The genetic architecture of fibromyalgia across 2.5M individuals

A multi-ancestry genome-wide association study meta-analysis of 2,563,755 individuals (54,629 cases and 2,509,126 controls) from 11 cohorts identified 26 risk loci for fibromyalgia, with the strongest association in the HTT gene. The study, published in Nature Medicine, found that fibromyalgia heritability is exclusively enriched in brain tissues and neural cell types, and showed strong genetic correlations with chronic pain, psychiatric, and somatic disorders, defining fibromyalgia as a central nervous system disorder.

Abstract Fibromyalgia is a common and debilitating chronic pain syndrome of poorly understood etiology. Here we conduct a multi-ancestry genome-wide association study meta-analysis across 2,563,755 individuals 54,629 cases and 2,509,126 controls from 11 cohorts, identifying 26 risk loci for fibromyalgia. The strongest association was with a coding variant in HTT , the causal gene for Huntington’s disease. Gene prioritization implicated the HTT regulator GPR52 , as well as diverse genes with neural roles, including DCC , DRD2 / NCAM1 , MDGA2 and CELF4 . Fibromyalgia heritability was exclusively enriched within brain tissues and neural cell types. Fibromyalgia showed strong, positive genetic correlation with a wide range of chronic pain, psychiatric and somatic disorders, including genetic correlations above 0.7 with low back pain, post-traumatic stress disorder and irritable bowel syndrome. Despite large sex differences in fibromyalgia prevalence, the genetic architecture of fibromyalgia was nearly identical between males and females. This study provides robust genetic evidence defining fibromyalgia as a central nervous system disorder, thereby establishing a biological framework for its complex pathophysiology and extensive clinical comorbidities. Similar content being viewed by others Main Fibromyalgia is a multifaceted syndrome encompassing chronic widespread musculoskeletal pain, fatigue, unrefreshing sleep, cognitive impairment and somatic symptoms. It frequently co-occurs with other pain conditions, irritable bowel syndrome, myalgic encephalomyelitis/chronic fatigue syndrome, autoimmune and neuropsychiatric disorders, and metabolic syndrome 1. Fibromyalgia is considered the prototypical central sensitization or nociplastic pain syndrome 2, whereby the central nervous system becomes sensitized to peripheral nociceptive input, leading to increased pain response to painful stimuli hyperalgesia and typically non-painful stimuli allodynia . Whether fibromyalgia has an autoimmune component is a matter of long-standing debate 3 /articles/s41591-026-04492-6 ref-CR3 . 4 /articles/s41591-026-04492-6 ref-CR4 While fibromyalgia has a partly genetic basis 5, attempts to find specific genetic loci have proven elusive . We searched for fibromyalgia genetic risk factors through combined genetic analysis of over 2.5 million individuals, identifying risk loci, prioritizing causal genes, mapping heritability-enriched tissues and cell types, quantifying genetic overlap with comorbidities, and assessing inter-sex genetic consistency. 6 ref-CR6 , 7 ref-CR7 , 8 ref-CR8 , 9 /articles/s41591-026-04492-6 ref-CR9 Results Associations with 26 variants establish genetic underpinnings of fibromyalgia We performed multi-ancestry genome-wide association study GWAS meta-analyses of fibromyalgia International Classification of Diseases, Tenth Revision ICD-10 code M79.7 in inpatient and/or primary care records across 11 cohorts Fig. 1 /articles/s41591-026-04492-6 Fig1 . We performed quality control and GWAS per cohort–ancestry pair, corrected for inflation using linkage disequilibrium LD score regression intercepts and conducted fixed-effects inverse-variance-weighted meta-analysis across cohorts and ancestries. Our primary meta-analysis encompassed 54,629 fibromyalgia cases 90% European and 2,509,126 controls, totaling 2,563,755 individuals Supplementary Table 1 /articles/s41591-026-04492-6 MOESM3 . Prevalence varied sixfold across cohorts, from 1.2% Estonian Biobank to 7.5% Michigan Genomics Initiative , with a median prevalence of 