Mice were randomly assigned to four groups (n=5 for each group): Male-CCI, Female-CCI, Male-Sham, and Female-Sham. Bilateral L4-5 DRGs were collected from mice on day 7 after bilateral sciatic CCI or sham surgery for scRNA-seq (Figure 1A, Figure 1—figure supplement 1A). In an animal behavior study conducted on day 6 after CCI, paw withdrawal frequencies to low-force (0.07 g) and high-force (0.4 g) mechanical stimulation at the hind paws (data averaged from both sides) were significantly increased (n=5/sex), as compared to the pre-injury frequency, indicating the development of mechanical hypersensitivity (Figure 1B). Paw withdrawal frequencies were not significantly changed after sham surgery (n=5/sex).

DRGs contain a large number of non-neuronal cells, including SGCs, Schwann cells, immune cells, and fibroblasts. Previous studies have shown that scRNA-seq is advantageous for differentiating neurons from these non-neuronal cells (Renthal et al., 2020; Wang et al., 2021). Here, by using Pirt^EGFPf mice, we established a highly efficient purification process to further enrich DRG neurons for scRNA-seq (Figure 1—figure supplement 1A).

After removing low-quality cells and doublets (Supplementary file 2), we recovered 3394 cells from the Male-Sham dataset; 5678 cells from the Female-Sham dataset; 2899 cells from the Male-CCI dataset; and 3681 cells from the Female-CCI dataset. Figure 1—figure supplement 1B showed the number of expressed genes in each dataset. We then utilized canonical correlation analysis embedded in Seurat 3.0 (1), a computational approach for minimizing experimental batch effect, to integrate cells from the four datasets for an unbiased cell clustering. The results showed that the integration worked well in our experiments, as clusters from each dataset aligned well regardless of different biological variations (Figure 1C). In total, we identified 16 neuronal clusters and one non-neuronal cluster (Figure 1D, Figure 1—figure supplement 1C).

The non-neuronal cluster had fewer genes and unique molecular identifier (UMI) counts than did neuronal clusters (Figure 1—figure supplement 1D). It expressed SGC marker genes Fabp7 and Apoe (Renthal et al., 2020; Wang et al., 2021) but showed minimal expression of the pan-neuronal marker Tubb3. In contrast, all 16 neuronal clusters showed strong expression of Tubb3 but had very low levels of Fabp7 (Figure 1—figure supplement 1E). Although this non-neuronal cluster showed marker genes known to be expressed in both Schwann cells and SGCs (e.g., Mpz, Mbp, Plp1), it did not express any Schwann cell-specific genes such as Mag, Prx, or Ncmap (Avraham et al., 2020). Therefore, this non-neuronal cluster included mainly SGCs. No other non-neuronal cluster was detected in our datasets, suggesting a successful enrichment of neurons.

We further validated the identities of the neuronal clusters by a list of known subtype markers (Figure 1E). Our findings confirmed the presence of 12 major standard neuronal clusters (Nppb⁺ non-peptidergic nociceptors [NP1], Mrgprd⁺/Cd55⁺ non-peptidergic nociceptors [NP2], Mrgpra3⁺/Cd55⁺ non-peptidergic nociceptors [NP3], Tac1⁺/Sstr2^- peptidergic nociceptors [PEP1-2], Tac1⁺/Sstr2⁺ peptidergic nociceptors [PEP3-4], Trpm8⁺ peptidergic nociceptors [PEP5], Trpv1⁺ peptidergic nociceptors [PEP6], Nefh⁺/Scn1b⁺ Aβ low-threshold mechanoreceptors [NF1, NF2], and Fam19a4⁺/Th⁺ low-threshold mechano-receptive neurons with C-fibers [cLTMR]). Importantly, we also identified four new CCI-ind1-4 clusters that were prominent in CCI groups but minimal in sham groups (Figure 1D and E, and Figure 2A). CCI-ind1-4 clusters expressed high levels of genes like Cryba, Tgfbi, Fgf3, Tnfrsf12a, Ecel1, and Sema6a, and were segregated from all standard clusters. Expression profiles of the top 50 marker genes in the 16 neuronal clusters are shown in Figure 1F and listed in Supplementary file 1.

