The impact of chronic pain on brain gene expression

Supplemental Digital Content is Available in the Text.

Chronic pain-associated brain gene expression is heterogeneous across cell types, largely distinct from other pain traits, and shows that basolateral amygdala microglia are a key cell type.

Abstract

Chronic pain affects one-fifth of American adults, contributing significant public health burden. Chronic pain can be further understood through investigating brain gene expression, potentially informing on brain regions, cell types, and gene pathways. We tested for differentially expressed genes (DEGs) in chronic pain, migraine, lifetime fentanyl and oxymorphone use, and with chronic pain genetic risk in 4 brain regions (dorsal anterior cingulate cortex [dACC], dorsolateral prefrontal cortex [DLPFC], medial amygdala [MeA], and basolateral amygdala [BLA]) and imputed cell type expression data from 304 deeply phenotyped postmortem donors, potentially highlighting variation relevant to factors such as predisposition to chronic pain development, mechanisms of chronic pain development and persistence, and indirect effects of chronic pain and associated treatment or medication, and substance use. We also investigated sex differences in chronic pain differential gene expression. At the brain region level, we identified 2 chronic pain DEGs: B4GALT2 and VEGFB in dACC. At the cell level, we found more than 2000 chronic pain cell-type DEGs, significantly enriched in microglia of the basolateral amygdala. The findings were enriched for mouse microglia pain genes, and for hypoxia and immune response pathways. Small amounts of cross-trait DEG overlap in migraine and chronic pain highlighted medial amygdala cells, and in chronic pain and oxymorphone use suggested the amygdala as a key region. Chronic pain differential gene expression was not significantly different between men and women. Overall, chronic pain-associated gene expression is heterogeneous across region and cell type, is largely distinct from that in pain-related factors and migraine, and our results highlight BLA microglia as a key brain cell type in chronic pain.

1. Introduction

Chronic pain affects roughly one-fifth of adults in the United States,^72,102 and chronic pain conditions (including low back pain and headache disorders) rank in the global top 10 noninfectious diseases contributing to disability-adjusted life years.⁹¹ Chronic pain is defined as pain lasting at least 3 months.^58,85 It is associated with a wide range of physical conditions and it may follow surgery, injury, or physical trauma.^1,24 Mechanisms of chronic pain development are not fully understood, and for many individuals, treatment is not effective, particularly long term.^8,11,41

Understanding the etiopathology of chronic pain requires that researchers investigate and compare the roles of predisposition to pain development, and effects stemming from experience of pain. Studies of predisposition including large genome-wide association studies (GWAS) have uncovered hundreds of trait-associated variants and genes,^21,34 and have suggested various central nervous system (CNS) pathways to be important in chronic pain.²⁶ However, these findings (as with GWAS findings more broadly) have yet to be translated into full mechanistic understanding of disease development or into actionable treatment. One step toward addressing this knowledge gap is through analysis at the gene expression and transcriptomic level, intermediate steps between genotype (GWAS findings) and phenotype (chronic pain). We and others have applied transcriptomic imputation approaches to translate GWAS findings into gene-tissue associations,^40,64 identifying brain regions with a putative role in chronic pain including hippocampus, cerebellum, amygdala, and frontal cortex among others. However, we do not yet know which brain regions and cell types are primarily involved with chronic pain development.

Rodent studies have identified >700 differentially expressed genes in brain and spinal cord, associated with a range of phenotypes including chronic pain, postsurgical pain, neuropathic pain, and spinal injury.^{3,4,68,99,105}

Meanwhile, human gene expression studies have focused on identifying genes associated with specific chronic pain conditions, including fibromyalgia,⁵⁵ osteoarthritis,¹² migraine,⁴⁴ sickle cell disease,⁵⁹ endometriosis,⁹² and others.^{22,23,33,36,81,101} However, sample sizes tend to be small, and typically do not include brain tissue. Here, we present the largest study of differentially expressed genes of brain tissue in chronic pain to date. We assess gene expression in postmortem brain tissues from 304 human donors, from 4 brain regions previously implicated in pain processing and chronic pain (basolateral amygdala [BLA], medial amygdala [MeA], dorsolateral prefrontal cortex [DLPFC], and dorsal anterior cingulate cortex [dACC]), with detailed phenotype data. We ask whether chronic pain has region- and cell type-specific transcriptomic signatures in the human brain. Defining chronic pain is complex, and it is vital to differentiate pain-associated transcriptomic signal from genes associated with predisposition, specific pain-causing conditions (such as migraine), or genes with expression changes resulting from use of pain-alleviating medications (such as opioids). Using donor phenotype and genotype data, we determine whether the transcriptomic signatures identified in our analyses arise from predisposition to chronic pain, experience of pain itself, or due to treatments or medications taken to alleviate pain (eg, fentanyl and oxymorphone).

2. Methods

2.1. Postmortem brain data set description

We obtained data for 304 postmortem human brains, donated at the time of autopsy through US medical examiner's offices, from the Veteran's Association (VA) National Posttraumatic Stress Disorder Brain Bank and Traumatic Stress Brain Research group (Table 1). Retrospective clinical diagnostic review of toxicology and next-of-kin interviews were performed in this previous study, and a wide range of clinical, anthropometric, psychiatric, and life history information, including on trauma history, drug use, and presence of chronic pain, was collected. One hundred fifteen total donors were female and 189 were male, with a mean age of death of 46.75 years.

