key: cord-0267610-85yur6fi
authors: Tansley, Shannon; Uttam, Sonali; Ureña Guzmán, Alba; Yaqubi, Moein; Pacis, Alain; Parisien, Marc; Rabau, Oded; Haglund, Lisbet; Ouellet, Jean; Santaguida, Carlo; Ragoussis, Jiannis; Zhang, Ji; Salter, Michael W; Diatchenko, Luda; Healy, Luke M.; Mogil, Jeffrey S.; Khoutorsky, Arkady
title: Single-cell RNA sequencing reveals time- and sex-specific responses of spinal cord microglia to peripheral nerve injury and links ApoE to neuropathic pain
date: 2020-12-10
journal: bioRxiv
DOI: 10.1101/2020.12.09.418541
sha: d7914ee8c7e1561b61a24a3988906ab482319ec9
doc_id: 267610
cord_uid: 85yur6fi

Activation of microglia in the spinal cord following peripheral nerve injury is critical for the development of long-lasting pain hypersensitivity. However, it remains unknown whether distinct microglia subpopulations or states contribute to different stages of pain development and maintenance. We show, using single-cell RNA-sequencing, that nerve injury induces the generation of a male-specific inflammatory microglia subtype, and demonstrate increased proliferation of microglia in males as compared to females. We also show time- and sex-specific transcriptional changes in different microglial subpopulations following injury. Apolipoprotein E (Apoe) is the top upregulated gene in microglia at chronic time points after nerve injury in mice and polymorphisms in the APOE gene in humans are associated with chronic pain. Single-cell analysis of human spinal cord microglia reveals a subpopulation with a disease-related transcriptional signature. Our data provide a detailed analysis of transcriptional states of mouse and human spinal cord microglia, and identify a previously unrecognized role for ApoE in neuropathic pain.

Peripheral nerve injury leads to neuropathic pain, a debilitating condition associated with spontaneous and light touch-induced pain, and accompanied by the activation and proliferation of microglia in the spinal cord dorsal horn 1,2 . Microglia are resident immune cells in the central nervous system (CNS) that constantly survey the environment. Damaged primary afferents release multiple molecules such as chemokines (CCL2, CCL21, CXCL1), signaling molecules (CSF1), and proteases (MMP-9) to induce microglia proliferation and activation, which is characterised by a transition from a ramified homeostatic microglia phenotype with multiple fine processes to an 3 amoeboid-like cell 3 . The "activated" microglia release a wide range of bioactive substances that signal to neuronal and non-neuronal cells to facilitate pain transmission, critically contributing to pain hypersensitivity [4] [5] [6] [7] . Microglia play important roles in both the development and maintenance phases of neuropathic pain as inhibition or depletion of microglia 1-2 weeks or three months postnerve injury alleviate the hypersensitivity 5, 8, 9 . The role of microglia in neuropathic pain is known to be sex-dependent 10, 11 . Microglia proliferation and morphological changes in response to peripheral nerve injury are present in both sexes; however, a functional role of microglia as critical drivers of neuropathic pain is observed in male but not in female animals [12] [13] [14] [15] [16] .

Progress in single-cell transcriptomic techniques has facilitated the study of microglia heterogeneity in different physiological and pathological processes, revealing that microglia in the brain exist in multiple transcriptional states tailored to different developmental stages, anatomical areas and pathologies [17] [18] [19] . Nine distinct transcriptional states of microglia have been described in the brain 17 , including several subpopulations of homeostatic microglia, microglia responding to injury (injury-responsive microglia; IRM), and to chronic pathologies including Alzheimer's disease (AD) (disease-associated microglia; DAM) 20 . The heterogeneity of microglia in the mouse and human spinal cord, however, remains ill-defined. Despite central roles of microglia in the development and maintenance of neuropathic pain 3 , it is unknown whether peripheral nerve injury leads to the generation of unique microglia transcriptional states that promote and maintain pain hypersensitivity.

In this study, we used single-cell RNA sequencing (scRNA-seq) to define transcriptional states of spinal cord microglia in mouse and human. In mice, by profiling a large number (188, 787) of cells of both sexes, we observed that nerve injury-induced changes in microglia differ significantly in acute and chronic phases of neuropathic pain. We detected sex-specific changes in 5 to the peak of microglia proliferation 1 . At day 14, hypersensitivity is fully established, representing a sub-chronic phase of pain development. The 5-month time point was selected to study a chronic phase of neuropathic pain. Microglia from the lumbar spinal cord of naive, sham surgery, and SNI mice (2 replicates per condition, 4 mice pooled per replicate) were FACS sorted (Extended Data Fig. 1) , gating for CD45 low CD11B high CX3CR1 high cells. Purified microglia were processed using the droplet-based 10X Genomics Chromium system, generating high-quality single cell transcriptomic data (Extended Data Fig. 2a-f ) from 188,787 sequenced microglia (~13,500 cells per condition after quality control) with a median number of ~1,900 genes per cell at an average sequencing depth of 43,000 reads/cell. Unsupervised clustering analysis for all cells under all conditions combined revealed 11 distinct clusters (Fig. 1e) . Cells in all clusters expressed canonical microglia genes such as Tmem119, Fcrls, P2ry12, Cx3Cr1, Trem2, and C1qa (Fig. 1f) .

