2.1 Preprocessing, quality control, and batch correction of scRNA-seq data

Raw sequencing reads were downloaded from the Sequence Read Archive (SRA) databases. The Cell Ranger (v3.1.0) pipeline²⁴ was used to align the datasets to the mouse genome (GRCm39) and create raw gene expression matrix. The raw gene expression matrix was filtered, normalized, and clustered using Seurat (v4.4.0).²³ Cells containing fewer than 500 features or more than 10% unique transcripts from mitochondrial genes were removed. After filtering, standard Seurat workflow (normalization, finding variable features, RunPCA, finding neighbors, finding clusters, and RunUMAP) was performed on each dataset. SCTransform v2^{25, 26} was used to normalize the datasets. RunPCA function in Seurat was used to decrease the linear dimensionality. After preprocessing the data, the datasets were merged and unintegrated analysis was performed to identify any batch effects. After confirming minimal batch effects, the Seurat data integration method was used to correct batch effects and perform comparative analysis across different experimental conditions.²³

2.2 Cell type annotation

Initially, the raw clustering of the integrated dataset resulted in 31 different clusters (cluster resolution 0.8). Expression patterns of canonical marker genes were used to define the various cell clusters. Clusters that had similar marker expression patterns were merged, which resulted in a final 11 distinct cell clusters. A list of markers used to identify each cluster can be found in Table S1.

2.3 Cell–cell communication analysis with CellChat

Cellular communication network analysis and visualization were performed using CellChat (version 2.1.2) (https://github.com/jinworks/CellChat) on the integrated datasets. To identify cytokine-mediated communications, we subset the database only for secreted factor signaling interactions. To predict cell–cell communication, CellChat identifies overexpressed ligands or receptor in one cell type and infer the ligand–receptor interactions if either ligands or receptors are overexpressed in other cell types. The “identifyOverExpressedGenes” function was used to identify genes that were overexpressed and then the gene expression data were projected onto protein–protein interaction network. To infer communication networks, we used “computeCommunProb” with the parameter “type = triMean” and filtered out cell–cell communications that were present in less than 10 cells using the “filterCommunication” function. We used “aggregateNet” to quantify the total number of communications in the integrated datasets. To reveal changes in sending or receiving signals for each cell type during regeneration, “netAnalysis_signalingRole_scatter” was used. The “computeCommunProbPathway” function was used to sum the communication probabilities of all ligand–receptors interactions for each signaling pathway. To compare the overall information flow between different timepoints, we used “rankNet” with “do.stat = TRUE”, and the paired Wilcoxon test was run to identify significant differences of the information flow between two timepoints. To identify upregulated ligand–receptor pairs during regeneration we compared each regenerating timepoint to 0 dpi using “identifyOverExpressedGenes” with the following parameters: “thresh.pc = 0.1”, “thresh.fc = 0.05”, and “thresh.p = 0.05”. To visualize upregulated ligand–receptor pairs, “netVisual_bubble” was used with “source.use” and “target.use” to specify the targets and sources of signals.

2.4 Data availability

All datasets analyzed in this study are published.^18-20 GEO accession information for the datasets can be found in Table S2. Codes and integrated datasets are available on reasonable request.

3.1 Integration of scRNA-seq datasets provides a high-resolution view of regenerating muscle transcriptomics

To comprehensively evaluate the transcriptomic profiles of mouse skeletal muscles at the single-cell level during homeostasis and regeneration, we curated 15 publicly available scRNA-seq datasets generated from uninjured and injured mouse tibialis anterior (TA) muscles at various timepoints^18-20 (Table S2). The raw data were aligned to the reference mouse genome with Cell Ranger²⁴ and processed with Seurat²³ to construct uniform manifold approximations and projections (UMAP) ²⁷ for all datasets. Unintegrated clustering analysis revealed that each of the datasets showed similar clustering behavior regardless of their sources and injury methods (Figure S1), suggesting that there was little batch effect. The datasets were then aggregated with Cell Ranger, generating four datasets for 0, 2, 5, and 7 days post injury (dpi), respectively. An estimated total of 92,066 cells were processed, and after filtering for high-pass feature counts 75,150 cells were integrated using Seurat (Figure 1A). Integration and unsupervised clustering by UMAP returned 11 distinct clusters, with transcriptomic profiles identifying with macrophages, FAPs, MuSCs, endothelial cells, smooth muscle cells, mature skeletal muscles, glial cells, neutrophils, tenocytes, B cells, and T cells (Figure 1B). A heatmap visualization of the top 10 most variably expressed genes in each cluster revealed a distinct transcriptional program and expression of known lineage-specific transcripts (Figure S2A). Macrophages constituted ∼37% of the total population based on the expression of the canonical macrophage markers Itgam, Csfr1 and Adgre1 (Figures 1C and S2B). FAPs expressing Pdgfra and Ly6a were the second most abundant cell population at ∼22%, whereas ∼5.7% of the total population (4335 cells) were assigned to MuSCs expressing the myogenic transcription factors Pax7 and Myod1 (Figures 1C and S2B).