Table 1.

Demographic information, veteran's association national posttraumatic stress disorder brain bank donors.

Trait	Trait status	Mean age at death	Mean PMI	Female, N (%)	Total N
Chronic pain	0	45.08	29.05	74 (34%)	220
	1	46.36	27.21	41 (49%)	84
Migraine	0	46.34	28.33	92 (36%)	256
	1	40.63	29.67	23 (48%)	48
Oxymorphone	0	45.47	28.66	106 (37%)	287
	1	44.94	26.65	9 (53%)	17
Fentanyl	0	46.02	28.87	99 (37%)	270
	1	40.75	26.11	15 (47%)	32
	NA	41.01	23.75	1 (50%)	2

0, assigned “control” by cohort phenotyping process; 1, assigned as “case” by cohort phenotyping process; fentanyl, lifetime fentanyl use; NA, insufficient information for assignment to case/control; Oxymorphone, lifetime oxymorphone use; PMI, postmortem interval.

Bulk tissue was sampled from 4 regions of each brain; the dACC, DLPFC, BLA, and MeA, totaling 1216 samples. Collection of brain tissue samples postmortem, genotyping, and producing gene expression data for this cohort is described in detail elsewhere.^37,77

2.2. Chronic pain phenotype

Chronic pain in this data set was encoded as present/absent (1/0) by researchers associated with the original data collection and may include a wide range of chronic pain conditions, locations of chronic pain on the body, and varying severity. Chronic pain presence/absence was derived from multiple data sources including health records and next-of-kin interview during retrospective clinical diagnostic review. Migraine, fentanyl lifetime use, and oxymorphone lifetime use were determined using the same approach.

2.3. Imputation of cell type-level gene expression data from bulk tissue

To obtain imputed cell type-level gene expression, we performed cell-type deconvolution and subsequent estimation of cell type-level gene expression. Because choice of reference panel affect cell-type imputation accuracy,⁸³ we used 2 reference panels. We imputed both cortical bulk tissues (DLPFC and dACC) with PsychENCODE reference cell types,^13,46,47 and both amygdala bulk tissues (BLA and MeA), using an amygdala-specific reference panel¹⁰³ (see Supplement, http://links.lww.com/PAIN/C320). Six cell types were available in both reference data sets (endothelial cells, microglia, inhibitory neurons, excitatory neurons, oligodendrocytes, and astrocytes) (see also Supplement, http://links.lww.com/PAIN/C320).

Imputation was performed in the same way across all tissues. First, using the bulk RNA-seq raw counts, we constructed a matrix of transcripts, using CIBERSORTx⁶² and the appropriate reference panel (cortex or amygdala) to estimate cell-type proportions for bulk tissue regions. bMind⁹³ was used to impute cell type-specific gene expression for a total of 6 cell types in each region (astrocytes, endothelial cells, excitatory neurons, inhibitory neurons, microglia, and oligodendrocytes).

2.4. Finding differentially expressed genes using bulk tissue and cell type-level expression data

We calculated surrogate variables using bulk tissue gene expression data and preserving for chronic pain, to capture sources of variability not directly measured in the study⁴⁹ using R package sva.⁴⁸ We then checked for significant correlation between these surrogate variables and our measured variables using Pearson correlation tests. We next removed variables with more than half of samples showing missing data, and then retained variables that were not correlated with at least 1 surrogate variable (representing variables that were not fully captured by surrogate variables). We then checked for collinearity among these retained variables, removing variables and rechecking correlations in a stepwise manner until a set of noncorrelated (noncollinear) variables remained. These uncorrelated-with-surrogate variables, noncollinear variables were included as covariates alongside our surrogate variables and chronic pain phenotype in a linear regression model to find DEGs in chronic pain (see Table S1, http://links.lww.com/PAIN/C320 for regression model per bulk tissue/cell-type analysis). We applied Bonferroni correction within-tissue (total 4 tissues) to our regression P value results. We then repeated DEG analysis steps (surrogate variable analysis, surrogate variable correlation check, collinearity check, and DEG regression analysis) for cell type-level gene expression data, applying Bonferroni correction within-tissue within cell type (total 24 sets) to our results.

We repeated DEG analyses for lifetime fentanyl use, lifetime oxymorphone use, and migraine in both bulk tissue and cell type. We conducted correlation tests among all lifetime opioid use variables—fentanyl was significantly (P < 0.05, Pearson correlation test) correlated with most other lifetime opioid use variables, except oxymorphone. Therefore, we chose fentanyl and oxymorphone for DEG analyses to investigate whether any chronic pain DEGs we found reflected opioid treatment or use as a result of chronic pain.

2.5. Sex differences in differentially expressed genes

We conducted additional DEG regression analyses to investigate potential sex differences in genes differentially expressed in chronic pain (see Supplement, http://links.lww.com/PAIN/C320).

2.6. Multiple test corrections

Unless stated otherwise, multiple test correction was performed in each DEG analysis using a Bonferroni correction for the number of genes tested. For bulk tissue analyses, we applied Bonferroni correction within-tissue, and for cell-type analyses, we applied Bonferroni multiple testing correction within-tissue within cell type. We also apply a secondary experiment-wise multiple test correction (see Supplement, http://links.lww.com/PAIN/C320).

