Exercise Engages Coordinated Neuron–Glia Signaling to Shape Spinal Cord Plasticity

Reviewer #1 (Public review):

Summary:

The authors integrated bulk proteomics, single-nucleus RNA sequencing, and cellular communication pipelines to map molecular changes in the mouse lumbar spinal cord following endurance training versus acute exhaustive exercise. This kind of data is currently missing in the literature for the healthy spinal cord; therefore, this work represents a useful resource for the community for the investigation of cellular mechanisms of exercise-induced neuroplasticity. The authors found that endurance training elicited robust plastic transcriptional changes in the glia, in genes involved in synaptic modulation, axon development, and intercellular signaling, with cell-specific differences. Acute exhaustive exercise triggered a more nuanced biphasic temporal response in metabolic and synaptic genes, which was different in trained versus sedentary mice. Although cholinergic neurons did not show robust gene expression changes, they were found to be central hubs for communication with glia, suggesting that their cues may act as upstream regulators of glial plasticity.

Strengths:

Nuclei fixation minimized unwanted RNA degradation and tissue processing-driven expression changes. However, in the text, it needs to be acknowledged that the fixation step was performed only after nuclei isolation, and not at the stage of spinal cord tissue collection. The time course study design allowed for the distinction of different temporal gene expression trajectories.

Weaknesses:

No clear indication of the number of biological replicates is given. No validation of the key findings with alternative methods is presented.

Some aspects of data analysis need to be clarified:

(1) Methods

a) Voluntary exercise: the authors should indicate whether the mice were singly housed, and, if not, clarify that the indicated mean km/day is an average of the mice in the cage.

b) The Authors should indicate more precisely which lumbar level of the spinal cord was used and the number of biological replicates.

c) The Authors should indicate the number of highly variable features and PCs (dims) used for Seurat and provide a QC metric table.

(2) Results and Figures

a) Bulk proteomic analysis: The authors used Pval-and not FDR- to assess differentially abundant proteins. Can the author indicate how many proteins passed a more stringent FDR cutoff? For GO analysis: the authors should indicate what background/reference was used.

b) Figure 1B and Figure S1B-C: the differences in total mass and relative lean mass are very subtle, even if statistically significant. This needs to be acknowledged in the relevant sentences.

c) Figure 2 and Figure S2E panels G and H are inverted.

d) Heatmaps in Figures 1F and 2 Figure 2E-F: some of the proteins and genes listed in the text are not present in the heatmaps (TIM22 and FABP4; Smap25 and Slc4a4). Please check the correspondence of the text with the heatmap, and indicate with an arrow the listed proteins and genes.

e) Page 9 "trained mice displayed a modest increase of oligodendrocytes 24h": from the plot, it looks to me like a decrease rather than an increase.

f) Figure 4 depicts expression changes in selected metabolism and synaptic activity-related genes: it would be useful to add a table as a supplementary file with expression data of all the synaptic and metabolic genes in addition to the ones that were selected.

Reviewer #2 (Public review):

Mansingh et al., investigate the impact of voluntary wheel training and acute physical exercise on the transcriptomic and proteomic profile of spinal cord tissues from young adult mice. They first describe the proteomic and transcriptomic differences between sedentary mice and mice provided with running wheels for voluntary exercise. They show that voluntary physical exercise induces changes at a transcriptional level as well as at a proteomic level, with most of these effects restricted to glial cells. They further analyze the putative cell interactions that are induced in the context of physical training and describe the specificity of transcriptional changes in the different cell populations. Using the same multi-omics pipeline, the authors assess dynamic changes in sedentary and trained mice 6 and 24 hours following a bout of physical exercise until exhaustion. Importantly, they demonstrate that the impact of this single bout to exhaustion is modified in mice that have access to running wheels compared with sedentary mice, with a reduced amplitude of the reaction and a faster resolution of changes caused by exercise until exhaustion.

Altogether, this study provides a useful description of the transcriptional changes at play following voluntary physical training and, importantly, uncovers the role of this training in shaping future transcriptomic reactions to a stressful bout of exercise until exhaustion. However, the conclusions of the manuscripts would be strengthened by the clarification of the methods, a better use of the proteomic data regarding the transcriptomic datasets, and a cross-validation of the main claims currently based solely on transcriptomic datasets.