A much higher percentage of the cell population was contained in the CCI-ind1-4 clusters from CCI groups than from sham groups (Figure 2A and B). These clusters also showed high expression levels of injury-induced genes such as Sprr1a and Atf3 (Figure 2C; Bonilla et al., 2002; Tsujino et al., 2000). In line with previous findings (Nguyen et al., 2017), Atf3 was also frequently detected in the sequencing of isolated cells from sham or uninjured groups. Comparatively, Sprr1a was more selectively expressed in injured neurons of CCI mice (Figure 2C). Thus, Sprr1a may represent a more specific marker of injured neurons than Aft3 in scRNA-seq (Bonilla et al., 2002).

Regeneration-associated genes, such as Anxa1, Flnc, Gadd45a, Inhbb, Itga7, Kif22, Plaur, Sema6a, Sox11, Tnfrsf12a, and Tubb6, were among the top 30 marker genes of CCI-ind1-4 clusters (Figure 2D; Chandran et al., 2016). Intriguingly, some of them may also be involved in neuropathic and inflammatory pain. For example, animal studies suggest that neurotensin (encoded by Nts) (Guillemette et al., 2012; Sarret et al., 2005) and annexin1 (encoded by Anxa1) may attenuate neuropathic pain and inflammatory pain, respectively (Pei et al., 2011; Zhang et al., 2021). Nnat and Sdc1 were exclusively increased in nociceptive DRG neurons after nerve injury (Chen et al., 2010; Murakami et al., 2015). Sox11 was identified by integrated bioinformatic analysis as a novel gene that is essential to neuropathic pain (Chen et al., 2021).

Gene ontology (GO) analysis of the top 50 marker genes showed that CCI-induced transcriptomic changes were important for both nerve regeneration (e.g., nervous system development, axon guidance, cellular response to nerve growth factor stimulus) and neuronal excitability (e.g., positive regulation of calcium ion import, response to pain, regulation of sodium ion transport, and the neuropeptide signaling pathway; Figure 2E).

Previous studies in different nerve injury models showed that no specific neuronal cluster was spared from injury. Injured neurons in all clusters lost their original subtype-specific marker genes beginning at day 1 after injury, and hence could no longer be categorized into original clusters (Hu et al., 2016; Nguyen et al., 2019; Renthal et al., 2020; Wang et al., 2021). Yet, these studies did not further examine the proportion of injured neurons in each cluster after injury. Our findings showed a large decrease of cell proportion in 8 of 12 standard neuronal clusters after CCI (Figure 3—figure supplement 1), suggesting that the injured neurons in these clusters were no longer categorized into their original clusters. Instead, they may be assigned to CCI-ind1-4 clusters (Figure 3A and B), as suggested by a previous study (Renthal et al., 2020). The remaining four clusters (NP1, PEP5, NF1, NF2) showed little decrease in cell proportion after CCI. Moreover, a portion of injured neurons (Sprr1a⁺) in these four clusters were still categorized into their original clusters after CCI (Figure 3A and B). Further analysis revealed two subpopulations of neurons in these clusters (Figure 3C–F). For example, in the NP1 cluster, the larger subpopulation showed high expression of Sprr1a and other injury-induced genes (Gal, Hspb1, Stmn4, Chl1, and Sox11) (Chandran et al., 2016), suggesting that these are injured neurons (Figure 3C). In contrast, the smaller subpopulation (Sprr1a^-) showed little or no expression of injury-induced genes, but expressed other genes highly, such as Nppb and Sst, which are NP1 subtype-specific marker genes (Figure 1E). We found similar results in PEP5, NF1, and NF2 clusters (Figure 3D–F).

Because a significant portion of injured neurons (Sprr1a⁺) in NP1, PEP5, NF1, and NF2 clusters maintained their identities after CCI (Figure 3), we were able to determine transcriptomic changes in these neurons by comparing them to uninjured neurons of the same clusters in the sham group. After CCI, 197, 67, 41, and 79 differentially expressed genes (DEGs) were generated from NP1, PEP5, NF1, and NF2 clusters, respectively (Supplementary file 3), with 19 shared DEGs (Supplementary file 4). Most were common regeneration-associated genes induced by nerve injury (Chandran et al., 2016), and some (Cacna2d1, Gal, Gap43, Gadd45a, Atf3, Sprr1a) were also significantly regulated under chronic pain conditions (LaCroix-Fralish et al., 2011; Perkins et al., 2014).