2.7. Pathway analyses of chronic pain differentially expressed genes using FUMA

We conducted gene set enrichment tests within functional mapping and annotation (FUMA)⁹⁵ using significant genes (P_Bonferroni < 0.05). We required at least 10 unique significant genes for inclusion in this analysis. Taking DEGs from cell-type analyses in each tissue (total N = 24 initial lists of significant genes), we included gene lists containing more than 10 unique genes. We used as background genes that were included in each of our DEG analyses per cell type, available in FUMA, and assigned an Ensembl gene ID. As the same genes are measured and tested across tissues and imputed cell types, this results in an initial background genes list of 17,550 genes per analysis.

We then repeated gene set enrichment analyses for traits oxymorphone, fentanyl, migraine, and polygenic risk score (PRS), and compared with chronic pain DEG-enriched pathways to investigate overlap at the pathway level.

2.8. Comparing cell-type differentially expressed genes and disease-associated microglia/activation response microglia and McGill transcriptomic pain signatures data

We next compiled a list of differentially expressed genes from the “transcriptomics pain signatures database” (TSPdb) maintained by the human pain genetics laboratory at McGill University,⁸⁸ where data are available for human whole blood, synovial fluid, and cartilage transcriptomics experiments. We retained genes from human experiments in whole blood (the majority of the human experiments in this database 1,260,387 [76% of experiments]), where the contrast investigated was “pain vs no pain,” where sequencing was through high throughput and not microarray, and where experiments had both men and women included. This resulted in a list of 203,984 experimental results comprising of 23,499 unique genes, 15,968 of which are also tested in each of our cell-type DEG analyses. We conducted a series of Fisher exact tests for enrichment of each list of chronic pain genes per cell type in this McGill database list, with the list of 15,968 shared genes as background. Note that it was not possible to directly test for DEG enrichment (ie, identical gene-tissue results) as opposed to enrichment for genes implicated in DEGs (regardless of tissue), as the only human tissues available in the McGill database are nonbrain.

We then compiled a list of genes from the same database from experiments on mice (Mus musculus) and in brain and nervous tissues (Medical Subject Headings [MeSH] beginning “A08.”) (2,940,318 experiments). We then again included results where contrast investigated was pain vs no pain, experiments used high throughput sequencing, and including both men and women (206,827 experiments). The final database subset contained results for 4 tissues: microglia (A08.637.400), brainstem (A08.186.211.132), spinal ganglia (A08.800.350.340), and sciatic nerve (A08.800.800.720.450.760). We conducted a series of Fisher exact tests for enrichment of each list of chronic pain genes per cell type in each set of database results per tissue.

We also compared our chronic pain microglia DEGs with disease-associated microglia (DAM) and activation response microglia (ARM) genes, using gene lists from publicly available information on DAMs and ARMs from 2 studies^43,74 compiled by Thrupp et al.⁸⁶ (Thrupp et al. Table S1, http://links.lww.com/PAIN/C320 “Genesets previous studies”) (360 genes, of which 349 were also tested in our analyses). We did not perform formal enrichment tests because it was not possible to discern the total number of genes shared as background between our analyses and the Thrupp et al. gene list experiments.

2.9. Polygenic risk score analyses

We performed analyses to uncover DEGs associated with polygenic risk of (proxying predisposition to) chronic pain, using GWAS summary statistics for multisite chronic pain (MCP).³⁹

First, using PLINK 1.9 and 2.0,⁶⁹ we performed recommended quality control (QC) procedure for PRSice2 PRS analyses.¹⁰ We removed duplicate and ambiguous (palindromic) SNPs from our GWAS summary statistics, as well as SNPs with a low minor allele frequency (MAF < 0.01) and low imputation quality (<0.8).

Next, we performed QC steps for genotype data associated with the postmortem brain data set. We excluded SNPs with MAF < 0.01, duplicate SNPs, SNPs with missing genotype call rates >0.01, and SNPs not in Hardy–Weinberg equilibrium (P < 1 × 10⁻⁶). We also removed samples with missing call rates greater than 0.01. Next, we linkage disequilibrium (LD)-pruned SNPs to obtain a set of SNPs in linkage equilibrium, using a window size 200 bp, step size 25 bp, and correlation threshold of 0.25 and PLINK –indep-pairwise. Then, we calculated sample heterozygosity using PLINK 2.0 –het function, and excluded samples with values greater than 3 SDs from the mean sample heterozygosity value. We then used PLINK 1.9 –check-sex function to flag samples where genetic sex did not match reported sex, which were then removed. We filtered for genetic relatedness using PLINK 1.9 –rel-cutoff at a threshold of π > 0.125 (third degree relatives). We performed principal component analysis (PCA) using PLINK PCA and the QC'd genotype data to obtain the first 20 genetic principal components for inclusion in the PRS calculation as covariates, to account for population structure (particularly differing genetic ancestry between source GWAS cohort and postmortem brain tissue donors).

We then used PRSice2 to calculate MCP-PRS for these N = 250 post-QC postmortem brain cohort participants. PRSice2 runs multiple PRS analyses at varying GWAS P value thresholds for inclusion, returning a best-fit PRS (measured by PRS-trait regression R2 value). All SNPs passing QC and associated with MCP were included in the PRS calculation. Differentially expressed gene analyses were then performed as previously described using cell type-level expression data, with the trait of interest being MCP-PRS value for each donor rather than chronic pain.

