Introduction

Positioned in the interstitial space between myofibers (Uezumi, et al., 2010), fibro-adipogenic progenitors (FAPs) interact with cellular components within skeletal muscle to ensure normal development, homeostasis, and regeneration of muscle tissue. During developmental myogenesis, embryonic FAPs expressing Osr1 contribute to limb muscle patterning by regulating the expression of extracellular matrix (ECM) genes that make up muscle connective tissue (Vallecillo-Garcia, et al., 2017). In young adults, FAPs are necessary for normal growth and long-term maintenance of skeletal muscle, which otherwise undergo progressive muscle atrophy in the absence of PDGFRα⁺ FAPs (Roberts, et al., 2013; Wosczyna, et al., 2019; Uezumi, et al., 2021). Upon muscle injury, FAPs proliferate in response to IL-4/IL-13 signals from eosinophils, and participate in clearing of necrotic debris via phagocytosis (Heredia, et al., 2013). Also, the proliferated FAPs regulate expansion and asymmetric commitment of muscle stem cells (MuSCs) via secreted factors such as WISP1, leading to robust de novo myofiber formation (Joe, et al., 2010; Lukjanenko, et al., 2019; Wosczyna, et al., 2019). Conversely, absence of FAPs or its functional decline with age cause premature differentiation of MuSCs upon injury, resulting in formation of smaller regenerated myofibers (Murphy, et al., 2011; Lukjanenko, et al., 2019; Wosczyna, et al., 2019). After sufficient regeneration of myofibers occurs, FAPs undergo apoptosis via TNF signaling from monocyte/macrophages, such that its numbers return to those of unperturbed muscles (Lemos, et al., 2015; Saito, et al., 2020). Failure to remove excess FAPs after muscle regeneration results in unwanted fibrosis, which compromises muscle function (Uezumi, et al., 2011; Uezumi, et al., 2014).

In addition to formation and maintenance of muscle tissue, FAPs also contribute to maturation and maintenance of the neural components within skeletal muscle. Previously, we reported that FAPs promote postsynaptic maturation of the neuromuscular junction (NMJ) through the BAP1/SMN axis during postnatal development (Kim, et al., 2022). Selective inactivation of Bap1 in FAPs results in dysfunctional NMJs, with sustained expression of the immature form of acetylcholine receptor subunit, AchRγ, in skeletal muscle (Kim, et al., 2022). Progressively, these mice exhibit denervation at the NMJ, retraction of motor axons, reduction of myelination and axon diameter, and eventually motor neuron loss, suggesting that FAPs prevent the dying-back loss of motor neurons (Kim, et al., 2022). Recently, we also reported defective presynaptic maturation and maintenance in mice with selective Smn downregulation in FAPs, again suggesting the role of FAPs in postnatal NMJ development (Hann, et al., 2023). In adult mice, BMP3B secretion by FAPs stabilize NMJs and Schwann cells by promoting the myelination program in Schwann cells, thereby directly contributing to the maintenance of neural components within skeletal muscle (Uezumi, et al., 2021). In the absence of FAPs or Bmp3b, mice exhibit muscle weakness and myofiber atrophy along with destabilization of Schwann cells and denervation at NMJs, which closely resemble the phenotypes observed in age-related sarcopenia (Uezumi, et al., 2021). Similarly, conditional deletion of Bap1 in FAPs in adulthood cause denervation at the NMJs and eventually loss of motor neurons, demonstrating the requirement of FAPs in maintaining the neuromuscular system (Kim, et al., 2022). Conversely, disturbance in the neural component can influence the behavior of FAPs; for instance, denervation is known to activate FAPs (Contreras, et al., 2016; Gonzalez, et al., 2017; Madaro, et al., 2018). This suggests that FAPs can somehow sense the anatomically distant peripheral nerve injury. However, the question of how FAPs are able to sense the distant peripheral nerve injury remains unanswered (Theret, et al., 2021). Furthermore, whether FAPs are actually able to exert beneficial effects on peripheral nerve regeneration remains elusive.

In accordance with its various functions, heterogeneity within FAPs began to be recognized with the advent of single-cell analysis technology (Contreras, et al., 2021). By profiling the expression levels of 87-selected genes in isolated singlets of FAPs, dynamic transitions between heterogeneous subpopulations of FAPs, identified by different expression levels of TIE2 and VCAM1, was observed during postnatal and regenerative myogenesis (Malecova, et al., 2018). The report showed that while activation of TIE2^high FAPs is observed in neonatal mice, activation of VCAM1⁺ FAPs is observed in injured muscles, suggesting distinct functional involvement of FAPs in the two different contexts of myogenesis (Malecova, et al., 2018). Additionally, single-cell RNA-sequencing (scRNA-seq) enabled identification of heterogeneity within FAPs based on the genome-wide transcriptome data (Lieberman, et al., 2021). In homeostatic adult muscle, two distinct subpopulations within FAPs have been identified, namely Dpp4⁺ and Cxcl14⁺ FAPs (Scott, et al., 2019; Oprescu, et al., 2020). Functionally, we reported that DPP4⁺ FAPs contribute to maturation and maintenance of the neuromuscular system via the BAP1/SMN axis (Kim, et al., 2022). In juvenile muscle, five different subpopulations within FAPs were characterized, each having different contexts of activation and differentiation potentials (Leinroth, et al., 2022). Osr1⁺ FAPs are precursor cells that can form all other subpopulations; Clu⁺ FAPs are most potent in mineralization; Adam12⁺ and Gap43⁺ FAPs are immune-responsive; and Hsd11b1⁺ FAPs respond to nerve transection (Leinroth, et al., 2022). The different activation cues and differentiation potentials in each subpopulation of FAPs suggest distinct roles those subsets can play in different contexts of skeletal muscle biology. Indeed, dynamic transcriptomic changes in FAP subpopulations in response to muscle injury (Scott, et al., 2019; De Micheli, et al., 2020; Oprescu, et al., 2020) or denervation (Nicoletti, et al., 2020; Proietti, et al., 2021; Lin, et al., 2022; Nicoletti, et al., 2023) have been identified in studies that implemented scRNA-seq analysis. Still, the question of how FAPs may sense the distant nerve injury and whether FAPs can beneficially contribute to nerve regeneration remain largely unknown.