To determine how the cell populations may change during the process of regeneration, we used the split plot function in Seurat to visualize single cells at their respective timepoints (Figure 1D). Uninjured muscles (0 dpi) were mostly composed of FAPs, endothelial cells, macrophages, and mature skeletal muscles. As expected, a transient increase in multiple immune cell populations after injury was observed: at 2 dpi and 5 dpi, immune cells comprised the largest portion of the total population, whereas at 7 dpi, their numbers returned to those of 0 dpi (Figures 1C and 1D). On the other hand, the numbers of FAPs and endothelial cells decreased drastically at 2 dpi and returned to their normal levels at 7 dpi (Figures 1C and 1D). In the uninjured state, ∼3.8% of the total cells were MuSCs, but at 2 dpi, very few MuSCs were detected (0.6%). The rarity of MuSCs at 2 dpi might be due to technical difficulties in their recovery from tissues with a high abundance of immune cells as previously suggested.^{18, 20} A drastic increase in the MuSC population was observed at 5 dpi and 7 dpi (Figures 1C and 1D) as expected.

To further validate the integrated datasets, we analyzed the temporal dynamics of gene expression by macrophages, FAPs, and MuSCs. As shown in Figure S3A, macrophages presented a transcriptional profile reflecting a pro-inflammation-to-anti-inflammation switch from early to later phase of regeneration as reported^{18, 20}: expression of Chil3, Ptgs2 and Cxcl3 at 2 dpi, and increased expression of Apoe, Ms4a7, Lsp1, and H2-Aa at 5 and 7 dpi. The temporal profile of gene expression in FAPs (Figure S3B) was consistent with a transition from homeostasis at 0 dpi (expression of ECM genes Dcn, Gsn, and Dpp4) to activation and remodeling as marked by the upregulation of Bgn, Mest, Postn, and Col1a2,^{18, 20} and returning to the homeostatic state at 7 dpi. Finally, the expression profile of MuSCs followed the anticipated progression from quiescence to activation, proliferation, and differentiation (Figure S3C). Taken together, these results fully validate our methods of data integration.

3.2 Cytokine expression across all cell types during muscle regeneration

To interrogate the integrated scRNA-seq data for cytokine expression, we set a threshold of 1%, that is, a cytokine is considered to be expressed in a cell type if it is detected in at least 1% of the cells in the cluster. We examined the expression of 393 mouse cytokines (https://www.informatics. jax.org/) (Table S3) and found that 183 of them were expressed in our integrated datasets (Table S3). The highest numbers of cytokines were expressed in FAPs (171), tenocytes (161), MuSCs (160), smooth muscle cells (145), glial cells (138), and macrophages (124). We found a significant overlap of cytokines that were expressed in various cell types (Figure 2A). For instance, of the 171 cytokines expressed by FAPs, 134 were also expressed by MuSCs, whereas 101 cytokines overlapped between MuSCs and macrophages.

To determine whether any cytokines were exclusively expressed in one cell type, we compared cytokine expression among macrophages, FAPs, MuSCs, endothelial cells, and the cell types that contributed most to the integrated datasets. A Venn diagram showed that MuSCs expressed 14 unique cytokines (6.5% of MuSC-expressed), whereas 24 (11.1%), 13 (6%), and 8 (3.7%) unique cytokines were observed in FAPs, macrophages, and endothelial cells, respectively (Figure 2B). Over the course of regeneration, the number of cytokines expressed in each cell type did not vary drastically (Figure 2C): FAPs and MuSCs expressed the highest numbers of cytokines at 2 dpi and 5 dpi, whereas macrophages expressed the most cytokines at 0 dpi; in tenocytes, higher numbers of cytokines were expressed at 5 dpi and 7 dpi.