2.5% All of Us . We found 26 independent genome-wide-significant P < 5 × 10−8 risk loci for fibromyalgia Table 1 /articles/s41591-026-04492-6 Tab1 and Fig. 2 /articles/s41591-026-04492-6 Fig2 , with largely consistent effect sizes across cohorts Supplementary Fig. 1 /articles/s41591-026-04492-6 MOESM1 . We performed multiple secondary meta-analyses: leave-one-cohort-out analyses to enable unbiased genetic risk prediction, European-only meta-analyses to reduce bias in LD-based analyses Extended Data Fig. 1 /articles/s41591-026-04492-6 Fig6 and sex-stratified meta-analyses Extended Data Figs. 2 /articles/s41591-026-04492-6 Fig7 and 3 /articles/s41591-026-04492-6 Fig8 , Supplementary Figs. 2 /articles/s41591-026-04492-6 MOESM1 and 3 /articles/s41591-026-04492-6 MOESM1 , and Supplementary Table 2 /articles/s41591-026-04492-6 MOESM3 . Consistent with known sex differences in fibromyalgia prevalence 10, the vast majority of cases 47,895 cases, 87.7% were female. While the lower number of male cases yielded less statistical power 2 loci in males versus 21 in females , the inter-sex genetic correlation in European-ancestry individuals was 1.03 P = 2.9 × 10 −16 , statistically indistinguishable from perfect correlation, and odds ratios for the 26 lead variants from the primary meta-analysis were similar in males and females Extended Data Fig. 4 /articles/s41591-026-04492-6 Fig9 . Thus, fibromyalgia’s genetic architecture is largely consistent between sexes. We prioritized causal genes via manual literature review, complemented by a deCODE genetic annotation pipeline that considers quantitative trait loci and coding variants in strong LD Supplementary Table 3 /articles/s41591-026-04492-6 MOESM3 and Methods /articles/s41591-026-04492-6 Sec12 and the FLAMES machine learning method. We discuss notable prioritizations here, many of which implicate brain function or brain disease. The top fibromyalgia association is with a coding variant in the Huntingtin gene The most significant association in the primary analysis was with rs149109767-A OR = 1.09, P = 2.2 × 10−12 , an inframe glutamic acid deletion in the Huntingtin HTT gene. HTT , the causal gene for Huntington’s disease HD , is ubiquitously expressed and involved in diverse cellular processes including signaling, transcriptional regulation, cell division, endocytosis, autophagy and axonal transport, and is required for healthy neurodevelopment 11. Our fibromyalgia variant is located in exon 58 of the 67-exon HTT gene, far from the exon 1 repeat-expansion region that causes HD, but is part of the HD-associated ‘A1’ haplotype found in Europeans . This variant shows the largest effect among the 26 lead variants: carriers have about 9% increased odds of fibromyalgia. 12 /articles/s41591-026-04492-6 ref-CR12 We investigated whether our fibromyalgia variant is associated with HD using a six-European-cohort HD GWAS meta-analysis available at deCODE N case/control = 184/1,198,665 . The top HD association rs71180116-CCAGCAGCAGCAGCAGCAGCAGCAGCAGCAGCAGCAG, OR = 44.3, P = 2.62 × 10−30 was not in strong LD with our lead fibromyalgia variant r 2 = 0.0055; Lewontin’s normalized disequilibrium coefficient D = 0.258 and showed no association with fibromyalgia OR = 1.027, P = 0.41 . While our fibromyalgia variant rs149109767-A did associate with HD risk OR = 2.118, P = 0.00354 , this association disappeared in the UK Biobank after conditioning on rs71180116 conditional OR = 1.328, P = 0.297 , an effect replicated in the smaller Icelandic GWAS. We found a second association with a possible HD link: rs6657341-A OR = 0.961, P = 7.1 × 10−9 , an intronic variant in RABGAP1L , ~150 kilobases kb upstream of its second-nearest gene GPR52 . While RABGAP1L has no clear link to fibromyalgia, GPR52 is a brain-specific orphan G-protein-coupled receptor being investigated as a drug target for HD, as it regulates HTT levels 13,14. Additional associations implicate pathways in pain processing Several associations implicated pain processing pathways. One involved rs11664242, an intergenic variant ~30 kb upstream of its nearest coding gene CELF4 . CELF4 encodes an RNA-binding protein