Here, using the scRNA-seq approach, we aimed to identify the response mechanism of FAPs to nerve injury, by uncovering its nerve injury-sensing mechanism and its potentially beneficial effect on nerve regeneration. To obtain a comprehensive scRNA-seq database of FAPs’ response to nerve injury, FAPs from both chronic, non-regenerating nerve injury (denervation)- and acute, regeneration-prone nerve injury (crush)-affected muscles were collected at different time points over the course of regeneration. As a result, distinct transcriptomic profiles of FAPs at different time points post the two types of nerve injuries were captured in single-cell resolution, from which the response mechanism of FAPs to nerve injury was identified and validated using mouse models. Specifically, we found that upon peripheral nerve injury, GDNF from Schwann cells can activate FAPs, which in turn express BDNF to promote remyelination during nerve regeneration. Our study suggests FAPs as an important player that actively participates in the nerve regeneration process, of which we believe should be considered in future studies aiming for improved understanding of the peripheral nerve regeneration process.

Results

Distinct response profiles of FAPs upon nerve crush injury versus denervation

To establish a comprehensive transcriptome database of nerve injury-affected FAPs at single-cell resolution, we performed scRNA-seq using FAPs isolated from sciatic nerve crush injury (SNC)- or denervation (DEN)-affected muscles at different time points over the course of regeneration (Figure 1—figure supplement 1A and B). Four time points (3, 7, 14, and 28 days post injury, hereafter dpi) along the regeneration process were chosen for analysis to capture the transcriptomes of FAPs at both early and late stages of regeneration. Selection of such time points was based on a previous report that showed the reinnervation process of the tibialis anterior (TA) muscle after SNC, where Wallerian degeneration was evident at 7 dpi, and reinnervation was mostly completed at 28 dpi (Magill, et al., 2007). Including the uninjured control, scRNA-seq data from a total of nine samples (Uninjured, SNC-3dpi, SNC-7dpi, SNC-14dpi, SNC-28dpi, DEN-3dpi, DEN-7dpi, DEN-14dpi, and DEN-28dpi) were obtained via the 10x Genomics platform (Figure 1—figure supplement 1A). Quality control and filtering of the sequenced cells yielded a total of 44,597 cells for further analysis, where 4955.2±1022.8 cells were captured from each sample. Prior to downstream analysis, integration (Hao, et al., 2021) of our scRNA-seq data with a publicly available scRNA-seq data of mononuclear cells from denervated muscles at 0, 2, 5, and 15 dpi (Nicoletti, et al., 2023) was carried out for data validation. Expectedly, most (97.7%) of the filtered cells in our scRNA-seq data clustered with the denervation-affected FAPs in the published scRNA-seq data, confirming the validity of the data produced in our study (Figure 1—figure supplement 1C and D).

Figure 1.

To look into the chronological transcriptomic changes that occur in FAPs in response to SNC or DEN on a global level, we analyzed our scRNA-seq data by samples. Visualization of the scRNA-seq data on uniform manifold approximation and projection (UMAP) plots showed similar changes in the early stages of regeneration (3 and 7 dpi) compared to uninjured FAPs, regardless of the type of injury (Figure 1A). However, as FAPs reached later stages of regeneration (14 and 28 dpi), SNC-affected FAPs returned to states similar to uninjured control, while DEN-affected FAPs stayed in the activated state (Figure 1A). The similarities and differences observed on the UMAP plots could also be found in the differentially expressed gene (DEG) analyses as well as in the hierarchical clustering analysis using those DEGs (Figure 1B–E, Figure 1—figure supplement 2A–C). As a result of pairwise DEG analyses comparing all nine samples, different numbers of DEGs were identified, of which correlated with the similarities between samples observed on UMAP plots (Figure 1A–E, Figure 1—figure supplement 2A and B). Hierarchical clustering of the nine samples using all unique DEGs identified from the pairwise comparisons showed clustering of uninjured control with SNC-14dpi and SNC-28dpi, suggesting FAPs’ return to homeostatic state (Figure 1—figure supplement 2C). On the other hand, DEN-14dpi and DEN-28dpi clustered with each other, but not with uninjured control, suggesting the chronic activation of FAPs in response to DEN as reported previously (Madaro, et al., 2018) (Figure 1—figure supplement 2C). Samples that captured the early responses of FAPs to nerve injuries (3 or 7 dpi) were clustered together by dpi rather than the type of injury, suggesting similar response profiles of FAPs to both types of nerve injuries (Figure 1—figure supplement 2C). Indeed, the numbers of DEGs between SNC-3dpi versus DEN-3dpi and SNC-7dpi versus DEN-7dpi were among the lowest identified in the pairwise analyses (Figure 1B and C, Figure 1—figure supplement 2A). Overall, the number of DEGs between SNC- and DEN-affected FAPs increased significantly with dpi, showing the bifurcation of FAP’s response to the different types of nerve injuries in the later stages of regeneration (Figure 1B–E).

In a previous study, chronic activation of the STAT3/IL-6 pathway in FAPs in response to DEN was reported (Madaro, et al., 2018). Indeed, Il6 was identified as one of the genes upregulated at all four time points in response to DEN compared to uninjured control in our data (Figure 1—figure supplement 2D and E). Although it did not pass the fold change (FC) threshold (FC >= 2) in the DEG analyses, expression of Stat3 also showed significant chronic upregulation in all DEN-affected FAPs as well (Figure 1—figure supplement 2F and G). In response to SNC, however, only transient upregulation of both genes was observed (Figure 1—figure supplement 2D–G).

To obtain biological insights on the different responses of FAPs to SNC versus DEN in the later stage of regeneration, we subjected the two lists of genes from Figure 1E (28 dpi, SNC_UP and DEN_UP) to gene set overrepresentation analysis (ORA) using g:Profiler (Kolberg, et al., 2023). From both sets of genes, pathways related to tissue regeneration/wound healing were enriched (Figure 1F and G, Figure 1—figure supplement 3A and B). In contrast, pathways related to immune cell recruitment, inflammation, and ECM regulation by collagen biosynthesis were enriched specifically in DEN-28dpi (Figure 1F, Figure 1—figure supplement 3A), which is consistent with the previous report that showed the direct contribution of FAPs to fibrosis in denervated muscles (Contreras, et al., 2016; Madaro, et al., 2018). Also, mild immune cell infiltration into affected muscles in response to DEN was previously described (Lin, et al., 2022; Nicoletti, et al., 2023); our results suggest the role of FAPs in chemotactic recruitment of immune cells in denervated muscles. In addition, the pathway ‘negative regulation of neuron apoptosis’ was enriched in DEN-28dpi, suggesting a prolonged attempt of FAPs to preserve neurons that must be alive for reinnervation of the denervated muscle (Figure 1F, Figure 1—figure supplement 3A). On the other hand, some pathways were exclusively enriched in SNC-28dpi, such as negative regulation of chemotaxis and prostaglandin metabolism (Figure 1G, Figure 1—figure supplement 3B). It is generally understood that after a successful tissue regeneration process, resolution of immune response allow for the tissue to return to homeostasis (Ortega-Gomez, et al., 2013; Aurora and Olson, 2014; Julier, et al., 2017); our results suggest the role of FAPs in regulating immune resolution near the end of nerve regeneration. Furthermore, the role of prostaglandin in peripheral nerve regeneration has recently been described (Forese, et al., 2020; Bakooshli, et al., 2023); our ORA results suggested that FAPs may also be involved in regulation of prostaglandin levels during peripheral nerve regeneration.

