Repopulated spinal cord microglia exhibit a unique transcriptome and contribute to pain resolution

SUMMARY

Microglia are implicated as primarily detrimental in pain models; however, they exist across a continuum of states that contribute to homeostasis or pathology depending on timing and context. To clarify the specific contribution of microglia to pain progression, we take advantage of a temporally controlled transgenic approach to transiently deplete microglia. Unexpectedly, we observe complete resolution of pain coinciding with microglial repopulation rather than depletion. We find that repopulated mouse spinal cord microglia are morphologically distinct from control microglia and exhibit a unique transcriptome. Repopulated microglia from males and females express overlapping networks of genes related to phagocytosis and response to stress. We intersect the identified mouse genes with a single-nuclei microglial dataset from human spinal cord to identify human-relevant genes that may ultimately promote pain resolution after injury. This work presents a comprehensive approach to gene discovery in pain and provides datasets for the development of future microglial-targeted therapeutics.

In brief

Donovan et al. demonstrate that depletion and repopulation, but not depletion alone, of CNS microglia contribute to complete pain resolution after peripheral injury. Key mouse microglial genes from these pro-resolution microglia are cross-referenced with a human spinal dorsal horn microglia dataset to reveal targets with high translational potential.

Graphical abstract

INTRODUCTION

The transition to chronic pain is still poorly understood, even though persistent pain affects one in three individuals.¹ Peripheral injury-induced neuronal hyperactivity leads to downstream engagement of microglia in the spinal cord,² which contributes to the initiation of pain.^3,4 This has been confirmed in several preclinical pain models in which pharmacologic or genetic reversal of microglial activation^5–7 or direct inhibition of spinal microglia with a Gi DREADD⁸ decreases nociceptive outcomes such as allodynia and thermal hyperalgesia. In addition, depletion of microglia using either the immunotoxin Mac1-saporin^9,10 or transgenic Cx3CR1-Cre^ERT2-eYFP;R26-iDTR^LSL mice¹¹ prevents or reverses existing pain at limited time points after nerve injury. Importantly, after genetic depletion, microglia fully repopulate the CNS within 14 days.^11,12 Therefore it is possible that either reactive microglia initiate the transition from acute to chronic pain or that repopulated microglia actively resolve pain. The reversal of pain after transient depletion of microglia offers an opportunity to study the contribution of microglia in a pain-resolution context.

While microglial “signature genes” have been identified, transcriptomic studies highlight that microglia are heterogeneous cells that are phenotypically and genetically shaped by their microenvironment,^13,14 particularly by their anatomic location.¹⁵ Importantly, the majority of these studies have been performed in distinct brain supraspinal regions^16,17 with few to date undertaking transcriptomic analysis of microglia from the mouse spinal cord^18,19 and even fewer performing such analyses in human spinal cord tissue.²⁰ In addition, transcriptomic analyses of CNS tissue in disease states involving neuroinflammation, such as Alzheimer’s disease,²¹ multiple sclerosis,²² and chronic pain,¹⁹ suggest heterogeneity in microglial responses to pathology. In-depth knowledge of the specific transcriptome of spinal cord microglia after pain-producing peripheral injury is therefore needed to identify markers for further targeted drug development for the treatment of chronic pain. In addition, because these data are often collected in rodent models, the use of single-cell/nucleus sequencing data from human spinal cord are necessary to improve the development of clinically translatable microglia-directed pain therapeutics.

In this study, we evaluated the temporal contribution of microglia to pain progression using targeted genetic microglia depletion with Cx3CR1-Cre^ERT2-eYFP;R26-iDTR^LSL mice in the context of peripheral injury. We used the clinically relevant tibial fracture and casting model of complex regional pain syndrome (CRPS), which has a distinctly timed acute-to-chronic pain transition. We discovered that microglia depletion and repopulation at the acute-to-chronic transition completely resolved pain and improved peripheral inflammation. We took advantage of this pain-resolution context to establish transcriptomic signatures of repopulated spinal cord microglia at the point of full repopulation, which coincided with behavioral improvements with regard to pain. Finally, we identified targets with human relevance by overlapping mouse genes of interest identified by weighted gene co-expression network analysis (WGCNA) with a human spinal cord single-nucleus microglia transcriptome dataset.

RESULTS

Morphologically distinct microglia repopulate the spinal cord 14 days after targeted genetic depletion

We utilized Cx3CR1-Cre^ERT2-eYFP;R26-iDTR^LSL mice, which allow for temporally controlled depletion of microglia without depletion of other myeloid-lineage cells (e.g., peripheral macrophages; Figure 1A). Tamoxifen was administered to express the diphtheria toxin (DT) receptor (DTR) in CX3CR1⁺ cells, then DT was administered 3–4 weeks later, giving time for peripheral macrophages to turn over and no longer express the DTR. While these mice have been previously characterized,¹² we independently confirmed that the Cx3CR1-Cre^ERT2-eYFP mouse was tamoxifen dependent in a parallel study.⁷ After DT-induced depletion, microglia repopulated to baseline numbers over the course of ~14 days (Figures 1B–1D), consistent with previous studies.^11,12,23 Interestingly, these repopulated microglia were morphologically distinct compared with baseline control microglia, with fewer, shorter branches and an overall lower complexity after depletion and repopulation even to 28 days post-repopulation (Figures 1C, 1E, and 1F).

Figure 1. Cx3CR1-Cre^ERT2-eYFP;R26-iDTR^LSL mice exhibit significant loss of microglia after diphtheria toxin (DT)-induced depletion, but these cells repopulate over time.

(A) Schematic representation of the strategy used to selectively deplete microglia, and no other myeloid-lineage cells, using the Cx3CR1-Cre^ERT2-eYFP; R26-iDTR^LSL mouse. Mice were injected with tamoxifen (TAM; 100 mg/kg daily × 5 days), allowing excision of the floxed STOP codon, resulting in all Cx3CR1⁺ cells expressing the DTR. After 21–24 days, high-turnover macrophages no longer express DTR, while slow-turnover microglia continue to express DTR. Treatment with DT (1,000 ng daily × 3 days) resulted in selective death of microglia.

(B) Lumbar spinal cord sections from Cx3CR1-Cre^ERT2-eYFP;R26-iDTR^LSL mice taken at multiple time points before (baseline) or after DT injection demonstrate a significant loss of Iba1⁺ microglia in the dorsal horn at day 1 post-DT by immunohistochemistry. Scale bars, 100 and 20 μm (insets).

(C) Representative Sholl analysis profiles of microglia demonstrate changes in morphology over time following repopulation. The start radius of the rings is 0.5 mm with a step size of 1 μm.

(D) Quantification of microglia number over time after depletion with DT. Four or five lumbar spinal cord dorsal horn sections were counted and averaged (n = 2–3 mice per group). **p < 0.01 by one-way ANOVA with Tukey’s post hoc test.

(E) Analysis of intersections at a given radial distance from the soma demonstrates a drop in the number of intersections in repopulating microglia, which recovers, although not quite to baseline, over 28 days. Fifty-six to 185 microglia were analyzed per time point, with n = 2–3 mice per group.

(F) Area under the curve (AUC) for number of intersections by radial distance from soma demonstrates a time-dependent increase after microglial depletion. **p < 0.01, ***p < 0.001, by one-way ANOVA with Dunnett’s post hoc test. Fifty-six to 185 microglia were analyzed per time point, with n = 2–3 mice per group.

(G and H) Microglia depletion alone does not affect (G) mechanical sensitivity or (H) latency to withdrawal on a 52.5°C hot plate in male or female Cx3CR1-Cre^ERT2-eYFP;R26-iDTR^LSL mice (n = 3–6/sex, comparison by unpaired t test, in all cases p > 0.05). ns, not significant. Data are represented as the mean ± SEM.

Microglia depletion at the acute-to-chronic transition drastically improves the pain trajectory

We have previously demonstrated that the tibial fracture and casting model of CRPS results in signs of peripheral inflammation (paw edema, increased paw temperature) lasting up to 4 weeks post-injury. In contrast, mechanical paw sensitivity (allodynia) persists for at least 20 weeks.^7,24 These findings suggest a peripheral-to-central (or acute-to-chronic) transition between 3 and 5 weeks post-injury. Importantly, microglia depletion alone did not affect baseline weight bearing, paw edema, or paw temperature (Figures S1A–S1C). We also did not observe any motor effects of microglia depletion or DT treatment (Figure S1D). In addition, microglia depletion alone had no effect on measures of nociception, including unchanged mechanical sensitivity (Figure 1G) and hot-plate latency (Figure 1H).

To establish the contribution of microglia to persistent pain, we first performed microglia depletion at the time of injury (day 0, acute phase) in the tibial fracture and casting model of CRPS (Figure 2A). When microglia were depleted in this early/preventative paradigm, both male and female injured mice with microglia depletion (injured-depleted) demonstrated a significant but partial attenuation of allodynia over the course of the 9-week testing period, compared with sex-matched injured-only controls (Figure 2B). The latency to withdrawal on a hot plate was significantly increased in females (p = 0.0304), but not males (p = 0.0780), in the injured-depleted groups (Figures S1E and S1F). Both male and female injured-depleted mice improved their weight bearing on the injured hindpaw at 4 weeks after injury (Figures S1G and S1H) and exhibited less edema (Figures S1I and S1J) and less warmth (Figures S1K and S1L) of the injured limb.

Figure 2. Microglia depletion at the acute-to-chronic phase results in a sustained change in the pain trajectory after peripheral injury.

(A) Schematic of experimental timeline for acute-phase (day 0) depletion of microglia.

(B) Mechanical threshold is decreased at the time of cast removal (3 weeks) in all groups; however, microglia depletion results in progressive, partial improvement in mechanical sensitivity in both males and females out to 9 weeks post-injury. n = 5–15 per group per sex. **p < 0.01, ***p < 0.001, ****p < 0.0001 by two-way ANOVA with Bonferroni’s post hoc test.

(C) Schematic of experimental timeline for acute-to-chronic-phase depletion of microglia.