and translational regulator predominantly expressed in excitatory neurons, regulating synaptogenesis during neurodevelopment by modulating translation of synaptic mRNAs 15. CELF4 reduces pain and sensory sensitivity by negatively regulating nociceptor excitability , and 16 /articles/s41591-026-04492-6 ref-CR16 CELF4 gene therapy is an investigational therapeutic strategy for chronic pain. Another association was found with rs6977550, an intergenic variant ~30 kb downstream of its nearest coding gene STK31 . STK31 encodes a testis-specific protein kinase with unclear relevance to fibromyalgia; however, the variant is ~400 kb upstream of NPY its sixth-nearest gene , an abundant neuropeptide that modulates appetite, circadian rhythm, anxiety and immune responses 17 and has both pro- and anti-nociceptive effects . 18 /articles/s41591-026-04492-6 ref-CR18 We also identified rs10166361, an intronic variant in ARHGAP15 , which encodes a Rho GTPase-activating protein that regulates cytoskeletal organization 19. In mouse models, ARHGAP15 knockout impairs cortical interneuron excitability causing subclinical seizures and hippocampal function causing memory impairment . The variant is also ~450 kb downstream of 19 /articles/s41591-026-04492-6 ref-CR19 , 20 /articles/s41591-026-04492-6 ref-CR20 KYNU the third-nearest gene , which encodes kynureninase, an enzyme that cleaves kynurenine into anthranilic acid . Kynurenine and its metabolites are involved in pain processing, and the kynurenine pathway is a proposed therapeutic target for pain 21 /articles/s41591-026-04492-6 ref-CR21 . Gene-based tests have linked 22 /articles/s41591-026-04492-6 ref-CR22 KYNU to depressive symptoms, often comorbid with chronic pain . 23 /articles/s41591-026-04492-6 ref-CR23 Other significant associations point to brain-related mechanisms One particularly notable brain association was with rs2734833, an intronic variant in DRD2 , a locus with psychiatric and sleep associations. DRD2 encodes the dopamine D2 receptor, which influences motivation and cognition 24, and is the primary target of most antipsychotic drugs. The variant lies ~150 kb downstream of NCAM1 , the fourth-nearest gene, which encodes a cell surface receptor CD56 with both nervous and immune roles. In the nervous system, it regulates neural cell adhesion, neuronal migration, neurite outgrowth, axon guidance and synaptic plasticity, and plays a role in memory . In the immune system, CD56 is the primary marker for natural killer cells and distinguishes their two main subtypes. It helps natural killer cells attach to target cells and trigger the release of cytotoxic granules that kill target cells 25 /articles/s41591-026-04492-6 ref-CR25 . 26 /articles/s41591-026-04492-6 ref-CR26 The second-most significant association was with rs9862795, an intergenic variant located ~8 kb upstream from CAMKV . CAMKV encodes a pseudokinase from the Ca2+/calmodulin-dependent protein kinase family that is predominantly brain-expressed and critical for dendritic spine maintenance, synaptic transmission and synaptic plasticity 27. The lead variant is also in strong LD r 2= 0.905; Supplementary Table 3 /articles/s41591-026-04492-6 MOESM3 with two variants in a pleiotropic tyrosine kinase receptor gene, MST1R , which regulates macrophage activation, inflammation and antiviral defence. Two other lead variants were rs809955, an intronic variant in MAML3 , and rs12587412, an intergenic variant ~40 kb downstream of its nearest coding gene MDGA2 . MAML3 encodes a transcriptional coactivator in the Notch signaling pathway 28. Notch signaling is vital for development and cell-fate decisions: in development, inhibition promotes neural differentiation, while in adulthood, activation influences neural stem cell regulation, neuronal survival, migration and synaptic plasticity . 29 /articles/s41591-026-04492-6 ref-CR29 , 30 /articles/s41591-026-04492-6 ref-CR30 MDGA2 is predominantly expressed in the brain and regulates synaptic organization , axonal migration 31 /articles/s41591-026-04492-6 ref-CR31 , BDNF/TrkB signaling 32 /articles/s41591-026-04492-6 ref-CR32 and excitatory neurotransmission 33 /articles/s41591-026-04492-6 ref-CR33 . 