To discover biological pathways behind the supposedly similar responses of FAPs to both SNC and DEN in the early phases of regeneration, we examined DEGs from comparing SNC-3dpi, SNC-7dpi, DEN-3dpi, and DEN-7dpi to uninjured controls. Many of the upregulated genes identified in the DEG analyses were shared amongst the four samples (Figure 1—figure supplement 3C). ORA using the shared upregulated genes revealed enrichment in pathways that were also enriched in DEN-28dpi compared to SNC-28dpi, such as immune cell recruitment and ECM regulation, supporting the idea that early-activated states within FAPs persist for a prolonged period in response to DEN (Figure 1F and H, Figure 1—figure supplement 3A and D). Collectively, analysis of our scRNA-seq data by samples revealed similar response profiles of FAPs to both SNC and DEN in the early stages of regeneration, which then bifurcated into chronic activation in response to DEN and return to homeostasis in response to SNC, showing correlative behaviors along with the degree of nerve regeneration and target muscle reinnervation.

Nerve injury-responsive subsets within FAPs

Although analysis of our scRNA-seq data on the population level provided general insights on how FAPs may respond to the different types of nerve injuries, the results could not provide us with sufficient clues on how FAPs can sense nerve injuries, or how they may directly contribute to nerve regeneration. Thus, we next analyzed our scRNA-seq data on the subpopulation level, hoping to distinguish subsets within FAPs that may be more relevant to the context of sensing and responding to nerve injury. To identify distinct subsets within nerve injury-affected FAPs, we applied unsupervised clustering to the merged scRNA-seq data of all nine samples following the Seurat-R workflow (Hao, et al., 2021). As a result, seven clusters with unique gene expression profiles were identified from the nerve injury-affected FAPs, with marker genes specifically expressed in each cluster (Figure 2A–C, Figure 2—figure supplement 1A).

Figure 2.

Interestingly, while clusters 4-7 showed little or no significant change in their proportions in response to nerve injury, clusters 1-3 exhibited dramatic changes upon nerve injury (Figure 2D and E). In particular, cluster 1 was mostly present in uninjured muscles or in muscles where reinnervation had occurred to at least some degree (SNC-14dpi and SNC-28dpi) (Magill, et al., 2007) (Figure 2D and E). In contrast, presence of clusters 2 and 3 were mutually exclusive to cluster 1, such that their appearances were transient in response to SNC and chronic upon DEN (Figure 2D and E). Based on this mutual exclusivity of the three clusters, we speculated that cluster 1 can sense and respond to nerve injuries, and that clusters 2 and 3 may have arisen from cluster 1 upon nerve injury.

To obtain clues on whether such changes between FAP clusters could have actually occurred in response to nerve injury, we first performed RNA velocity analysis using R package velocyto.R (La Manno, et al., 2018). RNA velocities on the UMAP plots predicted transcriptomic flow from cluster 1 to clusters 2 and 3 in the early stages of regeneration in both SNC- and DEN-affected FAPs, which was in line with our speculation (Figure 2—figure supplement 1B). Conversely, transcriptomic flow from clusters 2 and 3 back to cluster 1 was evident in SNC-affected FAPs in the later stages of regeneration (Figure 2—figure supplement 1B). However, RNA velocities in DEN-affected FAPs were represented as dots instead of arrows on clusters 2 and 3 in the later stages, suggesting an unchanging, chronic state of their transcriptomes, which is consistent with the chronic activation of FAPs in response to DEN (Madaro, et al., 2018) (Figure 2—figure supplement 1B). Additionally, hierarchical clustering of the seven FAP clusters using DEGs enriched in each cluster grouped clusters 1-3 together, supporting our speculation that clusters 2 and 3 originate from cluster 1 (Figure 2—figure supplement 1C). Recently, Hsd11b1-expressing FAPs were identified as the FAP subset that is specifically activated in response to nerve transection injury (Leinroth, et al., 2022). Since we speculated that cluster 1 in our scRNA-seq data can sense and respond to nerve injury, we examined the expressions of marker genes identified in the previous report – Hsd11b1, Mme, Ret, and Gfra1 (Leinroth, et al., 2022) – in our scRNA-seq data. Indeed, all four markers were enriched in cluster 1 in our data (Figures 3A and B, Figure 2—figure supplement 1D–F). Expressions of marker genes Hsd11b1 and Mme were also enriched in clusters 2 and 3, further supporting the idea that those clusters may have arisen from cluster 1 (Figure 2—figure supplement 1D–F). Together, these data suggest that clusters 1-3 are the dynamic interchanging subsets of FAPs that specifically respond to nerve injury, where cluster 1 senses nerve injuries in unperturbed muscles and clusters 2 and 3 arise from cluster 1 to respond to nerve injury.

Figure 3.

GDNF signaling pathway in the nerve injury-sensing mechanism by FAPs

Among the four marker genes expressed in cluster 1, Ret and Gfra1 are well-known as GDNF receptors, where Gfra1 directly binds GDNF, which in turn activates the receptor tyrosine kinase (RTK) Ret for downstream signal transduction (Jing, et al., 1996; Treanor, et al., 1996; Trupp, et al., 1996). Meanwhile, robust but specific expression of GDNF by Schwann cells in response to peripheral nerve injury has been reported (Hammarberg, et al., 1996; Hoke, et al., 2002; Arthur-Farraj, et al., 2012; Xu, et al., 2013; Proietti, et al., 2021). Accordingly, we presumed that FAPs, especially the Ret- and Gfra1-expressing cluster 1 cells, may sense the distant nerve injury by detecting GDNF secreted from Schwann cells. Notably, both Ret and Gfra1 were among the top 10 DEGs specifically enriched in cluster 1, suggesting that they can readily respond to GDNF (Figure 2C, Figure 3—figure supplement 1A). In addition, comparing the expression levels of Ret and Gfra1 in skeletal muscle-resident mononuclear cell populations isolated by fluorescence-activated cell sorting (FACS) revealed robust co-expression of both genes in FAPs, but not as much in other cell populations (Figure 3C and D). Thus, FAPs may be the main cell type within skeletal muscle that can respond to GDNF secreted by Schwann cells in case of a nerve injury.