GO analysis showed that these four clusters shared many common pathways, including those related to nervous system development, axon guidance, neuron projection development, microtubule-based process, neuropeptide signaling pathway, and cell differentiation (Figure 4A). In addition, we observed CCI-induced changes that affect neuronal excitability (e.g., downregulation of potassium and sodium channels, upregulation of calcium channel Cacna2d1, dysregulation of genes encoding neuropeptide, and G-protein-coupled receptors; Figure 4—figure supplement 1).

We further examined pain-related protein-protein interaction (PPI) networks within the pain interactome, a comprehensive network of 611 interconnected proteins specifically associated with pain (Jamieson et al., 2014). Examining 197 DEGs of the NP1 cluster revealed an interconnected network of 93 genes (Figure 4B). Among them, Calca, Prkca, Sst, and Tac1 were hub genes that were significantly downregulated after CCI, whereas Tgfb1 and Gal were hub genes that were significantly upregulated. Intriguingly, the top marker gene of the NP1 cluster, Sst, is also a key gene for neuropathic pain (Zhu et al., 2019); hence it may be an important new target for pain modulation.

When we examined an interconnected network of 44 genes from 67 DEGs in the PEP5 cluster, we identified Jun, Gal, Nts, and Tac1 as hub genes that changed significantly after CCI (Figure 4C). Examination of 41 DEGs from the NF1 cluster revealed an interconnected network of 69 genes (Figure 4D), with Calca, Npy, Jun, Gal, and Nts as hub genes. Similarly, an examination of 79 DEGs from the NF2 cluster revealed an interconnected network of 53 genes (Figure 4E), including hub genes Npy, Jun, Gal, Nts, and Tac1.

Among these hub genes, Calca and Tac1 were downregulated whereas Npy, Gal, Jun, and Nts were upregulated in multiple clusters. Functionally, Npy was recently identified as a key prognostic and therapeutic target of neuropathic pain (Tang et al., 2020), and Jun and Nts may play pivotal modulatory roles in both neuropathic pain and nerve regeneration (Zhao et al., 2020). For example, intrathecal administration of neurotensin (encoded by Nts) induced pain inhibition in animal models of neuropathic pain (Guillemette et al., 2012; Sarret et al., 2005). It is possible that the upregulation of Nts may represent a compensatory change after injury to limit the exaggeration of neuropathic pain. Future studies are warranted to delineate the roles of each of these shared hub genes in neuropathic pain pathogenesis.

Another goal of our study was to explore transcriptional changes in uninjured neurons (Sprr1a^-) of different clusters under neuropathic pain conditions. Strikingly, many DEGs were identified in Sprr1a^- neurons from a subset of clusters including NP1 (143), PEP5 (232), NF1 (95), and NF2 (132) after CCI (Figure 5A, Supplementary file 5). Comparatively, only a few DEGs were present in Sprr1a^- neurons from other clusters, suggesting subtype-specific transcriptional changes in uninjured neurons.

GO analysis showed that top pathways shared by Sprr1a^- neurons in NP1, PEP5, NF1, and NF2 clusters are related to protein biosynthetic processes, including translation, proton transport, ribosomal small subunit assembly, and transport (Figure 5B–F, Supplementary file 6). Pathways related to nerve regeneration and neuropathic pain were also found in these clusters but were much less significant. These findings suggest that these uninjured neurons in NP1, PEP5, NF1, and NF2 clusters may enter a ‘preparation’ state in response to nerve injury.