In addition, we calculated migraine-PRS-DEGs and compared with migraine trait DEGs (see Supplement, http://links.lww.com/PAIN/C320).

3. Results

3.1. Chronic pain affects the brain transcriptome in a region- and cell type-specific manner

To identify genes associated with chronic pain in cortical (dACC and DLPFC) and amygdala (MeA and BLA) brain regions, we performed a differential expression analysis comparing bulk tissue from individuals with and without histories of chronic pain. Two genes were significantly downregulated in the dACC in individuals with chronic pain compared with controls (Fig. 1A), B4GALT2 (P = 1.45 × 10⁻⁶, β = −0.936), and vascular endothelial growth factor B (VEGFB) (P = 1.2 × 10⁻⁷, β = −0.945).

Figure 1.

Chronic pain DEGs are found in the dACC and in the microglia of the basolateral amygdala in bulk and cell type-level analyses. (A) Bulk tissue chronic pain DEGs. Purple: significantly (P_Bonferroni < 0.05) downregulated; dotted line = DEG regression P value significance threshold for that bulk region. (B) Imputed cell-type proportions vary across bulk regions—note neurons only present in reference data for cortex (PsychENCODE) and oligodendrocyte progenitor cells (OPCs) only present in amygdala reference data (Yu et al); OPCs and Neurons both marked “Other” in this figure. (C) Chronic pain cell-type DEGs in microglia per region. Purple = significantly (P_Bonferroni < 0.05) downregulated, orange = significantly (P_Bonferroni < 0.05) upregulated, dotted line = P value significance threshold. For legibility, only the top 15 DEGs are labeled in cell type results panel. dACC, dorsal anterior cingulate cortex; DEG, differentially expressed gene; E × N, excitatory neuron; FC, fold change; InN, inhibitory neuron.

Because bulk tissue gene expression likely represents a combination of signal across cell types sampled in each region, we repeated DEG analyses for each brain region using deconvoluted, imputed cell type-level expression data, across 6 cell types (Fig. 1B; astrocytes, endothelial cells, excitatory neurons, inhibitory neurons, microglia, and oligodendrocytes). Two thousand ninety-three unique genes were significantly associated with chronic pain in these imputed cell types, with most associations (1810/2326, 77.8%) in BLA microglia (Fig. 1C, Fig. S1, http://links.lww.com/PAIN/C320). Given the relatively small proportions of microglia in these brain regions, these associations represent a significant enrichment over what might be expected by chance (P_binomial < 1 × 10⁻⁵⁰). There was also a significant enrichment of associations in medial amygdala endothelial cells (N DEGs = 254, P_binomial = 1.3 × 10⁻⁴²). The findings in amygdala region oligodendrocyte progenitor cells (OPCs) are presented in the Supplement (Table S2, http://links.lww.com/PAIN/C320).

3.2. Chronic pain associated genes are enriched in hypoxia response, ribosome component, and immunity and infection-related pathways

We explored potential functional impact of chronic pain DEGs through gene set enrichment analyses, including all cell-type DEG results with a sufficient number of unique significant genes; dACC microglia, dACC oligodendrocytes, BLA microglia, MeA microglia, MeA astrocytes, MeA endothelial, and MeA oligodendrocytes. Chronic pain-associated genes across these tissues and cell types were enriched for a total of 140 pathway enrichments comprising 101 unique pathways (P_Bonferroni = 1.53 × 10⁻⁶, adjusting across cell and tissue types, Table S3, Supplement, http://links.lww.com/PAIN/C320). We also performed gene set enrichment for cell types where at least 10 significant DEGs were found for traits oxymorphone, fentanyl, migraine, and PRS, and where significant (P < 1.53e-6 and minimum overlap 10 genes) gene set enrichment for these trait DEGs were found, and we assessed for overlap with chronic pain DEG-enriched pathways. Only certain migraine and oxymorphone analyses showed significant gene set enrichment, and of these, 7 of 367 migraine MeA microglia DEG-enriched gene sets overlapped with chronic pain DEG-enriched gene sets (see also Supplement, Table S11, http://links.lww.com/PAIN/C320).

3.3. Sex differences in chronic pain gene expression in the brain

We did not find evidence that association between chronic pain and gene expression is sex-dependent (Table S2, http://links.lww.com/PAIN/C320). Although there were no shared significant chronic pain DEGs between men and women, in secondary analyses of chronic pain DEGs in male-only subset vs female-only subset, DEG effect sizes were found to be significantly correlated between sexes across all genes tested (Table S3, http://links.lww.com/PAIN/C320).

3.4. Significant enrichment of chronic pain genes in transcriptomic experiment results for mouse central nervous system but not human whole blood

Chronic pain genes differentially expressed in the MeA microglia were significantly (Fisher P = 0.01) enriched for mouse spinal ganglia genes from the McGill TSPdb (Table 2). Chronic pain genes DEG in the MeA astrocytes, and in dACC microglia and endothelial cells were also significantly enriched for mouse microglia pain genes. There was no significant enrichment of mouse sciatic nerve or mouse brainstem genes in our chronic pain DEG findings.

Table 2.

Testing for enrichment of chronic pain genes in mouse brain and nerve tissue.

