome of interest is EC Usage, a binary measure of whether emergency contraception was used in the last 12 months. This is the dependent variable for analysis. Redundant as at the beginning you mentioned outcome of interest is….. Missing data For handling missingness in our data, a stratified approach based on missingness mechanisms and rates was followed and so on……….. The author used many approaches and it is difficult to keep track. So it is better to explain step by step and pros and cons of each process. Moreover, explain why this approach is best in this study Variables Page 12

Lots of category under one variable. Some category has very few observations. Justify the necessity. May be we can also show some cross-tabulation analysis result and report the p-value. Research Gap Page 19: The research goes beyond the correlational limitations of previous studies by utilizing predictive analytics to identify the modifiable factors and approximate their hypothetical effects What do you mean by correlational limitations? Moreover, over the report the previous studies were not mentioned in comparison to the authors current approaches. Sa add some recent references and explain the research gap. The Machine learning techniques are not new. So it is required to mention how those machine learning helps in your study as a novelty. All over the report there is a missing of synchronization and coherence of sentences. Moreover, the references, table titles etc are not space maintained. Abstract 1. SMOTE and SHAP 2. Conversely, recent reproductive events such as unintended pregnancy were linked to non-use. Static demographic factors showed poor predictive value. Findings highlight that knowledge gaps, not poverty or access, are key barriers to EC use. Tailored media campaigns and routine health counseling could enhance EC uptake. ML and XAI offer powerful tools for guiding targeted reproductive health interventions. 1. Did not mention what it is?

1. The message of these sentences are not coherent. I think author can check the whole paper from an English native reviewer.

R1:

Reviewer #4: I appreciate the authors' thoughtful revisions and detailed responses. Several of my earlier comments were addressed—specifically, the correction of Naive Bayes reporting errors, improved acknowledgment of sample size limitations, and removal of unsupported subgroup analyses. These are welcome improvements. However, key concerns about the internal consistency of results, causal interpretation of SHAP analyses, and overextension of policy recommendations remain unresolved.

First, while the outdated "11% precision" text has been removed, the confusion matrix values (TP=102, FP=180, FN=18) still do not correspond to the reported performance metrics. With these numbers, precision would equal roughly 0.36, not the 0.72 cited in Table 4. This suggests an ongoing internal inconsistency between the descriptive counts and the summary metrics. The lack of alignment raises continuing doubts about the reliability of the reported model performance.

Second, the manuscript still places heavy emphasis on accuracy values approaching 0.92–0.95 despite a highly imbalanced outcome (4.4% EC use). Although the authors state that AUC-ROC and recall were prioritized, the presentation continues to foreground accuracy, which is misleading in this context. No calibration or uncertainty measures (e.g., Brier score, calibration curve) have been added, leaving the reader without a sense of how well the predicted probabilities reflect actual risk.

Third, although the authors softened their language, the interpretation of SHAP values remains quasi-causal. The new statement—"counterfactual simulation using SHAP values … suggested that a 30% increase in EC knowledge could potentially increase utilization by approximately 12.7%", still presents SHAP outputs as if they represent real-world intervention effects. SHAP analysis identifies predictive associations within a model; it does not estimate the causal impact of changing a feature in the population. Likewise, subsequent phrases such as “integrating a predictive risk-scoring tool can help identify women at high risk” and “geographic machine learning modeling can optimize resource deployment” continue to frame the model as a validated operational tool. These remain prescriptive policy claims that move beyond what a cross-sectional, unvalidated predictive study can substantiate.

Finally, while the tone of the manuscript has improved, the discussion still reads as policy advocacy rather than analytical interpretation. Phrases like "representing a valuable public health gain”" and "can help optimize resource deployment" give the impression of proven effectiveness rather than exploratory modeling. A clearer distinction between predictive insights and causal or operational evidence is necessary for the study to maintain methodological integrity.

Authority of the author: I did a bit of research, and could not find any information as to the author's authority on this subject other then what is written in the her bio. Also, If there was a per-review by some type of running coach or fitness expert, that would in increase the credibility of this article.

Accuracy of text: According to the National Institute of Health and other sources, red blood cells have an average life span on 120 days not "20 days". nih.gov

Try and find every opportunity to get yourself up into higher altitudes.

High altitude training helps your body adapt to performing with less oxygen.

Currency of this article: This Information Is not out of date. I would consider this to be out of date if the article was written more than 8 years ago.

Research on this text: The high elevation at the this Olympics mainly affected the distance runners and swimmers, but it is believed that it actually helped athletes in the long jump, high jump, and pole vault.

Purpose of this text: To examine why athletes train at high altitudes and how you can apply this approach into your own running habits.

Research on this text: The world record Iron man runner Kristian Blummenfelt trains in Spain's Sierra Nevada Mountains. YouTube

The structure of this article is in the form a typical blog post.

This summarizes the main points and gives the reader a finishing statement to reflect upon.

Your body starts adapting to the altitude quickly, by producing more blood cells.

You should train at altitudes around 7000 feet above sea level in areas such as Flagstaff Arizona.

This is the main topic of this article.

NUT GRAF -confirmed by Alex

Kind of a BOWTIE ENDING (w/ Lede)

LEDE

TEXT: The line "Plant a city. sprout a chain" is symbolic saying that building a city is like planting a seed. "Sprout a chain" could mean starting connections and communities. Secondly, "Water it with maple stain" could be symbolic for the history and growth of Tacoma, hinting at its imperfections. "Grow a street, and dance around the rooty, broken open ground" I feel like this means to celebrating the imperfection, messiness and growth in Tacoma . The main message is that Tacoma is a growing city shaped by the people in it, despite its imperfections, its able to thrive through human connection.

