A skeletal muscle atlas shows neuromuscular junction adaptations to growth and atrophy

SUMMARY

The molecular bases underlying muscle atrophy, as occurs during disuse or aging, and activity-induced hypertrophy remain poorly understood. A major challenge has been defining the diverse cellular and niche environments within skeletal muscle, mostly composed of multinucleated myofibers. Here, we present a single-nucleus and single-cell transcriptomic atlas, coupled with spatial profiling, of mouse limb skeletal muscle in resting conditions and during experimentally induced atrophy or hypertrophy. We identify condition-dependent shifts in muscle-resident cell populations and fiber type-specific transcriptional responses. We also uncover extensive remodeling of the neuromuscular junction (NMJ), including the emergence of specialized synaptic myonuclei (SynM) and terminal Schwann cells (tSCs) associated with atrophic or hypertrophic states. High-resolution 3D imaging and spatial transcriptomics confirm these changes at the tissue level. Similar NMJ alterations are observed in denervated and exercised human muscle, supporting the translational relevance of this atlas for studying muscle plasticity and identifying therapeutic targets in muscle-related diseases.

Graphical Abstract

eTOC Blurb

Campanario et al. present a multimodal, spatially resolved transcriptomic atlas of skeletal muscle responding to neural and mechanical cues during atrophy or hypertrophy. Loss of neural input drives specialization of synaptic myonuclei and Schwann cells, highlighting neuromuscular junction plasticity and providing valuable resources for future studies in human muscle biology.

INTRODUCTION

Adult skeletal muscle can adjust its mass and function in response to diverse physiological and pathological stresses, including prolonged inactivity, increased workload, and aging. Skeletal muscle influences whole-body metabolism and systemic health¹, and loss of muscle mass and function (from disuse, disease, or aging) can have significant functional and clinical consequences.

Skeletal muscle consists predominantly of large, multinucleated myofibers that differ in their contractile and metabolic properties, typically classified as slow-twitch oxidative (type I) or fast-twitch glycolytic (type II). These myofibers are controlled by motoneuron inputs at the neuromuscular junction (NMJ)^2,3. In addition, skeletal muscle contains several less abundant mononucleated cell types, including muscle stem cells (MuSCs; also called satellite cells), fibro-adipogenic progenitors (FAPs), adipocytes, immune cells, vascular cells, and Schwann cells (SCs) that ensheathe motoneuron axons^4–7. Communication among these cells is essential for maintaining muscle homeostasis and for coordinating responses to injury, mechanical demand, or loss of neural input⁸. Whereas exercise-induced hypertrophy enhances muscle strength and performance, muscle atrophy associated with aging (sarcopenia) or denervation results in loss of mass and force, often leading to frailty, disability, and diminished quality of life^9,10.

In this study, we generated a comprehensive transcriptomic atlas of skeletal muscle spanning a continuum of atrophy and growth states. By integrating single-nucleus and single-cell RNA sequencing (sn/scRNA-seq), spatial transcriptomics, and whole-mount 3D confocal imaging, we delineate how resting muscle adapts to contrasting neural and mechanical stimuli. In particular, we characterize in unprecedented detail the cellular and molecular remodeling of the NMJ domain under conditions of atrophy and hypertrophy. Understanding these adaptive and degenerative processes may help to identify new targets for therapeutic interventions for conditions involving muscle atrophy.

RESULTS

Transcriptional landscape of a continuum of muscle in resting, atrophying or growing states

To investigate the molecular adaptations of adult skeletal muscle under opposing states of atrophy and growth, we isolated the plantaris muscle from mice assigned to three experimental conditions: resting (basal), atrophic following denervation (hereafter, Den), and hypertrophic following functional overload (OVL) (Figure 1A, left). Samples were collected at three or seven days (d) post-intervention (Figure 1A, right). Significant changes in muscle mass (Figure 1B) and myofiber cross-sectional area (CSA) were observed by 7d (Figures S1A and S1B), consistent with adaptive loss or gain of muscle tissue.

Figure 1. Transcriptomic atlas of atrophic and hypertrophic skeletal muscle.

(A) Schematic of the surgical procedures used to induce muscle atrophy (sciatic nerve transection, SNT) or hypertrophy (gastrocnemius excision, GNE) (left), and the timeline for plantaris muscle collection at 3 or 7 days after intervention (right). (B) Plantaris muscle weight (mg) normalized to body weight (g) (n = 19–22 mice). (C and D) UMAP visualization of all nuclei profiled by snRNA-seq (C) and the same dataset colored by condition (basal, 3 or 7 days after denervation [Den], and 3 or 7 days after overload [OVL]) (D). (E) Bubble plot showing selected marker genes used to identify the major nuclear populations. Color denotes average expression, and dot size represents the percentage of nuclei expressing each marker. (F) Relative abundance of each primary cell type across conditions and time points in the snRNA-seq atlas. (B) Two-tailed independent-samples t tests, p < 0.05. See also Figure S1–S2 and Table S1.

We performed whole-muscle single-nucleus RNA sequencing (snRNA-seq) on pooled plantaris samples from each group, generating a dataset comprising 75,343 nuclei, including 24,383 and 26,427 nuclei from the Den and OVL conditions, respectively. Unsupervised clustering identified all major cell types typically found in hindlimb muscle (Figures 1C and 1D), including myonuclei, fibro-adipogenic progenitors (FAPs), mural and vascular endothelial cells (ECs), muscle stem cells (MuSCs), immune cells, and Schwann cells (SCs), each defined by canonical lineage markers (Figure 1E). Within the myonuclear compartment, two smaller clusters distinct from the predominant “body” myonuclei (hereafter, “myonuclei”) were detected, corresponding to specialized domains of the myofiber—the neuromuscular junction (NMJ) and the myotendinous junction (MTJ) (Figures 1C, 1E, and 1F).

Across all conditions, myonuclei were the most abundant population, followed by FAPs, except for 3d OVL, where immune cells are more abundant (Figure 1F), likely reflecting mild myofiber damage and repair responses secondary to mechanical loading. Further, the proportion of MuSCs increased at both 3d and 7d, while that of myonuclei decreased, most prominently at 3d, indicating an early activation of mononuclear cell compartments in response to overload. Conversely, in Den muscle, the relative abundance of myonuclei decreased without a corresponding expansion of MuSCs, suggesting a net loss of myonuclei per myofiber as the fibers underwent progressive atrophy (Figure 1F). The proportion of FAPs was higher in both atrophic and hypertrophic conditions as compared to baseline (Figure 1F). Interestingly, the increased FAP abundance in Den muscle is consistent with previous reports of FAP accumulation after acute denervation¹¹.

Because interstitial cell types were underrepresented in our initial snRNA-seq dataset, we complemented the analysis with single-cell RNA sequencing (scRNA-seq) on dissociated plantaris muscle from the same experimental conditions. Fluorescence-activated cell sorting (FACS) was used to enrich for MuSCs, ECs, SCs, and other rare populations, following depletion of FAPs (Sca1+ CD31−) and immune cells (CD45+) (Figures S1C and S1D). This yielded 19,879 cells for scRNA-seq profiling (Figures S1E and S1F). Integration of the sn- and scRNA-seq datasets allowed us to map cell state changes across muscle growth and atrophy (Figures S1G and S1H; Table S1).

We identified distinct MuSC states based on expression of Pax7, Myod1, and Myog (Figure S1I), each enriched under specific muscle conditions (Figures S2A and S2B; Table S1). MuSCs from OVL muscle exhibited expression of proliferation- and differentiation-associated genes such as Birc5 or Mki67, and Myog or Tnni2, respectively; consistent with the presence of cells in S and G2/M phases (Figures S2A and S2B). In contrast, MuSCs from Den muscle remained in G1 phase and expressed activation-related genes such as Islr. Despite the absence of an overt proliferation or differentiation signature, these MuSCs also expressed extracellular matrix (ECM)-associated genes, including Dcn and Col8a1 (Figure S2B), suggesting that axonal loss triggers a distinct transcriptional program in MuSCs.

All major muscle-resident populations displayed differential gene expression patterns during growth and atrophy (Table S1). Among glial populations, both myelinating (Mpz+) and non-myelinating (Nrxn1+) SC markers were identified (Figure S1F). In Den muscle, SCs exhibited substantial transcriptional reprogramming (Figure S1J), including upregulation of genes associated with axonal regeneration and migration (Atf3, Ngfr, Tnc) (Figure S1K). In contrast, MuSCs, FAPs, and ECs showed greater transcriptional changes in OVL muscle (Figure S1J), reflecting enhanced cellular support for myofiber growth, ECM remodeling, and vascular expansion. Specifically, FAPs upregulated ECM and mechanosensory genes (Postn, Aspn, Adamtsl1, Piezo2), while ECs upregulated angiogenic factors (Mmp14, Lrg1) (Figure S1K).

In summary, integration of snRNA-seq and scRNA-seq datasets enabled the generation of a comprehensive transcriptomic atlas encompassing all major skeletal muscle cell types and states under resting, atrophic, and hypertrophic conditions (Figures S1G and S1H). This atlas provides an essential framework for understanding how diverse muscle-resident cells coordinate the adaptive and degenerative responses of skeletal muscle to mechanical and neural cues.

Heterogeneity in myonuclear responses to muscle atrophy and growth

The cluster of 41,240 myonuclei exhibited significant transcriptional changes highlighted by a notable shift in their UMAP embeddings, particularly in the muscle atrophic state (Figure 1D). To dissect the heterogeneity of myonuclei across the different conditions at a higher resolution, we re-clustered this population (Figure 2A and 2B; Table S2). As expected, type II fiber subsets (Myh1, Myh2, and Myh4) were predominant in the fast-twitch plantaris muscle in basal state, with no detection of the slow-twitch type I isoform Myh7 (Figure 2C; Figure S2C). In Den muscle, the percentage of fast-glycolytic myonuclei IIb (Myh4+) and IIx (Myh1+) decreased to 18% at 7d, while the relative proportions of type IIa myonuclei (Myh2+) remained relatively constant (Figure 2C).

Figure 2. Myonuclear heterogeneity and condition-specific transcriptional programs revealed by snRNA-seq.

(A and B) UMAP embedding of myonuclei clustered by transcriptional similarity (A) and colored by condition (basal, 3 or 7 days after denervation [Den], or 3 or 7 days after overload [OVL]) (B). (C) Relative proportions of each myonuclear subpopulation across basal, Den, and OVL conditions at 3 or 7 days. (D) Bubble plot of representative marker genes for each myonuclear subpopulation. Color indicates average expression, and dot size represents the percentage of nuclei expressing each gene. (E) Functional enrichment analysis of the myonuclear subclusters shown in (A). Color denotes the statistical significance of enriched pathways or categories. (F) Functional enrichment of genes specifically upregulated in Den or OVL conditions, grouped into pathways related to muscle growth, atrophy, and neuromuscular signaling. Color indicates statistical significance. (G) Computed atrophy (top) and hypertrophy (bottom) scores for individual myonuclei. Left: UMAP projections colored by score. Right: violin plots showing score distributions across conditions. (A–G) Data derived from snRNA-seq, n = 2 mice per condition. Gene lists used to calculate each score are provided in the Methods section. See also Figure S2 and Table S2.