Region	Cell type	Spinal ganglia	Microglia	Sciatic nerve	Brainstem
DACC	Astrocytes	1	1	0.17	1
	Endothelial	1	0.00	1	1
	ExN	0.64	1	0.23	1
	InN	0.56	1	0.14	1
	Microglia	0.46	0.01	0.16	1
	Oligodendrocytes	0.35	1	1	1
DLPFC	Astrocytes	1	1	1	1
	Endothelial	0.33	1	1	1
	ExN	1	1	1	1
	InN	1	1	1	1
	Microglia	1	1	1	1
	Oligodendrocytes	1	1	1	1
BLA	Astrocytes	1	1	1	1
	Endothelial	1	1	1	1
	ExN	1	1	1	1
	InN	1	1	1	1
	Microglia	0.16	0.05	0.11	1
	Oligodendrocytes	0.33	1	1	1
MeA	Astrocytes	0.26	0.01	0.23	1
	Endothelial	0.93	0.05	0.70	1
	ExN	0.46	1	1	1
	InN	1	1	1	1
	Microglia	0.01	1	0.77	1
	Oligodendrocytes	0.65	1	0.23	1

Fisher test P values from tests of enrichment of chronic pain genes (per tissue and cell type) within mouse brain and nerve tissue transcriptomics results for pain experiments (McGill TSPdb). Values ≤0.01 are indicated in bold.

BLA, basolateral amygdala; dACC, dorsal anterior cingulate cortex; DLPFC, dorsolateral prefrontal cortex; MeA, medial amygdala.

Sixty-seven genes identified as significantly differentially expressed in microglia were also up/downregulated in DAMs and/or ARMs (Table 3). Of these, 13 show concordant direction of effect across chronic pain, DAMs, and ARMs; 21 show concordant effect across chronic pain and DAMs; and 17 across chronic pain and ARMs. None of our chronic pain genes were enriched for pain results from human whole blood (Fisher P > 0.05).

Table 3.

Comparing microglia differentially expressed genes to disease-associated microglia/activation response microglia genes.

Region	Gene name	Z	KerenShaul_DAM	SalaFrigerio_ARM
BLA	ADAP2	−5.83	—	Down
	ADSSL1	6.498	Up	Up
	ANKRD55	−5.285	Up	—
	ATF3	5.277	Up	Up
	ATOX1	−4.949	Up	—
	BCL2A1	4.966	—	Up
	CD164	4.888	Down	Down
	FAU	−6.61	Up	Up
	FLT1	6.718	Up	—
	GCNT2	−6.231	—	Up
	GRN	−4.972	Up	—
	HIF1A	6.211	Up	Up
	LDHA	5.411	Up	Up
	MAFF	6.155	Up	—
	MGAT4A	5.232	—	Down
	NRP1	5.467	Up	—
	OLFML3	−6.164	—	Down
	PFDN5	−5.471	Up	—
	PGK1	5.726	Up	—
	PLAUR	5.44	Up	Up
	RAMP1	−5.089	Up	Up
	RHOB	4.97	Down	Down
	RPL13A	−5.665	Up	—
	RPL18	−5.831	Up	—
	RPL19	−5.568	—	Up
	RPL3	−5.315	Up	—
	RPL32	−6.286	Up	Up
	RPL4	−5.262	Up	—
	RPL41	−5.408	—	Up
	RPL6	−5.179	Up	—
	RPL7	−5.009	Up	—
	RPL8	−5.609	Up	—
	RPLP1	−6.13	Up	Up
	RPS11	−5.146	Up	—
	RPS14	−5.4	Up	Up
	RPS16	−5.54	Up	Up
	RPS18	−5.123	Up	Up
	RPS3	−6.556	Up	—
	RPS4X	−6.209	Up	Up
	RPS5	−5.935	Up	Up
	RPS9	−5.137	Up	—
	SALL1	−4.845	—	Down
	SCAMP2	−5.771	—	Down
	SLC11A1	6.67	Up	Up
	SLC16A3	6.479	—	Up
	SLC2A1	6.13	Up	Up
	SLC2A5	5.084	—	Down
	SPP1	5.559	Up	Up
	SSR4	−6.116	Up	—
	SUSD3	−5.587	—	Down
	TGFBR1	6.185	—	Down
	TLR2	6.27	Up	Up
	TMEM119	−5.638	Down	Down
	TREM2	−5.122	Up	Up
	USE1	−6.724	Up	—
dACC	CD84	−5.296	Up	—
	SELPLG	−5.854	Down	Down
	SUSD3	−5.851	—	Down
MeA	ATP6V0E1	−5.104	Up	—
	RPL32	−5.061	Up	Up
	RPL7	−4.997	Up	—
	RPS12	−4.955	Up	Up
	RPS21	−5.682	Up	Up
	RPS3	−5.33	Up	—
	SLC11A1	5.585	Up	Up
	TGFBR1	5.047	—	Down
	TLR2	5.253	Up	Up

Chronic pain microglia DEGs found to be upregulated or downregulated in DAMs and/or ARMs. Z = our DEG analysis beta/SE.

ARM, activation response microglia; BLA, basolateral amygdala; dACC, dorsal anterior cingulate cortex; DAM, disease-associated microglia; MeA, medial amygdala.