Both human and animal models suggest the presence of sex differences in pain sensitivity and chronic pain prevalence (Fillingim et al., 2009; Pieretti et al., 2016). The peripheral neuronal mechanisms underlying these sexual dimorphisms remain unclear, and few studies have compared transcriptional changes of DRG neurons at the single-cell level, especially under neuropathic pain conditions. X inactive-specific transcript (Xist) is a specific transcript expressed exclusively by the inactive X chromosome in female mammals (Borsani et al., 1991). Consistent with previous findings, we did not detect Xist expression in cells from our male mice (Figure 6—figure supplement 1A). Cell subtype distributions were similar between female and male mice in both sham and CCI groups (Figure 2B). Furthermore, Female-Sham and Male-Sham groups showed good correlation in cluster comparison, indicating a great similarity of transcriptional programs under physiologic conditions (Figure 6—figure supplement 1B). Nevertheless, our findings suggest sex differences in the transcriptional changes that occurred after CCI. When compared to the corresponding sham group, the Male-CCI group and the Female-CCI group exhibited 303 and 296 DEGs, within which there were 79 female-specific and 86 male-specific DEGs (Figure 6A, Supplementary file 7). Top pathways were further identified by conducting GO analysis of sex-specific and shared DEGs (Figure 6B–D).

Pearson correlation analysis was also performed based on the fold-change of the combined 382 DEGs. Most neuronal clusters showed a good correlation of DEGs between males and females (Figure 6E), indicating similarity and a minimal batch effect. Yet, the cLTMR cluster had a poor correlation. Intriguingly, Bohic et al. reported that deletion of Bhlha9, a transcription factor which is highly expressed in cLTMR, impaired thermotaxis behavior and exacerbated formalin-evoked pain only in male mice (Bohic et al., 2020). Furthermore, Calca, which is a nociceptor-specific gene that is highly upregulated in cLTMRs of male Bhlha9-null mice (Bohic et al., 2020), showed >2-fold differences between female and male cLTMRs in our dataset. These findings suggest that cLTMRs may play an important role in the sexual dimorphism of pain. From all clusters, 106 genes showed >2-fold differences between female and male CCI mice (Figure 6E, Supplementary file 8). The top 10 pathways from the GO analysis of these DEGs included nervous system development, neuropeptide signaling pathway, response to pain, sensory perception of pain, and response to estrogen (Figure 6F). Collectively, these findings suggest differential transcriptomic changes in DRG neurons after CCI between the two sexes.

Fgf3 is one of the top upregulated genes in CCI-ind1-4 clusters which express high levels of injury-induced gene Sprr1a as shown in our scRNA-seq study (Figures 1 and 2, Supplementary file 1). Previous studies suggested that Fgf3 may play a role in pain and neuropathy. For example, mice lacking NMDA receptor GluN1 in Schwann cells exhibited pain hypersensitivity and upregulated Fgf3 and Sprr1a expression in DRG neurons (Brifault et al., 2020). Accordingly, we conducted RNAscope in situ hybridization to further examine the expression and distribution of Fgf3 in DRG sections and compared findings between sham-operated and CCI mice. Both Fgf3 and Sprr1a signals were minimal in the sham group, but robustly increased after CCI (Figure 7A–C). In CCI mice, a large portion of Sprr1a⁺ cells were Fgf3⁺, suggesting the expression of Fgf3 in injured neurons. In contrast, Sprr1a^- neurons which express different cell subtype markers (Tac1, Nefh, Nppb) were rarely Fgf3⁺ in both CCI and sham groups (Figure 7D).

We next conducted an integration analysis of our scRNA-seq datasets and Renthal’s snRNA-seq datasets. The integration of six datasets (Renthal’s datasets: Naïve, Crush; our datasets: Male-Sham, Female-Sham, Male-CCI, Female-CCI) and twenty-two distinct clusters (Renthal’s datasets: cLTMR, NF, NP, PEP, SST; our datasets: SGC, NF1-2, NP1-3, PEP1-6, cLTMR, and CCI-ind1-4 clusters) were visualized by uniform manifold approximation and projection (UMAP) (Figure 7—figure supplement 1A and B). Feature heatmaps showed that Fgf3 expression was increased mainly in injury-induced new neuronal clusters (indicated by red circle) which were prominent in three injury datasets/groups (Crush, Male-CCI, Female-CCI), but were minimal in control groups (Naïve, Male-Sham, Female-Sham, Figure 7—figure supplement 1C). Collectively, evidence from in situ hybridization study and results from the integration analysis support our scRNA-seq findings, suggesting that Fgf3 mRNA expression was increased mainly in the injured neurons which were Sprr1a⁺ and negative for subtype markers after nerve injury.