Importantly, the clustering analysis revealed three distinct subpopulations of myonuclei (Myonuc1-3), which were more specific to the Den and the OVL conditions, and minimally present in the unperturbed basal state (Figure 2A–2C). The Myonuc1 cluster, primarily observed in Den muscle, was characterized by significant downregulation of myosin transcripts, upregulation of denervation-induced markers (e.g., Myog, Igfn1, and Ampd3), and activation of muscle wasting-related signaling pathways (Figure 2D and 2E; Table S2). Myonuc1 likely represent denervated myonuclei, based on the expression of key denervation-related genes (e.g., Runx1, Gadd45, Dlg2 or Scn5a), downregulation of metabolic and structural markers (e.g., Ckm, Ckmt2, Agbl1, Xirp2), and enrichment of signaling pathways such as MAPK, Hippo, and TGFβ, consistent with previously characterized denervated subpopulations by snRNA-seq (Table S2)^12,13. The Myonuc2 and −3 clusters, which are increased during OVL, showed enhanced expression of genes associated with myonuclear fusion (e.g., Egfr) and muscle adaptation (e.g., Atf3 and Flnc) (Figure 2D and 2E; Table S2). In addition, Myh4 expression was almost absent in Myonuc3, while Myonuc2 showed expression of the three classical Myh isoforms (Figure S2D). Minimal expression of Myh3 (encoding the embryonic myosin heavy chain isoform) was detected only in these new clusters (Figure S2E). In addition to the selected markers (Figure 2D), Myonuc2 differentially expresses Meg3, Celf2, Dclk1 or Auts2, while Myonuc3 is characterized by the expression of Scn2a, Xirp2 or Lmcd1 (Table S2), correlating with two previously described populations in response to muscle overloading¹⁴. Irrespective of the muscle state under which these emerging populations are more prevalent, they exhibit distinct gene expression signatures when comparing Den and OVL muscles (Figure S2F, Table S2). Interestingly, genes involved in nerve-muscle communication, such as those related to postsynaptic modification or the regulation of presynaptic cell functions, were enriched in both conditions (Figure 2F; Table S2). Overall, these results indicate a distinct myonuclear transcriptional profile influenced by muscle growth pressure and (loss of) neural inputs.

According to the muscle wasting or muscle growth in Den or OVL muscles respectively, Den myonuclei were enriched in atrophy-related genes, and OVL myonuclei, in hypertrophy-related genes (Figure 2G; Figure S2G). Notably, despite the opposite effects of the Den and OVL conditions on muscle mass, both processes resulted in NMJ molecular remodeling (Figure 2F). This may reflect a direct response or compensatory rearrangement of the nuclei at the neural synapse, to cope either with the loss of neural input or with the mechanical strain. Such changes could occur in muscle disuse, disease, aging, or after resistance exercise.

The specialized synaptic myonuclei respond differently to atrophy and growth demands

The myonuclei located at the NMJ, termed synaptic myonuclei (SynM), remain poorly characterized owing to their extreme scarcity (~3 myonuclei per NMJ). Even after pooling multiple plantaris muscle across all experimental conditions, SynM accounted for only ~1% of total myonuclei in our snRNA-seq dataset (Figure S3A), corresponding to 115, 119, and 95 SynM from basal, Den, and OVL muscle, respectively. Despite this low representation, re-clustering enabled us to examine the specific molecular changes occurring at the NMJ during muscle atrophy and hypertrophy (Figure 3A). SynM from Den muscle clearly segregated from both basal and OVL SynM in the UMAP space, reflecting their distinct transcriptional profiles (Figure 3A; Table S3).

Figure 3. Transcriptional remodeling of synaptic myonuclei during muscle atrophy and hypertrophy.

(A) UMAP visualization of re-clustered synaptic myonuclei (SynM) from snRNA-seq data in basal, denervated (Den), and overloaded (OVL) plantaris muscle. For each condition, SynM from muscles at 3 and 7 days post-intervention were pooled. (B and C) Functional enrichment analysis of genes upregulated in SynM after Den (B) or OVL (C). Color indicates the statistical significance of enriched pathways or categories. (D) Representative images of NMJ innervation in plantaris muscle across time points and interventions, showing postsynaptic NMJ region labeled α-bungarotoxin (BTX, green) and presynaptic terminals (SV2 and NF, magenta). Scale bar, 10 μm. (E) Quantification of the area of the NMJ occupied by the axon terminal (percentage overlap) (n = 36–84 NMJs). (F) Fragmentation analysis based on the number of discrete endplate fragments per NMJ (n = 64–173 NMJs). (G) NMJ size measured as endplate volume (μm³) from 3D reconstructions (n = 95–178 NMJs). (H) Representative images of myonuclear accretion at the NMJ following daily EdU administration during the indicated time points and interventions. Myonuclei are labeled with PCM1 (white), NMJs with BTX (red), and nuclei with DAPI (blue). Scale bar, 10 μm. (I and J) Quantification of the number of SynM (I) and EdU+ SynM (J) per NMJ in basal plantaris muscle and at 7 days after Den or OVL (n = 36–58 NMJs). (K and L) Feature plots (left) and violin plots (right) showing normalized expression of Scn5a (K) and Gpc4 (L) in SynM. (E–J) Data represent n = 3–6 mice. Bar plots show mean ± SD; box plots show median and quartiles. P values were determined using a two-tailed, unpaired t-test. See also Figure S3 and Table S3.

Den-specific SynM exhibited increased expression of atrophy-associated genes (Figure S3B, left), consistent with the overall loss of muscle mass (Figure 1B). These atrophy-related genes were also upregulated in myonuclei (mainly within Myonuc1; Figures 2A and 2G) under the same condition. In contrast, SynM from OVL muscle displayed enrichment for growth-associated transcripts (hypertrophy score) (Figure S3B, right), consistent with the gain of muscle mass (Figure 1B). Thus, NMJ myonuclei recapitulated the adaptive transcriptional signatures observed in non-synaptic myonuclei during muscle remodeling.

Functional enrichment analysis of SynM from basal, Den, and OVL muscle (Figures 3B and 3C; Table S3) revealed marked condition-specific gene programs. In Den SynM, genes associated with muscle contraction, calcium homeostasis, and sarcomeric structure (Ttn, Tnnt3) were downregulated (Figures 3B and S3C). Conversely, OVL SynM were enriched for genes and pathways linked to growth, including amino acid metabolism and Wnt signaling (Figures 3C and S3C), both of which regulate NMJ formation and receptor clustering^15,16. Genes related to neuronal migration were also specifically upregulated in OVL SynM (Figure 3C). Moreover, transcripts involved in nerve–muscle communication, postsynaptic organization, and glial or neuronal regulation were enriched in both Den and OVL SynM (Figures 3B and 3C), consistent with NMJ remodeling during both atrophy and hypertrophy (Figures 3D–3G).

Den SynM exhibited unique transcriptional changes not shared with OVL SynM as compared to basal muscle, including the upregulation of genes like Spp1 or Vat1l (Figure S3D), and the downregulation of synaptic genes such as Chrne and Colq (Table S3)¹³. OVL SynM had no uniquely regulated gene set but showed a tendency for upregulated Col13a1 and Dcaf8, similar to Den SynM (Figure S3D; Table S3). These findings highlight the sensitivity of SynM transcriptional programs to neural input.

To determine whether these transcriptional responses correlated with NMJ structural remodeling, we optimized CUBIC-based tissue clearing (Table 1)¹⁷ combined with immunofluorescence for high-resolution 3D whole-mount confocal imaging (Figure 3D). In Den muscle, over 92% of NMJs lacked detectable axon terminal contacts (average 7% nerve–acetylcholine receptor (AChR) overlap) (Figures 3E and S3E), whereas innervation in OVL muscle remained comparable to basal conditions. The number of NMJ fragments increased in both Den and OVL muscle (Figure 3F), although OVL NMJs were larger and remained innervated. Endplate volume decreased with atrophy but expanded during hypertrophy, underscoring distinct remodeling dynamics for each condition (Figure 3G). OVL NMJs also displayed an increased number of SynM (Figures 3H and 3I), consistent with the presence of newly accreted SynM in hypertrophic NMJs (Figure 3J). These data suggests that the general myonuclear accretion observed in growing myofibers at 7d OVL¹⁴ (Figures S3F and S3G) extends to the specialized NMJ domain (Figure 3J). In contrast, SynM number and incorporation of new myonuclei remained unchanged in Den muscle at 7d (Figures 3H–J and S3F–S3H). These morphometric analyses indicate that muscle atrophy leads to NMJ disruption through axonal loss, while hypertrophy preserves NMJ integrity and expands the endplate and SynM domains. Thus, both adaptive states involve dynamic yet distinct remodeling of the NMJ.

Table 1.

Reagents for CUBIC-based whole mount muscle clarification

Solution	Reagent	Concentration
CUBIC-1	Urea (Sigma, #51456)	25%
	80% (wt/wt) Quadrol (Sigma, #122262)	25%
	Triton-X100	1%
	Distilled water	49%
CUBIC-2	Urea (Sigma, #51456)	25%
	Triethanolamine (TEA) (Sigma, #90278)	25%
	Sucrose (Sigma, #S0389)	15%
	Distilled water	35%

Distinct SynM transcripts were identified for atrophic and hypertrophic conditions, including Scn5a and Gpc4, respectively (Figure 3K–L, Table S3). To further investigate the molecular organization of NMJs at the tissue level, we performed spatial transcriptomics on muscle sections from each condition. NMJ regions were confirmed by α-bungarotoxin (BTX) staining in consecutive sections (Figure 4A). Across the different conditions, NMJ-positive spots were enriched for postsynaptic genes from SynM (Figures 4B, 4C and 4D; Table S3) and presynaptic SC transcripts (Figure 4D; Table S3). In Den muscle, Chrng, which encodes the immature AChR subunit, was selectively detected in NMJ regions at 7d (Figures 4E and 4F), indicating degeneration-specific expression. Furthermore, Den muscle had a partial loss of NMJ identity, as evidenced by ectopic expression of synaptic genes (such as Chrna1) in NMJ-free regions (Figures 4E and 4F). In contrast, OVL muscle had elevated Chrna1 expression specifically within NMJ areas as compared to basal levels, consistent with NMJ growth and reinforcement during hypertrophy. Spatial mapping localized atrophy- and hypertrophy-associated SynM transcripts (Scn5a, Gpc4) (Figures 3K and 3L) to NMJ-positive regions (Figures 4G and 4H).