(D) Mechanical threshold is decreased at the time of cast removal (3 weeks) in all groups; however, microglia depletion results in sustained improvement in mechanical sensitivity in both males and females out to 9 weeks post-injury. n = 6–11 per group per sex. ***p < 0.001, ****p < 0.0001 by two-way ANOVA with Bonferroni’s post hoc test.

(E and F) Microglia depletion at the acute-to-chronic phase additionally results in improvement in thermal sensitivity, weight bearing, edema, and temperature changes in both (E) males and (F) females at 4 weeks post-injury. n = 6–11 per group per sex for hot plate, n = 4–5 per group per sex for unweighting, edema, and temperature. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 by two-tailed unpaired t test. Data are represented as the mean ± SEM.

Due to the incomplete reversal of allodynia in the early/preventive-phase depletion paradigm, as well as the clinical relevance of post-injury treatment, we next tested microglia depletion at 3 weeks post-fracture (acute-to-chronic phase; Figure 2C). Strikingly, we found that both male and female injured-depleted mice demonstrated a profound improvement in allodynia, with a return to preinjury baseline values over the course of the testing period (Figure 2D). The reversal of mechanical hypersensitivity was particularly remarkable as the mice remained allodynic immediately after DT treatment (when microglia were absent/depleted) and then exhibited a progressively improving pain trajectory that temporally paralleled full spinal cord repopulation of microglia. We additionally observed an increase in hot-plate latency in both male and female injured-depleted mice as well as improvement in weight bearing, edema, and warmth of the injured paw at 4 weeks in the injured-depleted mice (Figures 2E and 2F). Importantly, persistent depletion of microglia using high-dose PLX3397 chow for 4 weeks, starting at 3 weeks post-fracture (acute-to-chronic phase; Figure S2), did not change allodynia. These findings suggest that microglia repopulation, and not depletion itself, may trigger pain resolution and improvement in classic peripheral signs of CRPS (edema and temperature) in a post-injury/treatment paradigm.

Repopulated microglia exhibit a distinct transcriptome that may contribute to pain resolution

In our model, morphologically distinct microglia were repopulated by 14 days post-depletion, coincident with a sustained improvement in pain and peripheral inflammatory outcomes. To identify transcripts that may enable repopulated microglia to have a pro-resolution role in the context of peripheral injury, we isolated microglia from the lumbar spinal cord by fluorescence-activated cell sorting (FACS) at 5 weeks post-injury (14 days post-microglia depletion and repopulation) for subsequent RNA sequencing (RNA-seq) from four groups: uninjured mice with resident microglia (uninjured-resident microglia), uninjured mice with repopulated microglia (uninjured-repopulated microglia), injured mice with resident microglia (injured-resident microglia), and injured mice with repopulated microglia (injured-repopulated microglia).

We sorted microglia as follows: live (SYTOX blue negative), CD19⁻CD3⁻CD45^midCD11b⁺Cx3CR1-YFP⁺ (Figure S3A). To first confirm the specificity of our isolation method for CNS microglia, we used the publicly available ImmGen “My GeneSet” program to cross-reference genes in our male RNA-seq dataset with the ImmGen database of over 50 cell populations.²⁵ The top 100 genes expressed by log₂(cpm+1) in our datasets clearly defined a microglia-specific signature that could be differentiated from similar myeloid-lineage cells, including blood monocytes (Figure S3B). We next performed multidimensional scaling (MDS) and observed clustering of microglial gene profiles by treatment group, with microglia depletion and repopulation having the greatest effect (Figures S3C and S3D).

Differential gene expression analysis using DESeq2 was performed on the microglia transcriptome for four comparisons by sex: injured-resident microglia vs. uninjured-resident microglia (effect of injury alone), uninjured-repopulated microglia vs. uninjured-resident microglia (effect of microglia depletion and repopulation alone), injured-repopulated microglia vs. uninjured-repopulated microglia (effect of injury in the context of microglia depletion and repopulation), and injured-repopulated microglia vs. injured-resident microglia (effect of microglia depletion and repopulation in the context of injury). The analyses were performed at the transcript level for increased resolution, and then transcripts were mapped to their corresponding genes for data representation, as indicated. For females we identified 34,865 transcripts that mapped to 13,377 unique genes. For males we identified 28,345 transcripts that mapped to 11,999 unique genes. Consistent with datasets from other pain models,²⁶ the number of differentially expressed genes (DEGs) in sorted microglia in the injury-alone context (injured-resident microglia vs. uninjured-resident microglia) was limited. For females there were 4 upregulated DEGs and 8 downregulated DEGs, while for males there were 30 upregulated DEGs and 2 downregulated DEGs, none of which overlapped between sexes (Table S1 and Extended Data 1). Also consistent with prior datasets,¹² microglia depletion and repopulation resulted in many DEGs (uninjured-repopulated microglia vs. uninjured-resident microglia), indicating that repopulated microglia are distinctly different from microglia at baseline. Specifically, for females there were 251 upregulated DEGs and 375 downregulated DEGs, while for males there were 571 upregulated DEGs and 494 downregulated DEGs. In total, 99 upregulated DEGs and 72 downregulated DEGs overlapped between the sexes.

The effect of injury in the context of microglia depletion (injured-repopulated microglia vs. uninjured-repopulated microglia) was sex specific. For females there were 104 upregulated DEGs and 60 downregulated DEGs, while for males there were 92 upregulated DEGs and 7 downregulated DEGs. There were no downregulated DEGs that overlapped between the sexes, and just 1 shared upregulated DEG (Slc4a7) (Table S1). The transcriptome of repopulated microglia, therefore, depends on the context (injury or no injury) in which they differentiate.

While the above comparisons provide a resource and confirm the validity of our dataset, to characterize microglia that repopulate in the context of peripheral injury but no longer contribute to sensitization (pain-resolution context), we focused our subsequent analyses on the injured-repopulated microglia vs. injured-resident microglia comparison. We identified numerous DEGs, with several shared by both sexes. For females there were 293 upregulated DEGs and 338 downregulated DEGs, while for males there were 412 upregulated DEGs and 283 downregulated DEGs (Figure 3A and Extended Data 1). In total, 112 upregulated DEGs and 72 downregulated DEGs overlapped between the sexes.

Figure 3. Differential gene expression analysis of female and male spinal cord microglia after peripheral injury with or without microglia depletion and repopulation identifies sex-specific and sex-independent signatures of pain recovery.

(A) Differentially expressed microglial transcripts (DETs) and genes (DEGs) were identified using DESeq2 analysis in both males and females, with 112 sex-independent/shared upregulated DEGs and 72 sex-independent/shared downregulated DEGs. n = 4–7 mice per group per sex.

(B) Volcano plots depicting DEGs from injured-repopulated microglia compared with injured-resident microglia in females (left) and males (right). Highly regulated, sex-independent, and significant DEGs are labeled in red. Thresholds for the x axis are set at |log₂(FC)| > 2.0 (female) and >0.83 (male). Threshold for the y axis is set at −log₁₀(adjusted p value) > 1.3, equivalent to p-adjusted < 0.05, for both sexes.

(C and D) Significant gene ontology (GO) terms for (C) females and (D) males based on adjusted p < 0.05 for injured-repopulated vs. injured-resident microglia DEGs.

(E) GO terms for the shared DEGs between male and female injured-repopulated vs. injured-resident microglia. Size of circles represents number of GO terms combined into the parent GO term shown. Color of circles represents log₁₀ p value.

To evaluate the top regulated genes by sex in repopulated vs. resident microglia with and without injury, we generated volcano plots to establish fold change to significance comparisons (Figures 3B and S4). We noted that Apoe, previously implicated in microglial phenotypic responses to injury,²⁰ was highly upregulated in the injured-repopulated microglia in both sexes compared with injured-resident microglia. We highlighted additional genes in volcano plots that were highly differentially expressed in injured-repopulated microglia or had been previously implicated in the context of pain (Figure 3B, red text). Highly regulated female-specific genes included Gtse1, Tpx2, and Upk1b (downregulated) as well as Pilra, Clec12a, and Celf1 (upregulated). Highly regulated male-specific genes included Csf1r, Khdrbs3, and Fscn1 (all downregulated) as well as Cd72, Cd300lf, Tlr2, Slc2a6, Axl, and Cxcl13 (all upregulated). Highly regulated genes shared between the sexes included Serpinf1 (downregulated) as well as Apoe and Ms4a7 (upregulated) (Figure 3B, red text).

We next performed gene ontology (GO) term analysis on the DEGs from each comparison to identify biological pathways most strongly represented. In the female injured-repopulated microglia vs. injured-resident microglia comparison, several GO terms related to cell-cycle processes and regulation of cell-component organization were represented, as well as “regulation of response to stimulus” (Figure 3C and Extended Data 2). In the male injured-repopulated microglia vs. injured-resident microglia comparison, immune system processes were highly represented as were defense responses and responses to external stimuli/cytokines (Figure 3D and Extended Data 3). We next took all overlapping DEGs from male and female injured-repopulated microglia vs. injured-resident microglia comparisons and ran a GO analysis on these shared genes (Figure 3E and Extended Data 4). While broad GO terms remained highly represented (cytoplasmic translation, homeostatic process, cellular process), we also identified GO terms related to the immune system, including “inflammatory response,” “immune system processes,” and “response to external stimulus,” consistent with known microglial injury responses.⁷ Taken together, this differential gene expression analysis highlights that repopulated microglia are transcriptionally responsive to injury, with both sex-specific and sex-distinct gene changes, while the behavioral phenotype (i.e., pain resolution) is similar between sexes.

Construction of an unbiased gene co-expression network allows for discovery of genes contributing to pain resolution

While the DESeq2 analysis allowed us to identify individual DEGs, we next sought to discover biological processes and signaling pathways that better capture the functional difference between injured-resident microglia and the injured-repopulated microglia that contribute to pain resolution. To do this, we performed WGCNA, which clusters highly co-expressed genes into modules without the use of a priori knowledge (Figure S5). Importantly, gene expression in all samples is used in WGCNA, which allows for unbiased prediction of integral pathways associated with the cell states analyzed. Fifty-three unique modules were defined in females (Figures 4A and 4C) and 54 unique modules were defined in males (Figures 4B and 4C). Once these gene modules were constructed, the modules’ eigengenes (MEs), which are the first principal component of each module’s expression matrix, were then correlated by sex to each of the four microglial comparisons (“traits”) to define significant module-trait relationships (Figures 4A and 4B; also see Extended Data 5 for a full list of correlation coefficients and p values by trait).