Gene expression analysis identified genes uniquely expressed in microglia in clusters 4, 5, 6, 7, 8, Ontology (GO) analysis of cluster markers showed that microglia in clusters 7 and 8 are enriched 6 for DNA replication and cell cycle genes (Extended Data Fig. 3b ). Microglia in cluster 9 exhibit a transcriptional signature characterized by increased interferon and cytokine-mediated signaling (Extended Data Fig. 3c ). Cluster 9 microglia were absent in naive mice (0.035% of total) but their number was substantially increased 3 days post-SNI in males (4.9%) but not in females (0.03%) ( Fig. 2a, b, d and e) . Microglia belonging to cluster 9 were largely undetected at day 14 and 5 months post-SNI in both sexes ( Fig. 2a-b) , demonstrating their nerve injury-induced acute onset and male-specific origin. Cluster 9 microglia can be detected by the combination of two transcripts: the presence of Lgals1 and absence of Top2a. Microglia in cluster 10 uniquely express Cldn5 that encodes for the tight-junction, protein claudin-5. Claudin-5 is expressed in endothelial cells but also in vessel-associated microglia creating tight junction connections with endothelial cells to maintain the blood-brain barrier 23 . Cells in cluster 10 express canonical microglia genes, along with Cldn5, indicating that this small population (0.19% of total) represents vessel-associated microglia. Cluster 11 (0.58% of total) was composed of cells expressing canonical microglia genes (Fcrls, Trem2, Cx3Cr1, Tmem119, C1qa and P2ry12) along with unique expression of macrophage/monocyte markers (H2-Aa, Mrc1, Ccr2, Lyve1, Dab2, Mgl2, F13a1) (Supplementary   Table 1 and Extended Data Fig. 3d ). This subpopulation of cells might be composed of perivascular macrophages that express both macrophage and microglia genes. Perivascular macrophages, which are resident brain macrophages, originate from the same pool of yolk sac hematopoietic progenitors, migrate to the brain early in development, and self-renew, similar to microglia 24, 25 . All together, our data show that the majority of microglia in naive spinal cord exist in six transcriptional states (clusters 1-6), in addition to small populations (< 2% of total) of proliferating (cluster 7 and 8), and vessel-associated (cluster 10) microglia.

Three days following SNI, the number of microglia in clusters 7 and 8 increased dramatically from 0.5% in male naive mice (5.4% in sham) to 20.8% of all microglia in SNI mice (Fig. 2a, c, and d) .

Surprisingly, the increase in females was smaller, from 1.7% in naive mice (2.9% in sham) to 13 .2% in female SNI mice (Fig. 2b , c, and e), suggesting attenuated proliferation of microglia in females. To corroborate this finding, we labeled the proliferating microglia using a marker of proliferating cells, Ki67, co-stained with a microglia marker, Iba1. Quantification of Ki67-positive microglia in the dorsal horn spinal cord ( Fig. 2f-g) showed that the number of proliferating microglia was significantly higher in males as compared to females (Ki67-positive microglia per section; males: 68.8 ± 5.9; females: 46.6 ± 1.8, p=0.011, Fig. 2h ), confirming the scRNA-seq result. Notably, microgliosis (total number of Iba1-positive microglia) was not different between male and female mice (dorsal horn microglia per section, males: 82.8 ± 5.5; females: 76.4 ± 1.8, p=0.31, Fig. 2i ), suggesting that other mechanisms compensate for reduced proliferation in females.

To study transcriptional changes in microglia following peripheral nerve injury, we identified differentially expressed genes (DEGs) in SNI versus corresponding sham groups in each microglia cluster at three time points in male and female mice (Fig. 3a-e and Supplementary Table 3) . We detected a large number of DEGs in male and female microglia at day 3 post-SNI; this number decreased substantially on day 14 and 5 months in both sexes. Analyses of DEGs in each cluster 8 showed that the vast majority of changes in gene expression occurred in microglia from clusters 1-9 at day 3 post-SNI (Fig. 3c -e) and in clusters 1-6 at later time points.