3.5. Chronic pain and polygenic risk for multisite chronic pain are associated with differential expression in unique sets of genes

To assess whether chronic pain DEGs represent predisposition to chronic pain or are associated with downstream impacts of pain on brain gene expression, we sought to identify DEGs associated with polygenic risk for MCP at the cell-type level. We found 21 DEGs significantly associated with MCP-PRS (Table 4), none of which were identified as chronic pain DEGs in our previous analyses.

Table 4.

Genes associated with predisposition to chronic pain (polygenic risk score-differentially expressed genes).

Region	Cell type	Gene name	P	P bonferroni	Z
dACC	Oligodendrocytes	FAM13B	2.34 × 10−06	0.0436	−4.879
		FAM184B	1.12 × 10−07	0.0021	5.528
		CDH17	2.67 × 10−06	0.0499	4.849
		TRAF5	4.48 × 10−08	0.0008	5.715
		DOCK3	2.54 × 10−06	0.0475	4.86
		LAMB1	8.77 × 10−08	0.0016	5.579
		PLXDC2	5.86 × 10−07	0.0109	−5.182
		DUSP4	1.65 × 10−06	0.0308	−4.956
		NCOA5	6.59 × 10−09	0.0001	−6.092
		EPHA5	1.96 × 10−07	0.0037	5.413
		SCUBE3	2.57 × 10−06	0.0480	4.858
		PFKFB1	2.45 × 10−07	0.0046	5.367
		ANO10	1.06 × 10−06	0.0198	5.054
		C1QTNF7	2.07 × 10−08	0.0004	5.869
		NFKBID	2.51 × 10−07	0.0047	−5.361
		FZD4	6.33 × 10−07	0.0118	5.165
		MRPL48	1.15 × 10−06	0.0214	−5.037
		PDXDC2P	1.80 × 10−06	0.0336	4.938
		RP11.362A1.1	2.62 × 10−06	0.0489	4.854
MeA	Microglia	ACCS	1.85 × 10−06	0.0346	−4.92
		GPR173	2.47 × 10−06	0.0462	4.856

dACC, dorsal anterior cingulate cortex; MeA, medial amygdala; P_bonf, Bonferroni-adjusted P value.

3.6. Migraine and chronic pain differentially expressed genes significantly overlap in medial amygdala endothelial cells

To distinguish genes associated with chronic pain from genes associated with specific chronic pain conditions, we tested for migraine DEGs in bulk and imputed cell-type data. Although no genes were significantly associated with migraine in bulk tissue, 942 DEGs were associated at the cell-type level, with an overrepresentation in microglia (754/942 DEGs, P_binomial < 2 × 10⁻¹⁶; albeit in the MeA rather than BLA), in dACC oligodendrocytes (52/942, binomial P = 0.027), and in MeA endothelial cells (117/942, binomial P = 3.3 × 10⁻²⁵). Twelve DEGs were associated with both migraine and chronic pain (Table S4, http://links.lww.com/PAIN/C320), representing a significant overlap overall (Fisher P = 0.004), with this enrichment being driven solely by DEGs in MeA endothelial cells (hypergeometric P = 0.02, Table S5, http://links.lww.com/PAIN/C320).

3.7. Chronic pain associations are not driven by opioid use

Chronic pain gene associations may also be confounded by long-term use of medications taken to address pain. To test this, we identified DEGs associated with lifetime fentanyl (fentanyl DEGs) and oxymorphone use (oxymorphone DEGs), and compared both single-gene and genome-wide associations with our chronic pain signatures.

At the bulk level, we identified 1 gene significantly downregulated in individuals with a lifetime history of oxymorphone use in MeA (GPR158.AS1; P = 1.17 × 10⁻⁰⁶). This DEG was not a chronic pain DEG in bulk tissue. At the cell-type level, we identified 744 significant oxymorphone DEGs, of which 13 were also previously associated with chronic pain (Table S6, http://links.lww.com/PAIN/C320, Fig. 2). Oxymorphone DEGs significantly overlapped with chronic pain DEGs (hypergeometric P < 0.05) in both BLA and MeA endothelial cells and microglia (Table S7, http://links.lww.com/PAIN/C320).

Figure 2.

Cell-type DEG overlap across all traits (lifetime oxymorphone use, lifetime fentanyl use, multisite chronic pain PRS, chronic pain, and migraine) tends to be low. DEG, differentially expressed gene; MCP-PRS, multisite chronic pain-polygenic risk score.

No genes were significantly associated with fentanyl use in bulk tissue. Cell type-specific analyses identified 17 significant fentanyl DEGs (Table S8, http://links.lww.com/PAIN/C320); however, none of these DEGs overlapped with chronic pain.

4. Discussion

4.1. Microglia represent a key cell type in chronic pain

We found most significant associations in BLA microglia (77.8% of cell-level associations, [P_binomial < 1 × 10⁻⁵⁰]). We also show chronic pain genes to be significantly enriched in mouse microglia, with partial overlap with genes associated with DAMs and ARMs.^74,97,98

Microglia have previously been implicated in imaging studies of low back pain,⁵³ in the transition from acute to chronic pain in rodent models,⁶⁷ in chronic visceral pain,⁵⁴ and in general pain-related neuroplasticity.³⁵ Studies in rodents found that inhibition of BLA microglia increased antinociceptive effects of opioid drugs⁴⁵ and reduced pain-induced depressive behaviour in a bone cancer pain model.³⁸

4.2. Impact of pain-related factors and conditions on gene expression in the brain

Importantly, our study compares different factors involved in pain. There may be key transcriptional differences associated with living with chronic pain, predisposition to developing chronic pain, and indirect results of chronic pain (eg, opioid use). Distinct genes may also be differentially expressed in chronic pain compared with more specific chronic pain conditions such as migraine.