Figure 4. Weighted gene co-expression network analysis (WGCNA) reveals module-trait relationships that are distinct by sex and condition.

(A and B) WGCNA identified (A) 53 unique modules in females and (B) 54 unique modules in males, which were then correlated with each of the four microglial comparisons (“traits”). Module-trait relationships that had significant correlations are marked with an asterisk, with heatmaps depicting the strength of the relationship (blue, negative correlation with the trait; red, positive correlation with the trait). Arrows indicate modules selected for further analyses.

(C) The number of significant modules per sex as well as total number of transcripts within each module.

WGCNA identified networks of genes that characterized the pain-resolution context with 21 significant modules in females and 8 significant modules in males in the injured-repopulated microglia vs. injured-resident microglia comparison (Figure 4C). To identify gene clusters strongly associated with pain resolution, we selected modules of interest for further analysis based on a combination of their absolute trait correlations across the four trait comparisons (Figures 4A and 4B) and the number of differentially expressed transcripts (DETs) from the DESeq2 analysis represented in each module (Extended Data 6). We thus selected four modules (female, blue and midnightblue; male, darkseagreen1 and turquoise), all of which had a strong correlation to the injured-repopulated vs. injured-resident trait comparison (Figures 4A and 4B, arrows) and had high representation of DETs (Extended Data 6). To define highly represented biological processes and pathways, GO analysis and Ingenuity pathway analysis (IPA) were performed on all transcripts within each module, and genes highly influencing these pathways were further identified. Finally, we established the top 20 genes within each module with the highest gene significance vs. module membership, a measure of the strength of the correlation of the individual gene with the module eigengene (Figure 5). These genes are central to module construction because they correlate with a large number of genes within the module, acting as a major node (red/blue data points), and thus serve as natural candidates for further investigation (Figure 5).

Figure 5. Modules of interest for each sex highlight networks of genes that may contribute to pain resolution.

(A) Gene ontology (GO) terms identified for female module blue.

(B) Gene significance vs. module membership plot demonstrates correlation of member genes to the module eigengene. The 20 most extreme genes are highlighted in either red, for positive fold change, or blue, for negative fold change, in the injured-repopulated microglia group vs. the injured-resident microglia group.

(C) List of top 20 most significant genes in gene significance vs. module membership correlation for female module blue.

(D) GO terms for female module midnightblue.

(E) Gene significance vs. module membership plot for female module midnightblue.

(F) Top 20 most extreme genes for gene significance vs. module membership correlation for female module midnightblue.

(G) GO terms for male module darkseagreen1.

(H) Gene significance vs. module membership for male module darkseagreen1.

(I) Top 20 most extreme genes for gene significance vs. module membership correlation for male module darkseagreen1.

(J) GO terms for male module turquoise.

(K) Gene significance vs. module membership for male module turquoise.

(L) Top 20 most extreme genes for gene significance vs. module membership correlation for male module turquoise. Size of circles represents number of GOterms combined into the parent GO term shown. Color of circles represents log₁₀ p value.

For the female module containing the highest number of DETs, blue, we identified GO terms related to “regulation of response to stimulus,” consistent with the identified IPA pathway “clathrin-mediated endocytosis signaling” (Dnm2, Clta, Ap2b1), which is upstream of actin polymerization. In addition, the module is enriched for key genes in the “IL-12 signaling” pathway (Map2k3, Mapk12, Nfkb1) involved in regulation of pro-inflammatory tumor necrosis factor (TNF) signaling as well as nitric oxide production.²⁷ Interestingly, these key genes are all downregulated in our dataset, indicating reduced inflammation in injured-repopulated microglia. Regulation of plasma lipoprotein levels was also highly represented by GO analysis, and IPA included “LXR/RXR activation” (Rxry, Apoe, Lyz2), a pathway that affects lipoprotein clearance and cholesterol homeostasis (Figure 5A and Extended Data 7).^28–30 For this module, six transcripts for the Apoe gene demonstrated positive significance (Figures 5B and 5C). In addition, “phagosome maturation” (Ctss, Ctsh, Tuba1a), an essential process following phagocytosis of debris and apoptotic cells, was identified by IPA (Extended Data 7). In line with this, a natural candidate identified in the blue module, Cd300lf, enhances phagocytosis³¹ (Figures 5B and 5C). Overall, this module suggests that repopulated microglia have an enhanced ability to clear neuronal debris following injury and suppress inflammation.

In the female module midnightblue, GO terms identified represented broad categories such as “macromolecule modification” and “positive regulation of biological processes” (Figure 5D). IPA highlighted more specific pathways and included “acute phase response signaling” (IL6st, Serpinf1, Trf), with key genes downregulated and involved in inflammatory IL-6 and TNF signaling (Extended Data 7). In line with this, Serpinf1, an identified natural candidate downstream of IL-6 signaling, was significantly downregulated (Figures 5E and 5F). Overall, this module represents reduced inflammatory pathway signaling, which may contribute to the pain-resolution effects of injured-repopulated microglia.

In the male module darkseagreen1, significantly enriched GO terms (Figure 5G) were related to “regulation of protein transport” and “protein localization,” with related enriched IPA pathways of “calcium signaling” (Arcan3, Prkaca, Asph), which can change protein location within the cell,³² and “biosynthesis of phosphatidylcholine and choline” (Pcyt1a), whose levels determine physical properties that affect fluidity and protein localization in the cell membrane (Figure 5G and Extended Data 7). One top 20 gene identified in darkseagreen1 was Prkaca, which encodes the catalytic subunit of protein kinase A. Other enriched GO terms are related to pre-mRNA processing (Wdr33), which directly controls translation in the cell. Another top 20 gene found in this module was Maf1, a transcriptional regulator (Figures 5H and 5I). Overall, the module darkseagreen1 emphasizes that these injured-repopulated microglia tightly regulate protein expression, localization, and function in the cell, in particular, the membrane.

In the male turquoise module, GO terms identified included “immune system process,” with IPA pathways such as “neuroinflammation signaling” (Irf7, Tlr2, Tnf, Cybb, Ccl2, Ccl5, IL12) and “TREM1 signaling” (Cd40, Itgax, Casp1, IL18, IL1β). In addition, identical to the female blue module, male turquoise was enriched for the GO term “regulation of response to stimulus,” with related IPA pathways “LXR/RXR activation” (Rxry, Apoe, Cd36, Il1rn) and “phagosome maturation” (Cybb, Ctss, Ctsh, Lamp1, Rab7) (Figure 5J and Extended Data 7). Specifically, the genes represented in this pathway are related to late phagosome maturation, when lysosomal fusion forms the phagolysosome. One of the top 20 genes in this module was Gm2a, a gene encoding a co-factor of lysosomal enzymes. Further, two cathepsins (Ctss and Ctsh) with relevance to pain³³ were in the top 20 gene list in turquoise (Figures 5K and 5L). Ctss and Ctsh proteins are enriched in actively phagocytic cells, have lysosomal activity, and modify neuronal spine density extracellularly.³⁴ The turquoise module in males captures several biological functions similar to those identified in the female modules of interest, pointing toward a sex-independent phenotype of repopulating microglia active in phagocytosis and stress responses.

To compare the shifts in gene expression in males and females captured by the WGCNA modules of interest, we plotted the fold change of transcripts from our selected modules in the injured-repopulated vs. injured-resident microglia comparison (Figure S6A) and observed a strong correlation between sexes (R² = 0.6802). In contrast, when the fold change comparison was expanded to include all transcripts expressed in both sexes (Figure S6B), there was no correlation (R² = 0.0714). This suggests that WGCNA identified a shared transcriptomic shift in repopulated microglia in both males and females that may contribute to the pain-resolution phenotype observed.

Human dorsal horn spinal microglia exhibit heterogeneous cell states

Since our overall goal was to identify microglial targets for pain relief in humans, we next isolated human spinal cord dorsal horn microglia to cross-reference their transcriptomes with the generated mouse datasets. We performed single-nucleus RNA-seq (snRNA-seq) on nuclei isolated from dorsal lumbar spinal cord from either a male or a female postmortem human donor with no known pain history (Table S2). We isolated dorsal horn nuclei and enriched for microglia using negative selection sorting to provide an unbiased sample of microglia nuclei for sequencing (Figure 6A). The male and female snRNA-seq datasets were then integrated, and uniform manifold approximation and projection (UMAP) of the 6,591 nuclei revealed seven clusters representing unique cell types. These clusters were defined using a human brain single-nuclei RNA-sequencing dataset³⁵ and included microglia, oligodendrocytes, oligodendrocyte precursor cells (OPCs), astrocytes, endothelial cells, neurons, and neuronal stem cells (NSCs) (Figures 6B and 6C). The negative selection strategy enriched microglial nuclei to ~40% of the sample, compared with ~3.7% collected from whole mouse spinal tissue in other studies³⁶ (Figure 6B). Microglial nuclei clusters expressed known marker genes described in previous studies, including ITGAM, CTSS, CSF1R, and C1QA (Figure 6D).²⁰ One small nuclei cluster in the upper-left quadrant of the UMAP was determined to have high expression of T cell markers (and lacked microglial gene expression) and was thus eliminated before further subclustering of microglia (Figures 6C and 6D). A total of 2,316 microglial nuclei (dashed box in Figure 6C) were reclustered using Seurat,³⁷ revealing six distinct microglia subclusters based on unique gene expression profiles (Figure 6E, Table S3, and Extended Data 8). GO analysis was then used to identify microglia subcluster cell states (Figure 6F and Extended Data 9). As expected from human postmortem tissue, we detected a small cluster, cluster 6 (97 nuclei), of dying microglia with the top GO term “cell death.” We also detected a small cluster, cluster 5 (81 nuclei), that represented proliferative microglia with top terms such as “cell cycle,” “DNA metabolic processes,” and “chromosome organization.” Notably, cluster 5 expressed cluster-unique gene MKI67 (adjusted p = 6.02E–134), the gene encoding Ki-67, as well as NDC80, TOP2A, and POLQ, each a prominent marker of proliferation (Figure S7A and Extended Data 8). We identified an additional subcluster, cluster 4 (81 nuclei), with a signature similar to that of reactive microglia, with GO terms such as “cell activation” and “regulation of cell motility” and a cluster-unique term, “inflammatory process,” with significant expression of key genes, such as TLR2 and CYBB (also known as NOX2), both potent neuroinflammatory-associated genes expressed by microglia^27,38 (Figure S7B). In addition, a top DEG in cluster 4 was AC083837.1 (adjusted p = 9.96E–105), a long noncoding RNA induced in a human microglial cell line (HMC3) in response to lipopolysaccharide (LPS) treatment³⁹ (Figure S7B). Importantly, we further identified a large nuclei cluster that represented homeostatic microglia (cluster 1, 730 nuclei), with cluster-unique terms “myeloid cell homeostasis” and “regulation of supramolecular fiber organization,” as well as “regulation of cell adhesion” and “cell junction organization.” Anti-inflammatory or homeostatic genes such as TGFBI, ITGAX, and IGF1R were significantly expressed with positive fold changes, while inflammatory-associated genes, such as TLR2 and CYBB, had negative fold changes in this homeostatic cluster 1 compared with other clusters (Figures S7B and S7C and Extended Data 8). The final two clusters (cluster 2, 857 nuclei; cluster 3, 470 nuclei) had both homeostatic and reactive features, with terms shared by homeostatic cluster 1, such as “immune system process” and “regulation of cell adhesion,” in addition to terms shared by clusters 4 and 6, such as “cell activation” and “cell death” (Figure 6F).