Interestingly, we observed changes in the expression of numerous ribosomal proteins (small ribosomal subunit (Rps) and large ribosomal subunit (Rpl)) in microglia following SNI, an effect which was significantly more pronounced in males than females ( Fig. 3f and Extended Data Table 4 ) and in clusters 1-6 combined (Fig. 3g) . We found that in males, nerve injury induces a strong immune response at day 3 (major GO category: cellular response to cytokine stimulus, inflammatory response, apoptotic process), whereas in females inflammatory processes were less pronounced and top categories included ATF-6mediated unfolded protein response and receptor-mediated endocytosis (Fig. 3g) . At day 14 in males, mitochondrial ATP production and other metabolic processes, including lipid metabolism and trafficking, were the top categories, whereas genes involved in regulation of endocytosis/phagocytosis, cell adhesion, and lipid catabolism were affected in females. At 5 months post-SNI, male microglia showed changes in genes related to mitogen-activated protein kinase (MAPK) signaling, whereas female microglia exhibited changes in metabolic processes and responses to lipopolysaccharide (LPS). Microglia in male-specific cluster 9 at day 3 post-SNI exhibited strong inflammatory responses and showed changes in genes related to mitochondrial ATP production and phagocytosis (Fig. 3h ).

Sexual dimorphism of microglia involvement in neuropathic pain has been extensively studied, revealing that microglia are necessary for the development of hypersensitivity in males 9 but not in females 3, 10, 11 . To study the differences in responses of male and female microglia to peripheral nerve injury, we compared DEGs between two sexes. Microglia in males showed more DEGs than females at day 3 post-SNI ( Fig. 3a-b) , and only a fraction of DEGs was shared between the sexes at each time point (Fig. 3i) . Surprisingly, correlational analysis between DEGs in male and female microglia in each of the first six clusters (1-6) on day 3 revealed no or negative correlations (Extended Data Fig. 5a 6 ). This finding is consistent with male-specific upregulation of p38 phosphorylation 15 , which is central to the activation of microglia following nerve injury 16, 26 . Il1b and Tnf also showed malespecific upregulation at day 3 post-SNI, whereas Tspo was increased in both sexes (Extended Data Previous studies have identified several populations of microglia exhibiting unique transcriptional responses to pathological and context-specific physiological conditions. Therefore, we compared, using Fisher's exact test, SNI-induced changes in the microglia transcriptome to previously described transcriptional signatures (Fig. 3j ). IRM were identified in the subcortical white matter in response to lysolecithin (LPC)-induced demyelination, a common model of multiple sclerosis (MS) 17 . DAM were found in the brain of a mouse model of AD. Axon tractassociated microglia (ATM) were identified as a transient population detected in the pre-myelinated developing brain in the axon tracts (corpus callosum) 17 . Our analysis revealed that the transcriptional landscape in male microglia in clusters 1-6 was similar to IRM at all time points post-SNI (day 3, day 14 and 5 months) and showed mild similarities to DAM and ATM at day 14 ( Fig. 3j and Supplementary Table 5 ). In contrast, changes in gene expression in female microglia were less similar to the assessed transcriptional signatures, exhibiting similarities to IRM, DAM and ATM microglia mainly at day 14 post-SNI in clusters 1, 3, 4, 5 and 6 ( Fig. 3j) . Interestingly, microglia in male-specific cluster 9 at day 3 post-SNI exhibited a transcriptional profile which was highly similar to IRM as compared to other clusters (Fisher's exact test: p=5.8x10 -68 , Fig. 3j and Supplementary Table 5 ).

One of the top genes in both the IRM and DAM transcriptional signatures is that coding for apolipoprotein E (ApoE), which is involved in lipid metabolism and trafficking and implicated in AD pathophysiology 27 . Our analyses revealed that Apoe is the most abundant transcript in cluster 5 spinal cord microglia in both sexes (Fig. 1h, Fig. 4a and Supplementary Table 2 ). Following SNI in males, Apoe mRNA levels showed no change in expression at day 3 after nerve injury but were significantly upregulated at day 14 (top upregulated gene in clusters 1 and 3 and significantly upregulated in clusters 2, 4 and 6) and 5 months (top upregulated gene in clusters 1-6) ( Fig. 4a-b) .

Similarly, Apoe remained unchanged at day 3 post-SNI in females and was upregulated at day 14 and 5 months (top upregulated gene in clusters 1-6 at both time points) (Fig. 4c) . To study if upregulation of Apoe transcript leads to changes in ApoE protein levels, we performed immunostaining against ApoE in the spinal cord after SNI. ApoE levels in microglia, labelled by Iba1, were low at day 3 post-SNI, but increased significantly at day 14 and 8 months (Fig. 4d-e) . 11 These findings demonstrate that ApoE is persistently upregulated in microglia following peripheral nerve injury and suggest that ApoE might have a role in chronic responses of microglia to peripheral nerve injury.