Figure 4. Identification of NMJ-positive regions with associated synaptic myonuclei signatures in atrophic and hypertrophic muscles by spatial transcriptomics.

(A) Top: spatial feature plot showing the normalized summed expression of synaptic myonuclei (SynM) marker genes. Scale bar, 500 μm. Middle: Cropped consecutive longitudinal section stained for postsynaptic AChRs (BTX, white) and nuclei (DAPI, blue). Scale bar, 50 μm. Bottom: final annotation of NMJ-positive spatial spots based on the overlap between SynM marker expression and BTX staining in the adjacent section. Color represents normalized counts. Scale bar, 500 μm. (B) Heatmap showing the average log₂ fold-change (FC) of differentially expressed genes (DEGs; adjusted p < 0.05) between NMJ-positive and NMJ-negative spots for each condition. Expression values are displayed as normalized counts ranging from low (dark blue) to high (yellow), with dark blue indicating zero expression. Common differentially expressed synaptic genes are colored in red. (C) Venn diagram of marker genes associated with NMJ-positive spots in basal, denervated (Den), and overloaded (OVL) muscle. (D) Bar plot showing the distribution of predicted cell types contributing to each spatial spot in a representative sample. (E and F) Spatial feature plots (E) and violin plots (F) showing expression levels and spatial distributions of Chrng (top) and Chrna1 (bottom) in NMJ-positive and NMJ-negative spots. Scale bar, 200 μm. Color indicates log-normalized counts. (G and H) Spatial feature plots (G) and violin plots (H) illustrating expression levels and distribution of Scn5a (top) and Gpc4 (bottom) in NMJ-positive and NMJ-negative spatial transcriptomics spots. Scale bar, 200 μm. Color represents normalized counts. (A–H) Data derived from spatial transcriptomics, n = 2 mice per condition. See also Table S3.

Thus, SynM possess a distinct and dynamic transcriptional identity that reflects and supports NMJ remodeling during opposing physiological states. Their adaptive gene programs are tightly linked to neural input, underscoring a bidirectional communication between the postsynaptic myonuclei and the presynaptic cellular components of the NMJ.

A previously unrecognized terminal Schwann cell population emerges during muscle atrophy

In addition to SynM, the NMJ comprises two additional cellular components that are essential for its function: the presynaptic motor neuron and the supporting glial cells known as terminal Schwann cells (tSCs). This non-myelinating glia envelops the axon terminal and interface with the myofiber endplate¹⁸. To trace SCs, we used the glial-restricted Plp1-CreER driver crossed with the tdTomato (Tom) reporter allele (Plp1^CreER; R26^tdTomato, hereafter Plp1^Tom). In these muscles, the number of Tom+ tSCs did not significantly decrease at 7d Den (Figures 5A and 5B). In contrast, the number of Tom+ tSCs increased at 7d OVL (Figure 5B), likely reflecting enhanced tSC proliferation (Ki67+ cells) during muscle growth (Figure 5C), consistent with the increased NMJ volume observed under the same condition (Figure 3G).

Figure 5. Schwann cell heterogeneity in response to muscle atrophy and hypertrophy.

(A) Representative NMJ images from Plp1-tdTomato (Plp1^Tom) plantaris muscle in basal, denervation (Den), and overloading (OVL) conditions, showing Schwann cells (SCs, Tom+, red), proliferating cells (Ki67+, white), NMJs (BTX, green), and nuclei (DAPI, blue). Scale bar, 10 μm. (B) Number of terminal SCs (tSCs) per NMJ across time points after intervention (n = 73–140 NMJs). (C) Percentage of NMJs containing proliferating tSCs (Ki67+ tSCs) at each time point (n = 36–68 NMJs). (D) Representative images of NMJ fragmentation in Plp1^iDTR mice after 7 days of diphtheria toxin (DT) treatment in basal plantaris muscles. Scale bar, 10 μm. (E) Distribution of the number of NMJ fragments in Plp1^WT and Plp1^iDTR mice (n = 76–111 NMJs). (F) Representative transverse sections of plantaris muscle from Plp1^WT and Plp1^iDTR mice stained for laminin (cyan) and DAPI (magenta). Scale bar, 100 μm. (G) Frequency distribution of fiber cross-sectional area (CSA) with a fitted Gaussian distribution curve of plantaris myofibers from Plp1^WT and Plp1^iDTR mice. (H) UMAP visualization of SC subpopulations from integrated sc/snRNA-seq data. (I) Proportions of the SC subpopulations identified in (H) across basal, 3- or 7-day Den, and 3- or 7-day OVL conditions. (J) Representative RNAscope images showing Ajap1+ and Inhba+ nuclei. (K and L) Percentage of NMJs containing Ajap1+ nuclei (K) or Inhba+ nuclei (L) (n = 39–68 NMJs). (M) Proportions of NMJs containing nuclei with single, double, or no expression of Ajap1 and Inhba (n = 39–68 NMJs) (N) Spatial transcriptomics: NMJ-positive and NMJ-negative spots (top) and normalized summed expression of aSC1 marker genes (bottom). Color indicates normalized counts. Scale bar, 200 μm. (O) Violin plots showing expression of aSC1 markers in NMJ and non-NMJ spatial spots. (B–G) Data represent n = 3–5 mice, (K–M) Data represent n = 3 mice. Bar plots show mean ± SD; box plots show median and quartiles. P values were determined using a two-tailed, unpaired t-test. See also Figures S5–S6 and Table S4.

To directly assess the role of SCs in NMJ homeostasis and muscle maintenance, we selectively depleted these cells in vivo using a diphtheria toxin (DT)/diphtheria toxin receptor (DTR) system under the control of Plp1¹⁹. After seven days of intraperitoneal DT administration, tSC numbers within the NMJ were markedly reduced (Figures S4A and S4B). This targeted depletion induced fragmentation of NMJ structures even in resting muscle (Figures 5D and 5E) and led to a measurable decrease in muscle size (Figures 5F and 5G), indicating that loss of tSCs triggers denervation-like effects in otherwise innervated muscle. The importance of SCs was also addressed in Den and OVL muscles. For instance, reduction in the number of tSCs aligned with the increased NMJ fragmentation in SC-depleted muscles during atrophy or hypertrophy induction, when compared to their control counterparts (Figure S4C–S4E). As in basal conditions, this targeted depletion led to a decrease in muscle size (Figure S4F and S4G). Together, these findings highlight the essential role of glial cells in preserving NMJ integrity and supporting overall muscle function, not only in maintaining homeostasis, but also in muscle adaptation to distinct muscle states.

Transcriptomic analyses revealed that the canonical tSC cluster contributed minimally at all Den time points (Figures 5H and 5I). We observed the emergence of activated SCs (aSCs), which expressed developmental and repair-associated genes such as Ngfr (Figure S5A)^20–22. Each condition was characterized by distinct aSC subclusters: aSC1 and aSC3 predominated in Den muscle, whereas aSC2 was characteristic of OVL muscle (Figures 5I, S5A, and S5B; Table S4). Myelinating SCs (mSCs) disappeared after denervation due to axonal loss, and their abundance also declined transiently in OVL muscle at early time points (Figure 5I). Among the aSC subsets, aSC3 was the most abundant in Den muscle. It expressed genes typically upregulated during Wallerian degeneration, including Atf3²³, as well as genes implicated in migration (Slit2)²⁴ and phagocytosis (Lgals3, Axl)^25,26 (Figures S5A and S5C; Table S4).

The aSC1 subcluster expressed the highest levels of regeneration-associated genes^22,27, such as Olig1, Gap43, and Tnc, followed by the aSC3 and then the aSC2 subclusters (Figure S5C; Table S4). Moreover, aSC2 was enriched in collagen-encoding genes (Table S4) and likely contributes to extracellular matrix (ECM) remodeling during muscle hypertrophy. Similarly, aSC1 exhibited strong expression of ECM-related transcripts such as Wisp1, Col7a1, and Tnc (Figures S5A and S5C). We confirmed the presence of aSCs in Den NMJs via GAP43 immunostaining (Figures S5C and S5D). Certain tSC markers, including Col20a1 and Bche, were also expressed by aSC1 cells (Figure S5A), suggesting that aSC1 may represent an activated tSC-like population arising during denervation.

To test this hypothesis, we performed RNAscope analyses on resting and 7d-treated muscle. Col20a1 transcripts are a marker of non-myelinating SCs and tSCs^28,29 and were detected in 80–90% of NMJs across all conditions (Figures S6A and S6B). The aSC1-specific marker Inhba, encoding the β-subunit of activin A, was predominantly expressed at NMJs in Den samples, where 90% of NMJs contained double-positive Col20a1+ Inhba+ SCs (Figure S6C). This confirms that the aSC1 population localizes to the NMJ and functions as activated tSCs. Notably, Inhba+ tSCs persisted at 15d in Den muscle (Figures S5E–S5G)²², suggesting a sustained role for these cells during prolonged denervation.

Expression of the tSC marker Ajap1^28,30 in OVL muscle was comparable to that in basal NMJs, whereas its expression in Den NMJs dropped sharply to <5% of basal levels (Figures 5J and 5K). Conversely, the proportion of NMJs containing Inhba+ cells increased dramatically upon Den (Figures 5J and 5L). Dual mRNA detection of Ajap1 and Inhba revealed an inverse relationship: ~80% of Den NMJs contained Inhba+ cells, but few exhibited double Ajap1+ Inhba+ expression (Figure 5M). The Sostdc1 transcript, a marker of the aSC2 subcluster, did not significantly increase in NMJs at 7d OVL (Figures S6D–S6F), confirming that aSC2 cells do not specifically localize to the NMJ. Spatial transcriptomics further validated these findings: pan-SC and tSC markers were localized to NMJ-positive areas (Figures S6G and S6H), and aSC1-specific markers were enriched in Den muscle NMJ-positive regions but not in basal or OVL conditions (Figures 5N and 5O). Consistent with RNAscope results, aSC2 enrichment was not spatially enriched in NMJ-positive areas in hypertrophic muscle (Figures S6I and S6J).

These data identify a previously uncharacterized population of activated tSC-like cells (aSC1) that arise during denervation through dedifferentiation of mature SCs. In this activated state, these cells localize to the NMJ and express factors linked to neurotrophic support and ECM remodeling, likely reflecting the loss of axonal signaling. These changes contribute to NMJ remodeling and help maintain muscle homeostasis under atrophic conditions.