To further investigate the relevance of GDNF signaling in the nerve injury-sensing mechanism by FAPs, we subjected lists of DEGs enriched in clusters 1-3 to ORA. Genes enriched in cluster 1 returned pathways ‘GDNF receptor signaling pathway’, ‘regulation of cellular response to growth factor stimulus’, and ‘positive regulation of peptidyl-tyrosine phosphorylation’, where tyrosine residues on the RTK Ret is known to be phosphorylated upon activation (Jing, et al., 1996; Treanor, et al., 1996; Trupp, et al., 1996) (Figure 3—figure supplement 1B). The two latter pathways were also found in the ORA results using DEGs enriched in clusters 2 and 3, supporting the idea that those clusters originate from cluster 1 upon nerve injury (Figure 3E, Figure 3—figure supplement 2A and B). The pathway ‘positive regulation of ERK1/2 cascade’ was enriched in cluster 2, suggesting the involvement of GDNF-Ret-Ras-ERK signaling cascade within the MAPK signaling pathway in this cluster (Airaksinen and Saarma, 2002; Sariola and Saarma, 2003; Kanehisa, et al., 2023) (Figure 3G, Figure 3—figure supplement 1C and 2A). In addition to ORA, we predicted upstream transcription factors (TFs) that could have regulated the expressions of the DEGs enriched in the two activated FAP subsets, clusters 2 and 3, using TRRUST (Han, et al., 2018). As a result, TFs Fos, Jun and NF-κB (Nfkb1, Nfkb2) were predicted from both lists of DEGs, all of which are known to act downstream of the GDNF/Ret-induced MAPK signaling pathway (Fielder, et al., 2018; Kanehisa, et al., 2023) (Figure 3F and G, Figure 3—figure supplement 1C, Supplemental tables 1 and 2). Collectively, robust and specific co-expression of GDNF receptors in cluster 1, together with the prediction of RTK activation and involvement of GDNF signaling pathway downstream TFs in clusters 2 and 3, suggests that GDNF signaling could be the mechanism by which FAPs sense the distant nerve injury, where local Schwann cells act as the GDNF source upon nerve injury.

Bdnf expression in FAPs by both endogenous and exogenous GDNF

Next, to discover how FAPs may contribute to nerve regeneration, we screened the list of genes enriched in clusters 2 and 3 that were predicted to be downstream of the GDNF signaling pathway to identify candidate effector genes. From the TRRUST analysis results, 44 genes were identified to be regulated by either Fos, Jun, or NF-κB (Figure 4A, Supplemental tables 1–3). Since FAPs themselves do not constitute the neural components within skeletal muscle, we reasoned that secreted factors from FAPs would most likely exert beneficial effect on the regenerating nerves. Also, considering the effector gene’s potential function in supporting nerve regeneration, we presumed that it regulates neurons or glial cells. Moreover, we anticipated that expression of the effector gene would be limited to the context of nerve injury and regeneration, since the vast majority of FAPs are in a quiescent state in unperturbed adult musclesScott, et al., 2019. Thus, we applied the following criteria to narrow down our candidate gene list: (1) genes that are known to code secreted proteins, (2) genes that are known to regulate neurons or glial cells, and (3) genes that are expressed exclusively in activated FAPs in response to nerve injury (Figure 4A, Supplemental table 3). Unexpectedly, after filtering out genes that did not fit the three criteria, only Bdnf remained in our candidate gene list that could act as the effector secreted by FAPs upon nerve injury to support nerve regeneration (Figure 4A). Expression patterns of Bdnf in FAPs upon nerve injury showed transient upregulation in SNC-affected FAPs, whereas chronic expression of Bdnf was observed in DEN-affected FAPs, showing correlation with its potential requirement during nerve regeneration (Figure 4C). The expression of Bdnf was mostly limited to cluster 2 (Figure 4B), where pathway analysis and TF prediction suggested the involvement of GDNF-Ret-Ras-ERK-Fos signaling cascade in this subset of FAPs (Figure 3F and G, Figure 3—figure supplement 1C and 2A). Accordingly, we hypothesized that FAPs secrete BDNF in response to GDNF from Schwann cells upon nerve injury, to actively take part in the regeneration process.

Figure 4.

To validate our hypothesis in vivo, we first examined the expression profiles of Gdnf in Schwann cells and Bdnf in FAPs at early time points in response to SNC, using Plp1^CreER; Rosa^tdTomato mice to specifically label and hence isolate Schwann cells (Doerflinger, et al., 2003) (Figure 4D). Expectedly, we could observe sequential upregulation of Gdnf and Bdnf from Schwann cells and FAPs, respectively, where Gdnf levels peaked at 1 dpi in Schwann cells, followed by a gradual increase of Bdnf expression in FAPs, which peaked at 3 dpi (Figures 4D and E). In addition, to validate the sufficiency of GDNF signaling in inducing Bdnf expression in FAPs in vivo, we injected recombinant mouse GDNF protein into the TA and the two gastrocnemius (GA) muscles (lateral and medial GA), from which FAPs were FACS-isolated 48 hours post injection to investigate the expression of Bdnf (Figure 4F). Compared to PBS control, intramuscular injection of GDNF sufficiently induced Bdnf expression in FAPs, even in the absence of a nerve injury (Figure 4F and G). Together, we suggest that FAPs can respond to nerve injury via the GDNF-BDNF axis, since recombinant GDNF protein could sufficiently induce Bdnf expression in FAPs without nerve injury, and sequential upregulation of Gdnf and Bdnf could be observed from Schwann cells and FAPs following nerve injury, respectively.