Persistent upregulation of microglial ApoE after nerve injury in mice prompted us to study the link between APOE and chronic pain in humans. Polymorphisms in APOE are associated with several chronic diseases of the CNS; for example, it is a major genetic risk factor for late-onset AD 27 .

APOE has 3 major haplotype-based allelic variants: ε2, ε3, and ε4 (Fig. 5a) . APOE-ε4 confers increased risk 28 , whereas APOE-ε2 confers decreased risk 29 for AD, as compared to the common APOE-ε3 allele. We assessed the association of APOE-ε2 and APOE-ε4 variants with the report of pain at several body sites in humans in the large UK Biobank cohort (Fig. 5b) . Strikingly, we found that whereas APOE polymorphisms were not associated with acute pain (Fig. 5c and Supplementary and knee pain (OR =1.48, p=0.03) (Fig. 5d) . Interestingly, meta-analysis of conditions with a neuropathic component (back, hip, knee and neck/shoulder pain) showed a significant negative correlation with APOE-ε4 (neuropathic; OR=0.64, p=6.4x10 -5 ) and a positive correlation with APOE-ε2 (neuropathic; OR=1.37, p=1.4x10 -4 ), whereas no correlation was found with other conditions. Collectively, our data indicate that, in direct opposition to AD, carriers of APOE-ε4 have a decreased risk whereas carriers of APOE-ε2 have an increased risk to develop chronic pain.

To elucidate the role of spinal cord microglia in chronic pain and other spinal cord pathologies such as MS and amyotrophic lateral sclerosis (ALS) 30 , and translate this knowledge to humans, it is imperative to study human spinal cord microglia heterogeneity and transcriptional states. Table 7) , all expressing canonical microglia markers TREM2, and C1QA. In addition to the main population of microglia, we observe a cluster of cells expressing mixed microglia-oligodendrocyte genes (cluster 9) and a minor population of T/NK cells (cluster 10). Cluster 1 is the largest microglia cluster, containing >15% of total cells analyzed. This cluster is characterised by high expression of genes associated with 'homeostatic' microglia 31 such as CSF1R, BHBLE41, FCGR1A, CTSS and GPR34. Microglia in cluster 3 are characterized by expression of chemokines CCL3 and CCL4, and the zinc finger containing transcription factors early growth response protein 1 and 2 (EGR1 and EGR2). A recent study of 1,180 human cortical microglia derived from surgically resected temporal lobe tissue identified a similar cluster of cells characterized by high expression of CCL2, CCL4, EGR2 and EGR3, with the authors denoting these cells as 'pre-activated' 19 . Prior to attaining a mature 13 surveillant phenotype, pre-microglia express high levels of EGR1 and are closely associated with synaptic pruning and neural maturation 32 . Microglia in cluster 3 also express high levels of the complement component C3 which is associated with mediating synapse elimination and the developmental of precise synaptic connectivity 33, 34 . Collectively, the profile of spinal cord microglia in cluster 3 is that of cells interacting with and potentially modulating neuronal structure and function, as suggested by associated GO terms (Fig. 6c) . Interestingly, gene expression in This analysis represents the first report of primary human spinal cord-derived microglia transcriptomic signatures. We observe a strong correlation between human and mouse microglia ( Fig. 6e) , particularly in human clusters 1-3 and 6 ( Fig. 6f) . The most striking observation from these data is the identification of cluster 4, which displays remarkable similarity to previously identified transient developmental (ATM) and disease-associated signatures (IRM/DAM) in the mouse brain.

In this study, we analysed the heterogeneity of microglia in the mouse and human spinal cord, and investigated microglial responses to peripheral nerve injury in mice at three different time points and both sexes by transcriptionally profiling a large number of microglia using scRNAseq.

Single cell transcriptional analyses revealed that mouse and human spinal cord microglia exist in numerous heterogeneous subpopulations. Unsupervised clustering analysis of gene expression revealed that the majority of microglia in mice exist in six clusters (1-6), constituting ~98% of all microglia in naive animals. Small populations (less than 1% each) of vessel-associated microglia (cluster 10) and perivascular macrophages (cluster 11) were also detected. Peripheral nerve injury resulted in the appearance of three additional clusters: two subtypes of proliferating microglia (clusters 7 and 8) and IRM-like male-specific microglia (cluster 9). In human spinal cord, eight distinct microglia subpopulations were detected, consistent with a recent study showing that human microglia, collected from brain tissue, exhibit greater heterogeneity as compared to other species 18 . Intriguingly, in the human spinal cord, we detected a large subpopulation of microglia (cluster 4) exhibiting a transcriptional profile highly similar to the previously described mouse disease (IRM/DAM)/early development (ATM) signatures. The identification of IRM/DAM/ATM in tissue isolated from 'healthy' individuals may be indicative of an undiagnosed neurological condition, or suggest that this population, while not observed in the mature mouse brain, may develop in the normally aged human spinal cord.