We also did a correlation analysis between our CCI-ind1-4 clusters and neuronal clusters identified in Renthal’s datasets (7 days after sciatic nerve crush) and found that the CCI-ind1-4 clusters showed a good correlation with cLTMR1, cLTMR2, and NP in their datasets (Figure 7—figure supplement 2).

We then conducted in vitro calcium imaging to examine whether attenuating the upregulated Fgf3 expression may alter DRG neuronal response to capsaicin, a TRPV1 agonist that induces heat pain, in CCI mice. Lumbar DRG neurons from sham and CCI mice were transfected with siRNA specifically targeting Fgf3 (siFgf3, 0.2 nmol) or non-targeting siRNA (siNT, control) in culture. Previously we have demonstrated that the transfection efficiency of small RNA oligos using nucleofection was over 90% (Jiang et al., 2015). Quantitative PCR study showed that Fgf3 mRNA level in siNT-transfected DRG neurons was significantly higher in CCI group (n=3) than that in sham group (n=2, Figure 7E), which is consistent with our scRNA-seq and in situ hybridization results. Compared to siNT, siFgf3 transfection significantly decreased the upregulated Fgf3 mRNA level in CCI group (Figure 7E).

Because the expression of Fgf3 was quite low in sham group, we conducted in vitro calcium imaging on cultured DRG neurons from CCI mice. In CCI group, the intracellular calcium rise evoked by capsaicin (0.3 μM, bath application) was significantly increased in neurons transfected with siFgf3, as compared to siNT (Figure 7F).

In summary, our findings in a well-established animal model of neuropathic pain share some similarities with recent findings in transection injury models, including the loss of marker genes in injured neurons and the emergence of new, injury-induced clusters. Importantly, we demonstrated that subtype-specific transcriptomic changes occurred in both injured and uninjured neurons of NP1, PEP5, NF1, and NF2 clusters after CCI. Furthermore, we also provided novel evidence at single-cell level that transcriptomic sexual dimorphism may occur in DRG neurons after nerve injury, and cLTMR may play a pivotal role in sex-specific pain modulation. Lastly, by examining Fgf3 as proof-of principle, our RNAscope in situ hybridization study provided further evidence supporting findings from scRNA-seq analysis, and in vitro calcium imaging study offered novel insights of the functional implication of Fgf3. New knowledge gained from current work and other recent scRNA-seq studies may provide important rationales for developing cell-subtype-specific and sex-based therapies that can optimize neuropathic pain inhibition and sensory nerve regeneration. For instance, an ideal therapy would target top DEGs in uninjured neurons in NP1, PEP5, NF1, and NF2 clusters for pain control, but would not interfere with important transcription programs needed for the regeneration of injured neurons.

The Pirt^EGFPf mouse strain (C57BL/6 background) was a gift from Dr. Xinzhong Dong in the Solomon H Snyder Department of Neuroscience, School of Medicine, Johns Hopkins University. The axonal tracer farnesylated enhanced green fluorescent protein (EGFPf) replaced the entire open reading frame of Pirt and the EGFPf expression is under the control of the endogenous Pirt promoter (Kim et al., 2008). We bred the homozygous mutant with WT mice to get heterozygous for experiments. Adult mice (7–8 weeks of age) of both sexes were used. Genotypes of the mice were determined by PCR. Mice were housed three to five per cage and given access ad libitum to food and water. All animal experiments were conducted in accordance with the protocol approved by the Institutional Animal Care and Use Committee of Johns Hopkins University.