Communication networks between synaptic myonuclei and terminal Schwann cells are major drivers of muscle atrophy and growth in mouse and human

How the different cellular components of the NMJ communicate is still poorly understood. We considered whether this intercellular communication differs across states of muscle plasticity. The transcriptional signatures identified in tSCs and SynM under Den conditions suggested that these two populations engage in reciprocal signaling relevant to NMJ repair and axonal guidance. To explore this possibility, we used CellChat³¹ to infer ligand–receptor (L–R) communication between SynM and the heterogeneous SC populations, excluding subpopulations based on minimal cell numbers and histological validation (Table S4). The overall number of predicted L–R interactions was highest under OVL conditions, followed by Den (Figure 6A, top). As expected, tSCs displayed more L–R interactions with SynM than myelinating SCs (mSCs). Moreover, the number of tSC–SynM interactions was higher in OVL muscle than in basal conditions (Figure 6A, bottom). In Den muscle, the aSC1 subpopulation (representing the tSC-like cells localized at the NMJ of atrophic muscle) displayed more interactions with SynM than canonical tSCs, although fewer than the aSC3–SynM pair (Figure 6A, bottom), likely reflecting differences in subpopulation size (Figure S5B).

Figure 6. Shared communication features between synaptic myonuclei and Schwann cells in mice and humans.

(A) Number of predicted total ligand–receptor (L–R) interactions between synaptic myonuclei (SynM) and SCs (top), or between SynM and each SC subtype (bottom) in basal, denervated (Den), and overloaded (OVL) muscle. (B) Relative information flow (sum of interaction-probability differences) for all L–R pairs across basal, Den, and OVL conditions. (C and D) Predicted BMP (C) and TGFβ/activin (D) signaling networks across conditions. Nodes represent cell types, and edges represent inferred interactions; edge width denotes interaction probability. (E) Bubble plot showing ligands and receptors from the BMP and TGFβ/activin pathways for SynM, terminal Schwann cells (tSCs), and aSC1 cells under each condition. (A–E) Data derived from snRNA-seq and scRNA-seq, n = 2 and n = 10–16 mice per condition, respectively. (F) Cross-sectional sections from human PAN muscle biopsies with signs of acute denervation (Human Den). Top: H&E staining. Bottom: Esterase histochemistry. Arrows represent esterase-positive fibers with angular atrophy. Scale bar, 500 μm. (G) Heatmap of denervation- and TGFβ/activin-associated genes in human control and denervated muscle samples (PAN muscles). Expression values range from low (blue) to high (red), with white indicating zero expression. (F and G) Data represent n = 5 denervated PAN patients and n = 8 healthy controls. (H) Representative RNAscope images showing INHBA and CHRNE transcripts in human control and Den muscle. Data represent n = 3 human samples per group. Scale bar, 10 μm. (I) Heatmap showing denervation-related genes, INHBA (aSC1 marker), and atrophy-associated genes in a publicly available microarray dataset from human muscle biopsies obtained 2 or 5 days after spinal cord injury (SCI 2d and SCI 5d, respectively)⁴². Color represents log₂ fold change (Log₂FC) relative to control samples. (J) Heatmap of hypertrophy-related pathways and RFLNA (aSC2 marker) in human muscles early after an exercise bout (24h). Color represents normalized expression. (K) Heatmap of hypertrophy-related pathways and RFLNA (aSC2 marker) in late stages of exercise training (6 weeks). (J and K) Data represent n = 3 exercised and n = 3 control human samples. Color represents normalized expression. See also Table S4.

Although many signaling pathways were shared, each condition showed distinct sets of L–R interactions (Figure 6B). In Den muscle, signaling was enriched for tenascin C (Tnc)-mediated extracellular matrix (ECM) remodeling, semaphorin-related pathways (involved in axon guidance³²), and TGFβ-related signaling (Figure 6B, left). Among the SynM-derived ligands, Spp1—previously shown to promote SC survival³³—was prominent. Interactions involving Lama2 further suggested a role in maintaining the tSC-like phenotype of aSC1 cells and in establishing a permissive environment for glial processes, reminiscent of its role in sustaining Büngner bands during nerve injury and axon regeneration^34,35.

Given that BMP and TGFβ/activin signaling are known regulators of muscle mass^8,36–38, we examined these pathways in more detail. In OVL muscle, SC–SynM interactions were enriched in Wnt and BMP signaling, which are implicated in myofiber growth, NMJ development, and ECM remodeling³⁹ (Figure 6B, right). In this hypertrophic condition, tSCs exhibited increased expression of transcripts encoding BMP ligands, which could potentially interact with SynM (Figure 6C). Conversely, in Den muscle, we observed upregulation of profibrotic factors, including TGFβ and activins, which may modulate SynM–aSC1 communication and were absent in basal conditions (Figure 6D).

Comparative analysis of ligand–receptor expression revealed that SynM from basal or OVL muscle displayed similar expression profiles, with overall higher expression in OVL, particularly for Acvr1 and Bmpr2 (Figure 6E). In contrast, the aSC1 subpopulation from Den muscle showed marked upregulation of Inhba expression. Correspondingly, SynM from Den muscle exhibited globally increased expression of TGFβ and activin receptors, including Acvr2a, Tgfbr1, and Tgfbr2 (Figure 6E). Collectively, these results indicate that TGFβ signaling is strongly enhanced between SynM and activated tSCs during muscle atrophy, suggesting that SynM may receive pro-atrophic cues from these specialized SCs. This communication likely transduces the loss of axonal input to the rest of the myonuclei within the muscle fiber. Given that the TGFβ/activin pathway negatively regulates muscle mass and promotes fibrosis⁴⁰, dysregulation of this signaling axis may contribute to NMJ-associated myonuclear dysfunction and muscle wasting.

Finally, to assess whether these findings in mouse are evolutionarily conserved and extend to humans, we analyzed human muscle samples from patients with evidence of acute denervation due to neuropathy in the context of polyarteritis nodosa (PAN), a form of medium vessel vasculitis that frequently causes nerve injury⁴¹ (Figures 6F and 6G, Human Den). Muscle biopsies from these patients showed signs of acute denervation, indicated by the presence of esterase-positive angular atrophic fibers (Figure 6F). Bulk RNA-seq from these samples revealed upregulation of canonical denervation markers—GADD45A, RUNX1, and CHRNG—along with the SC marker NGFR (Figure 6G). The same pro-atrophic signaling signatures identified in mice were conserved in human biopsies, including increased expression of TGFBR1, TGFBR2, and INHBA (the murine aSC1 marker) (Figure 6G). INHBA transcripts were detected at NMJ regions after acute denervation, identified by CHRNE expression (Figure 6H). INHBA expression was also detected in public datasets of individuals at early stages of spinal cord injury (SCI) (Figure 6I)⁴² mirroring the early activation of aSC1 in Den mouse muscle (Figures 5I, S5G). Conversely, bulk RNA-seq of human muscle subjected to acute exercise bouts revealed upregulation of BMP-related genes (BMP1, BMPR2) (Figure 6J). Consistent with the mouse OVL model, the human homolog for the aSC2 marker Fam101a (RFLNA) was upregulated early after the exercise bout (Figure 6J). These transcriptional changes were maintained at later stages of exercise training (Figure 6K).

These findings indicate that key components of the denervation-induced signaling network are conserved between mice and humans, highlighting this pathway as a promising therapeutic target. The localization of INHBA expression at human NMJs provides spatial support for the involvement of this signaling axis. Although comparing single-nucleus transcriptomic data (in mice) with bulk RNA-seq (in humans) has inherent limitations, such as the lack of cellular resolution, the observed transcriptional trends were consistent with our predictions. Overall, these data reinforce the translational relevance of mouse models for studying denervation responses and suggest that similar molecular mechanisms operate in human muscle.

DISCUSSION

To our knowledge, this study presents the first comprehensive atlas that simultaneously integrates both muscle atrophy and growth states within the same muscle type. This reference atlas of skeletal muscle adaptation provides a unified and detailed view of how muscle tissue responds to neural and mechanical stress over time, offering unprecedented cellular and molecular resolution. Periods of disuse—whether due to aging, immobilization, or disease— or exercise induce significant and opposing changes in muscle mass. This plasticity is reflected in alterations in myofiber composition, myonuclear characteristics, and the phenotypes of resident muscle cells. For instance, muscle undergoing atrophy or growth displays distinct subsets of MuSCs with specific myogenic states, FAPs with differential fibrogenic traits, and Schwann cells (SCs) with either myelinating or non-myelinating roles.

Changes in the myonuclear compartment include the loss or gain of specific nuclear subtypes, as well as the emergence of myonuclear subpopulations that can occur globally or in a fiber type–specific manner. Strikingly, NMJ-associated nuclei (SynM) exhibit transcriptional behaviors that closely resemble extrasynaptic myonuclei, despite their functional specialization. Both nuclear populations display classic signatures of muscle atrophy under denervated conditions and markers of muscle growth after overload. SynM specifically upregulates Scn5a, which encodes the cardiac sodium channel NaV1.5, shortly after denervation. Previous studies have reported that denervation induces the re-expression of NaV1.5, leading to tetrodotoxin-resistant action potentials in muscle fibers^43,44. Our data suggest that this re-expression originates locally at the NMJ. Additionally, the upregulation of Gpc4 in SynM from hypertrophic muscle, together with its expression in Chrna1-enriched regions during embryonic development, suggests that Gpc4 may be co-regulated with other NMJ-associated genes and participate in signaling pathways essential for synapse formation and maturation⁴⁵. These molecular responses likely represent compensatory or causative mechanisms contributing to muscle growth, atrophy, or both.

From a technical standpoint, our study overcomes major limitations of traditional bulk RNA-seq approaches, which cannot resolve transcriptional changes from specific NMJ components. While single-cell and single-nucleus transcriptomics can introduce assumptions or clustering artifacts, and particularly for rare populations such as SynM, our integrated approach yields meaningful biological insights into the molecular adaptations of this scarce myonuclear subpopulation across distinct muscle states. By combining single-nucleus and single-cell transcriptomics with spatial transcriptomics and 3D imaging, we mapped the molecular dynamics of both pre- and post-synaptic elements of the NMJ throughout muscle growth and atrophy, capturing changes in morphology, topology, and gene expression.

Our data reveal that the neuromuscular synapse is highly responsive to mechanical loading and unloading, driving coordinated adaptations in both muscle growth and atrophy. Previous studies have investigated SC behavior following nerve injury^21,22; here, we show that, in addition to SynM changes, both atrophy and growth induce distinct and opposing molecular responses in pre-synaptic SCs. We identify a newly activated tSC-like population that appears exclusively in denervated muscle. This NMJ-localized subpopulation expresses components of the activin signaling pathway, suggesting a potential pro-atrophic role. Sustained Inhba expression in tSCs at prolonged denervation stages implies a contribution to both the initiation and progression of chronic muscle atrophy. Computational modeling (CellChat) indicates that crosstalk between activated tSCs (aSC1) and denervated SynM is primarily mediated by TGFβ/activin signaling, potentially triggering muscle wasting. These findings support the hypothesis that muscle atrophy may originate at the NMJ, where sub-synaptic nuclei receive pro-atrophic cues from nearby activated Schwann cells in response to axonal disruption. Although RNAscope validation confirmed the presence of Inhba+ tSCs (aSC1) at atrophic NMJs, the proposed enhancement of TGFβ signaling between SynM and activated tSCs remains to be directly demonstrated.