We found, in accordance with previous studies, a robust inflammatory response in male microglia at acute phase (day 3) following peripheral nerve injury. A large number of genes showed differential expression, many related to immune functions. At day 14 and 5 months post-SNI, the number of DEGs was significantly reduced, and metabolic processes, including energy production and lipid metabolism, were predominantly induced. These data indicate that following initial immune response to injury, there is a transcriptional shift from immuno-reactive microglia towards a metabolically altered state. This phenomenon has been observed in macrophages in response to damage 37 and in microglia in other chronic brain pathologies such as AD and MS 38 , revealing that following prolonged activation, microglia shift from mitochondrial oxidative phosphorylation to glycolysis (Warburg effect) for ATP production 39, 40 , and upregulate genes involved in lipid, cholesterol and lipoprotein metabolism 41 .

In females, acute inflammatory responses were less pronounced as compared to males, and predominant processes included endocytosis, cellular metabolism, and cell adhesion. Interestingly, we found that a large number of ribosomal proteins (Rpl and Rps) showed increased expression selectively in male but not in female microglia at day 3 post-SNI. Induction of ribosomal proteins might enhance the microglia's capacity to produce new proteins, which are required during the activation state, via increased ribosomal biogenesis and mRNA translation 42 .

We found that changes in gene expression in response to peripheral nerve injury differ significantly between males and females at day 3 post-SNI but show similarities at later time points. Correlation analysis between DEGs in male and female microglia revealed negative or no correlation between males and female at day 3 post-SNI, but weak-to-moderate correlation at day 14 and 5 months. This is consistent with the divergent acute responses of male and female microglia, wherein microglia in males show stronger immune activation as compared to females.

We observed several additional sex differences of note. A subpopulation of microglia (cluster 9) with a marked inflammatory profile was detected exclusively in males at day 3 post-SNI, but not in females and not at later time points. These microglia, which constitute ~4.5% of all microglia at day 3 in males, show a pronounced IRM signature. We also found that microglia proliferate less in females as compared to males at day 3 post-SNI despite equal microgliosis.

Altered expression of genes related to apoptosis in male microglia at day 3 post-SNI raises the possibility that apoptosis and loss of male microglia following their activation is one of the mechanisms accounting for equal microgliosis in males and females.

Our scRNA-seq analysis revealed a robust, across-cluster upregulation of Apoe mRNA levels in microglia at late (day 14 and 5 months) but not acute (day 3) time points after nerve injury ( Fig. 4) . Apoe was the top upregulated gene in numerous clusters at day 14 and 5 months.

Remarkably, we found a strong link between APOE and chronic pain in humans by demonstrating that polymorphisms in APOE are associated with chronic, but not acute human pain states (Fig.   5 ). Interestingly, meta-analysis revealed an association between APOE polymorphisms and chronic pain states with a neuropathic component (back, hip, knee, and neck/shoulder pain), whereas no association was detected for pain without a neuropathic component. APOE-ε4 confers a decreased risk and APOE-ε2 an increased risk to develop specific chronic pain conditions. This is opposite to the direction of associations found in individuals with AD, indicating that carriers of APOE-ε4 have increased risk to develop AD, but decreased risk for developing chronic pain.

ApoE is the most abundant apolipoprotein in the CNS, and is involved in trafficking and metabolism of lipoproteins and cholesterol 27 . ApoE may promote the efflux of intracellular cholesterol which accumulates in microglia following phagocytosis of dying cells and myelin debris. Lipids in microglia are used as precursors of many inflammatory mediators and as a source of energy, via oxidative metabolism. Therefore, the regulation of lipid metabolism by ApoE may have a direct effect on microglia inflammatory functions and energy production.

In summary, our datasets and analyses represent a comprehensive characterization of spinal cord microglia heterogeneity, providing the basis for elucidating the roles of specific microglia subpopulations and transcriptional states in distinct processes underlying chronic pain and other spinal cord pathologies (e.g., MS and ALS). Detection of a male-specific microglia subpopulation (cluster 9) with a marked inflammatory profile might lead to a better understanding of sex-specific mechanisms and development of targeted therapeutic approaches. The switch from an immune to an altered metabolic state of microglia is a potential mechanism underlying chronic and clinically relevant phases of neuropathic pain. Upregulation of Apoe, a gene that we show is strongly associated with human chronic pain conditions with neuropathic component, might be a central mechanism in this switch, and its further examination might generate important insights into the microglia-dependent mechanisms of neuropathic pain. 