Contralateral changes may develop after unilateral CCI of the sciatic nerve, including spontaneous pain and mechanical hypersensitivity in hind paws (Paulson et al., 2002; Wilkerson et al., 2020), as well as gene expression in the spinal cord and DRGs (Jancálek et al., 2010). Because transcriptional changes in contralateral DRGs may differ from those on the ipsilateral side, we performed bilateral sciatic CCI to allow the pooling of bilateral DRG tissues for sequencing and to avoid sample variations. The bilateral CCI model has been validated in previous studies, which showed that animals displayed prolonged cold and mechanical hypersensitivities in hind paws but did not exhibit the asymmetric postural or motor influences of unilateral CCI or behavior changes (Dai et al., 2014; Datta et al., 2010; Vierck et al., 2005).

Adult Pirt^EGFPf mice were randomly assigned to undergo bilateral CCI surgery or sham surgery. All procedures were performed by the same experimenter to avoid variation in technique. CCI of the sciatic nerve was performed as previously described (Guan et al., 2010; Li et al., 2017). Briefly, mice were anesthetized with 2% isoflurane, and a small incision was made at the mid-thigh level. The sciatic nerve was exposed by blunt dissection through the biceps femoris. The nerve trunk proximal to the distal branching point was loosely ligated with three nylon sutures (9–0 nonabsorbable monofilament; S&T AG) placed approximately 0.5 mm apart until the epineuria was slightly compressed and hindlimb muscles showed minor twitching. The muscle layer was closed with a 4–0 silk suture and the wound was closed with metal clips. Pirt^EGFPf mice developed mechanical hypersensitivity in both hind paws, as indicated by a significant increase in paw withdrawal frequency (PWF) to von Frey filament stimulation (Figure 1B).

Animals were allowed to acclimate for a minimum of 48 hr before any experimental procedures. Hypersensitivity to punctuate mechanical stimuli was assessed by the PWF method using two von Frey monofilaments (low-force, 0.07 g; high-force, 0.4 g). Each von Frey filament was applied perpendicularly to the mid-plantar area of each hind paw for ~1 s. The left hind paw was stimulated first, followed by the right side (>5 min interval). The stimulation was repeated 10 times at a rate of 0.5–1 Hz (1–2 s intervals). If the animal showed a withdrawal response, the next stimulus was applied after the animal resettled. PWF was then calculated as (number of paw withdrawals/10 trials)×100%.

Bilateral L4-5 DRGs were collected from mice at day 7 after bilateral sciatic CCI or sham surgery. Male and female mice from the same litter were subjected to the same surgery (sham or CCI). Bilateral L4-5 DRGs were collected from each mouse and DRGs from five mice of the same group (20 DRGs in total) were pooled as one sample for sequencing. The four groups included Female-Sham, Male-Sham, Female-CCI, and Male-CCI. DRGs were dissected out, digested with 1 mg/mL type I collagenase (Thermo Fisher Scientific) and 5 mg/mL dispase II (Thermo Fisher Scientific) at 37°C for 70 min (10 DRGs/tube), and disassociated into single cells in Neurobasal medium containing 1% bovine serum albumin (BSA). Cells were filtered through a 40 µm cell strainer and centrifuged at 500×g for 5 min. The pellet was resuspended in 500 μL of Triple Express (Thermo Fisher Scientific) and digested at 37°C for 2 min. The reaction was stopped with 1 mL of fetal bovine serum. The cell suspension was laid onto 5 mL of Neurobasal medium containing 20% Percoll and centrifuged at 1400 RPM for 8 min. The upper layer was removed first and then the lower layer. The pellet was washed with 2 mL of Neurobasal medium containing 1% BSA and centrifuged at 500×g for 5 min. The pellet was resuspended with 500 μL of Neurobasal medium containing 1% BSA. The GFP⁺ cells were sorted into 500 μL of Neurobasal medium containing 1% BSA and centrifuged at 500×g for 5 min to remove most supernatant. The pellet was resuspended with the remaining supernatant to a concentration of ~1000 cells/μL.

The single-cell suspensions were further processed with Chromium Next GEM Single Cell 3′ GEM, Library & Gel Bead Kit v3 (PN-1000094) according to the manufacturer’s instructions to construct the scRNA-seq library. All libraries were sequenced with the Illumina NovaSeq platform. The raw sequencing reads were processed by Cell Ranger (v.2.1.0) with the default parameters. The reference genome was mm10.