To identify cytokines differentially expressed during regeneration, analysis was performed for 2, 5, and 7 dpi in comparison with 0 dpi. When compared to homeostatic muscles, 85, 112, and 61 cytokines were upregulated at 2, 5, and 7 dpi, respectively (Table S4). Among the upregulated cytokines, 35 appeared at all regenerating timepoints, whereas 9 and 22 cytokines were upregulated at 2 and 5 dpi, respectively (Figure 2D). We further examined differential cytokine expression in MuSCs, FAPs, and macrophages (Table S4); expression profiles of representative genes are shown in Figures 2E–2G. It is evident that within each cell type different genes were upregulated at different timepoints of regeneration. Similarly, dynamic patterns of cytokine expression throughout regeneration were also observed for other cell types (data not shown).

3.3 Extensive and dynamic cell‒cell communications in homeostatic and regenerating muscles

To define the intercellular communication networks facilitated by secreted factors, we performed CellChat²⁸ analysis, which included ∼1300 secreted factor-mediated interactions (CellChatDB v2). We chose to use the stringent trimean method to identify only strong and highly significant interactions. Our analysis revealed a total of 4141 distinct cell–cell interactions that were predicted to be mediated by secreted factors during regeneration (Figure 3A). The total number of cytokine-mediated interactions were highest at 5 dpi (1492), whereas the lowest number of interactions (610) were observed at 0 dpi (Figure 3B). We further zoomed in on the interactions among MuSCs, FAPs, and macrophages (Figure 3C). Predicted interactions between macrophages and FAPs increased more than two-fold at 2 dpi from 0 dpi and further increased at 5 dpi. Indeed, 5 dpi saw the largest number of interactions involving macrophages with either FAPs or MuSCs. This number decreased drastically at 7 dpi, consistent with the idea that infiltrating macrophages mostly subsided at this stage of regeneration. The total number of interactions involving MuSCs (with both macrophages and FAPs) was also the highest at 5 dpi. The interactions were bi-directional at every timepoint; there were substantial numbers of both macrophage–MuSC interactions and MuSC–macrophage interactions, and similarly for macrophage–FAP and FAP–MuSC interactions (Figure 3C, arrows indicating signaling direction).

In addition to the number of interactions described above, we also used CellChat to analyze the interaction strength. Based on this parameter, FAPs were the major contributors to both sending signals (expressing cytokine) and receiving signals (expressing receptor) in homeostatic muscles (0 dpi, Figure 4A), whereas smooth muscle cells and tenocytes were also major signal senders. At 2 dpi (Figure 4B) macrophages were the major outgoing source of signals, whereas B cells received the most signals. At 5 dpi (Figure 4C), FAPs again became the major signal senders as well as receivers, followed by macrophages and tenocytes, whereas neutrophils and B cells were also major signal receivers. MuSCs also markedly increased their contributions to both sending and receiving signals at 5 dpi. Interestingly, at 7 dpi (Figure 4D) macrophages and B cells received the most signals, whereas tenocytes and MuSCs sent the most signals.

3.4 Cell type-specific cytokine signaling before and during muscle regeneration

To assess changes in specific signaling pathways over the course of regeneration, we examined information flow, which is defined as the sum of communication probabilities among all pairs of cell groups in the inferred networks.²⁸ Pathways that underwent any significant change during regeneration were included in the analysis. As shown in Figure 5, signaling by some cytokines appeared at 2 dpi only (ncWNT, IL6, and ANNEXIN), 5 dpi only (IFN-1, SLIT, and GDF), or both 2 and 5 dpi (IL10, KLK, and OSM). Some pathways were present at all the regeneration timepoints (2, 5, and 7 dpi) but not at 0 dpi, including CCL, FGF, MK, RANKL, ANGP, and PARs. VEGF signaling was unique, as the only pathway appearing exclusively in homeostatic muscles.

We then performed analysis of the outgoing (expressing cytokine) and incoming (expressing receptor) signaling by each of the 11 subpopulations of cells in homeostatic (0 dpi) and regenerating muscles (2, 5 and 7 dpi). As shown in Figure 6, we found some signaling to be constitutive and others to display temporal dynamics throughout regeneration. Take MuSC for example (see red boxes in Figure 6). Of the outgoing signals (Figure 6A), MIF signaling was present at all timepoints, including 0 dpi, whereas BMP signaling was restricted mostly to homeostatic muscles. Strong PDGF signaling from MuSCs appeared at 2 dpi and it was mostly gone by 7 dpi. Several other outgoing pathways were present in MuSCs at 5 dpi, including TGFβ, SPP1, and IGFBP. Of the signals that MuSCs were predicted to receive (Figure 6B), PTN stood out as a constitutive signal at all timepoints (0, 2, 5, and 7 dpi). Incoming TWEAK signaling was strong at all timepoints of regeneration but absent at 0 dpi. IGF signaling was also received by MuSCs at all timepoints, including 0 dpi, but the signaling strength was further increased at 5 and 7 dpi. Interestingly, incoming FGF signal was present exclusively at 7 dpi.