4.3. Genetic predisposition to chronic pain is transcriptomically distinct from experience of chronic pain

In contrast to chronic pain DEGs, DEGs for polygenic risk of multisite chronic pain were found primarily in dACC oligodendrocytes (19/21 PRS-DEGs). Previous studies showed that oligodendrocytes may contribute to chronic pain development⁵⁶ and that oligodendrocyte ablation can independently induce central pain in rodents.³¹ Therefore, our results suggest one mechanism in heightened predisposition to developing chronic pain involves dACC oligodendrocytes without glial cell involvement.

Polygenic risk score-differentially expressed genes included genes previously associated with pain and cardiac phenotypes in the literature, including Fam13b (associated with mouse cardiac phenotypes),⁸⁰ TRAF5 (vascular inflammation and atherosclerosis),⁸¹ and FAM184B (heart disease and hip pain).⁸⁹ This aligns with established links between chronic pain and heart conditions,⁷¹ including in our own previous work.⁴⁰ Other MCP-PRS–associated genes have previous evidence of association with pain phenotypes, including DOCK3,⁷⁰ LAMB1,⁵¹ DUSP4,⁵⁰ and C1QTNF7.³⁰ Several genes were associated with nerve injury and neuropathies, including NCOA5,⁹ EPHA5,⁵ FZD4,⁸⁴ and MRPL48,⁸⁷ and NFKBID.¹⁶

4.4. Opioid and chronic pain differentially expressed gene overlap is small and concentrated in the amygdala

None of the chronic pain DEGs overlapped with lifetime fentanyl use DEGs, and 13 DEGs overlapped between chronic pain and lifetime oxymorphone use. This overlap represented significant general enrichment (Fisher P = 0.0001), concordant direction of effect, and was driven by 4 specific amygdala cell types (each showing hypergeometric P < 0.05); BLA microglia, BLA endothelial cells, MeA microglia, and MeA endothelial cells. We did not explicitly examine opioid-induced hyperalgesia, but these results in chronic pain–opioid use overlap suggest relevant genes and cell types in this condition.

Overlapping DEGs included those implicated in tumor necrosis factor response and previously linked to Lyme disease (YBX3),^29,73 encoding cytokines (CCL2),³² Raine syndrome and bone mineralization (FAM20C),^52,94 autocrine signaling and lipid storage (HILPDA),¹⁵ connective tissue biogenesis (LOXL2),⁷⁶ neurogenesis (RASF10),⁷⁸ and Alzheimer disease, Parkinson disease, schizophrenia, and posttraumatic stress disorder (SERPINA3).^7,75,82,104 YBX3 gene expression in the brain has also been previously linked to posttraumatic stress disorder.²⁸

4.5. Migraine and chronic pain differentially expressed gene overlap is small and driven by amygdala endothelial cells

Pathophysiology of migraine is not completely understood,^2,60 and there may be distinct mechanisms underlying condition-specific processes and potential tissue damage vs condition-associated chronic pain. We found small (12 DEGs) but significant overlap driven by MeA amygdala endothelial cells.

Endothelial dysfunction has been previously implicated in migraine, although it is not definitively known if this is a cause or consequence.⁶⁵ Our findings suggest that transcriptomic changes at endothelial cells and migraine-related pain specifically are linked, and that nonpain migraine symptoms may be associated with other cell types.

Furthermore, specific genes differentially expressed in endothelial cells in both migraine and chronic pain (Table S13, http://links.lww.com/PAIN/C320) tend to play a role in bone growth, immune system function, and fatty acid metabolism. These genes include GLUD1P3, a pseudogene possibly related to glutamate metabolism, and KDM4B that encodes a protein involved in histone demethylation and associated with intellectual disability disorders²⁵ and general brain development—this protein also interacts with oxygen homeostasis regulating proteins and hypoxia-induced genes. Hypoxia has been shown to trigger migraine attacks²⁷ and has been linked to chronic pain development with obstructive sleep apnea.⁴² In this study, we also found that hypoxia response pathways were enriched for chronic pain DEGs.

Another gene, LIFR, is downregulated in both migraine and chronic pain. Loss of function LIFR mutations cause Stuve–Wiedemann syndrome, a syndrome involving dysfunction in bone growth regulation and the autonomic nervous system.⁷⁹ It may be that relatively lower expression of LIFR in migraine and chronic pain involves similar pathways. Migraine has also been associated with autonomic nervous system dysfunction,⁶⁶ as has chronic pain.^61,96

NRBF2, downregulated in both chronic pain and migraine, encodes a protein involved in autophagy and fatty acid metabolism,¹⁰⁰ and associated with childhood-onset dystonia.¹⁴ Finally, TNFRSF6B, a tumor necrosis factor superfamily member, is upregulated in both migraine and chronic pain—mutations in this gene are associated with spondyloenchondrodysplasia with immune dysregulation (including immunodeficiency and autoimmune conditions such as thrombocytopenia) and dyskeratosis congenita,⁸⁰ which often involves progressive bone marrow failure and associated immune dysfunction. In our previous work, we also saw a link between gene expression in the amygdala that contributes to chronic pain development, anemias, and primary thrombocytopenia,⁴⁰ and the findings here suggest that this relationship extends to pain in migraine.