COLOR: This poster consists of orange, red, faded teal, and a little of magenta. Orange means warmth, creativity and enthusiasm. Red means, passion, energy and courage. Faded Teal means calmness, rejuvenation and clarity. Lastly magenta it means passion, creativity and emotional balance.

an interesting way of reframing the issue

none of the following explanations are as parsimonious as the simplest which is: "monotheists are just smarter because they are further along in intellectual development toward atheism." religion among the most primitive people isn't even characterized by gods — they have earthen spirits in the trees and soil. then you get a panoply of gods in the civs of antiquity, then a supreme One Being, before genuine materialism, etc.

This is the reason why idle games took off, something simple that fills in a little gap.

There are many ways to participate and video games have many ways to do so as well and the networks that create connect to specific social networks depending on the games.

How one plays a game says a lot more than the gameplay itself I've seen my cousins and nephews exhibit unusual behavior with games.

Its true I played many a games when I was younger and got many things back emotionally, psychologically, and physically.

There is always a media people will blame and it is always the next media medium, its all to make answers by blaming the medium to be the cause.

Its more than just male-dominated identity and lowbrow there is a lot more subjects, genres, and target audiences that have flooded the market.

Game play has its effect on people depending on content, pace, and the subject as a whole.

Factors like these have influence depending on what media is being utilized.

When it comes to a persons age were they are at in life can determine where they will learn more effectively genre wise.

There are game genres that instill certain cognitive features when utilized depending on the genre.

It is good that people are now looking further into video games now instead of surface level.

Sometimes people focus on things that should not be the concern over actual concerns.

The various genres can teach a lot of knowledge to people as long as the lessons teach the knowledge effectively.

Newer gaming mediums have been even implemented in newer ways with certain organizations even building lessons into them.

This was an image pushed by many and is still talked about to this day just not as bad now than back then.

TEXT: In this poster the line "Holy city of both body and mind"portrays Tacoma as a living being with both strength and is full of ideas. Then it goes on to say "You are the well-crafted city. We are your true sons and daughter." This signifies that Tacoma was intentionally made, where citizens of tacoma almost feel like Tacoma is a family, that they belong in. Lastly "A sacredness of love and death, dwells in your noise and smoky break." This line means that tacoma is filled with history and other memories as well, where both life and the dead linger in the city. The text repeats Tacoma, which points empathizes on it, along with different text fonts. The overall message is that tacoma is a vibrant, lively city of both physical and intellectual strength, where residents feel a deep sense of belongingness where history, life, and emotion coexist.

COMPOSITION: There is this big arm and hand holding a ranch. This just builds significance to Tacoma being well crafted, which is written on the bottom of the arm, in cursive. Below the image is the text its smaller than the image. The background color of the whole poster is this white light blue mix.

¿Cómo puede el enfoque de Cardumem, basado en los metatools y la programación en Lua/YueScript, influir en la forma en que gestionamos y preservamos el conocimiento dentro de las comunidades académicas y sociales, especialmente en contextos del Sur Global? Me cuestiono cómo podríamos usar este tipo de metaherramientas para fortalecer la memoria de comunidades indígenas, campesinas o urbanas, sin imponerles estructuras externas. Tal vez la clave esté en que Cardumem no busca imponer un modelo, sino abrir la posibilidad de construir nuestras propias maneras de registrar y compartir el saber, desde el Sur, con nuestras voces y en nuestros propios lenguajes digitales.

Effective engagement is a skill in its own right and when done right teaches a whole lot better.

Everyone's lives are different and this is always the case and not everyone doesn't have the same opportunities like everyone else.

lower-income communities have all suffered in various communities as higher-income and other schools get prior attention.

The factors outside of school are just as important as the ones in school and will continue to be major issues in certain communities.

They are troubling and will continue to be issues as the years going on and this will continue to be bad in certain areas as time goes on.

They will always be there for anything and of course they will continue on.

They're only available when the tech companies demand it but never guaranteed now more than ever.

It has become financially challenging as you needed a degree for better jobs which are no longer guaranteed for courses that are more expensive now more than ever.

These modern markets can really be confusing and terrible as the new future looks more uncertain for those who had expectations with new unfortunate results.

This is true even in my family as they pressure us to get degrees for better jobs that may not be there anymore, this is what I worry.

This is a thought provoking concept- is the archivist self serving in their work? Who should archiving benefit? The archivist or the communities represented in the archives?

The author here is making a very interesting statement- the intent of the digitalization of these files is questional

Ungemah realized that the most significant lesson she learned from her students was how to confront inequality head-on. They helped her understand the structural issues within education: despite their efforts, many students from disadvantaged backgrounds remain systematically excluded from higher education. A teacher's awakening stems not only from professional training but also from the shared reality of education experienced alongside students. Students are not merely learners; they are also the ones who reveal the truth about the education system.

Can't be good for capitalism.

It shouldn't surprise me but it does. But percentage of students who selfrevealed is 5.8%

Accountability doesn't necessarily ensure that organizations will take ownership of their AI-generated outcomes. It's a bit of a stretch to say it ensures that, I think it would be better if he said 'encourages'. A better argument would be establishing an AI ethical code of conduct.