31 /articles/s41591-026-04492-6 ref-CR31 MDGA2 haploinsufficiency is associated with autism spectrum disorder . 33 /articles/s41591-026-04492-6 ref-CR33 Another neurodevelopmental association was with rs62100765, an intronic variant in DCC , which was the sole coding gene within 1 megabase Mb . DCC encodes a transmembrane receptor that guides midline-crossing axons during neurodevelopment 34 and regulates myelin structure and axon domain organization . 35 /articles/s41591-026-04492-6 ref-CR35 We also found an association with rs6748639, an intergenic variant located ~100 kb upstream of the nearest coding gene, SRD5A2 , and ~350 kb upstream of SPAST , the fifth-nearest gene and the one prioritized by FLAMES. SRD5A2 is one of three enzymes that convert testosterone to the more potent androgen dihydrotestosterone, which is critical to normal male sexual development: biallelic loss causes 5α-reductase deficiency, which can result in female-appearing external genitalia 36. Some studies report an inverse association between testosterone level and pain perception . An alternative causal gene candidate at this locus, 37 /articles/s41591-026-04492-6 ref-CR37 SPAST , encodes the neuronally expressed protein spastin, which regulates microtubule length and disassembly ; haploinsufficiency is the most common cause of hereditary spastic paraplegia 38 /articles/s41591-026-04492-6 ref-CR38 . 39 /articles/s41591-026-04492-6 ref-CR39 Several associations implicated genes linked to intellectual disability and developmental disorders. One was with rs359243, an intergenic variant located in a gene-dense region ~200 kb downstream from its nearest coding gene, the transcription factor BCL11A . Besides its canonical role in hematopoiesis and hemoglobin regulation, BCL11A has neurodevelopmental roles, and haploinsufficiency causes intellectual disability and autism 40. Another developmental disorder association was with rs34261763, an intronic variant in BPTF , which encodes a subunit of the developmentally essential NURF chromatin remodeling complex 41. BPTF haploinsufficiency causes a neurodevelopmental disorder with intellectual disability, developmental and speech delay, microcephaly, and facial and limb dysmorphia . 41 /articles/s41591-026-04492-6 ref-CR41 We also found an association with rs11167957, an intronic variant in PPP2R2B , a brain-enriched subunit of the serine/threonine phosphatase PP2A ref. 42 . CAG repeat expansion in PPP2R2B causes spinocerebellar ataxia 12, an autosomal dominant speech and motor disorder , and 43 /articles/s41591-026-04492-6 ref-CR43 PPP2R2B missense variants are linked to intellectual disability with developmental delay . The variant is also ~550 kb upstream of the orphan brain-specific G protein-coupled receptor gene 42 /articles/s41591-026-04492-6 ref-CR42 GPR151 its 5th-nearest gene , which mediates neuropathic pain via microglial activation and neuroinflammation . 44 /articles/s41591-026-04492-6 ref-CR44 , 45 /articles/s41591-026-04492-6 ref-CR45 Finally, we found an association with rs1788821, an intronic variant in NPC1 . NPC1 encodes a transmembrane protein vital for intracellular cholesterol trafficking, and biallelic loss of NPC1 function causes Niemann–Pick disease type C1, characterized by lysosomal lipid accumulation, neuroinflammation, abnormal dendrite growth, neurofibrillary tangles and neuroaxonal dystrophy 46. The lead variant is in strong LD with missense variants in NPC1 and an autophagy-related gene , 47 /articles/s41591-026-04492-6 ref-CR47 RMC1 r 2= 0.998 and 0.990, respectively; Supplementary Table 3 /articles/s41591-026-04492-6 MOESM3 . Phenome-wide associations of lead variants point to substantial pleiotropy We cross-referenced the 26 lead variants and their high-LD proxies r 2 0.8 with the GWAS Catalog. A total of 20 traits were associated with at least 4 of the 26 lead variants Fig. 3 /articles/s41591-026-04492-6 Fig3 top and Supplementary Table 4 /articles/s41591-026-04492-6 MOESM3 . The most frequent association was with educational attainment 10 variants , followed by body mass index 8 , type 