Scrublet with default parameters was used first to remove single-cell doublets. After doublet removal, we filtered out cells with fewer than 1500 genes expressed and cells with more than 10% mitochondrial UMI counts. The rigorous filtering was intended to remove smaller residual non-neuronal cells such as SGCs. After doublet removal and quality control, we applied Seurat’s integration workflow to correct possible batch effects for the remaining cells of the four datasets. Before the integration, the four datasets were transformed into four individual Seurat objects with standard steps including ‘CreateSeuratObject’, ‘NormalizeData’, and ‘FindVariableFeatures’. Subsequently, we used ‘FindIntegrationAnchors’ with the top 3000 variable genes to locate possible anchors among the four datasets. Next, ‘IntegrateData’ was used to merge the four individual datasets. After the integration, default clustering steps embedded in Seurat were performed with 20 principal components (PCA) and 3000 variable genes. The steps included scaling normalized UMI counts with ‘ScaleData’, dimensional reduction with ‘RunPCA’, building a k-nearest neighbor graph with ‘FindNeighbors’, and finding clusters with the Louvain algorithm by ‘FindClusters’. Finally, we visualized identified clusters with 2D UMAP by ‘RunUMAP’. To find the most conserved markers in every cluster, we used ‘FindConservedMarkers’ and show the top 50 markers in Figure 1F and Supplementary file 1. To evaluate similarities between identified single-cell clusters, we applied unsupervised hierarchical clustering with the pairwise Pearson correlation using 3000 variable genes (Figure 1—figure supplement 1B). To find DEGs between clusters or conditions, we used ‘FindMarkers’ with padj <0.05 and Log₂fold-change >0.5 as the thresholds.

We used the expression level of Sprr1a to divide neuronal clusters into two subpopulations (Sprr1a⁺and Sprr1a^-). Sprr1a⁺represent cells with normalized UMI >1 and Sprr1a^- represent cells with normalized UMI <1.

GO analysis was conducted with DAVID (Huang et al., 2009a; Huang et al., 2009b). We used p-value = 0.05 as the threshold to find enriched GO terms such as biological processes. The gene clusters were visualized with the ClusterProfiler package in R.

A PPI network was drawn with the igraph R package based on the list of reported 1002 PPIs involved in pain (Jamieson et al., 2014). The genes in the DEG lists that had no connections (receive or send) in the PPI network were filtered out. Removed nodes also filtered out any of their edges.

RNAscope fluorescence in situ hybridization experiment was performed according to the manufacturer’s instructions, using the RNAscope Multiplex Fluorescent Reagent Kit v2 (ACD, Advanced Cell Diagnostics, Newark, CA) for fresh frozen tissue. Briefly, lumbar DRGs (L4-5) were dissected at 7 days after sham surgery or CCI, frozen, and sectioned into 12 μm sections using a cryostat. In situ probes against the following mouse genes were ordered from ACD and multiplexed in the same permutations across quantified sections: Sprr1a (Cat No. 426871-C2), Fgf3 (Cat No. 503101), Tac1 (Cat No. 410351-C3), Nefh (Cat No. 443671-C3), Nppb (Cat No. 425021-C3). High-resolution images of 10 Z-stack were obtained using a ×40 oil immersion objective on Zeiss LSM 800 confocal microscope (Zeiss, Oberkochen, Germany). Cells were considered positive if at least three puncta were observed (Peirs et al., 2021).

To transfect RNA oligos into DRG neurons, the dissociated neurons from lumbar DRGs were centrifuged to remove the supernatant and resuspended in 100 μL of Amaxa electroporation buffer for mouse neuron (Lonza Cologne GmbH, Cologne, Germany) with siRNAs (0.2 nmol per transfection). siFgf3 (sense: CAGAGACCUUGGUACGUGUtt; Antisense: ACACGUACCAAGGUCUCUGgg). Suspended cells were then transferred to a 2.0 mm cuvette and nucleofected with the Amaxa Nucleofector apparatus. After electroporation, cells were immediately mixed to the desired volume of prewarmed culture medium and plated on precoated coverslips or culture dishes. After neurons fully attached to the coverslips or culture dishes (4–6 hr), the culture medium was changed to remove the remnant electroporation buffer.