Remyelination by FAP-derived BDNF during peripheral nerve regeneration

Although BDNF is known to function in processes such as axon elongation (Oudega and Hagg, 1999; English, et al., 2013), survival of neurons (Ghosh, et al., 1994; Baydyuk and Xu, 2014), and myelination by Schwann cells (Zhang, et al., 2000; Xiao, et al., 2009), the role of BDNF secreted by FAPs in nerve regeneration is unknown. To find out how FAP-derived BDNF can contribute to nerve regeneration, we produced conditional knockout (cKO) mice where Bdnf is specifically inactivated in mesenchymal progenitors including FAPs, by crossing Prrx1^Cre mice (Logan, et al., 2002; Kim, et al., 2022; Leinroth, et al., 2022; Hann, et al., 2023) with Bdnf-floxed (Bdnf^fl) mice (Prrx1^Cre; Bdnf^fl/fl, hereafter cKO) (Figure 5—figure supplement 1A). Inactivation of Bdnf in FAPs in the cKO mice was confirmed on both genomic DNA and mRNA levels (Figure 5—figure supplement 1B and C). Though Cre expression in Prrx1-expressing cells occur from embryonic day 9.5 (Logan, et al., 2002), no visible phenotypes were observed in the postnatal, juvenile and adult cKO mice compared to littermate controls (hereafter, Ctrl). However, upon SNC in the right hindlimb, cKO mice displayed a delay in nerve regeneration compared to Ctrl, measured by compound muscle action potential (CMAP) amplitude and latency via electromyography (EMG) on the GA muscle (Figure 5A–5D, Figure 5—figure supplement 1D and E). At 4 weeks post injury (wpi), nerve regeneration in both Ctrl and cKO mice showed insufficient recovery in the right, injured GA compared to the left, uninjured GA, where lower amplitude and prolonged latency in CMAP was observed (Figure 5B–D). However, at 6 wpi, while CMAP amplitude and latency in the left and right GAs became comparable in Ctrl mice, recovery of such values were stalled at levels comparable to 4 wpi in the cKO mice (Figure 5B–D). By 12 wpi, electrophysiological functions of the injured nerves became statistically comparable to that of its contralateral counterpart in the cKO mice, indicating a delayed regeneration in the cKO mice (Figure 5B–D).

Figure 5.

Generally, decrease in CMAP amplitude and prolonged CMAP latency can be explained by two main causes: axonal loss and defective myelination (Mallik and Weir, 2005; Chung, et al., 2014). Since complete regeneration of the injured nerves on the electrophysiological level could be achieved after a sufficient period of time in the cKO mice (Figure 5B–D), we presumed that axonal loss would not have occurred, since it would result in permanent defects by loss of motor units. Instead, we thought that defective myelination could have occurred in the cKO mice, considering the fact that BDNF is already known to promote remyelination during peripheral nerve regeneration (Zhang, et al., 2000; Zheng, et al., 2016), and that defective myelination alone can affect both CMAP amplitude and latency (Mallik and Weir, 2005). Thus, we investigated the effect of conditional Bdnf inactivation in FAPs on regenerative myelination by examining the sciatic nerves from Ctrl versus cKO mice at 6 wpi, when the delayed functional recovery of the injured nerves in the cKO mice was prominent (Figure 5A–D). Toluidine blue staining of the semi-thin sections of injured sciatic nerves revealed significantly reduced myelin thickness in the cKO mice compared to Ctrl mice (Figure 5E). Indeed, higher G-ratio values were calculated from cKO mice compared to Ctrl, confirming the reduced myelination in the regenerating nerves in cKO mice (Figure 5F, Figure 5—figure supplement 1F and G). This decrease in myelin thickness was independent from axon diameter, which were comparable in both Ctrl and cKO mice, implying that no axonal loss or defect had occurred in the cKO mice compared to controls (Figure 5G). Taken together, our results revealed the direct involvement of FAP-derived BDNF in the remyelination process during peripheral nerve regeneration, such that inadequate levels of Bdnf expression in FAPs caused delayed remyelination and hence delayed nerve regeneration in the cKO mice.

Implication of FAP-derived BDNF in the age-related delay in nerve regeneration

Finally, to seek clinical relevance of Bdnf expression in FAPs during nerve regeneration, we compared the expression levels of Bdnf in adult (5-6 months) versus aged (24 months) mice in FAPs post SNC. At 7 dpi, Bdnf expression was significantly reduced in the aged mice compared to adult mice (Figure 6A and B). Such a difference could be one of the factors that lead to the delayed or failed regeneration of injured nerves in the elderly compared to healthy, young adults, suggesting the clinical role of FAPs in a timely, successful peripheral nerve regeneration process.

Figure 6.

Discussion

Traumatic injury to the peripheral nerve has severe consequences, including lifelong paralysis of the injured limb that can compromise the quality of life significantly (Grinsell and Keating, 2014). Thus, understanding the regeneration process of the peripheral nerves is fundamental for treating the potentially devastating injury. Previously, several cellular components in and outside the injured nerve were discovered to actively participate in the regeneration process, including the injured neurons (Hanz, et al., 2003), glial cells (Arthur-Farraj, et al., 2012), immune cells (Mueller, et al., 2001; Kalinski, et al., 2020), and nerve-resident mesenchymal cells (Parrinello, et al., 2010; Toma, et al., 2020), via diverse mechanisms so that the nerve can regain its function (Scheib and Hoke, 2013). In this study, we investigated the response mechanism of muscle-resident FAPs to both acute and chronic peripheral nerve injury via scRNA-seq, and revealed that this population of cells can also actively take part in the nerve regeneration process. Here, we discovered that muscle-resident FAPs can recognize the distant nerve injury by sensing GDNF secreted by Schwann cells. Though GDNF secretion by Schwann cells in response to nerve injury had previously been recognized (Hammarberg, et al., 1996; Hoke, et al., 2002; Arthur-Farraj, et al., 2012; Xu, et al., 2013; Proietti, et al., 2021), we identified FAPs as a major target cell population of GDNF within skeletal muscle, based on the enriched expression of GDNF receptors. In depth, exploiting the technical advantage of scRNA-seq, we suggested that a subset of FAPs, named cluster 1 in this study, can sense the local GDNF by expressing GDNF receptors Ret and Gfra1, and that upon GDNF sensing, cluster 1 FAPs turn into clusters 2 and 3 to contribute to the nerve regeneration process. Specifically, we discovered that FAPs, especially the F2rl1-expressing cluster 2, express Bdnf in response to nerve injury and/or GDNF, which in turn was shown to promote the remyelination process by Schwann cells during nerve regeneration, using our cKO mouse model (Figure 6C). Since epineurial and perineurial, but not endoneurial mesenchymal cells share their origins with the muscle-resident FAPs and therefore are Prrx1-positive (Joseph, et al., 2004; Carr, et al., 2019), the possibility that the delayed remyelination observed in our cKO mice could be due to the combined effect of Bdnf depletion in both muscle- and nerve-resident mesenchymal cells cannot be eliminated. To resolve such issue, further investigations using muscle-resident FAP-specific Cre mouse lines is required, but such line is currently unavailable as no such specific marker has been found. Nevertheless, our findings show that muscle-resident FAP-derived BDNF is indeed important for nerve regeneration, since intramuscular injection of recombinant BDNF can sufficiently accelerate the nerve regeneration process (Zheng, et al., 2016). Conversely, intramuscular injection of BDNF-neutralizing antibodies can sufficiently delay nerve regeneration (Zheng, et al., 2016). Thus, endogenous supply of intramuscular BDNF by the muscle-resident FAPs in our Ctrl mice would likely have supported remyelination by Schwann cells, while the lack of such BDNF supply by FAPs in our cKO mice would have resulted in delayed remyelination. Also, we found that while muscle-resident FAPs robustly express both GDNF receptor genes, neither epineurial nor perineurial mesenchymal cells express significant levels of Ret and Gfra1 (Supplemental figure 1), implying that the GDNF-BDNF axis found in this study could be valid uniquely in muscle-resident FAPs. Collectively, we suggest that muscle-resident mesenchymal progenitors can directly contribute to nerve regeneration via the GDNF-BDNF axis (Figure 6C), of which was previously unidentified; we suggest that in future studies regarding peripheral nerve regeneration, active participation of this intramuscular mesenchymal population should be taken into consideration. In that process, our scRNA-seq data may provide valuable insights, which may lead to additional discoveries on FAP’s contributions to nerve regeneration by other mechanisms, and/or provide clues on the cell-to-cell communications of FAPs with other cell types that may lead to facilitation of the regeneration process.