Figure 6. Single-nucleus RNA sequencing (snRNA-seq) of human microglia uncovers gene targets identified in repopulated microglia.

(A) Diagram depicting experimental flow of isolation and enrichment of microglial nuclei from human spinal cord dorsal horn gray matter and subsequent 10× Chromium sequencing.

(B) Percentages of cell-type-specific nuclei captured and sequenced, with microglia enriched to 40% of total nuclei.

(C) Uniform manifold approximation and projection (UMAP) plot showing distribution of human spinal cord single-nucleus clusters by cell type. Dashed box indicates final microglial clusters that were further subclustered in (E).

(D) Microglial-specific cell clusters are identified using known microglial gene markers.

(E) Six distinct microglial nucleus clusters were identified after subclustering using Seurat.

(F) Gene ontology (GO) analysis showing top 15 GO terms that capture differential microglia cell states.

(G) Dot plot of selected top mouse genes that were significantly differentially expressed between human microglial subclusters (clusters 1–6). Top genes that overlap in male and female mouse datasets are shown in the gray zone, top female genes in the purple zone, and top male genes in the green zone.

(H) UMAP plots showing genes significantly highly expressed in the human homeostatic cluster 1 and also upregulated in injured-repopulated mouse microglia. Adjusted p values from differential cluster-expression analysis are shown in parentheses below the gene name. Scale bar represents relative gene expression.

Identification of human-relevant homeostatic gene targets detected in repopulated mouse microglia

To identify human-relevant microglia-expressed genes found in our injured-repopulated microglia vs. injured-resident microglia mouse dataset, we first selected mouse transcripts that met a DESeq2 cutoff of p-adjusted < 0.05. We then selected for the transcripts within our top WGCNA modules of interest, ranked in descending order of their differential expression fold change and adjusted p value. Last, we mapped the transcripts to their corresponding genes, and the resulting mouse gene list contained 12 top female genes, 29 top male genes, and 30 overlap genes (Table S4). These mouse genes were then screened across the human microglial nucleus clusters to identify cluster-specific expression patterns. Twenty-one of these mouse genes were significantly differentially expressed across the human microglia clusters (adjusted p < 0.05) (Figure 6G). We identified one gene, KHDRBS3 (also known as SLM2), that had significantly lower expression, specifically in homeostatic human cluster 1, and was also downregulated in mouse injured-repopulated microglia. In addition, five genes (CD300LF, CD72, CTSS, RNF213, and GM2A) had significantly higher expression in human cluster 1, and no other cluster, indicating possible involvement in homeostatic functions in human microglia (Figures 6G and 6H). Interestingly, all these genes were upregulated in mouse injured-repopulated microglia (Table S4), and thus may serve as relevant microglial targets to promote spinal cord homeostasis and pain resolution.

DISCUSSION

In this study, we established the importance of microglia at multiple time points after pain-producing peripheral injury, identified transcriptomic signatures of spinal cord microglia associated with a pain-resolved state, and screened top gene candidates with those expressed by microglia subclusters obtained from human spinal cord transcriptional profiling. We applied a genetic depletion strategy using Cx3CR1-Cre^ERT2-eYFP;R26-iDTR^LSL mice to temporally and selectively deplete microglia without affecting peripheral circulating monocytes/macrophages.^7,12 We find that depletion of microglia at the acute-to-chronic pain transition (at 3 weeks post-injury) is sufficient to resolve existing allodynia and thermal hyperalgesia in both sexes. Importantly, the onset of pain resolution coincided with spinal cord microglial repopulation, not microglia depletion. Furthermore, persistent microglia depletion from week 3 to 7 post-injury using the drug PLX3397 did not reverse the pain trajectory, indicating that microglial loss is not sufficient for pain resolution. We performed RNA-seq followed by bioinformatic analysis using DESeq2 and WGCNA. WGCNA is a powerful technique that has historically been applied to population-level studies of publicly available datasets.^40,41 Our study is unique in applying WGCNA in the context of pain transcriptomic analysis of microglia. Using this in-depth, unbiased approach, we identified both sex-dependent and independent microglial transcriptional signatures related to the uninjured, injured, or pain-resolved states. Given the limitations of using mice as a model for human disease, we next generated an snRNA-seq dataset from human dorsal horn lumbar spinal cord microglial nuclei. We characterized the heterogeneity of microglia in this anatomically limited region, including six subclusters with unique transcriptomes, some of which shared similarities with previously identified subpopulations in the brain.⁴² Finally, we cross-referenced candidate mouse microglia genes linked to the pain-resolved state with this human snRNA-seq dataset and identified genes of high translational interest based on their subcluster expression profiles.

Microglia are canonically associated with the acute phase/initiation of pain after peripheral injury²; our data using genetic microglia depletion suggests they may also contribute to the transition from acute to chronic pain and perhaps even to pain resolution. Previous studies have demonstrated that microglial activation is necessary⁸ and sufficient⁴³ to initiate acute pain. Microglia depletion studies using pharmacologic or genetic tools show that microglia also contribute to the initiation of pain progression in models of nerve injury.^9–11 We find that preventive depletion of microglia (at the time of injury) was insufficient to completely reverse allodynia. However, the extent of improvement was similar in magnitude to that observed after early microglia depletion in a high-frequency stimulation model of pain² and in spinal nerve transection.¹¹ This lack of full recovery may be due to nociceptor hyperactivity or high levels of acute nociceptor-derived inflammatory signals acting on the spinal cord environment immediately post-injury.⁴⁴

Of note, consistent with our own findings, several studies^2,11,45,46 reported equivalent improvements in pain outcomes in both sexes with microglial manipulation. Taken together, these results suggest that microglia themselves are unlikely to be sex-specific contributors to pain progression, but rather microglial-expressed receptors such as TLR4^7,47 or chemokines such as CCR2² may be sex specific.

The current work temporally extends the involvement of microglia to the acute-to-chronic phase, with microglia depletion at 3 weeks post-injury significantly and persistently improving the pain trajectory to baseline levels. In contrast, Peng et al.¹¹ demonstrated that early depletion of microglia delayed the development of allodynia, while late depletion of microglia only transiently improved allodynia. The reason for this discrepancy may relate to the type of pain model used (direct nerve injury vs. fracture and immobilization) or the exact timing of microglia depletion (week 1 vs. week 3 post-injury). In contrast to the commonly used nerve-injury models,^48,49 the tibial-fracture model of CRPS exhibits a clearly timed transition from acute to chronic pain and triggers both inflammatory and neuropathic mechanisms, mimicking the clinically relevant human condition.^24,50 Interestingly, microglia depletion at the acute-to-chronic transition was also sufficient to reduce peripheral inflammation as evidenced by decreased paw edema, even though macrophages were not depleted. This suggests a role for microglia in neurogenic inflammation. Previous research has shown that sensitized afferent neurons can signal from the spinal cord and modulate peripheral edema and inflammation.⁵¹ Microglia may contribute to this response through the release of neuromodulators such as adenosine, cytokines, and chemokines that lead to neuronal sensitization.⁵²

In our study, pain resolution after microglial depletion at the acute-to-chronic phase coincided with the time frame of complete repopulation of CNS microglia and did not occur with persistent microglia depletion. This outcome suggests that repopulated microglia may reestablish homeostasis within the pain circuit. Indeed, newly repopulated microglia can function to repair synapses in the hippocampus as well as promoting recovery following brain injury.^53,54 These cells also retain the ability to respond to environmental cues, and our WGCNA analysis highlights that they remain highly active cells, engaged in surveillance and phagocytosis.^55,56 The possibility that repopulated microglia exhibit a phenotype that is distinct from resident microglia is further supported by transcriptome analyses from prior published work showing clear differences between repopulated and “control” brain microglia.^23,56 One major finding of the current work is that the environment (either injured or uninjured) in which microglia repopulate also influences their phenotype. This is perhaps unsurprising, given that microglia are dynamic, multidimensional cells constantly surveying their environment and responding to stimuli with a range of state changes.^57,58 An urgent question remains regarding whether these heterogeneous microglial states may have functional significance in disease progression or resolution, in particular, for pain.