Mice were deeply anesthetized with isoflurane and subjected to bilateral spared nerve injury (SNI) or sham surgeries. SNI surgery was performed as described previously 21, 22 . Briefly, the sciatic nerve was exposed after making an incision on the skin on the lateral surface of the mouse thigh and sectioning through the biceps femoris muscle. Two of the three terminal branches of the sciatic nerve, the tibial and common peroneal nerves, were tightly ligated with 7.0 silk (Covidien, S-1768K) and 2-4 mm of the nerve distal to the ligation were removed, avoiding any disturbance of the sural nerve. The muscle and skin were closed in separate layers using coated Vicryl (Ethicon, J489G). For the sham surgeries, the sciatic nerve, as well as its three branches, were exposed but left intact.

Behavioral studies 19 All experiments took place during the light cycle, no earlier than 09:00 h and no later than 16:00 h. Naive mice were placed in custom-made Plexiglas cubicles (5.3x8.5x3.6 cm) on a perforated metal floor and were habituated for at least 1 hour before testing. The up-down method of Dixon was used to estimate 50% withdrawal thresholds using calibrated von Frey nylon monofilaments (Stoelting Touch Test). Filaments were applied to the plantar surface of the hind paw for 3 seconds and responses were recorded. At least 2 consecutive measures were taken on each hind paw at each time point and averaged. Although data were collected on both hind paws, only data from the left hind paw (subjected to SNI) are presented, as no significant genotype effects on the contralateral paw were observed. Mice were tested for mechanical sensitivity using von Frey fibers at baseline and 3 days, 14 days, and 5 months after surgery to quantify mechanical allodynia. 

Raw sequencing data for each sample was converted to matrices of expression counts using the Cell Ranger software provided by 10X Genomics (version 3.0.2). Briefly, raw BCL files from the Illumina HiSeq were demultiplexed into paired-end, gzip-compressed FASTQ files using Cell Ranger's mkfastq. Using Cell Ranger's count, reads were aligned to the GRCm38 (mm10) mouse reference genome, and transcript counts quantified for each annotated gene within every cell. The resulting UMI count matrices (genes × cells) were then provided as input to Seurat suite (version 3.1.0). To filter out low-quality cells, we defined a window of a minimum of 500 and a maximum of 4000 detected genes per cell. Cells with more than 5% of the transcript counts derived from mitochondrial-encoded genes were further removed.

To identify sex-mediated differences in response to injury, all microglia samples were merged and analyzed using reciprocal PCA with reference-based integration as part of the Seurat single-cell analysis package 43, 44 . Here we specify one or more of the datasets as the 'reference'

for integrated analysis, with the remainder designated as 'query' datasets. In short, PCA was performed separately for each dataset, following normalization, variable feature selection, and scaling. Integration anchors were identified using both male and female naïve samples as reference (FindIntegrationAnchors function with parameter reduction = "rpca").

Clustering and visualization of the integrated dataset were performed using Uniform Manifold Approximation and Projection (UMAP), with the first 15 principal components at a resolution of 0.3. Differential expression analysis between two conditions was performed using Tissue was processed as previously described 45 . Briefly, the meninges were carefully removed, and the tissue crosscut into small pieces of 1-2 mm 3 . A single cell suspension was obtained following mechanical and enzymatic (trypsin) dissociation. Myelin was removed using an isotonic Percoll gradient. Glia cells were located between the myelin and red blood cell layers, this layer was collected, washed and resuspended in ice-cold PBS. Finally, a solution of cells (1000 cells per 23 microliter) was sent for library preparation and high-throughput sequencing through Génome Québec. Library preparation was carried out as follows. In accordance with the Single Cell 3' Reagent Kits v2 User Guide (CG0052 10x Genomics), a single-cell RNA library was generated using the GemCode Single-Cell Instrument (10x Genomics, Pleasanton, CA, USA) and Single 

To compare transcription between species, we extracted one-to-one orthologs using BioMart software suite 46 . Human cells expressing either TMEM119, ITGAM or CX3CR1 were included in the analysis. Mouse cells were subsampled to equal number of human cells, resulting in 5,456 cells for each species. Mouse and human microglia samples were analyzed separately using Seurat (version 3.1.0). Briefly, canonical correlation analysis (CCA) was performed to identify shared sources of variation to produce anchors across the datasets, following SCTransform normalization.