To analyze the mRNA expression in DRG neurons, total RNA was isolated using PicoPure RNA Isolation Kit (Thermo Fisher Scientific) following the manufacturer’s manual. RNA quality was verified using the Agilent Fragment Analyzer (Agilent Technologies, Santa Clara, CA). Two-hundred ng of total RNA was used to generate the cDNA using the SuperScript VILO MasterMix (Invitrogen, Waltham, MA). Ten ng of cDNA was run in a 20 μl reaction volume (triplicate) using PowerUp SYBR Green Master Mix to measure real-time SYBR green fluorescence with QuantStudio 3 Real-Time PCR Systems (Thermo Fisher Scientific). Calibrations and normalizations were performed using the 2^-ΔΔCT method. Mouse Gapdh was used as the reference gene. Mouse Fgf3 (#MP204816) and Gapdh (#MP205604) primers were purchased from OriGene Technologies (Rockville, MD). Fgf3: (forward primer: GCAAGCTCTACTGCGCTACCAA; reverse primer: CACTTCCACCGCAGTAATCTCC). Gapdh (forward primer: CATCACTGCCACCCAGAAGACTG; reverse primer: ATGCCAGTGAGCTTCCCGTTCAG).

Experiments were conducted as that in our previous studies (Liu et al., 2009). Lumbar DRGs from 4-week-old mice that underwent sham surgery or CCI were collected in cold DH10 20 (90% DMEM/F-12, 10% fetal bovine serum, penicillin [100 U/mL], and streptomycin [100 μg/mL] Invitrogen, Waltham, MA) and treated with enzyme solution (dispase [5 mg/mL] and collagenase type I [1 mg/mL] in Hanks’ balanced salt solution without Ca²⁺ or Mg²⁺) for 35 min at 37°C. After trituration, the supernatant with cells was filtered through a Falcon 40 µm (or 70 µm) cell strainer. Then, the cells were spun down with centrifugation and were resuspended in DH10 with growth factors (25 ng/mL NGF; 50 ng/mL GDNF), plated on glass coverslips coated with poly-D-lysine (0.5 mg/mL; Biomedical Technologies Inc, Madrid, Spain) and laminin (10 μg/mL; Invitrogen), cultured in an incubator (95% O₂ and 5% CO₂) at 37°C, and used within 48 hr. Neurons were loaded with Fura-2-acetomethoxyl ester (Molecular Probes, Eugene, OR) for 45 min in the dark at room temperature (Liu et al., 2009). After being washed, cells were imaged at 340 and 380 nm excitation for the detection of intracellular free calcium.

We conducted the integration analysis of our scRNA-seq datasets and Renthal’s snRNA-seq datasets (GSE154659). We extracted DRG cells from naïve mice (seven replicates: GSM4676529, GSM4676530, GSM4676533, GSM4676534, GSM4676535, GSM4676536, GSM4676537) and mice at day 7 after sciatic nerve crush injury (four replicates: GSM4676564, GSM4676565, GSM4676531, GSM4676532) from Renthal’s datasets. We only considered the cells with gene count between 200 and 12,000, and mitochondrial DNA less than 10%. The remaining 17,207 cells were integrated with our datasets using default parameters described in Seurat’s integrated workflow. The first 30 PCA were used to build UMAP for visualizing the integration result.

We extract cells (four replicates: GSM4676564, GSM4676565, GSM4676531, GSM4676532) from mice at day 7 after sciatic nerve crush injury from Renthal’s datasets and then used Seurat’s default workflow to create its Seurat single-cell object. We further used shared variable genes between our dataset and their datasets to calculate Pearson correlation.

Statistical analyses were performed with the Prism 8.0 statistical program (GraphPad Software, Inc). The methods for statistical comparisons in each study are given in the figure legends. Comparisons of data consisting of two groups were made by Student’s t-test. Comparisons of data in three or more groups were made by one-way analysis of variance (ANOVA) followed by the Bonferroni post hoc test. Comparisons of two or more factors across multiple groups were made by two-way ANOVA followed by the Bonferroni post hoc test. Two-tailed tests were performed, and p<0.05 was considered statistically significant in all tests.