Since microglia repopulation did not result in resensitization of the periphery, and repopulated microglia were morphologically distinct from control microglia, our model allowed for identification of genes with potential to contribute to pain resolution. The intriguing possibility that specific populations of microglia may contribute to pain resolution is supported by a seminal study characterizing CD11c⁺ (Itgax) spinal cord microglia as necessary and sufficient to the resolution of pain after peripheral nerve injury in both sexes.⁴⁶ Aside from high expression of Itgax, Igf1, and Apoe, this antinociceptive subpopulation also highly expressed the receptor tyrosine kinase Axl. Importantly, we also find upregulation of Itgax, Apoe, and Axl in both male and female repopulated microglia isolated from the pain-resolution group, suggesting that these repopulated microglia may share properties with the pain-resolving microglia identified in the above study. Another microglia-expressed gene that has received recent attention is Apoe, which is involved in trafficking and metabolism of cholesterol and lipoproteins. Apoe was elevated in resident microglia after injury²⁰ as well as disease-associated microglia in other contexts.^59,60 The elevated level of Apoe in a diversity of microglial states, spanning reactive to resolving, suggests the gene may in fact be a nonspecific marker for highly active rather than disease-promoting microglia, especially in the context of pain. In support of this interpretation, we detected widespread expression of APOE in our human microglial nucleus clusters, a finding supported by a recent human spinal cord sequencing study.⁶¹ Additional studies directly testing the function of Apoe in microglia in the context of pain will be necessary to implicate it as a pain-regulating gene.

For translatable therapies to be developed, it is necessary to validate mouse-identified gene targets using human tissue samples. To address this, we profiled single nuclei of human spinal microglia taken specifically from the lumbar gray matter (dorsal horn). We thus focused on microglia in a pain-relevant region, which is crucial, given that gene signatures and cellular states are defined by the microglial environment.^21,62,63 We performed negative selection (removing astrocytes, neurons, and oligodendrocytes) to enrich the sample 10-fold for microglia, while also capturing those not expressing canonical positive-selection markers (e.g., Iba1 or Cx3CR1). This enhanced our ability to characterize human microglial states, which are known to be more heterogeneous compared with other species, including the mouse.⁶⁴ Using this region-specific, negative-selection approach, we uncovered six microglial clusters that were spatially distinct by UMAP and displayed unique gene-expression patterns. Of note, we observed one large spatially distinct cluster by UMAP (cluster 1) that contained homeostatic gene expression signatures and lacked reactive ones.

To concentrate our gene identification on those with human relevance, we screened top candidate mouse genes for expression within human microglia clusters. We identified several genes significantly upregulated in repopulated mouse microglia from the pain-resolution group, also significantly, and specifically, expressed by human homeostatic cluster 1. CD300LF (the human conserved ortholog of the mouse gene cd300f⁶⁵) encodes a member of the CD300 cell-surface glycoprotein family that functions to positively regulate phagocytosis of apoptotic cells. This function is increased in the spinal dorsal horn microglia subpopulation shown to promote pain resolution following injury.^31,46 Specifically, CD300LF encodes an inhibitory receptor for myeloid cells that negatively regulates TLR signaling via MYD88.⁶⁶ In addition, overexpression of CD300LF has a neuroprotective effect after acute brain injury,⁶⁷ and microglia isolated from mice resistant to cerebral malaria express high levels of cd300lf and inhibit inflammation.⁶⁸ Another gene identified was CD72, whose protein product is a receptor for semaphorin 4D and contributes to microglial-mediated CNS inflammation.^69,70 Last, we identified GM2A, which encodes a small, secretable glycolipid transport protein and stimulates the degradation of ganglioside GM2 in the lysosome. Elevated levels of GM2A in culture reduce neurite extension and neuronal firing rate, which may be beneficial in the context of pain and dorsal horn circuitry.⁷¹

In this study, we took advantage of a pain-resolution context to profile repopulated spinal cord microglia and identify genes that may promote homeostatic features of microglia and ultimately promote pain recovery after injury. Future studies will evaluate whether these repopulated microglia directly contribute to the resolution of pain, similar to previously described CD11c⁺ microglia.⁴⁶ Our findings that repopulated microglia are transcriptomically distinct from resident microglia after injury suggest they contribute in a unique manner to the spinal cord circuit environment. Interrogating the individual genes identified in this study that overlap with human microglia subpopulations, many of which have no known function in microglia, will be an exciting next step to confirm their potential as therapeutic targets for analgesic development.

Limitations of the study

The main limitation of this study is that we did not pursue further confirmation of identified microglial pro-resolution genes as contributors to pain. These important studies will be the subject of future work. We performed the isolation of microglia for sequencing studies from whole lumbar spinal cord, not only the injured dorsal horn. As a result, it is possible that microglia from the ventral horn, or contralateral (uninjured) side, may have attenuated the number of DEGs between groups. Importantly, this would result in detection of significance only in the subset of genes with larger-magnitude differences, which may ultimately better represent genes with essential contributions to the pro-resolution microglial phenotype. Regarding the influence of sex on the contribution of microglia to pain, we have fully addressed this in our mouse work, which was powered independently by sex. For confirmation of microglial genes in human samples, we were limited to just one male and one female sample and therefore cannot address sex similarities or differences in this portion of the study.

STAR★METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Dr. Vivianne Tawfik, vivianne@stanford.edu

Materials availability

This study did not generate any new or unique reagents.

Data and code availability

• Transcriptomic data are included as Extended Data Tables. All transcriptomic data have been deposited as raw FASTQ files in GEO database accession number (GSE249210). All other data supporting the findings of this study are available from the corresponding author upon reasonable request.

• No original code was generated for this study. All code used for analysis was properly cited in the Methods.

• Any additional information required to reanalyze the data reported in this work paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Mice

Adult male and female mice 10–12 weeks old at the start of the experiments were housed 2–5 per cage and maintained on a 12-hour light/dark cycle in a temperature-controlled environment with ad libitum access to food and water. Male mice weighed approximately 25 g at the start of the study, and female mice weighed approximately 20 g at the start of the study. Mice used in this study: wild type C57BL/6J mice (Jax stock #00664), CX3CR1CreERT2-EYFP (Jax stock #021160),¹² Rosa26-loxP-stop-DTR (R26iDTR, Jax stock #007900). To specifically and conditionally ablate central CX3CR1 cells (microglial depletion), we crossed homozygous CX3CR1CreERT2-EYFP with homozygous R26iDTR/iDTR mice to generate double heterozygous Cx3CR1-CreERT2-eYFP;R26-iDTRLSL mice. Mouse genotypes from tail samples were determined using real time RT-PCR with specific probes designed for each gene (Transnetyx, Cordova, TN). For extended microglia depletion, C57BL/6J wildtype male mice from 10–12 weeks were used (Jax stock #00664).

Human samples

Samples for the human nuclei sorting were from human donor lumbar spinal cord tissue. The tissue was obtained from one Caucasian male donor (age 43) and one Asian female donor (age 49), both of whom died from cardiovascular events.

Study approval

All procedures were approved by the Stanford University Administrative Panel on Laboratory Animal Care and the Veterans Affairs Palo Alto Health Care System Institutional Animal Care and Use Committee in accordance with American Veterinary Medical Association guidelines and the International Association for the Study of Pain. Human post-mortem spinal cord was obtained in collaboration with Donor Network West and received Stanford University Institutional Review Board exemption.

METHOD DETAILS

Drugs and route of administration

Tamoxifen (Sigma, #T5648) was dissolved in corn oil at a concentration of 25 mg/mL. Mice were injected intraperitoneally (i.p.) at 100 mg/kg daily for 5 days. Diphtheria toxin (DT) from Corynebacterium diphtheriae (Sigma, #D0564–1MG) was dissolved in sterile water to a concentration of 10 mg/mL and 0.1 mL (1000 ng) was administered i.p. daily for 3 days. Control subjects for all behavioral studies were mice without the R26^iDTR allele: Cx3CR1-Cre^ERT2-eYFP; R26-iDTR^+/+ that received tamoxifen (TAM) and diphtheria toxin (DT) with the same dosing and administration schedule as experimental mice. For extended microglia depletion, mice were fed AIN-76A rodent diet formulated by Research Diets to contain 75 mg PLX3397/kg (low PLX, Chemgood) or 660 mg PLX3397/kg (high PLX),⁸⁴ starting immediately after uncasting (week 3 post-injury). Mice were maintained on this diet until the end of the experiment. Mice were monitored for the first 24 hours to ensure normal food consumption. Control mice were fed 2018 Teklad Global 18% Protein Rodent Diets.

Tibial fracture/casting model of complex regional pain syndrome (CRPS)

Mice were anesthetized with isoflurane and underwent a closed right distal tibia fracture followed by casting. The right hind limb was wrapped in gauze and a hemostat was used to make a closed fracture of the distal tibia. The hind limb was then wrapped in casting tape (ScotchCast^™ Plus) from the metatarsals of the hind paw up to a spica formed around the abdomen to ensure that the cast did not slip off. The cast over the paw was applied only to the plantar surface with a window left open over the dorsum of the paw and ankle to prevent constriction when post-fracture edema developed. Mice were inspected throughout the post-operative period of cast immobilization to ensure that the cast was properly positioned. At 3 weeks post-fracture, mice were briefly anesthetized, and casts were removed with cast shears. For behavioral assessment, mice were tested beginning three days after cast removal at 3 weeks until 9 weeks post-fracture, as indicated in each section below. CRPS model generation and behavioral testing were conducted following well established methods for evaluating mouse behavior in the tibial fracture-casting model of CRPS.^85,86

Behavioral testing

To ensure rigor in our findings and avoid the contribution of experimenter sex to our behavioral data, experimenters were female or there was a female scientist’s lab coat in the room during acclimation and testing.⁸⁷ In vivo behavioral testing was performed in a blinded fashion and mice were allocated to experimental group by cage for all behavioral assays. All testing was conducted between 7:00 am – 1:00 pm in an isolated, temperature- and light-controlled room. Mice were acclimated for 30 – 60 minutes in the testing environment within custom clear plastic cylinders (4” D) on a raised metal mesh platform (24” H). Mice were randomly placed in a cylinder; after testing, mouse identification numbers were recorded on the data sheet.

Mechanical nociception assays

To evaluate mechanical reflexive hypersensitivity, we used a logarithmically increasing set of 8 von Frey filaments (Stoelting), ranging in gram force from 0.007 to 6.0 g. These were applied perpendicular to the plantar hind paw with sufficient force to cause a slight bending of the filament. A positive response was characterized as a rapid withdrawal of the paw away from the stimulus filament within 4 s. Using the up-down statistical method,⁸⁸ the 50% withdrawal mechanical threshold scores were calculated for each mouse and then averaged across the experimental groups. Mechanical nociception testing was performed at Weeks 3, 4, 5, 7, and 9 post-fracture.