Here, we have primarily investigated the roles of Ret-expressing and F2rl1-expressing FAPs in sensing and responding to nerve injury, respectively. However, possibilities that other subpopulations can exert distinct beneficial effects on nerve regeneration remain largely unexplored. For example, although identified as a nerve injury-relevant subpopulation in this study, specific contributions of the Alkal2-expressing cluster 3 FAPs to nerve regeneration are yet to be discovered. Identification of effector genes such as Bdnf from this subpopulation may lead to the discovery of an additional mechanism by which FAPs can contribute to nerve regeneration. Although other subpopulations (clusters 4-7) did not exhibit dramatic fluctuations in population percentage, some of the genes expressed specifically in those subpopulations showed patterns that followed the regeneration process (Supplemental figure 2). Such gene expression patterns suggest subpopulation-specific functions even in the less dynamic subpopulations within FAPs in peripheral nerve regeneration, and require further investigations to reveal their possible contributions.

The role of BDNF in peripheral nerve regeneration has been identified previously, where it was found to promote both intrinsic axonal regeneration in neurons as well as remyelination by Schwann cells (Zhang, et al., 2000; Zheng, et al., 2016). In such circumstances, various cellular sources of BDNF have been identified. Bone marrow transplantation of wild-type cells into Bdnf heterozygotic knockout mice revealed the involvement of bone marrow-derived cells in expressing Bdnf that can promote nerve regeneration in the sciatic nerve (Takemura, et al., 2012). Schwann cells themselves are cellular sources of BDNF during nerve regeneration (Wilhelm, et al., 2012). In this study, we showed that BDNF from FAPs can also promote myelination of the regenerating axons post injury, suggesting FAPs as an additional cellular source of BDNF in peripheral nerve regeneration. The existence of cellular sources of BDNF other than FAPs such as Schwann cells would provide an explanation for the delayed, but not failed, remyelination in our Bdnf cKO mice, where complete regeneration had occurred after a sufficient amount of time, despite the lack of Bdnf expression in FAPs. Still, ablation of Bdnf in FAPs displayed significant delays in the remyelination process during nerve regeneration, suggesting their requirement in the timely regeneration process of injured nerves. Meanwhile, scRNA-seq data of all mononuclear cells from denervated muscles (Nicoletti, et al., 2023) suggested expression of Bdnf in tenocytes and pericytes in addition to Schwann cells and FAPs (Supplemental figure 3A and B). Although this may imply the involvement of those cellular components in providing BDNF for nerve regeneration, expression of Bdnf in such cells were not altered as significantly in response to nerve injury as in FAPs or Schwann cells (Supplemental figure 3C). Moreover, while Ret-expressing FAPs and Schwann cells are known to be in proximity to the NMJ or relevant nerve regeneration sites (Leinroth, et al., 2022), such locational enrichment of tenocytes and pericytes have not been reported so far. Thus, it is likely that FAPs, together with Schwann cells, are the main sources of BDNF within skeletal muscle that can act in the remyelination process during peripheral nerve regeneration.

Aging is one of the well-known factors that can slow down the nerve regeneration process (Verdú, et al., 2008; Maita, et al., 2023). Since multiple cell types are known to participate in this process (Scheib and Hoke, 2013), determination of the cell types that can cause age-related delays in nerve regeneration is important for the development of therapeutic approaches targeting the relevant cell types. Of note, previous research emphasized the importance of niche factors, rather than the intrinsic regenerative capacity of the injured neurons, in the age-related decline in nerve regeneration (Painter, et al., 2014). Specifically, inability of Schwann cells to adopt repair cell phenotypes has been pointed out as one of the age-related changes (Painter, et al., 2014; Wagstaff, et al., 2021). In addition, age-related changes in immune cells, especially macrophages, were suggested as causal factors that delay nerve regeneration (Buttner, et al., 2018). In particular, chronic inflammatory phenotypes were shown to interfere with the remyelination process by Schwann cells (Buttner, et al., 2018), and failed macrophage infiltration in the early stages of regeneration resulted in defective Wallerian degeneration and myelin debris clearing (Scheib and Hoke, 2016). In addition to the involvement of Schwann cells and immune cells, we have identified muscle-resident FAPs as an additional cellular component that can contribute to nerve regeneration, by promoting remyelination via BDNF secretion. Surprisingly, expression of Bdnf by FAPs was significantly reduced in aged mice compared to adult mice, suggesting the clinical relevance of FAP’s involvement in the age-related delay in peripheral nerve regeneration. We believe that further studies on age-related changes in FAPs may provide valuable clues to understanding clinical observations from aged individuals, which can lead to development of additional therapeutic strategies that include FAPs as target cells in treating both young and aged nerve injury patients.