2 diabetes 7 , pain intensity 6 and insomnia 6 . Notable psychiatric associations included depressive symptoms 4 variants , schizophrenia 3 , depression 2 and suicide attempt 1 . Remaining associations spanned pain, immune and behavioral traits for example, multisite chronic pain, white blood cell count, neuroticism , and anthropometric and metabolic traits for example, weight, visceral adipose tissue, metabolic syndrome . We also performed a phenome-wide association study PheWAS , associating the 26 lead variants with 330 diseases, leveraging preexisting GWAS meta-analyses of the Million Veteran Program, FinnGen and the UK Biobank ‘MVP–Finngen–UKBB’ . After Bonferroni correction across the 330 diseases and 26 variants tested, 20 diseases were associated with 3 or more lead variants Fig. 3 /articles/s41591-026-04492-6 Fig3 middle and Supplementary Table 5 /articles/s41591-026-04492-6 MOESM3 . These diseases largely fell into three categories: pain and pain-associated traits dorsalgia, spondylosis, joint pain, limb pain, spinal stenosis, cervicalgia , metabolic syndrome and its comorbidities type 2 diabetes, obesity, sleep apnea, hypertension, coronary heart disease, gout, lipoprotein and lipid disorders, hypercholesterolemia, cholelithiasis and immune disease autoimmune hypothyroidism, asthma . The most common association was dorsalgia back pain, 13 of 26 variants associated ; the second-most common was gastroesophageal reflux disease 9 variants . Applying the same approach to 124 drug prescription GWAS from FinnGen Fig. 3 /articles/s41591-026-04492-6 Fig3 bottom and Supplementary Table 6 /articles/s41591-026-04492-6 MOESM3 , the most frequent associations were with analgesics 14 variants and anti-inflammatory and anti-rheumatic drugs 12 variants . A total of 11 variants were associated with both categories, reflecting prescription of nonsteroidal anti-inflammatory drugs included in both categories for pain. A further 10 drug categories were associated with at least 3 lead variants, including psychoactive drugs muscle relaxants, psycholeptics, psychoanaleptics , treatments for metabolic syndrome and its comorbidities diabetes drugs, lipid-modifying agents, renin–angiotensin agents and immune modulators anti-asthmatics, anti-histamines . These results show that the top genetic risk variants for fibromyalgia are highly pleiotropic, broadly influencing risk for many of the disorder’s most common comorbidities and likelihood of prescription of the drugs used to treat them. Fibromyalgia heritability is exclusively enriched in neural tissues and cell types We used LD score regression applied to specifically expressed genes LDSC-SEG to quantify fibromyalgia heritability enrichment near genes with enriched expression in each of 53 tissues in the Genotype-Tissue Expression GTEx project, and each of 119 cell types in the ~20 million whole-mouse PanSci single-cell atlas Fig. 4 /articles/s41591-026-04492-6 Fig4 and Supplementary Tables 7 /articles/s41591-026-04492-6 MOESM3 and 8 /articles/s41591-026-04492-6 MOESM3 . We found strong and exclusive enrichment within the nervous system: all 5 GTEx tissues with Bonferroni-significant enrichment were brain regions cortex, caudate, frontal cortex, putamen and anterior cingulate cortex , and 12 of 13 enriched cell types were neuronal, with the 13th pulmonary neuroendocrine cells being neuron-like. The strongest cell-type association was with neurons from the dentate gyrus P = 1.3 × 10−6 , a hippocampal region critical for contextualizing sensory experience, including pain. Significant enrichments were also found in enteric neurons P = 7.9 × 10−6 and diverse interneurons, cortical projection neurons and striatal neurons. Even when grouping cell types by lineage with the goal of increasing power, the only significant lineage was neural P = 2.3 × 10−4; P 0.05 for all other cell types; Supplementary Table 9 /articles/s41591-026-04492-6 MOESM3 . Collectively, these results implicate a broad network of neurons across the central and peripheral nervous systems, including sensory-processing neurons, in the etiology of fibromyalgia. Pervasive genetic correlations