Thermal nociception assays

To evaluate thermal-induced reflexive responses, we used the hotplate test (plate temperature was set to 52.5°C). Mice were placed on the plate and the latency (seconds) to the first appearance of a reflex response (flinch) was recorded as a positive reflex withdrawal response. A maximal cut-off of 45 s was set to prevent tissue damage. Only one exposure to the hotplate was applied on a given testing session to prevent behavioral sensitization that can result from multiple noxious exposures. Measurements were made at 5 weeks post-fracture.

Unweighting

An incapacitance device (IITC Inc Life Science, Woodland Hills, CA) was used to measure hind paw unweighting. Mice were manually held in a vertical position over the apparatus with the hind paws resting on separate metal scale plates, and the entire weight of the mouse was supported on the hind paws. The duration of each measurement was 6 seconds, and 6 consecutive measurements were taken at 60-second intervals. Six readings were averaged to calculate the bilateral hind paw weight-bearing values. Unweighting was measured at baseline and then again at 4 weeks post-fracture. Data were analyzed as a ratio between the right hind paw weight divided by the sum of right and left hind paw values [2R/(R + L)] × 100%).⁸⁹

Paw edema

Hind paw edema was determined by measuring the hind paw dorsal-ventral thickness over the midpoint of the third metatarsal with a LIMAB laser measurement sensor (LIMAB, Goteborg, Sweden) while the mouse was briefly anesthetized with isoflurane. Temperature and hind paw thickness data were analyzed as the difference between the fracture side and the contralateral intact side and averaged across experimental groups. Paw edema was measured at 4 weeks post-fracture. Experimenters were blinded to study group.

Temperature measurements

The temperature of the hind paw was measured using a fine-gauge thermocouple wire (Omega, Stamford, CT). Temperature testing was performed over the hind paw dorsal skin between the first and second metatarsals (medial), the second and third metatarsals (central), and the fourth and fifth metatarsals (lateral). The measurements for each hind paw were averaged for the mean paw temperature. Data were expressed as the average difference between the ipsi- and contralateral hind paw within an experimental group. Paw temperature was measured at 4 weeks post-fracture. Experimenters were blinded to study group.

Basso Mouse Scale for locomotor activity

We performed in-depth analysis of motor function of mice after microglia depletion using the Basso Mouse Scale (BMS) which evaluates multiple aspects of locomotion including plantar stepping, coordination, paw position, trunk stability and tail position. Two researchers, blinded to genotype, observed mice in an open field and assessed the score, per described criteria.⁹⁰

Immunohistochemistry

Mice (12 – 30 weeks) were transcardially perfused with 10% formalin in PBS. The spinal cord (lumbar cord L3 – L5 segments) were dissected from the mice and cryoprotected in 30% sucrose in PBS and frozen in O.C.T. (Sakura Finetek, Inc.). Spinal cord sections (40 μm) were prepared using a cryostat (Leica Biosystems) and incubated in blocking solution (5% normal donkey serum and 0.3% Triton X-100 in PBS) for 1 h at room temperature followed by incubation with primary antibodies at 4°C, overnight. The following primary antibodies were used: rat anti-CD11b (Bio-Rad, #MCA711G, 1:500), and rabbit anti-Iba1 (Wako, #019–19741, 1:500). Tissues were washed with 1X PBS 3 times and sections were incubated with appropriate secondary antibody conjugated to AlexaFluor (1:1000) in 1% normal donkey serum, 0.3% Triton X-100, and 1X PBS for 2 h at room temperature. Sections were mounted with Fluoromount G with DAPI medium (ThermoFisher, #00–4959-52). Images were collected with a Keyence BZ-X800 fluorescent microscope (Keyence) using the sectioning module to remove non-focused light using 20x (NA 0.75) or 60x (NA 0.95) objectives. Four to five lumbar spinal cord dorsal horn sections were counted in a blinded manner and averaged for 2–3 mice per group.

Sholl analysis for microglial morphology

Lumbar spinal cord was sliced on a cryostat at 40 μm and microglia were labeled by immunohistochemistry using the rabbit anti-Iba1 antibody (Wako, #019–19 741, 1:500). Dorsal horn localized microglia were imaged using a Keyence BZ-X800 fluorescent microscope (Keyence) using the sectioning module (1D Slit method) to remove non-focused light using a 40X objective. Each z-stack was taken at the same exposure using 43 slices with a 0.3 μm step size. Images were imported into ImageJ/FIJI⁷² and maximum intensity projections of dorsal horn images were generated. Region of interest (ROI) was randomly selected around in focus individual microglia. Unsharp Mask was applied using set radius to 3 pixels and weight of 0.6 followed by the de-speckle tool was to remove non-specific background. The images were made binary and ‘remove outliers’ was used at threshold of 50. The Eraser tool was then used to manually eliminate Iba1+ processes within the frame that were clearly not associated with the microglial cell analyzed. Between 27–92 microglia per animal per time-point (Baseline: 185 microglia, Day 1: 74 microglia, Day 3: 117 microglia, Day 14: 87 microglia, Day 28: 56 microglia) were analyzed by Sholl analysis plugin in ImageJ/FIJI with a start radius of 0.5 μm and a step size of 1 μm. All images were analyzed in the same manner.

Spinal cord cell dissociation

Mice were anesthetized with 120 mg/kg ketamine and 5mg/kg xylazine and perfused with 15 mL ice-cold Medium A (50 mL 1x HBSS without Ca2+ or Mg2+ (Gibco, 14185052), 750 μL 1M HEPES (Gibco, #15630080), 556 μL 45% glucose (Corning, #25–037-CI)). Lumbar spinal cords were isolated and placed in 2 mL Medium A + 80 μL Dnase I (12,400 units/mL, Sigma, 11284932001) until all samples were dissected. Samples were dounce-homogenized, passed through a 100 μm strainer, washed with 5 mL Medium A, and spun down by centrifugation at 340 × g for 7 minutes at 4°C. Supernatant was removed, and pellets were resuspended in 6 mL 25 % Standard Isotonic Percoll (Percoll: GE Healthcare, #17–5445-02 with 10% 10x PBS) in Medium A. The suspension underwent centrifugation for 20 minutes at 950 × g at 4°C to remove myelin. Supernatant was discarded then washed with 5 mL Medium A and spun down by centrifugation at 340 × g for 7 minutes at 4°C. Cells were resuspended in FACS buffer (5 mM EDTA in 1% BSA in 1x PBS).

Flow cytometry

The following antibodies were used for flow cytometry: e450-conjugated anti-CD3 (Thermo Fisher Scientific Inc., #48–0031-82), e450-conjugated anti-CD19 (Thermo Fisher Scientific Inc., #48-0193-82), PE-Cy7-conjugated anti-CD45 (Biolegend, #103114), APC-conjugated anti-CD11b (Biolegend, #101212). Dead cells were stained with 1:1000 SYTOX Blue (Invitrogen, #S34857). Cell suspension was then spun down at 300 × g for 5 minutes at 4°C. The FACS buffer supernatant was suctioned off, and cells were resuspended in fresh FACS buffer. Before staining, all samples were pre-blocked with 1:100 anti-CD16/CD32 (BD Pharmingen, #553142) for 5 minutes at room temperature. Afterward, all antibodies were added to the samples at 1:100 along with 1:1000 Sytox Blue and placed on ice for 30 minutes. Samples were spun down for 5 minutes at 400 × g at 4°C, suctioned and replaced with fresh FACS buffer. Suspension was then spun down for 5 minutes at 400 × g at 4°C. Supernatant was suctioned off, then cells were again resuspended in FACS buffer and passed through 35 mm filter into polystyrene tubes. Samples were sorted using a BD Aria II (BD Biosciences). Flow cytometry analysis was done using FlowJo v10.6.2. Single cells were gated for live (SYTOX blue negative), CD19− CD3− CD45^mid then CD11b⁺ and Cx3CR1-YFP⁺ to isolate microglia from spinal cord.

Preparation of samples for sequencing

For male samples, whole RNA was isolated from FACS microglia using the Rneasy^® Plus Micro kit (Qiagen, Cat No. 74034) then cDNA was synthesized using the SMART-Seq^® v4 Ultra^® Low Input RNA kit (Takara Bio USA Inc. #634888). Libraries were prepared using the Nextera^™ DNA Flex Library Prep (Illumina Cat No. 20015828) and sequencing performed by the Stanford Genomics Service Center on a HiSeq4000 using a paired end 2×150bp unstranded protocol. For the female samples, microglia were isolated by FACS into Trizol LS (Invitrogen, #10296028) and sent to Azenta Life Sciences (formerly Genewiz) to isolate total RNA using a routine protocol with glycogen added in the precipitation step followed by sequencing on an Illumina HiSeq using paired end 2×150bp unstranded protocol.

DESeq2 analysis

Raw sequencing data were analyzed using the nf-core/rnaseq v3.5 RNA-Seq analysis pipeline (https://nf-co.re/rnaseq)⁷⁶ with nf-core helper tools v2.2 (https://github.com/nf-core/tools)⁷⁶ which was run using STAR and RSEM with isoform (transcript) counts and extensive quality control. The Nextflow v22.01.0 workflow tool (https://github.com/nextflow-io/nextflow)⁷⁷ facilitated the running of various pipeline tasks across a compute cluster in a very portable manner. FastQC v0.11.9 (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/)⁷⁸ was used to run quality control checks on raw FASTQ sequence data. The wrapper tool Trim Galore v0.6.7 (https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/)⁷⁹ that employs Cutadapt v3.4⁸⁰ and FastQC v0.11.9 was run to consistently apply quality and adapter trimming to FASTQ files. Alignment to the GENCODE GRCm38 release M25 reference was done using STAR v2.7.6a and the aligned transcripts were quantitated by RSEM v1.3.1 (http://deweylab.github.io/RSEM/).⁸¹ MultiQC v1.11 (https://multiqc.info/)⁸² helped to aggregate results from the various bioinformatics analyses across the many samples into a single report. Differentially expressed transcripts were detected from raw transcript counts produced by the STAR-RSEM pipeline using DESeq2 v1.34.0 (https://bioconductor.org/packages/release/bioc/html/DESeq2.html)⁸³ running in R v4.1.3.