Materials and Methods

Animals

C57BL/6J (RRID:IMSR_JAX:000664), Prrx1^Cre (RRID:IMSR_JAX:005584), Bdnf^fl (RRID:IMSR_JAX:004339), Plp1^CreER (RRID:IMSR_JAX:005975) and Rosa^tdTomato (RRID:IMSR_JAX:007914) mice were all obtained from The Jackson Laboratory. To generate Prrx^Cre; Bdnf^fl/flmice, Prrx^Cre/+; Bdnf^fl/+ males and Bdnf^fl/flfemales were crossed to avoid germline recombination in the female reproductive cells, and littermates were used as controls. To generate Plp1^CreER; Rosa^tdTomato mice, Plp1^CreER/+ mice were crossed with Rosa^tdTomato/+ mice, and the line was kept by breeding Plp1^CreER/+; Rosa^{tdTomato/tdTomato} males and females. Only mice that had the Plp1^CreER allele was used. Primers used for genotyping are listed in Supplemental table 4. All mice were bred on the B6 background, except for the Prrx^Cre; Bdnf^fl/fl mice and littermates that were kept in the mixed B6, 129S4 and BALB/c background. All mice were housed in a specific-pathogen-free (SPF) animal facility, with 12-h light/12-h dark cycle at room temperature (RT, 22℃) and 40∼60% humidity, and were fed with normal chow diet and water ad libitum. Tamoxifen administration, sciatic nerve crush injury, denervation and intramuscular GDNF injections were all given to 3 to 4-month-old adult mice. When comparing aged mice versus adult mice, 24-month-old and 5 to 6-month-old mice were used, respectively. In all cases except for Plp1^CreER; Rosa^tdTomato mice, male mice were used for the experiments. No sex-specific differences were observed in experiments using Plp1^CreER; Rosa^tdTomato mice. All experimental procedures were approved by the Institutional Animal Care and Use Committtee at Seoul National University and were carried out according to the guidelines provided.

Tamoxifen administration

To label Schwann cells, 3-month-old Plp1^CreER; Rosa^tdTomato mice were administered orally with tamoxifen (20 mg/ml in corn oil, 160 mg/kg body weight; Sigma-Aldrich) 3 times every other day.

Sciatic nerve injury

Mice were deeply anesthesized via intraperitoneal injection of Avertin (32 mg/ml, ∼800 mg/kg; Sigma-Aldrich), and the incision site on the posterior side of the right hindlimb was shaved and depilated using surgical clippers and hair removal cream. After cleansing the incision site with 70% ethanol, incision was made on the skin with surgical scissors and the biceps femoris muscle was punctured open with fine-tip forceps to expose the sciatic nerve. For sciatic nerve crush injury, the exposed nerve was crushed with fine forceps for 30 seconds at the site just proximal to where the tibial, peroneal and sural nerves branched out from the sciatic nerve. For denervation, ∼5 mm of the sciatic nerve proximal from the crush injury site was cut and removed. The punctured biceps femoris muscle and skin were then sutured, and the incision site was sterilized with povidion-iodine.

Intramuscular injection of GDNF

Tibialis anterior muscle and the two gastrocnemius muscles (GA, lateral and medial) were each injected with 10 μl of either PBS or recombinant mouse GDNF (10 μg/ml, Sigma-Aldrich) using 31-gauge insulin syringes without damaging any innervating nerves. The injected muscles were then dissected 48 hours post injection for isolation of FAPs and further analysis.

Isolation of FAPs, Schwann cells and others

Isolation of muscle-resident FAPs, Schwann cells and others was performed according to a previously reported protocol (Liu, et al., 2015) with minor modifications. Muscles indicated in each experiment were dissected, finely chopped with surgical scissors and washed with 10% horse serum (Gibco), DMEM (HyClone) for further dissociation. Enzymatic dissociation was carried out in 10% horse serum, DMEM containing collagenase II (800 U/ml, Worthington Biochemical) and dispase II (1.1 U/ml, Gibco) for 40 minutes at 37℃ with mild agitation, and mechanical dissociation was performed by tituration of the dissociated solution with a 20-gauge needle ten times. After filtering the solution through a 40 μm strainer, dissociated mononuclear single cells were stained with the following antibodies: APC anti-mouse CD31, APC anti-mouse CD45, PE anti-mouse TER-119, Biotin anti-mouse CD106 (Vcam1) (Biolegend) and FITC anti-mouse Ly-6A/E (Sca1) (BD Pharmingen); 7-aminoactinomycin D (7-AAD, Sigma-Aldrich) was added to stain dead cells. Gating strategies used for the isolation of each cell type were as follows: FAPs, 7-AAD^-Ter119^-CD31^-CD45^-Vcam1^-Sca1⁺; MuSCs, 7-AAD^-Ter119^-CD31^-CD45^-Vcam1⁺Sca1^-; lineage-positive cells, 7-AAD^-Ter119^-CD31⁺ or 7-AAD^-Ter119^-CD45⁺; double-negative cells, 7-AAD^-Ter119^-CD31^-CD45^-Vcam1^-Sca1^-. For Schwann cells, 7-AAD^-tdTomato⁺ cells were sorted from tamoxifen-administered Plp1^CreER; Rosa^tdTomato mice.

scRNA-seq library construction and sequencing

FAPs were isolated from sciatic nerve crush injury-affected or denervated muscles on days 3, 7, 14 and 28 post injury using wild type B6 mice, so that a total of nine samples, including uninjured control, were collected for library generation. For each sample, isolated FAPs were pooled from two mice. Chromium Next GEM Single Cell 3ʹ Kit v3.1 (10x Genomics) was used according to the manufacturer’s instructions for the nine collected FAP samples, and target cell number for recovery was set to 5,000 in each sample. Sequencing of the libraries was carried out using HiSeq X Ten (Illumina).

Computational analysis of scRNA-seq data

Sequenced reads were aligned to the mouse reference genome mm10 using CellRanger v3.1.0 (10x Genomics), and aligned reads were transformed into gene-cell count matrices using velocyto v0.17 (La Manno, et al., 2018) to obtain count matrices for both spliced and unspliced mRNAs. Output loom files were then loaded with R package SeuratWrappers v0.3.1, and were preprocessed and analyzed using R package Seurat v4.3.0 (Hao, et al., 2021) for downstream analysis. The preprocessing steps for quality control included doublet filtering, live cell filtering and removal of non-FAPs as previously described (Kim, et al., 2022). All nine sample data were merged and normalized for dimensionality reduction, where top 8 principal components from the principal component analysis using 5,000 variable genes were selected for 2-dimensional UMAP embedding and visualization. Unsupervised clustering of cells was achieved through FindNeighbors and FindClusters functions in Seurat R package. To identify DEGs in the pairwise comparisons of scRNA-seq samples, the FindMarkers function in Seurat R package was used with the following parameters: fold change >= 2, pseudocount.use = 0.01, min.pct = 0.01, adjusted p value < 0.05. For identification of DEGs in each cluster, FindAllMarkers function was used with the parameters: fold change >= 1.5, pseudocount.use = 0.01, min.pct = 0.02, adjusted p value < 0.05. For hierarchical clustering, R package pheatmap v1.0.12 was used. For RNA velocity analysis, R package velocyto.R v0.6 was used following the instructions provided by the developer (La Manno, et al., 2018). For the prediction of upstream regulatory transcription factors using lists of DEGs enriched in the selected FAP clusters, the web-based tool TRRUST v2 (Han, et al., 2018) was used. For color mapping of the MAPK signaling pathway, online version of the KEGG Mapper – Color (Kanehisa, et al., 2023) was used.