with comorbid disorders We computed genetic correlations between the European-only meta-analysis and each of 855 diseases in FinnGen, yielding 337 Bonferroni-significant genetic correlations Fig. 5 /articles/s41591-026-04492-6 Fig5 and Supplementary Table 10 /articles/s41591-026-04492-6 MOESM3 . The strongest genetic overlap was with pain and musculoskeletal disorders. The strongest signals arose from conditions characterized by widespread or syndromic pain, including cervicobrachial syndrome genetic correlation r g = 0.87, P = 9.2 × 10−65 , myalgia r g = 0.82, P = 9.4 × 10−45 , hypermobility syndrome r g = 0.79, P = 8.4 × 10−13 and soft tissue disorders r g = 0.73, P = 5.0 × 10−190 . We also observed substantial genetic correlations with a range of common localized pain presentations, from joint pain r g = 0.81, P = 1.7 × 10−160 and low back pain r g = 0.75, P = 1.1 × 10−137 to trochanteric bursitis r g = 0.82, P = 3.1 × 10−21 and migraine r g = 0.61, P = 2.2 × 10−73 . Fibromyalgia showed substantial genetic correlations with mental and behavioral disorders, with a magnitude for some traits, such as post-traumatic stress disorder r g = 0.78, P = 6.6 × 10−16 , that rivaled those of pain conditions. Strong correlations were also evident for dissociative disorders r g = 0.63, P = 5.0 × 10−6 , somatoform disorder r g = 0.67, P = 8.8 × 10−26 and depression r g = 0.63, P = 4.1 × 10−118 , alongside more modest correlations with generalized anxiety disorder r g = 0.46, P = 3.8 × 10−20 and insomnia r g = 0.47, P = 2.0 × 10−43 . Significant genetic overlap extended to other commonly co-occurring conditions, including disorders of the digestive, genitourinary and respiratory systems. Among digestive disorders, we observed strong signals for irritable bowel syndrome r g = 0.70, P = 2.1 × 10−35 and functional dyspepsia r g = 0.67, P = 9.1 × 10−24 . We also identified correlation with genitourinary conditions such as polycystic ovarian syndrome r g = 0.59, P = 8.3 × 10−39 and the respiratory condition chronic laryngitis r g = 0.75, P = 1.1 × 10−6 . Genetic correlations with autoimmune disorders were comparatively modest. The highest correlations in this category were with psoriatic arthropathies r g = 0.41, P = 5.4 × 10−17 and Sjögren’s syndrome r g = 0.39, P = 1.7 × 10−6 . Other rheumatological conditions with significant genetic correlations included rheumatoid arthritis r g = 0.33, P = 3.1 × 10−16 , psoriasis r g = 0.29, P = 6.0 × 10−15 and autoimmune hypothyroidism r g = 0.21, P = 4.2 × 10−17 . Given the genetic correlations observed with asthma r g = 0.51, P = 2.5 × 10−57 and rheumatoid arthritis RA and the long-standing debate on whether fibromyalgia is an auto immune disease 4, we explored genetic correlations with subtypes of these diseases. Asthma can be broadly classified as T2 high or T2 low, defined by high or low levels of T2 molecules such as eosinophils, with half of individuals with asthma falling into each category . While the mechanisms driving T2-high typical asthma are increasingly understood, much less is known about T2-low asthma 48 /articles/s41591-026-04492-6 ref-CR48 . Using GWAS results from T2-high and T2-low asthma at deCODE 48 /articles/s41591-026-04492-6 ref-CR48 , we found much stronger genetic correlation between fibromyalgia and T2-low 49 /articles/s41591-026-04492-6 ref-CR49 r g= 0.44, P = 3.5 × 10 −24 relative to T2-high r g= 0.27, P = 7.4 × 10 −14 asthma, consistent with fibromyalgia’s lack of genetic correlation with eosinophil levels r g= 0.04, P = 0.1 . Similarly, RA is classified as seronegative or seropositive based on the absence or presence of anticitrullinated protein antibodies and rheumatoid factor, with seronegative RA considered less severe than seropositive RA. We found that fibromyalgia is more genetically correlated with the seronegative form of RA r g = 0.41, P = 2.7 × 10−15 than with seropositive RA r g = 0.24, P = 3.0 × 10−10 , an association also observed clinically 50. Thus, fibromyalgia is genetically correlated with the more heterogeneous and less well-defined forms of asthma and RA, further distinguishing fibromyalgia’s genetic architecture