Finally, our findings at the murine NMJ were confirmed in human muscles from patients with a neuropathy related to polyarteritis nodosa (PAN), a form of medium vessel vasculitis that frequently causes nerve injury, and that presented evident signs of acute denervation. This confirmation was extended to muscles from patients early after spinal cord injury. This highlights the potential therapeutic targets for promoting muscle maintenance in front of nerve injuries. Despite sex-related heterogeneity in our human bulk RNA-seq datasets the expected transcriptional patterns were consistently detected, indicating that our findings are robust to sex differences.

This atlas can help to clarify how the NMJ remodels in conditions such as neuromuscular diseases, aging, or inactivity. The discovery of dynamic and rare NMJ subpopulations underscores the potential for targeting neuromuscular connectivity—through physical activity, lifestyle modification, or pharmacological intervention—to prevent, slow, or reverse muscle wasting. Consistent with this, both age-related muscle atrophy and trained muscle expression datasets in mice and humans exhibit dysregulation of NMJ-related genes, reinforcing the relevance of our findings^46–48.

Our study demonstrates the remarkable plasticity of the NMJ and its specialized cellular components across the spectrum of muscle growth and atrophy. By identifying rare NMJ-associated myonuclear and Schwann cell subtypes, we reveal mechanisms through which pre- and post-synaptic elements coordinate to maintain muscle mass and function. This work establishes a robust foundation for future studies in human muscle biology and offers a valuable resource for investigating therapeutic strategies to modulate NMJ function and combat muscle wasting.

Limitations of the Study:

The main findings of our study rely on transcriptomic and computational analyses, which may introduce biases, especially for rare cell populations. While some findings were validated by immunofluorescence or RNAscope, further functional evidence for key signaling interactions is still needed. Despite that most insights come from mice, key observations were confirmed in both murine and human samples, underscoring the relevance of our findings across species. Altogether, further studies are required to define the direct causal contribution of these molecular changes and muscle plasticity.

RESOURCE AVAILABILITY

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Pura Muñoz-Cánoves (pmunozcanoves@altoslabs.com).

Materials availability

Further information for resources and reagents should be directed to and will be fulfilled by the lead contact. This study did not generate new unique reagents.

Data and code availability

Raw single nuclei and single cell RNA-seq, spatial transcriptomics and bulk RNA-seq data are deposited on NCBI Gene Expression Omnibus (GEO) under accession numbers GSE294765, GSE311567 and GSE312393. Further information of data and code should be directed to the lead contact.

STAR★Methods

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Mouse models

C57BL/6J wild-type (WT) mice of both sexes were obtained from Charles River Laboratories or from an in-house breeding colony at the Centro Nacional de Investigaciones Cardiovasculares (CNIC). Mice were housed under standard conditions with a 12-h light/dark cycle and ad libitum access to a chow diet. All procedures were approved by the CNIC Institutional Animal Care and Use Committee and the regional authorities of the Madrid and Catalan Governments. Experiments were carried out in young adult mice (3–4 months old) on a C57BL/6 background, using sex- and age-matched littermates with equal numbers of males and females in each group. Mice were randomly assigned to experimental groups; no blinding was used, and sample size was not predetermined by statistical methods. Plp1-CreER; Rosa26-LSL-tdTomato (Plp1^Tom) mice were generated by crossing Plp1-CreER (JAX #005975) with Rosa26-LSL-tdTomato (JAX #007909). Cre recombination was induced by daily intraperitoneal injections of tamoxifen (TMX; 2 mg per 25 g body weight; Sigma, 20 mg/mL in corn oil) for four consecutive days, and surgical interventions were performed seven days after the final TMX dose. Plp1-CreER; Rosa26-iDTR (Plp1^iDTR) mice were generated by crossing Plp1-CreER (JAX #005975) with Rosa26-iDTR (JAX #007900). Cre activity was induced using the same TMX regimen, and after three days mice received daily intraperitoneal injections of diphtheria toxin (DT; 450 ng per 25 g; Sigma) for seven days. After this period, plantaris muscles were collected for basal conditions. For atrophic and hypertrophic muscles, mice were subjected to denervation or overloading, and plantaris muscles were collected 7 days post-intervention.

Genotyping of mice

For PCR genotyping, the following primers were used:

• ROSA (R316) 5’-GGAGCGGGAGAAA TGGATATG-3’;

• ROSA (R883) 5’-AAAGTCGCTCTGAGTTGTTAT-3’;

• ROSA (R4982) 5’-AAG ACCGCGAAGAGTTTGTC-3’;

• CRE pblast (Fw) 5’-CCCGCAGAACCTGAAGATGT-3;

• CRE pblast (Rv) 5’-CAGCGTTTTCGTTCTGCCAA-3’;

• tdTomato (IMR9020) 5’-AAGGGAGCTGCAGTGGAGTA-3’;

• tdTomato (IMR9021) 5’- CCGAAAATCTGTGGGAAGTC-3’;

• tdTomato (IMR9103) 5’- GGCATTAAAGCAGCGTATCC-3’;

• tdTomato (IMR9105) 5’- CTGTTCCTGTACGGCATGG-3’;

• iDTR (Rv, IMR8052) 5’-GCGAAGAGTTTGTCCTCAACC-3’;

• iDTR & WT (Fw, IMR8545) 5’-AAAGTCGCTCTGAGTTGTTAT-3’,

• WT (Rv, IMR8546) 5’-GGAGCGGGAGAAATGGATATG-3’.

Human samples

Denervated human muscle: Muscle samples were collected from patients with polyarteritis nodosa (PAN) and muscle denervation (4 females and 1 male) and from healthy comparators (5 females and 3 males) at the Muscle Research Unit, Internal Medicine Service, Hospital Clinic (Barcelona, Spain). The muscle tissue was routinely frozen in isopentane cooled with liquid nitrogen and stored at −80 °C. Routine staining techniques included hematoxylin and eosin (H&E) and esterase staining, which confirmed the acute denervation phenotype. Samples were shipped on dry ice to the NIH (Bethesda, USA) where all library preparation and sequencing were performed under uniform conditions. Exercised human muscle: Percutaneous muscle biopsies from the vastus lateralis with the sampling site located between 15 and 25 cm from the midpatella were obtained from 3 young healthy male volunteers after an eccentric exercise protocol (24 h after completion of an eccentric exercise session consisting in downhill running on a treadmill (−15%) for 40 min at 60% VO2max), and from 3 male healthy volunteers at rest. Individuals were not subjected to previous exercise programs. The study was approved by the University of Valencia Ethical Committee (Reference: UV-INV_ETICA-1351487), and the experiments conformed to the Declaration of Helsinki. In addition, muscle biopsies from vastus lateralis were obtained from three healthy young women that participated in a six-week resistance training program to induce muscle hypertrophy (leg press and leg extension exercises, performed twice per week). Training intensity was individualized based on a one-repetition maximum (1RM) test conducted at baseline for each exercise. Throughout the intervention, intensity was maintained within 70–100% of 1RM, while volume was progressively increased by raising the number of sets from 4 to 9 per exercise. The number of repetitions per set (8–15) and rest intervals (1.5–3 minutes) remained constant. The study was conducted following the Declaration of Helsinki and received ethical approval from the Medicines Research Ethics Committee of the University Clinical Hospital of Valencia, Spain (License reference: 2021/361). All participants provided written informed consent after a detailed explanation of the study’s purpose and procedures.

METHOD DETAILS

Denervation-induced muscle atrophy

For denervation experiments, mice were anesthetized with isoflurane and received an intramuscular injection of buprenorphine (0.1 mg/kg) in the contralateral quadriceps prior to surgery. Hair was clipped from the hindlimb, and the area was disinfected with chlorhexidine. A small incision was made along the mid-lateral thigh, parallel to the femur, to expose the sciatic nerve. The nerve was transected over a ~3 mm segment, and the skin was closed with 6/0 silk sutures. Mice were euthanized, and plantaris muscle was collected at the indicated time points.

Overloading-induced muscle hypertrophy

Mice were anesthetized with isoflurane, and buprenorphine (0.1 mg/kg, intramuscular) was administered in the contralateral quadriceps before surgery. Mice were placed in the prone position, and hair was clipped from the hindlimb undergoing the procedure. The surgical site was prepared with chlorhexidine solution. A longitudinal skin incision was made along the midline of the hindlimb overlying the gastrocnemius muscle, and connective tissue was gently removed. To induce compensatory hypertrophy in the plantaris and soleus muscle, the distal tendons of the medial and lateral gastrocnemius muscle were sectioned under a stereoscope. The transected gastrocnemius muscle was then reflected and sutured near its proximal origin. Finally, the skin was sutured with 6/0 silk. At the corresponding time point, mice were euthanized, and muscle were harvested.

In vivo EdU incorporation (proliferation assays)

Control and experimental mice subjected to surgical interventions were administered daily ethynyl-deoxyuridine (EdU; Invitrogen A10044; 25.5 mg/kg, i.p.) until the experimental endpoint. Basal control mice received daily EdU injections for 7 days. Experimental mice received daily EdU for either 3 or 7 days, depending on the protocol. At the designated time points, mice were euthanized, and plantaris muscle was collected for analysis. EdU staining was performed according to manufacturer’s protocol (ThermoFisher, #C10337)

Muscle sample processing for histology

At the designated time points after intervention (3 or 7 days), mice were euthanized and plantaris muscle was dissected. For spatial transcriptomics and RNAscope, muscle was embedded in OCT compound (Sakura #4583) and rapidly frozen in isopentane (Sigma #277258) cooled in a liquid nitrogen bath. For three-dimensional imaging, EdU incorporation tracking, and immunostaining in fluorescent reporter mice, tissue was prefixed for 2 h in 2% PFA (Cole-Parmer #CF100400–1A) at 4°C, washed in PBS, incubated in 30% sucrose at 4°C overnight, and then frozen in isopentane cooled with liquid nitrogen. All samples were stored at −80°C until use.

CUBIC-based tissue clarification

Pre-fixed frozen muscle was trimmed on a cryostat until a thickness of 200–500 μm was obtained. OCT was removed by washing in PBS, and permeabilization and blocking were performed simultaneously by incubating tissue in 1% Triton X-100 and 5% BSA in PBS for 2 h at room temperature. When mouse primary antibodies were used, M.O.M. blocking solution was added to the permeabilization/blocking mixture. Whole mounts were incubated with primary antibodies diluted in 5% BSA at 4°C for up to 3 days when required. For NMJ labeling, biotin-conjugated α-bungarotoxin was included in the primary antibody solution. Tissues were then washed and incubated with fluorophore-conjugated secondary antibodies and/or fluorophore-conjugated streptavidin diluted in 5% BSA for 2 h at room temperature; DAPI was added to the secondary antibody solution for nuclear staining. After PBS washes, whole mounts were fixed in 4% PFA for 30 min at room temperature, washed again, and incubated overnight in CUBIC-1 solution⁴⁹. Following an additional PBS wash, tissues were incubated in 30% sucrose for 2 h and then transferred to CUBIC-2 overnight at room temperature. Whole mounts were mounted between two coverslips sealed with plumber’s putty, using CUBIC-2 as the mounting medium. Reagents used to prepare CUBIC-1 and CUBIC-2 solutions are listed in Table 1. Primary and secondary antibodies are listed in Key Resources Table.