Gene set overrepresentation analysis

To identify pathways enriched in the lists of DEGs, web-based version of g:Profiler (Kolberg, et al., 2023) was used with the following parameters: organism – Mus musculus; ordered query – YES; data sources – GO biological process without electronic GO annotations, Reactome, and WikiPathways; advanced options were set to default. Visualization of the results were done as previously described (Reimand, et al., 2019), using the Cytoscape v3.10.1 (Shannon, et al., 2003) application with tools EnrichmentMap v3.3.6 (Merico, et al., 2010) and AutoAnnotate v1.4.1 (Kucera, et al., 2016).

RNA extraction and qRT-PCR

Total RNA extraction and reverse transcription was carried out for isolated FAPs, Schwann cells, MuSCs and others using TRIzol™ reagent (Invitrogen) and ReverTra Ace™ qPCR RT Master Mix (Toyobo) reagents respectively, following the manufacturer’s instructions. qPCR was performed using ORA™ SEE qPCR Green ROX L Mix (HighQu) reagent, with gene-specific primers listed in Supplemental table 4. Quantitative analysis of mRNA levels were done using the 2^-ΔΔCt method with β-actin (Actb) as the housekeeping gene for normalization.

Electromyography and CMAP measurement

Intraperitoneal injection of Avertin (32 mg/ml, ∼800 mg/kg) for anesthetization of mice was carried out prior to CMAP measurement. Stimulation of the sciatic nerve was achieved by placing stimulating electrodes subcutaneously on either side of the sciatic notch and applying supramaximal stimuli (∼70 mA) at a rate of 1 pulse per second with the duration of 0.1 ms, using Isolated Pulse Stimulator Model 2100 (A-M Systems). Recording electrode was placed carefully on the GA muscle subdermally without puncturing the muscle, with the reference electrode placed near the Achilles tendon and ground electrode placed on the tail for data recording using Data Recorder IX-RA-834 (iWorx). CMAP amplitude was determined by the absolute difference between potentials of positive and negative peaks, and CMAP latency was determined by the delay from stimulus peak to the beginning of response peak. Three individual measurements were taken from each animal’s GA muscle, and average values were used as representatives for statistical analysis.

Toluidine blue staining of sciatic nerves

Sciatic nerve distal to the injury site was dissected at 6 weeks post injury for analysis. The dissected nerves were fixed in 4% paraformaldehyde dissolved in Sorensen’s phosphate buffer (0.1M, pH 7.2) at 4℃ overnight, followed by procedures described previously (Kim, et al., 2022) with minor modifications for semi-thin sectioning. Briefly, fixed samples were washed with 0.1M sodium cacodylate buffer (pH 7.2), post-fixed with 1% osmium tetroxide in 0.1M sodium cacodylate buffer (pH 7.2) for 1 hour at RT, washed with distilled water (DW) and stained with 0.5% uranyl acetate at 4℃ overnight. Stained samples were then washed with DW and dehydrated using serial ethanol and propylene oxide. Samples were then embedded in Spurr’s resin (Electron Microscopy Sciences), and semi-thin sections (500 nm) were prepared with a diamond knife on an ultramictorome EM UC7 (Leica). The sections were dried down on glass slides for staining and light microscopy. Toluidine blue staining were done using 1% toluidine blue solution containing 1% sodium borate on a slide warmer (70℃), and images were obtained with the light microscope EVOS FL Auto 2 (Thermo Fisher Scientific) for analysis.

G-ratio quantification

Semi-automated quantification of naked and myelinated axon diameters were carried out using an ImageJ plugin for g-ratio quantification (Goebbels, et al., 2010), where axon diameters and G-ratios were quantified and calculated, respectively. G-ratios were calculated as [naked axon diameter]/[myelinated axon diameter].

Quantification and statistical analysis

All statistical analyses were performed using Prism v5.01 (GraphPad) and R v4.2.1. Continuous variables were tested for normal distribution with Shapiro-Wilk test, and F-test was used to check for equal variance. For comparison of significant differences in multiple groups, one-way analysis of variance(ANOVA) followed by Bonferroni’s pairwise post hoc test was applied. For comparison of two groups, unpaired t-test was used for data with normal distribution and equal variance; Welch’s t-test was used for data with normal distribution and unequal variance; and Mann-Whitney U test was used for non-normally distributed data. ANCOVA was applied to test for differences between slopes of linear regression lines. Two-way ANOVA was applied to compare two groups with two variables. For comparison of gene expression levels between scRNA-seq data, Wilcoxon rank sum test was applied as a default in functions FindAllMarkers and FindMarkers within the R package Seurat v4.3.0. For RT-qPCR, the average of triplicate technical values were used for each biological replicate. All error bars represent mean ± SD. P value of less than 0.05 was considered statistically significant at the 95% confidence level. The number of technical and biological replicates and statistical analysis used in each experiment are indicated in the figure legends.

Data availability

Single-cell RNA-sequencing data produced in this study have been deposited at GEO under accession number GSE250436 and are publicly available as of the date of publication.

Acknowledgements

We thank Dr. Jong-Eun Park for his helpful comments on the analysis of our scRNA-seq data. Prrx1^Cre mice, Bdnf^flmice, and Plp1^CreER mice were obtained via Korea Mouse Phenotyping Center, deposited by Dr. Rho Hyun Seong, Dr. Yun-Hee Lee, and Dr. Myunghwan Choi, respectively. Also, we would like to acknowledge technical support from Center for Research Facilities, Biological Sciences Department, Seoul National University (SNU) and SNU National Instrumentation Center for Environmental Management. Library construction and sequencing of the scRNA-seq data was performed by Macrogen Inc.. Experimental schemes and graphical summary in this paper were created using BioRender. This work was supported by grants from the National Research Foundation of Korea (NRF-2020R1A5A1018081, NRF-2022R1A2C3007621) funded by the Korean government (MSIT).