Clustering and visualization of the integrated dataset were performed using UMAP, using FindClusters and RunUMAP functions. 24 In order to compare global expression profiles of microglia across species, we aggregated all cell counts within each scRNA-seq sample to generate a pseudo bulk sample, and then analyze the pooled data by using approaches designed for bulk RNA-seq. For all downstream analyses, we excluded lowly-expressed genes with an average read count lower than 10 across all samples, resulting in 10,672 genes in total. Raw counts were normalized using edgeR's TMM algorithm 47 and were then transformed to log2-counts per million (logCPM) using the voom function implemented in the limma R package 48 . We tested for significantly differentially expressed genes using the lmfit function and nominal p-values were corrected for multiple testing using the Benjamini-Hochberg method. 

Genetic analyses for the APOE gene were conducted in the UK Biobank 52, 53 . Analyses were restricted to the "White British" ancestry, after genotyping quality control performed by the UKB detailed in the resource: https://biobank.ctsu.ox.ac.uk/crystal/crystal/docs/genotyping_qc.pdf.

Alleles for SNPs rs7412 and rs429358 were extracted using bgenix (bioArxiv 308296; https://doi.org/10.1101/308296) and qctool (https://www.well.ox.ac.uk/~gav/qctool_v2/). The dosage data were converted into hard calls, using a threshold of 0.1 via PLINK 54 . Allelic combinations were assigned one of three possible haplotypes: ε2 (rs7412-T, rs429358-T), the ancestral haplotype ε3 (rs7412-C, rs429358-T), or ε4 (rs7412-C, rs429358-C). Association tests between haplotypes and pain phenotypes were performed using the haplo.cc function in the haplo.stats R computer package (https://cran.r-project.org/package=haplo.stats), with minimum counts of haplotypes set to 100, in an additive genetic inheritance model, and relative to the ancestral haplotype. Age, age squared, sex, recruitment sites, genotyping array, and first 40 principal genetic components were used as co-variables. Meta-analyses used an inverse standarderror based analytical strategy, as suggested by METAL 55 . Phenotype assignments for Alzheimer's diseases derived from UK Biobank field 20002 "Non-cancer illness code, self-reported" code 1263 "dementia/alzheimers/cognitive impairment", as well as main (field 41202) and secondary (field 41204) ICD10 diagnoses: G30 "Alzheimer's disease" and sub-codes (G300, G301, G308, G309) for a total of 470 cases and 384054 controls. Cases for chronic pain at a body site were defined as participants answering "yes" to the following question: "Have you had [ for acute pain at a body site were defined as participants that answered "no", which implied that they had answered "yes" for the question "In the last month have you experienced any of the following that interfered with your usual activities?" (UKB field 6159). Body sites included: headaches (n=43,310), facial (n=3,882), neck/shoulder (n=28,407), stomach/abdominal (n=14,918), back (n=32,209), hip (n=9,967), knee (n=18,316), and widespread (n=1,038). Control subjects were those that answered: "none of the above" at field 6159 (n=163,825).

All results are expressed as mean ± SEM. Statistical tests were made using a one-way ANOVA followed by between-group comparisons using Tukey's post hoc test or unpaired t-test with p < 0.05 as significance criteria (GraphPad Prism 7.03). Fisher's exact test was performed in R, using fisher.test function, to evaluate the statistical significance of the overlap between two gene lists. CTSS  FCGR1A  CD14  CSF1R  BHLHE41  GPR34  DNAJB1  HSPA1A  JUN  HSPA1B  HSPB1  CCL3  TMEM107  CCL4  C3  EGR2  EGR1  IL1RN  IL1B  SPP1  FABP5  FTL  CTSD  CTSB  LPL  CD63  GPNMB  CD9  APOE  AIF1  S100A8  APOC2  CCND1  BIN1  MCL1  GPR183  NR4A2  MAP3K8  HBEGF  ARL6IP1  LGALS1  CALM3  PSME2  TSPO  TUBB  PCNA  DEK  CENPM  CARHSP1 C2 C1 Table 7 for a complete list of clusters markers. c, Alluvial plot depicting the most affected biological processes for each cluster. Upregulated genes of each cluster were used for the analysis. Ribbon thickness indicates the number of genes per biological term. P value for each term is shown in bracket. d, Heatmap showing the enrichment for IRM, DAM, and ATM gene expression signature in cluster markers for each cluster (Fisher's exact test). White indicates no significant enrichment (p > 0.05). Shades of blue (for p values < 0.05) indicate the significance of the overlap between two gene lists. See Supplementary Table 8 

Characterization of cell proliferation in rat spinal cord following peripheral nerve injury and the relationship with neuropathic pain

A quantitative analysis of the microglial cell reaction in central primary sensory projection territories following peripheral nerve injury in the adult rat

Microglia in Pain: Detrimental and Protective Roles in Pathogenesis and Resolution of Pain

p38 mitogen-activated protein kinase is activated after a spinal nerve ligation in spinal cord microglia and dorsal root ganglion neurons and contributes to the generation of neuropathic pain