(1) In this study, the housing strategy used is key as it will impact both the proteome and transcriptome of cells in the central nervous system. It can be difficult to measure the running activity of individual mice if they are not housed individually. Yet, individual housing has a major impact on the nervous system and notably on glial cells. Therefore, a better description of the housing strategy for the sedentary and trained group during the 6 weeks of training is required.

(2) In the first part of the paper that uses the results from the first set of multi-omics data, the protocol used is not clear. From Figure 1A, it seems that the mice went through a max performance test before and after the 6-week period in which the two groups had different life experiences (voluntary running versus sedentary). Since in the methods the maximal test protocol is effectively an exercise until exhaustion, it is difficult to understand why the authors defined this first experiment as the one allowing them to test "molecular remodeling in the spinal cord at rest". Also, it is not clear how long after the max performance test the tissues were collected. If indeed the mice went through the max endurance test before tissue collection, it is not a condition at rest, and this first protocol in some way looks like a duplication of a subpart of the second experiment. If mice did not go through this max performance test, it needs to be clarified both in the text and in the figure.

(3) One of the strengths of this study is its multi-omics approach assessing changes at both transcriptomic and proteomic levels. Yet, the use by authors of the proteomic datasets is minimal, and there are no comments on how the proteomic and transcriptomic datasets support each other. Changes at the transcriptional level do not necessarily translate into changes at the protein level. Therefore, it would improve the quality of the study if authors could use the bulk proteomic data in relation to the transcriptomic dataset. The fact that the proteomic datasets do not provide the identity of the cells from which the changes originate should not prevent authors from putting them in perspective with transcriptomic results.

(4) None of the results from the single-nucleus RNA sequencing are cross-validated with, for instance, in situ hybridizations. It would improve the strength of the claim if some findings, in particular regarding the dynamics of the changes 6 vs 24h after exhaustion bout, were cross-validated.

(5) Although the authors note as a limitation that cholinergic neurons were underrepresented in their dataset, since one of the main claims of the manuscript relates to them, it calls for some additional comments on the identity of the cholinergic neurons present in their dataset. There are different populations of spinal cholinergic neurons with very different functions. It would greatly improve the strength of this result if the authors could identify which cholinergic neurons show these changes (or at least which proportion of the different cholinergic population is present in their datasets). For instance, which proportion of cholinergic neurons are expressing classical markers of motor neurons versus markers of cholinergic interneurons (for instance, from the V0c population).

Reviewer #3 (Public review):

Summary:

Mansingh et al. used single-nucleus transcriptional and bulk proteomic profiling to characterize how gene expression changes in the lumbar spinal cord of adult, healthy mice after training (voluntary wheel-running exercise) and acutely after forced treadmill exercise. They found (1) that training was associated with a number of differentially expressed proteins, (2) training was associated with cell-type specific changes in transcription, notably glial cells had the highest numbers of differentially expressed genes, and (3) that trained mice had blunted transcriptional response to an acute exercise bout compared to sedentary mice.

Strengths:

The characterization of the changes to the proteome and the transcriptome associated with exercise will undoubtedly be a useful resource for scientists interested in the effects of exercise on central nervous system gene expression and may inspire mechanistic follow-up studies. The authors nicely use pathway and intercellular communication analyses to distill the complex dataset into key trends.

Weaknesses:

Weaknesses of this paper include two aspects of the analyses that underexplored the rich dataset. The analysis fails to explicitly compare the proteome and transcriptome results. Do the differentially expressed proteins correspond to the differentially expressed genes? If so, in which cell types? If not, why not? Comparison of the GO terms from the proteome dataset and the GSEA terms from the single-nucleus transcriptome dataset suggests that the same gene families were not identified in both data sets. I expect that integrating analyses across these datasets would help make the study truly multi-omic and highlight which expression changes are the most abundant and consistent across approaches. Second, the authors emphasize that related studies do not account for inter-individual variability in both the introduction and discussion. This aspect of the authors' dataset is also underexplored - the transcriptomic data appear to be pooled across animals, and only a single panel shows protein expression from individual animals (Fig. 1F). Is the variability in Figure 1F explainable by the amount of running on the wheel?