Weighted gene co-expression network analysis

Weighted gene co-expression network analysis (WGCNA) was used to identify clusters (modules) of highly correlated genes. Uninjured-Resident, Injured-Resident, Uninjured-Repopulated, and Injured-Repopulated replicates from female and male mice were analyzed using WGCNA v1.70.3 package in R v4.1.3.⁷⁵ Variance stabilization transformed RNA-Seq expression data were loaded and genes and samples with too many missing values, if any, were removed. Samples were then clustered based on their Euclidean distance to see whether there were any obvious outliers. Traits of interest including Uninjured-Repopulated microglia vs. Uninjured-Resident microglia, Injured-Repopulated microglia vs. Uninjured-Repopulated microglia, Injured-Repopulated microglia vs. Injured-Resident microglia, and Injured-Resident microglia vs. Uninjured-Resident microglia were visualized to see how they related to the sample dendrogram. A step-by-step network and module detection approach was then employed. The soft thresholding power to which co-expression similarity was raised to calculate adjacency was chosen based on the criterion of approximate scale-free topology. Adjacencies were calculated based on the chosen soft thresholding power. To minimize effects of noise and spurious associations, the adjacency was transformed into Topological Overlap Matrix (TOM) following which the corresponding dissimilarity was calculated. Hierarchical clustering was employed to produce a dendrogram of genes. The Dynamic Tree Cut R library that implements dynamic branch cutting methods was used to detect clusters (modules) of highly co-expressed genes in the dendrogram. Modules whose expression profiles were very similar were merged after calculating eigengenes of modules and clustering them on their correlation. A cut height of 0.25 corresponding to a correlation of 0.75 was used to merge modules and assign module colors. In the next step modules that were significantly associated with above-mentioned traits of interest such as Uninjured-Repopulated microglia vs. Uninjured-Resident microglia, Injured-Repopulated microglia vs. Uninjured-Repopulated microglia, Injured-Repopulated microglia vs. Injured-Resident microglia, and Injured-Resident microglia vs. Uninjured-Resident microglia were identified by quantifying the module-trait relationships by correlation values and p-values. Next, associations of individual genes with each trait of interest were quantified by defining gene significance (GS) as the absolute value of the correlation between the gene and the trait. For each module, a quantitative measure of module membership (MM) was defined as the correlation of the module eigengene and the gene expression profile which allowed quantification of the similarity of all genes to every module. The GS and MM measures were used to identify genes that have a high significance for each trait of interest as well as high module membership in interesting modules. Finally, a table was created to show gene annotation, module color, gene significance for traits of interest with p-values, and module membership with p-values in all modules.

Gene ontology and signaling pathway analysis of gene sets

Gene sets of interest were analyzed via the g:profiler website⁷³ to identify enriched Gene Ontology Terms and KEGG Pathways based on a threshold of Benjamini-Hochberg FDR > 0.05. Gene Ontology terms were visualized with GO-Figure! Version 1.0.1⁷⁴ utilizing the 2021-02-01 release of the Gene Ontology (http://release.geneontology.org/2021-02-01/index.html) and the 2022-03-02 release of the UniProt GO Associations.⁹¹ The pathway analysis was executed through the use of QIAGEN IPA (QIAGEN Inc., https://digitalinsights.qiagen.com/IPA).⁹²

Isolation of postmortem human microglia nuclei

Human lumbar spinal cord was removed immediately after organs were removed for transplant from either male or female donor (see Table S2). The lumbar segment was further dissected into three smaller segments, and immediately frozen on dry ice and stored at −80°C in Eppendorf or conical tubes. Lumbar spinal cord was then embedded in OCT and sectioned at 80 μm thickness (40–50 total sections per sample was used). Dorsal horn was collected by microdissection and immediately homogenized in Nuclei Lysis buffer on ice (10 mM Tris-HCl pH 8.0, 250 mM sucrose, 25 mM KCl, 5 mM MgCl2, 0.1% Triton-x 100, 0.5% Rnasin^® Ribonuclease Inhibitors (Fisher/Promega, #PRN2615), 1X Protease inhibitor, 0.1 mM DTT). Samples were then filtered through a 40 μm nylon cell strainer (BD#352340) pre-wet with Nuclei Lysis buffer in a 15 mL conical tube and another 2 mL of lysis buffer was used to rinse the filter. The homogenate was then centrifuged at 900 × g for 10 min at 4°C. The resulting pellet was resuspended in 6 mL of sterile 25% Standard Isotonic Percoll (SIP) to remove myelin and centrifuged at 2000 × g for 20 min at 4°C without brake. The myelin layer was discarded and the nuclei pellet was resuspended in 1 mL of Staining buffer (1XPBS, 0.8%BSA, 0.5% Rnasin plus). The nuclei suspension was transferred to an Eppendorf tube and centrifuged at 900 × g for 10 min at 4°C. Supernatant was discarded and the pellet was resuspended in Staining buffer containing the Human BD Fc Block^™ antibody (BD, 564220) and incubated with agitation for 15 min at 4°C in a cold room. Samples were then incubated with rabbit anti-NeuN Alexa Fluor^® 647 (1:1000; Abcam, #ab190565), rabbit anti-Sox10 Alexa Fluor^® 647 (1:5000; Abcam, ab270151), rabbit anti-Sox9 Alexa Fluor^® 555 (1:1000; Millipore, #AB5535-AF555), and DAPI (1:1000, Thermo Fisher Scientific Inc, #62248). Samples were incubated with agitation for 30 mins at 4 °C (protected from light) and centrifuged at 900 × g for 10 minutes at 4°C. Samples were then resuspended in 1mL Staining buffer and centrifuged at 900 × g for 10 minutes at 4°C. Nuclei were resuspended in 400–500 μl of Staining buffer and filtered through the lid of 35mm BD FACS tube before sorting on a BD Aria II (BD Biosciences) machine. DAPI only positive nuclei were sorted away from nuclei co-positive with any other signal to negatively select for and enrich nuclei sample for microglia. Nuclei were then centrifuged at 200 × g for 1 min using gentle brake and acceleration and repeated two additional times. Supernatant was removed to leave ~60 μl to resuspend nuclei.

Single-nuclei RNA sequencing

Nuclei were counted on a hemocytometer and estimated nuclei/μl were calculated for loading onto a 10X Genomic Chromium single cell chip (10x Genomics). Reverse transcription and subsequent library preparations were performed using the Chromium Next GEM Single Cell 3ʹ Kit v3 by the Stanford Functional Genomics Facility per manufacturer’s instructions. Samples were then sequenced to an average depth of 97,000–120,000 reads per nuclei on an Illumina NovaSeq6000 sequencer.

snRNA-seq data processing

Cell Ranger (version 6.0.0) was used for barcode processing, unique molecular identifiers filtering, gene counting, and sample mapping to the reference transcriptome (GRCh38). The filtered UMI count matrix was provided from Cell Ranger as an input in the Seurat workflow (version 4.0).³⁷ Nuclei with less than 200 and greater than 6,000 (for the male sample) or 8,000 (for the female sample) detected genes based on read depth were filtered out to eliminate low quality nuclei and multiplets. Nuclei with more than 5% mitochondrial genes detected were further removed. After filtering, a total of 6,591 nuclei were obtained, 2,782 nuclei for the male sample and 3,809 nuclei for the female sample.

We applied the anchor-based strategy in the Seurat workflow to integrate samples from different batches. To merge two samples, both male and female count matrices were log-normalized in Seurat’s function ‘NormalizeData’ and the top 2,000 shared highly variable genes were found using the function ‘FinderVariableFeatures’ with “selection.method = ‘vst”’. The integration anchors were then identified using the canonical correlation analysis from the function ‘FinderIntegrationAnchors’. Based on the anchors, two datasets were integrated using the function ‘IntegrateData’.

Principal component analysis (PCA) was performed on the integrated dataset. Based on the first 30 principal components (PC), the uniform manifold approximation and projection (UMAP) was used for dimension reduction. A shared nearest neighbor (SNN) graph was built with the first 30 PCs using Seurat’s function ‘FindNeighbors’. The nuclei were clustered using a Louvain algorithm applied by the function ‘FindClusters’.

Identification of microglia nuclei

Cell types were annotated based on the reference dataset³⁵ using Seurat’s label-transferring approach. The identified nuclei clusters were comparable to the unsupervised clustering results under the resolution parameter 0.04. We compared the clusters from reference-based annotation and unsupervised clustering analysis to determine the subset of microglia-specific nuclei.

We found a sub-cluster of the reference-annotated microglia (cluster 4 from the unsupervised clustering results) highly expressed T cell signature genes (THEMIS, SKAP1). These nuclei were excluded in further analysis. The final microglia nuclei were selected as the intersection of the reference-annotated microglia cells and the nuclei from clusters 1 and 3 in unsupervised clustering analysis (boxed in Figure 6C). It was confirmed that microglia marker genes (ITGAM, CTSS, CSF1R, C1QA) were highly expressed in these clusters (boxed in Figures 6C and 6D). Using these criteria, 2,316 nuclei were classified as microglia and isolated for further analysis.

Sub-cluster annotation for microglia nuclei

Using the isolated microglia nuclei, we applied the unsupervised clustering analysis under a series of resolution parameters to investigate biological meaningful subclusters. Microglia were clustered using resolution 0.06 to form 4 clusters. Based on differential gene expression patterns, cluster 0 was further sub-clustered into three additional clusters (cluster 2, 3 and 6). After this regrouping, we ended with 6 total microglia nuclei subclusters (Figure 6E). Differential expression analysis was performed from the function ‘FindAllMarkers’ in the Seurat package with filtering parameters ‘min.pct = 0.25’ and ‘logfc.threshold = 0.25’.