Overall, while current migraine treatments already include immunomodulating agents (eg, calcitonin gene related peptide monoclonal antibodies), our findings suggest that other potential avenues by which immune dysfunction can increase pain in migraine and so other possible treatment targets.

In addition, VEGF, encoded by VEGFA, is a “protective angiogenic factor” in migraine.⁶⁵ VEGFB (in contrast to VEGFA) is “dispensable” in new blood vessel growth in normal tissue but may play a more essential protective role for vasculature during neurodegenerative disease progression and stroke.⁹⁰ VEGF has also been implicated in rodent and human studies of neuroplasticity, stress response, and antidepressant action.^17–20,63 Our results showed VEGFB was downregulated in bulk dACC tissue in chronic pain, indicating VEGFB may represent a protective angiogenic factor in a general chronic pain context outside of migraine.

4.6. Sex differences in gene expression in chronic pain

We did not find evidence that chronic pain DEGs are sex-dependent (Table S2, S3, http://links.lww.com/PAIN/C320). While sex, gender, and interplay between these multifactorial concepts are important in chronic pain, our findings suggest that chronic pain brain gene expression is broadly similar between men and women.

4.7. Limitations

While a large amount of phenotypic data for each donor are available (160+ variables on lifestyle, trauma, psychiatric and medical diagnoses, and lifetime substance and medication use),^6,57 which allows us to thoroughly account for sources of variance in gene expression that are not due to chronic pain or other trait-of-interest status, chronic pain status is a broad and dichotomized variable. Opioid use phenotypes were similarly broad, and lifetime use did not delineate between prescribed/nonprescribed, and current or past use. When comparing lifetime fentanyl use and lifetime oxymorphone use, “case” numbers are higher for fentanyl, but far fewer DEGs are found. This may reflect different clinical uses of these drugs, with fentanyl as a more discrete rather than chronic exposure, making effects on gene expression more difficult to capture postmortem and potentially many years after a brief exposure.

Our sample size (304 donors) is relatively low for PRS analyses, however, paired with the large base GWAS (N > 380,000), is considered acceptably powered.¹⁰ In addition, while the MCP phenotype may not exactly match that of the “chronic pain” noted in the donor cohort, it is a broad chronic pain phenotype significantly genetically and phenotypically correlated with a range of chronic pain conditions,³⁹ while also being the most well-powered GWAS of a broad chronic pain phenotype available.

Although we analyse potential sex–chronic pain interactions in DEGs and find no sex interaction effects, we are likely underpowered to fully explore potential sex-specific differences in our male/female subsets.

Finally, there are no human brain tissue experiments available in the McGill TSPdb for direct comparisons with our results, although we explored enrichment in human whole blood and mouse brain and nervous system.

5. Conclusions

Our study is the largest to examine the impact of chronic pain across brain regions through differential expression analysis of brain tissue. Uniquely, our sample is almost perfectly balanced in reported sex in donors with chronic pain. Our results suggest that chronic pain affects gene expression in a heterogeneous way. In particular, our results suggest that BLA microglia are a key cell type in chronic pain, and that migraine, opioid use, genetic predisposition to chronic pain, and chronic pain are all largely distinct at the brain transcriptome level. Points of overlap, although small, can inform potential shared mechanisms to inform treatment, including on tailoring of treatment to stage in chronic pain development.

Conflict of interest statement

J.H.K. has consulting agreements (less than US $10,000 per year) with the following: Aptinyx, Inc. Biogen, Idec, MA, Bionomics, Limited (Australia), Boehringer Ingelheim International, Epiodyne, Inc., EpiVario, Inc., Janssen Research & Development, Jazz Pharmaceuticals, Inc., Otsuka America Pharmaceutical, Inc., Spring Care, Inc., Sunovion Pharmaceuticals, Inc.; is the cofounder for Freedom Biosciences, Inc.; serves on the scientific advisory boards of Biohaven Pharmaceuticals, BioXcel Therapeutics, Inc. (Clinical Advisory Board), Cerevel Therapeutics, LLC, Delix Therapeutics, Inc., Eisai, Inc., EpiVario, Inc., Jazz Pharmaceuticals, Inc., Neumora Therapeutics, Inc., Neurocrine Biosciences, Inc., Novartis Pharmaceuticals Corporation, PsychoGenics, Inc., Takeda Pharmaceuticals, Tempero Bio, Inc., Terran Biosciences, Inc.; has stock options with Biohaven Pharmaceuticals Medical Sciences, Cartego Therapeutics, Damona Pharmaceuticals, Delix Therapeutics, EpiVario, Inc., Neumora Therapeutics, Inc., Rest Therapeutics, Tempero Bio, Inc., Terran Biosciences, Inc., Tetricus, Inc.; and is editor of Biological Psychiatry with income greater than $10,000.

Supplemental digital content

Supplemental digital content associated with this article can be found online at http://links.lww.com/PAIN/C320.

Supplementary Material

jop-166-e689-s001.pdf ^{(8.9MB, pdf)}

jop-166-e689-s002.pdf ^{(310.7KB, pdf)}