3.5 Communication between MuSCs and other cell types via PDGF, TGFβ, IGF, and TWEAK signaling

Further expanding the analysis above to predict the cell populations that receive or send signals from or to MuSCs, we highlight four signaling pathways: PDGF, TGFβ, IGF, and TWEAK (Figure 7). Known to promote myoblast proliferation and inhibit myogenic differentiation in vitro,¹⁷ and to be expressed by MuSCs in vivo,²⁹ the role of MuSC-produced PDGF in muscle regeneration is unclear. Our analysis revealed that PDGF signaling from MuSCs occurred exclusively during regeneration and not in homeostatic muscles, and that FAPs were the major receivers of PDGF signal although several other cell types were also involved in the receiving (Figure 7A). Similar to PDGF, TGFβ also promotes myoblast proliferation and inhibits myogenic differentiation in vitro,¹⁷ as well as negatively regulates myocyte fusion in vivo.³⁰ Moreover, similar to PDGF, what role MuSC-sourced TGFβ plays in muscle regeneration is not known. We found prevalent TGFβ signaling from MuSCs to FAPs and tenocytes, as well as autocrine signaling, at 5 dpi (Figure 7B). Early activation of autocrine TGFβ signaling aligns with the notion of promoting MuSC proliferation and inhibiting differentiation, but TGFβ signaling from MuSCs to FAPs and tenocytes warrant future investigation.

Studies across diverse animal and cellular models indicate that IGFs play crucial roles in regulating skeletal muscle development as well as homeostasis, hypertrophy, and regeneration of adult muscle tissues.^31-33 Despite the long-standing notion of autocrine IGF signaling in muscle cells based on numerous in vitro studies,¹⁷ evidence for a muscle cell-derived IGF function in vivo is still absent. Our analysis suggested that MuSCs predominantly served as a receiver rather than a sender of IGF signals, mostly from tenocytes and FAPs at all timepoints of regeneration and in homeostatic muscles, and also from macrophages at 5 and 7 dpi (Figure 7C). The TNF family member TWEAK has been reported to promote myoblast proliferation while inhibiting terminal differentiation in vitro.³⁴ Expression of TWEAK increased upon muscle injury,^{34, 35} and it was suggested that this expression mostly came from macrophages.³⁴ Systemic knockout of TWEAK in mice led to enhanced muscle regeneration,³⁵ but mice deficient in Tn14, a receptor of TWEAK, displayed impaired muscle regeneration upon injury, presumably due to a perturbed inflammatory response.³⁴ A cell type-specific function of TWEAK in vivo is not reported. Surprisingly, our results indicated that TWEAK signaling to MuSCs came from smooth muscle cells at all timepoints with the highest strength at 7 dpi, as well as from FAPs at 0 dpi and from tenocytes at 2 dpi (Figure 7D). These predictions of informational flow can guide future investigations to yield novel insights in the regulation of muscle regeneration.

3.6 Cytokine-receptor signaling to MuSCs during muscle regeneration

Finally, we were interested in the signaling to MuSCs upregulated in regenerating muscles and the sources of the signaling, particularly from FAPs, macrophages, tenocytes, or MuSCs (autocrine). Ligand‒receptor interactions significantly upregulated at 2, 5, and 7 dpi from 0 dpi are shown in Figure 8. The number of upregulated ligand–receptor interactions with high probability was at the lowest at 2 dpi and highest at 5 dpi. Among the upregulated ligand‒receptor pairs at 2 dpi communication from macrophages to MuSCs mediated by Spp1-Cd44 had the highest probability (Figure 8A), and this signaling remained at 5 dpi while also appeared as autocrine signaling in MuSCs (Figure 8B). Interestingly, even though overall Ptn signaling appeared to be constitutive (Figure 6B), specific Ptn-receptor signaling received by MuSCs were upregulated during regeneration, especially Ptn-Ncl signaling from tenocytes at all regeneration time points (Figure 8A–C) and also from FAPs at 5 dpi (Figure 8B). The large number of highly upregulated signaling to MuSCs at 5 dpi (Figure 8B) potentially offer new insights into the myogenic process in vivo. The significant reduction in the number of upregulated ligand‒receptor pairs at 7 dpi (Figure 8C) is not surprising, as new myofiber formation has reached full force by that time. It is plausible that the signaling pathways at 7 dpi are mainly engaged in myofiber maturation and/or clearance of immune cells.