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit anti-PCM1 (dilution 1:100)	Sigma Aldrich	Cat# HPA023370
AF647 Rabbit anti-Laminin (dilution 1:200)	Novus biologicals	Cat# NB300-144AF647
AF750 Rabbit anti-Laminin (dilution 1:100)	Novus biologicals	Cat# NB300-144AF750
Rabbit anti-Ki67 (dilution 1:100)	Abcam	Cat# ab15580
Mouse anti-SV2 (dilution 1:200)	DSHB	Cat# AB_2315387
Mouse anti-2H3 (NF) (dilution 1:200)	DSHB	Cat# AB_531793
AF488 Goat anti-Rabbit (dilution 1:300)	ThermoFisher Scientific	Cat# A-11008
AF568 Goat anti-Rabbit (dilution 1:300)	ThermoFisher Scientific	Cat# A-11036
AF647 Goat anti-Rabbit (dilution 1:300)	ThermoFisher Scientific	Cat# A-21245
AF647 Goat anti-Mouse (dilution 1:300)	ThermoFisher Scientific	Cat# A-21235
AF750 Goat anti-Mouse (dilution 1:100)	ThermoFisher Scientific	Cat# A-21037
PE-Cy7-conjugated anti-CD45 (dilution 1:200)	Biolegend	Cat# 103114
PE-Cy7-conjugated anti-Sca1 (dilution 1:200)	Biolegend	Cat# 108114
BV605-conjugated anti-CD31 (dilution 1:200)	Biolegend	Cat# 102427
Biological samples
Denervated human muscle biopsies	Hospital Clinic (Barcelona, Spain)	N/A
Exercised human muscle biopsies	University of Valencia	N/A
Chemicals, peptides, and recombinant proteins
Biotin-conjugated a-bungarotoxin (dilution 1:100)	Alomone Labs	Cat# B-100-B
AF488-conjugated a-bungarotoxin (dilution 1:100)	Biotium	Cat# 00005
AF488 Streptavidin (dilution 1:200)	ThermoFisher Scientific	Cat# S32354
AF594 Streptavidin (dilution 1:200)	ThermoFisher Scientific	Cat# S32356
AF647 Streptavidin (dilution 1:200)	ThermoFisher Scientific	Cat# S32357
DAPI (dilution 1:500)	ThermoFisher Scientific	Cat# 62248
Topro3 (dilution 1:10000)	Invitrogen	Cat# T3605
Urea	Sigma Aldrich	Cat# 51456
Quadrol	Sigma Aldrich	Cat# 122262
Triton-X100	Sigma Aldrich	Cat# T8787
Triethanolamine (TEA)	Sigma Aldrich	Cat# 90278
Sucrose	Sigma Aldrich	Cat# S0389
Ethynyl-deoxyuridine (EdU)	Invitrogen	Cat# A10044
4% Paraformaldehyde (PFA)	Cole-Parmer	Cat# CF100400-1A
Bobine Serum Albumin (BSA)	Sigma Aldrich	Cat# A7030
M.O.M. blocking solution	Vector Laboratories	Cat# MKB-2213-1
Liberase	Roche	Cat# 5401020001
Dispase	Sigma Aldrich	Cat# D4693
Fetal bobine serum (FBS)	Gibco	Cat# 26-170-043
UltraPure BSA	ThermoFisher Scientific	Cat# AM2616
Hematoxylin	Sigma Aldrich	Cat# MH16
Eosin	Sigma Aldrich	Cat# HT110216
TRIzol	ThermoFisher Scientific	Cat# 15596026
Opal 570	Akoya Biosystems	SKU FP1488001KT
Opal 690	Akoya Biosystems	SKU FP1497001KT
Wheat germ agglutinin (WGA)	Biotium	Cat# 29077
Critical commercial assays
RNAscope Multiplex Fluorescent Reagent Kit v2	ACD Bio-Techne	Cat# 323270
Click-IT EdU Cell Proliferation Kit for Imaging, Alexa Fluor 488 dye	ThermoFisher Scientific	Cat# C10337
RNeasy Mini Kit (50)	Qiagen	Cat# 74104
Chromium Nuclei Isolation Kit	10X Genomics	Cat# 1000494
Qubit 1X dsDNA High Sensitivity (HS) Assay Kits	ThermoFisher Scientific	Cat# Q33231
Single Cell 3′ Reagent Kit v3.1 (dual index)	10X Genomics	Cat# PN-1000268
NEBNext Ultra™ II RNA Library Prep Kit for Illumina	NEB	Cat# E7775
Visium Spatial Tissue Optimization Slide & Reagent Kit	10X Genomics	Cat# PN-1000193
NEBNext Poly(A) mRNA Magnetic Isolation Module	NEB	Cat# E7490
Ultra II Directional RNA Library Prep Kit for Illumina	NEB	Cat# E7760
Watchmaker mRNA Library Prep Kit v1.1.0823	Watchmaker Genomics	Cat# 7BK0001
SMART-Seq mRNA LP kit	Takara Bio	Cat# 634768
Deposited data
scRNA-seq, snRNA-seq, bulk RNA-seq and spatial transcriptomics	This paper	GSE294765, GSE311567 and GSE312393
scRNA-seq data from Schwann cells in denervated muscles at different time points	Nicoletti et al.22	GSE221736
Microarray data from human muscles with spinal cord injury	Urso et al.42	GSE21496 and GSE21497
Experimental models: Organisms/strains
Mouse: C57BL/6J	Charles River Laboratories	https://www.criver.com/products-services/find-model/c57bl6-mouse?region=3611
Mouse: Plp1-CreER	The Jackson Laboratory	Cat# 005975
Mouse: Rosa26-LSL-tdTomato	The Jackson Laboratory	Cat# 007909
Oligonucleotides
Cell Multiplexing Oligos (CMOs)	10X Genomics	https://www.10xgenomics.com/support/universal-three-prime-gene-expression/documentation/steps/sequencing/cell-multiplexing-oligo-sequences-cell-ranger
Inhba-C2	ACD Bio-Techne	Cat# 455871-C2
Ajap1-C3	ACD Bio-Techne	Cat# 549641-C3
Col20a1-C1	ACD Bio-Techne	Cat# 455771
INHBA-C2	ACD Bio-Techne	Cat# 415111-C2
CHRNE-C3	ACD Bio-Techne	Cat# 1242771-C1
Software and algorithms
Fiji	Open platform	https://fiji.sc/
BioRender	Web based	https://www.biorender.com/
GraphPad Prism 10	GraphPad	https://www.graphpad.com/scientific-software/prism/
CellChat (v2.1.2)	Jin et al.31	http://www.cellchat.org/
Cell Ranger (v7.1.0)	10X Genomics	https://www.10xgenomics.com/support/software/cell-ranger/downloads/previous-versions
MyoSight	Babcock et al.50	https://github.com/LyleBabcock/MyoSight/tree/master
Dropkick (v1.2.7)	Heiser et al.51	https://pypi.org/project/dropkick
Solo (v1.3)	Bernstein et al.52	https://github.com/calico/solo
Seurat (v5.0.1)	Hao et al.53	http://www.satijalab.org/seurat
SoupX	Young and Behjati54	https://github.com/constantAmateur/SoupX
Harmony (v1.2.0)	Korsunsky et al.55	https://github.com/immunogenomics/harmony
SpaceRanger (v2.0.0)	10X Genomics	https://www.10xgenomics.com/support/software/space-ranger/downloads/previous-versions

Immunofluorescence in muscle cryosections

Prefixed plantaris muscle was cryosectioned at 20 μm thickness and permeabilized in 0.5% Triton X-100 in PBS for 25 min. Sections were incubated with M.O.M. blocking solution (Vector Laboratories) for 30 min, followed by 5% BSA (Sigma A7030) in PBS for 1 h at room temperature. Slides were then incubated with primary antibodies at 4°C overnight. For NMJ staining, biotin-conjugated α-bungarotoxin (1:100) was added to the primary antibody solution. After PBS washes, sections were incubated with fluorophore-conjugated secondary antibodies and/or fluorophore-conjugated streptavidin (1:200) and labeling dyes diluted in 5% BSA; nuclei were counterstained with DAPI (Invitrogen, 1:500). Following final washes, tissue sections were mounted with Fluoromount-G (SouthernBiotech). Primary and secondary antibodies are listed in Key Resources Table.

RNAscope multiplex fluorescence in situ hybridization

RNAscope multiplex fluorescent in situ hybridization was performed using the RNAscope Multiplex Fluorescent Reagent Kit v2 (Advanced Cell Diagnostics) following the manufacturer’s instructions with minor modifications. Sections were fixed in 4% PFA for 1 h at 4°C and treated with Protease Plus for 7 min at 40°C. Plantaris muscle from wild-type mice was dissected, freshly frozen in liquid nitrogen, and processed into 10-μm longitudinal sections. Probe hybridization was carried out by incubating sections with target probes for 2 h at 40°C. Dual-probe sets were used, with Inhba-C2 (455871-C2, ACD Bio-Techne) included in both sets as a common probe, paired with either Ajap1-C3 (549641-C3, ACD Bio-Techne) or Col3a1-C1 (455771, ACD Bio-Techne). The Inhba-C2 probe was conjugated to Opal 690, whereas Ajap1-C3 and Col3a1-C1 were conjugated to Opal 570 (Akoya Biosystems). NMJs were identified by staining with wheat germ agglutinin (WGA; Biotium 29077; 1:200) and DAPI for 30 min at room temperature. After washing, sections were mounted with ProLong Gold Antifade Mountant (Thermo Fisher). Confocal images were acquired on a Zeiss LSM-980 microscope using a Plan-Apochromat 40×/1.4 NA oil-immersion objective. NMJ endplates were defined by WGA staining, and quantification of Inhba+, Ajap1+, and Col3a1+ cells was performed using Fiji.

Digital image acquisition

Confocal images of whole-mount preparations and RNAscope sections were acquired using a Zeiss LSM 700 or LSM 980 confocal microscope equipped with an LD LCI Plan-Apochromat 25×/0.8 Imm Korr DIC M27 objective. For three-dimensional analyses, image stacks were collected over a 426.78 × 426.78 μm field of view with a depth of 50–100 μm. Higher-magnification images for analysis of synaptic myonuclei (SynM) at the NMJ were obtained using a Plan-Apochromat 40×/1.3 Oil DIC M27 objective. A Zeiss Axioscan 7 microscope with a 20× dry objective was used for lower-magnification imaging, including longitudinal and cross-sectional sections. Fluorophores (3–4 per experiment) were excited with 405, 488, 568, 633, and 733 nm laser lines as appropriate. Images were processed in Fiji, where global brightness and contrast adjustments were applied uniformly to reduce background.