QUANTIFICATION AND STATISTICAL ANALYSIS

Cohort sizes were determined based on historical data from our laboratory using a power analysis to provide >80% power to discover 25% differences with p<0.05 between groups to require a minimum of 4 animals per group for all behavioral outcomes, and 2 animals per group for Sholl analyses with >30 microglia counted per animal per time-point. All experiments were randomized by cage and performed by a blinded researcher. Researchers remained blinded throughout histological, biochemical, and behavioral assessments. Groups were unblinded at the end of each experiment before statistical analysis. Data distribution was assumed to be normal but this was not formally tested. Data are expressed as the mean ± SEM. Statistical analysis was performed using GraphPad Prism version 8.4.1 (GraphPad Software) or R, as described in STAR Methods. Data were analyzed using a Student’s t tests, or ordinary one-way with Dunnett’s or Tukey’s post-hoc test, or two-way analysis of variance with a Bonferroni post-hoc test, as indicated in the main text or figure captions, as appropriate, with complete statistical analyses detailed in Table. The “n” for each individual experiment is listed in the figure legends. No data were excluded from analyses.

Extended Data

Extended data Table 1.

Differentially expressed transcripts and genes for each comparison in both sexes.

NIHMS1970887-supplement-Extended_data_1.xlsx ^{(105.3KB, xlsx)}

Extended data Table 2.

Gene ontology terms enriched by differentially expressed genes for each comparison in females.

NIHMS1970887-supplement-Extended_data_2.xlsx ^{(215.4KB, xlsx)}

Extended data Table 3.

Gene ontology terms enriched by differentially expressed genes for each comparison in males.

NIHMS1970887-supplement-Extended_data_3.xlsx ^{(125.8KB, xlsx)}

Extended data Table 4.

Gene ontology terms enriched by differentially expressed genes that overlapped between sexes for each comparison.

NIHMS1970887-supplement-Extended_data_4.xlsx ^{(79.2KB, xlsx)}

Extended data Table 5.

Correlation coefficients and p values for the module-trait heatmaps.

NIHMS1970887-supplement-Extended_data_5.xlsx ^{(33.7KB, xlsx)}

Extended data Table 6.

Distribution of differentially expressed transcripts in WGCNA modules.

NIHMS1970887-supplement-Extended_data_6.xlsx ^{(16.7KB, xlsx)}

Extended data Table 7.

Ingenuity pathway analysis of full list of genes in modules of interest and the top 20 genes by gene significance and module membership in each module.

NIHMS1970887-supplement-Extended_data_7.xlsx ^{(26.1KB, xlsx)}

Extended data Table 8.

Differentially expressed genes in microglia subclusters of human single-nucleus RNA-seq dataset.

NIHMS1970887-supplement-Extended_data_8.xlsx ^{(387.1KB, xlsx)}

Extended data Table 9.

Gene ontology terms enriched by differentially expressed gene microglia subclusters of the human single-nucleus RNA-seq dataset.

NIHMS1970887-supplement-Extended_data_9.xlsx ^{(347.4KB, xlsx)}

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
e450-conjugated anti-CD3	Thermo Fisher Scientific	Cat#48–0031-82, RRID:AB_10735092
e450-conjugated anti-CD19	Thermo Fisher Scientific	Cat#48–0193-82; RRID:AB_2734905
PE-Cy7-conjugated anti-CD45	Biolegend	Cat#103114; RRID:AB_312979
APC-conjugated anti-CD11b	Biolegend	Cat#101212; RRID:AB_312795
anti-CD16/CD32	BD Pharmingen	Cat#553142; RRID:AB_394657
rabbit anti-NeuN Alexa Fluor® 647	Abcam	Cat#ab190565; RRID:AB_2732785
Human BD Fc Block™ antibody	BD Pharmingen	Cat#564220; RRID:AB_2869554
rabbit anti-Sox9 Alexa Fluor® 555	Millipore	Cat#AB5535-AF555; RRID:AB_2239761
rabbit anti-Sox10 Alexa Fluor® 647	Abcam	Cat#ab270151, RRID:AB_2927700
rabbit anti-Iba1	Wako	Cat#019–19741; RRID:AB_839504
rat anti-CD11b	Bio-Rad	Cat#MCA711G; RRID:AB_323167
Biological samples
Human donor lumbar spinal cord tissue	Donor Network West	N/A
Chemicals, peptides, and recombinant proteins
Tamoxifen	Sigma	Cat#T5648
Diptheria Toxin	Sigma	Cat#D0564–1MG
Pexidartinib (PLX3397)	Chemgood	Cat#C-1271
AIN-76A Rodent Diet With 660 mg PLX3397/kg	Research Diets	Custom
AIN-76A Rodent Diet With 75 mg PLX3397/kg	Research Diets	Custom
2018 Teklad Global 18% Protein Rodent Diets	Inotiv	https://www.inotivco.com/rodent-natural-ingredient-2018-diets
O.C.T. Compound	Sakura Finetek, Inc.	Cat#4583
Fluoromount G with DAPI medium	ThermoFisher	Cat#00–4959-52
Dnase I (12,400 units/mL)	Sigma	Cat#11284932001
1x HBSS without Ca2+ or Mg2+	Gibco	Cat#14185052
Percoll	GE Healthcare	Cat#17–5445-02
SYTOX Blue	Invitrogen	Cat#S34857
Trizol LS	Invitrogen	Cat#10296028
Rnasin® Ribonuclease Inhibitors	Fisher/Promega	Cat#PRN2615
DAPI	Thermo Fisher Scientific Inc	Cat#62248
Critical commercial assays
SMART-Seq® v4 Ultra® Low Input RNA kit	Takara Bio USA Inc.	Cat#634888
Rneasy® Plus Micro kit	Qiagen	Cat#74034
Chromium Next GEM Single Cell 3ʹ Kit v3	10x Genomics	Cat#PN-1000128
NexteraTM DNA Flex Library Prep	Illumina	Cat#20015828
Deposited data
Massively Parallel Drop Seq Database	Habib et al.35	https://www.nature.com/articles/nmeth.4407#accession-codes
Newly repopulated spinal cord microglia exhibit a unique transcriptome and contribute to pain resolution (Superseries GSE249210, includes mouse and human)	This manuscript	https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE249210
Experimental models: Organisms/strains
C57BL/6J mice	Jackson Lab	Cat#00664
CX3CR1CreERT2-EYFP	Jackson Lab	Cat#021160
R26iDTR	Jackson Lab	Cat#007900
Oligonucleotides
All mice were genotyped at weaning by a commercial vendor	Transnetyx	https://www.transnetyx.com/
Software and algorithms
FlowJo v10.6.2	BD Biosciences	https://www.flowjo.com/solutions/flowjo/downloads/previous-versions
ImageJ/FIJI	Schneider et al.72	https://imagej.nih.gov/ij/
Seurat	R Package	https://satijalab.org/seurat/
GraphPad Prism	GraphStats	https://www.graphpad.com/
CellRanger	10x Genomics	https://support.10xgenomics.com/single-cell-gene-expression/software/pipelines/latest/using/tutorial_in#download
g:profiler	Raudvere et al.73	https://biit.cs.ut.ee/gprofiler/gost
Gene Ontology (2021–02-01 release)	Gene Ontology archive	http://release.geneontology.org/2021-02-01/index.html
GOA UniProtKB (2022–03-02 release)	Gene Ontology Annotation Database	https://www.ebi.ac.uk/GOA/downloads
GO-Figure! Version 1.0.1	Reijnders and Waterhouse74	https://gitlab.com/evogenlab/GO-Figure
Qiagen IPA	Qiagen	https://digitalinsights.qiagen.com/products-overview/discovery-insights-portfolio/analysis-and-visualization/qiagen-ipa/?utm_source=QDI_GA_IPA&cmpid=QDI_GA_IPA&gclid=Cj0KCQjw4NujBhC5ARIsAF4Iv6fIyfOIfTZssiSa8Fb8kGB7p6c47wPFdgBZjl4w9ULpo1X26bYKx8YaAvjxEALw_wcB
WGCNA v1.70.3	Langfelder and Horvath75	https://horvath.genetics.ucla.edu/html/CoexpressionNetwork/Rpackages/WGCNA/
Dynamic Tree Cut	Reijnders and Waterhouse74	https://cran.r-project.org/web/packages/dynamicTreeCut/index.html
nf-core/rnaseq v3.5 RNA-Seq analysis pipeline	Ewels et al.76	https://nf-co.re/tools
nf-core helper tools v2.2	Ewels et al.76	https://nf-co.re/tools
Nextflow v22.01.0 workflow tool	Di Tommaso et al. 77	https://github.com/nextflow-io/nextflow/releases
FastQC v0.11.9	Andrews, S. 78	https://github.com/s-andrews/FastQC/releases
Trim Galore v0.6.7	Krueger et al. 79	https://github.com/FelixKrueger/TrimGalore
Cutadapt v3.4	Martin M.80	https://cutadapt.readthedocs.io/en/v3.4/
STAR v2.7.6a	Github, Inc.	https://github.com/alexdobin/STAR
R Studio	R v4.1.3	https://www.rstudio.com/
RSEM v1.3.1	Li and Dewey 81	https://deweylab.github.io/RSEM/
MultiQC v1.11	Ewels et al. 82	https://multiqc.info/
DESeq2 v1.34.0	Love et al. 83
Other
cell strainer	BD Biosciences	Cat#352340
von Frey filaments	Stoelting	Cat#58011
Incapacitance device	IITC Inc Life Science	Cat#600MR
LIMAB laser measurement sensor	LIMAB	https://www.limab.com/industries/laser-sensors/
fine-gauge thermocouple wire	Omega	Cat# IEC-TFCI-003–15M
Cryostat	Leica Biosystems	Cat# CM3050S
Keyence BZ-X800 fluorescent microscope	Keyence	https://www.keyence.com/
BD Aria II	BD Biosciences	https://www.bdbiosciences.com/en-us/products/instruments/flow-cytometers/research-cell-sorters
Illumina NovaSeq6000 sequencer	Illumina	https://www.illumina.com/systems/sequencing-platforms/novaseq.html

Highlights.

• Selective depletion and repopulation of microglia completely reverse pain-like sensitivity

• Reversal in pain trajectory coincides with microglial repopulation, not depletion

• Repopulated microglia transcriptomes show active environmental surveillance after injury

• Micro glial genes are validated by intersection with human lumbar spinal cord microglial genes