Inhibition of microglial activation attenuates the development but not existing hypersensitivity in a rat model of neuropathy

P2X4 receptors induced in spinal microglia gate tactile allodynia after nerve injury

Microglial Modulation as a Target for Chronic Pain: From the Bench to the Bedside and Back

DREADDed microglia in pain: Implications for spinal inflammatory signaling in male rats

Spinal microglia are required for long-term maintenance of neuropathic pain

Sex-dependent mechanisms of chronic pain: A focus on microglia and P2X4R

Qualitative sex differences in pain processing: emerging evidence of a biased literature

Different immune cells mediate mechanical pain hypersensitivity in male and female mice

Microglial P2X4R-evoked pain hypersensitivity is sexually dimorphic in rats

Spinal cord Toll-like receptor 4 mediates inflammatory and neuropathic hypersensitivity in male but not female mice

Spinal inhibition of p38 MAP kinase reduces inflammatory and neuropathic pain in male but not female mice: Sex-dependent microglial signaling in the spinal cord

Intrathecal administration of antisense oligonucleotide against p38alpha but not p38beta MAP kinase isoform reduces neuropathic and postoperative pain and TLR4-induced pain in male mice

Single-Cell RNA Sequencing of Microglia throughout the Mouse Lifespan and in the Injured Brain Reveals Complex Cell-State Changes

Cross-Species Single-Cell Analysis Reveals Divergence of the Primate Microglia Program

Spatial and temporal heterogeneity of mouse and human microglia at single-cell resolution

A Unique Microglia Type Associated with Restricting Development of Alzheimer's Disease

Spared nerve injury: an animal model of persistent peripheral neuropathic pain

Spared nerve injury model of neuropathic pain in the mouse: a behavioral and anatomic analysis

Dual microglia effects on blood brain barrier permeability induced by systemic inflammation

Early Fate Defines Microglia and Non-parenchymal Brain Macrophage Development

High-Dimensional Single-Cell Mapping of Central Nervous System Immune Cells Reveals Distinct Myeloid Subsets in Health

Activation of p38 mitogen-activated protein kinase in spinal hyperactive microglia contributes to pain hypersensitivity following peripheral nerve injury

Apolipoprotein E and Alzheimer disease: pathobiology and targeting strategies

Gene dose of apolipoprotein E type 4 allele and the risk of Alzheimer's disease in late onset families

Protective effect of apolipoprotein E type 2 allele for late onset Alzheimer disease

Spinal Cord Microglia in Health and Disease

Identification of a unique TGF-beta-dependent molecular and functional signature in microglia

Transcriptional control of microglia phenotypes in health and disease

Microglia sculpt postnatal neural circuits in an activity and complementdependent manner

The classical complement cascade mediates CNS synapse elimination

New Insights into the Function of the Immunoproteasome in Immune and Nonimmune Cells

Single-cell sequencing reveals dissociation-induced gene expression in tissue subpopulations

Metabolic regulation of macrophages in tissues

A Breakdown in Metabolic Reprogramming Causes Microglia Dysfunction in Alzheimer's Disease

The Metabolism of Tumors in the Body

Succinate Dehydrogenase Supports Metabolic Repurposing of Mitochondria to Drive Inflammatory Macrophages

Lipid and Lipoprotein Metabolism in Microglia

Persistent Alterations in Microglial Enhancers in a Model of Chronic Pain

Integrating single-cell transcriptomic data across different conditions, technologies, and species

Spatial reconstruction of single-cell gene expression data

Isolating, culturing, and polarizing primary human adult and fetal microglia

Mapping identifiers for the integration of genomic datasets with the R/Bioconductor package biomaRt

A scaling normalization method for differential expression analysis of RNA-seq data

limma powers differential expression analyses for RNA-sequencing and microarray studies

Enrichr: interactive and collaborative HTML5 gene list enrichment analysis tool

Enrichr: a comprehensive gene set enrichment analysis web server 2016 update

Profiler--a web-based toolset for functional profiling of gene lists from large-scale experiments

UK biobank data: come and get it

UK biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age

PLINK: a tool set for whole-genome association and population-based linkage analyses

METAL: fast and efficient meta-analysis of genomewide association scans

Neuronal atlas of the dorsal horn defines its architecture and links sensory input to transcriptional cell types

Extended Data Fig. 7 Extended Data Fig. 7 . Information on rapid autopsy human spinal cord material. a, Rapid autopsy human spinal cord tissue is provided from the level of the first lumbar nerve (L1) caudally to the conus terminalis. Meningeal layer is removed in preparation of tissue for digestion. b, Donor demographics and sequencing parameters.