Imaging analysis

Myofiber cross-sectional area (CSA) was quantified in Fiji using the MyoSight plugin⁵⁰. Li thresholding was applied with particle size limits set to 200–3000 μm². Myofibers not detected by the automated pipeline were manually outlined and added to the ROI Manager. Feret diameter and CSA measurements were exported as Excel files for subsequent analysis.

NMJ size and innervation were quantified in Fiji. For three-dimensional analysis of NMJ volume, images were preprocessed using Gaussian blur and background subtraction, and an appropriate threshold was applied in 3D Manager to generate a binary mask. For two-dimensional analysis of NMJ area, preprocessed image stacks were converted to maximum-intensity projections and thresholded to obtain masks corresponding to endplate shape. The innervated area was calculated as the proportion of the NMJ mask that overlapped with the presynaptic terminal mask.

Cell isolation by flow cytometry for scRNA-seq

For scRNA-seq, 10–16 plantaris muscle samples from 13–15-week-old male and female mice per condition were pooled, manually minced, and digested in DMEM containing Liberase (0.1 mg/g muscle weight; Roche 5401020001) and 0.3% dispase (Sigma D4693–1G) for 1.5 h at 37°C in a shaker incubator. The resulting homogenate was filtered, pelleted, and frozen in FBS (Gibco 26-170-043) with 10% DMSO (Sigma C6164) for storage at −80°C. For sorting, cells were thawed, resuspended in PBS with 1% goat serum (FACS buffer), and stained with PE-Cy7–conjugated anti-CD45 (BioLegend 103114; 1:200) and anti-Sca-1 (BioLegend 108114; 1:200) for lineage-negative selection (Lin−), and BV605–conjugated anti-CD31 (BioLegend 102418; 1:200) for endothelial cell (EC) identification. Cells were pelleted and incubated with the appropriate Cell Multiplexing Oligos (CMOs; 10X Genomics) for 5 min at room temperature, washed, and resuspended in 0.05% UltraPure BSA (ThermoFisher AM2616) in calcium- and magnesium-free PBS. Cells from all conditions were pooled and sorted on a BD FACSAria II, collecting sorted cells into 0.05% UltraPure BSA. ECs were isolated based on CD31+ staining, and from the CD31− fraction, Sca1−CD45− (Lin−) cells were collected to enrich for muscle stem cells (MuSCs) and Schwann cells (SCs). Cell concentration was measured using an automated cell counter (Bio-Rad TC20), and samples were adjusted as needed to reach the optimal range for loading onto the 10X Chromium chip. Cell suspensions were loaded into the 10X Chromium Controller using the Single Cell 3′ Reagent Kit v3.1 (dual index) according to the manufacturer’s instructions. After library construction, sequencing was performed on an Illumina NextSeq platform.

Nuclei isolation from plantaris muscle and snRNA-seq library preparation

Plantaris muscle was isolated immediately after euthanasia and snap-frozen in liquid nitrogen for storage at −80°C. Nuclei were isolated using the Chromium Nuclei Isolation Kit following the manufacturer’s instructions with minor optimizations for skeletal muscle. Briefly, two plantaris muscle samples from 13–15-week-old male and female mice were mechanically disrupted with a pestle in Lysis Buffer, and the homogenate was transferred to a Nuclei Isolation Column and centrifuged at 16,000 g for 20 s at 4°C. The column was discarded, and the flowthrough was collected in a 1.5 mL LoBind Eppendorf tube to facilitate pellet recovery. Centrifugation parameters were modified to improve nuclei yield: nuclei were pelleted at 700 g for 5 min at 4°C before Debris Removal Buffer was added, and again at 700 g for 10 min at 4°C after incubation. Nuclei were resuspended in Wash Buffer and subjected to a final centrifugation step (700 g for 5 min at 4°C). The optimal final resuspension volume was 70 μL when starting from two plantaris muscle samples. Nuclei concentration was measured using a Bio-Rad TC20 automated cell counter, and suspensions were adjusted as needed to reach the recommended loading concentration for the 10X Chromium system. A small aliquot was stained with DAPI and examined under a fluorescence microscope to confirm nuclei integrity, rounded morphology, predominance of singlets, and absence of residual debris. Approximately 15,000 nuclei were loaded into the 10X Chromium Controller using the Single Cell 3′ Reagent Kit v3.1 according to the manufacturer’s protocol. Libraries were sequenced on an Illumina NextSeq platform.

Single-cell and single-nucleus data preprocessing

Single-cell RNA-seq data were processed using Cell Ranger (v7.1.0) and mapped to the GRCm39 reference genome. Dropkick (v1.2.7) was used to identify empty droplets and low-quality barcodes⁵¹, and Solo (v1.3) was applied to demultiplex CMO hashtags and remove doublets⁵². Single-nucleus RNA-seq data were processed with Cell Ranger (v7.1.0) using the same reference genome. For both single-cell and single-nucleus datasets, downstream analyses were performed in Seurat (v5.0.1)⁵³.

Quality control

For scRNA-seq data from Sample A, cells with >40,000 RNA counts, <700 or >7,000 detected features, >6% mitochondrial reads, or <1.5% ribosomal reads were removed. For Sample A2, cells with >60,000 RNA counts, <1,200 or >7,500 detected features, >5% mitochondrial reads, or <1.5% ribosomal reads were excluded. For snRNA-seq data, ambient RNA was removed using SoupX, with the contamination factor (rho) set to the lower of the automatically estimated value or 0.1⁵⁴. Doublets were removed using DoubletFinder, applying a well-specific estimated doublet rate based on the nearest 10X multiplet-rate prediction. After doublet removal, nuclei with >20,000 RNA counts, <300 or >5,000 detected features, or >5% mitochondrial reads were excluded.

Batch integration, clustering and cell type annotation

scRNA-seq data from Samples A and A2 were aggregated and processed in Seurat using NormalizeData and FindVariableFeatures (1,500 features). Cell cycle effects were regressed out with ScaleData, and the top 30 principal components (PCs) were computed using RunPCA. Integration across CMO-labeled samples was performed with Harmony (v1.2.0) using the top 30 PCs⁵⁵, followed by UMAP embedding with RunUMAP. Clustering was performed using FindNeighbors and FindClusters with a resolution of 0.03. Cell type identities for all scRNA-seq and snRNA-seq datasets were assigned based on canonical skeletal muscle marker genes. Data from snRNA-seq were processed in Seurat with NormalizeData and FindVariableFeatures (2,000 features), followed by cell cycle regression using ScaleData. The top 40 PCs were computed with RunPCA. Samples were integrated with Harmony using the top 40 PCs, and UMAP embeddings were generated with RunUMAP. Clustering was performed using FindNeighbors and FindClusters with a resolution of 0.1.

Integration of single-cell and nucleus RNA-seq data

Integration of the snRNA-seq and scRNA-seq datasets was performed by identifying shared top variable features between the two technologies and supplementing these with markers for nuclei-derived populations absent from the scRNA-seq data. The top 4,000 variable features from each dataset were intersected and combined with marker genes from the myonuclear, NMJ, and MTJ clusters. Marker genes for these clusters were defined using Seurat’s FindMarkers with an adjusted p-value < 0.001, log₂ fold change > 3.0, and expression in >20% of nuclei within the cluster. Using this combined feature set, cell cycle effects were regressed out with ScaleData, and the top 30 principal components were computed using RunPCA. Integration across all samples was then performed with Harmony using the top 30 PCs, followed by UMAP projection using RunUMAP. Cell type annotations obtained from the individual scRNA-seq and snRNA-seq analyses were transferred to the integrated dataset.

Analysis of mononucleated cell types

Mononucleated cell types represented by at least 30 cells or nuclei in every condition in both the single-cell and single-nucleus datasets were selected. Endothelial cells (ECs), fibro–adipogenic progenitors (FAPs), muscle stem cells (MuSCs), and Schwann cells (SCs) met this criterion. Differentially expressed genes for Den and OVL were computed for each cell type and for each modality (single-cell or single-nucleus) using Seurat’s FindMarkers, with basal cells used as the reference group. Significant genes were defined as those with adjusted p-value < 0.05, |log₂ fold change| > 0.15, and expression in at least 5% of cells or nuclei within the enriched group. Genes that were significant in both the single-cell and single-nucleus datasets were retained as the final gene set.

Sub-clustering analysis of myonuclei, NMJ, and Schwann cell clusters

Higher-resolution analyses were performed by subsetting specific populations and re-running the processing workflow. For the myonuclear, NMJ, and MTJ populations, nuclei belonging to any of these clusters were subset from the snRNA-seq dataset. Data were processed in Seurat using NormalizeData and FindVariableFeatures (1,000 features). Cell cycle scores and mitochondrial read percentages were regressed out with ScaleData. The top 10 principal components (PCs) were computed with RunPCA, and integration across samples was performed with Harmony using the top 10 PCs. UMAP embeddings were generated using RunUMAP, and clustering was performed with FindNeighbors and FindClusters at a resolution of 0.25. For NMJ-specific analysis, nuclei assigned to the NMJ cluster were subset and processed with NormalizeData and FindVariableFeatures (1,000 features). Cell cycle effects were regressed out with ScaleData. The top 20 PCs were computed with RunPCA, followed by Harmony integration using the top 20 PCs. UMAP embeddings were generated using RunUMAP. For Schwann cell analysis, the integrated snRNA-seq and scRNA-seq datasets were subset to Schwann cells and nuclei. A small Pax7+ subcluster was removed prior to downstream processing. Data were processed with NormalizeData and FindVariableFeatures (500 features). Features differentially expressed between single-cell and single-nucleus Schwann cell profiles (adjusted p-value < 0.1) were excluded, resulting in 283 retained variable features. Cell cycle scores and mitochondrial percentages were regressed out using ScaleData. The top 10 PCs were computed with RunPCA, and Harmony was used to integrate across technologies using these PCs. UMAP embeddings were generated with RunUMAP, and clustering was performed with FindNeighbors and FindClusters at a resolution of 0.2. For analysis of SCs from the dataset of Nicoletti et al.²², raw scRNA-seq data were downloaded from GEO (GSE221736) and processed with Cell Ranger (v7.1.0) using the GRCm39 reference genome. Downstream processing was performed in Seurat (v5.0.1)⁵³. Cells with >50,000 RNA counts, <200 or >6,000 detected features, or >20% mitochondrial reads were removed. Doublets were removed with DoubletFinder using a 5% estimated doublet rate. Data were processed with NormalizeData, FindVariableFeatures (1,000 features), and ScaleData. Dimensionality reduction was performed with RunPCA and RunUMAP using the top 30 PCs. Clustering was carried out with FindNeighbors and FindClusters at a resolution of 0.2. Schwann cells were identified based on canonical marker expression. After subsetting to SCs, subclustering was performed using the same workflow, with 1,200 variable features, 20 PCs, and a clustering resolution of 0.25.

Atrophy and hypertrophy scores

To evaluate the degree of atrophy or hypertrophy across conditions, curated gene sets representing each state were compiled from the literature and scored in the single-cell and single-nucleus RNA-seq data. The atrophy gene set included Runx1, Gadd45a, Myog, Ankrd1, Fbxo32, Chrng, and Ctsl. The hypertrophy gene set included Flnc, Mybpc2, Postn, Klhl41, Tpm1, Irs1, Creb5, Atp8a1, Gnas, Cast, Mcu, Nr3c1, Tpm2, Esr1, Itgb6, Palld, Taco1, Col1a2, Apoe, Lama2, Gatm, and Trim63. Expression scores for each gene set were computed using Seurat’s AddModuleScore function.

Ligand-receptor analysis of NMJ and Schwann cells

To infer intercellular communication between NMJ nuclei and Schwann cells, CellChat (v2.1.2)³¹ was used. Clusters were required to contain at least 15 cells to be included in the analysis, and the aSC1 cluster from the OVL condition was excluded based on histological evidence indicating that this population was not clearly present. Ligand–receptor interactions and associated signaling pathways were inferred using the computeCommunProb and computeCommunProbPathway functions.

Visium spatial transcriptomics

Sections were collected directly onto a Visium slide (10X Genomics PN-1000193), ensuring that each section was positioned within the 8 × 8 μm fiducial frame. Consecutive cryosections were collected onto Superfrost Plus slides for subsequent immunofluorescence staining. The Visium slide was fixed in 100% methanol at −20°C for 30 min, followed by hematoxylin staining for 6 min (Sigma-Aldrich, #MH16) and eosin staining for 1 min (Sigma-Aldrich, #HT110216). Slides were mounted in 30% glycerol with a coverslip, and brightfield images of each section, including the fiducial frame, were acquired on a Leica SP8 system. Sections on the reference slide were fixed in 4% PFA for 5 min at room temperature and stained with α-bungarotoxin–Alexa 488 (1:100) and DAPI (1:500) in PBS for 5 min. Immunofluorescence images from the reference slide were acquired using a Zeiss LSM780 confocal microscope.

Visium pre-processing

Visium data were processed using SpaceRanger (v2.0.0) mapped to the mm10–2020-A reference genome. Seurat (v4.4.0) was used for downstream processing. Spots with nCount < 1,000 or nFeature < 500 were removed. Data were then processed using Seurat’s NormalizeData and FindVariableFeatures workflows, with 2,000 variable features selected.

Visium cell type annotation

Spots were annotated using the single-cell atlas from McKellar et al. (2021)⁵⁶ and the BayesPrism (v2.0) framework with minor modifications. NMJs were first identified by extracting published NMJ calls^7,57,58. Using the myogenic/myofiber RData object from McKellar (56), cells were clustered with 30 PCs and a resolution of 0.2; clusters lacking any previously reported NMJs were removed, leaving predominantly myonuclei. Clustering was repeated, and the cluster retaining the highest number of known NMJ labels and expressing the highest levels of Chrne was designated as NMJ. Remaining McKellar cells were grouped into broader categories, and informative markers were identified using Seurat’s FindAllMarkers. This marker list was filtered using the get_cell_type_model function from McKellar et al.,⁵⁶ with an additional requirement of pct.1 > 0.5 and absolute logFC > 1 before being used as input for BayesPrism. Bar plots were generated using ggplot2 (v3.4.4). To classify Visium spots as NMJ or non-NMJ, spots were first clustered using Vesalius (v1.0.1) with the following settings: buildImageArray (resolution = 100), equalizeHistogram (sleft = 15, sright = 15), iterativeSegmentation.array (method = c(“box”,”iso”), colDepth = 6), and isolateTerritories.array (captureRadius = 0.035, minBar = 5); colDepth, captureRadius, and minBar were specifically tuned for this dataset. Only results with tile = 1 and cc = 1 were retained. Spots predicted by BayesPrism to contain ≥1% NMJ were initially classified as NMJ. In parallel, serial sections were processed for immunofluorescence using α-bungarotoxin–488 and DAPI to identify NMJs, and the resulting images were aligned to the Visium slide. Spots were excluded as NMJ if the majority of NMJs within a given Vesalius cluster did not spatially correspond to NMJ structures observed in the matched immunofluorescence section.

Visium differential analysis

NMJ and non-NMJ spots were filtered to retain only those with nCount > 5000 and nFeature > 1500. Differential genes for each individual slide were identified with FindMarkers using MAST with a minimum percentage of 0.2 and a log fold change threshold of 0.2. Results were filtered if the adjusted p-value was below 0.01.

Statistical analysis of Visium data

Figure captions indicate the sample size for each experimental group, with a minimum of three biological replicates unless noted otherwise. Statistical analyses, excluding sequencing-related analyses, were performed using GraphPad Prism. Quantitative data displayed as histograms are presented as mean ± standard error of the mean, with error bars indicating variability; violin plots represent the median and quartiles. Descriptive statistics were calculated from group-averaged values. Inter-group comparisons were performed using two-tailed independent-samples t tests, with statistical significance defined as p < 0.05.

Human RNAscope multiplex fluorescence in situ hybridization

RNAscope multiplex fluorescent in situ hybridization was performed using the RNAscope Multiplex Fluorescent Reagent Kit v2 (Advanced Cell Diagnostics) following the manufacturer’s instructions, with minor modifications as used for mouse cryosections. Denervated human muscle sections were hybridized with INHBA-C2 (415111-C2, ACD Bio-Techne) and CHRNE-C3 (1242771-C1, ACD Bio-Techne). The INHBA-C2 probe was conjugated to Opal 690, and the CHRNE-C3 probe was conjugated to Opal 520 (Akoya Biosystems). Sections were counterstained with DAPI for 30 min at room temperature, washed, and mounted with ProLong Gold Antifade Mountant (Thermo Fisher). Confocal imaging was performed on a Zeiss LSM 980 system using a Plan-Apochromat 40×/1.4 NA oil-immersion objective. NMJ endplates were identified based on CHRNE probe signal.

Human sample and library preparation for RNA-seq

For the analysis of PAN and healthy human muscles, bulk RNA-seq was performed on frozen muscle biopsy specimens as previously described at the NIH (Bethesda, USA)^59,60. RNA was extracted using TRIzol, and libraries were prepared using either the NeoPrep system following the TruSeq Stranded mRNA Library Prep protocol (Illumina, San Diego, CA) or the NEBNext Poly(A) mRNA Magnetic Isolation Module combined with the Ultra II Directional RNA Library Prep Kit for Illumina (New England BioLabs, #E7490 and #E7760). For the analysis of exercised and resting muscles, sequencing libraries were prepared from muscle biopsies following RNA extraction (Qiagen). Libraries were generated using the Watchmaker mRNA Library Prep Kit v1.1.0823 (for 24 h samples) or SMART-Seq mRNA LP kit (Takara Bio #634768) (for 6 weeks samples) and sequenced on an Illumina NextSeq 2000 platform (paired-end, 51 bp read length, ~250 million reads per run).

Human bulk RNA-seq data processing and analysis

For RNA-seq analysis of the human PAN and healthy samples, sequencing reads were demultiplexed using bcl2fastq (v2.20.0) and preprocessed with fastp (v0.21.0)⁶¹. Gene abundances were quantified using Salmon (v1.5.2)⁶². Counts were normalized using the Trimmed Mean of M values (TMM) method implemented in edgeR (v3.34.1)⁶³ for downstream visualization. Normalized counts were gene-scaled for heatmap representation. For the RNA-seq analysis of the human exercise and control muscles, sequencing data was processed using the Nextflow nf-core pipeline^64,65. Raw reads were trimmed using Trimmomatic (v.0.6.10) and quantified using Salmon (v.1.10.3) with the reference genome GRCh38^62,66. Counts were summarized using tximport (v1.30.0) and transformed using variance stabilizing transformation via DESeq2^67,68. Transformed counts were gene-scaled for heatmap representation.

Human spinal cord microarray data processing and analysis

Microarray data from human muscles with spinal cord injury were obtained from the Gene Expression Omnibus (GEO) series GSE21496 and GSE21497 as raw Affymetrix CEL files⁴². Detection calls for each probeset, expression summarization and between-array normalization were carried out with the MAS5.0 algorithm using the affy package⁶⁹. MAS5-normalized expression values were then transformed to the log₂ scale, with a lower bound of 1 applied to avoid taking logarithms of extremely low or zero intensities. Probes were re-annotated to current gene symbols, and low-quality or non-informative probes were excluded. Gene-wise expression patterns across Control, SCI (2d and 5d) groups were visualized as boxplots.

QUANTIFICATION AND STATISTICAL ANALYSIS

The number of NMJs and the number of animals used for each panel was shown in the figure legend. The data was collected from at least 3 different mice except for snRNA-seq data and spatial transcriptomics, where 2 mice were used. Sample sizes were not pre-determined based on statistical power calculations, and no randomization techniques were used. Mice were randomly assigned to experiments. No blinding was performed. Variation was indicated using SD. To assess the statistical significance, we use two-tailed unpaired Student’s t-tests in GraphPad Prism (version 10) between control and perturbed muscles, and p-value < 0.05 was considered statistically significant. For transcriptomic analysis, significant genes were defined as those with adjusted p-value < 0.05, |log₂ fold change| > 0.15, and present in at least 5% of cells or nuclei. For spatial transcriptomics, differential genes were defined as those with adjusted p-value < 0.01, |log₂ fold change| > 0.2, and with a minimum percentage of 0.2. Throughout the figures, asterisks indicate the significance of p-values: *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Supplementary Material

1

Document S1. Figures S1–S4.

2

Table S1. DEGs across skeletal muscle-resident populations captured by sc- and snRNA-seq (related to Figure 1)

3

Table S2. DEGs and functional enrichment analysis from the myonuclear subpopulations (related to Figure 2)

4

Table S3. DEGs and functional enrichment analysis of SynM and NMJ-positive spots (related to Figure 3 and Figure 4)

5

Table S4. DEGs across SC subpopulations and significant SC–SynM interaction pairs (related to Figure 5 and Figure 6)

Highlights.

• A transcriptomic atlas reveals muscle adaptations to atrophy and hypertrophy in mice

• Distinct SynM gene programs drive NMJ remodeling across both physiological states

• Denervation activates tSCs (Inhba+) at the NMJ, signaling to SynM via TGFβ/activin

• Denervated and exercised human muscles exhibit similar NMJ-associated changes