from that of well-characterized, peripherally driven inflammatory and autoimmune disease. Collectively, this landscape of genetic correlations highlights the profound pleiotropy of fibromyalgia genetic risk factors. This apparent paradox—widespread somatic correlations despite a genetic architecture strongly enriched in neuronal cell types—is resolved when considering that many of the abovementioned conditions are themselves rooted in the central nervous system. A partially shared, centrally mediated genetic etiology could drive their frequent co-occurrence with fibromyalgia. Genetic risk prediction is modestly effective at patient stratification To enable fibromyalgia genetic risk prediction, we constructed polygenic risk scores PRSs for the European and multi-ancestry leave-UK Biobank-out meta-analyses Data Availability . We evaluated the performance of the multi-ancestry leave-UK Biobank-out PRS within the UK Biobank. The PRS showed modest predictive ability, commensurate with fibromyalgia’s modest observed-scale heritability of 10.4% 95% confidence interval CI 9.8% to 11.0% . Performance was highest among European-ancestry UK Biobank participants area under the receiver-operating curve AUC = 0.59 , with attenuated South Asian AUC = 0.55 and African AUC = 0.55 accuracies Extended Data Fig. 5a /articles/s41591-026-04492-6 Fig10 reflecting the mostly European composition of our samples. Despite modest AUCs, the PRS effectively stratified individuals by risk, particularly Europeans in which power and predictive accuracy were greatest. Compared with the middle quintile, European-ancestry participants in the highest PRS quintile had a fibromyalgia odds ratio of 1.5 95% CI 1.4 to 1.7 , while those in the lowest quintile had an odds ratio of 0.63 95% CI 0.57 to 0.68; Extended Data Fig. 5b /articles/s41591-026-04492-6 Fig10 . This risk stratification translated to marked differences in disease prevalence between the lowest and highest PRS quintiles Extended Data Fig. 5c /articles/s41591-026-04492-6 Fig10 : among European-ancestry participants, prevalence increased from approximately 1.0% in the lowest quintile to 2.4% in the highest. Discussion In GWAS meta-analysis of 2.5 million individuals, we identified 26 genetic risk loci for fibromyalgia, providing robust characterization of its genetic architecture. These findings establish a firm biological basis for a long-debated condition. Fibromyalgia heritability is enriched in the brain and in neuronal cell types, consistent with the central sensitization model of fibromyalgia. The strongest cell-type association was with hippocampal dentate gyrus neurons, involved in episodic memory and contextualizing sensory inputs such as pain; impairment could contribute to fibromyalgia’s cognitive impairment or ‘brain fog’. The second-strongest association was with enteric neurons, which regulate gastrointestinal motility and sensation, supporting involvement of the gut–brain axis in the comorbidity with irritable bowel syndrome. However, genes specifically expressed in closely related cell types often overlap, limiting the confidence with which enrichments can pinpoint precise neuronal populations. The specific genes implicated at risk loci further refine this model. Multiple candidate causal genes, including HTT , DCC , MDGA2 and CELF4 , have neural roles. Lead variants overlap extensively with brain-related traits in our GWAS Catalog and MVP–Finngen–UKBB PheWAS analyses, including pain traits such as multisite chronic pain MAML3 , DCC , OLFM4 , ANAPC4 and dorsalgia CAMKV/MST1R , MAML3 , BCL11A , DCC , DCAF5 , CNNM2 , RABGAP1L / GPR52 , ZKSCAN2 , ANAPC4 , ARHGAP15 , DRD2 / NCAM1 , MDGA2 , CELF4 , sleep traits such as insomnia CAMKV/MST1R , DCC , OLFM4 , CNNM2 , MVK , NUDT12 and canonical psychiatric disorders such as depression DCC , NUDT12 and schizophrenia CNNM2 , STK31 / NPY , PPP2R2B / GPR151 . In addition, risk loci overlapped with long COVID 51 BPTF and myalgic encephalomyelitis/chronic fatigue syndrome 52 /articles/s41591-026-04492-6 ref-CR52 OLFM4 , RABGAP1L/GPR52 , two poorly characterized disorders with a hypothesized neural basis, albeit with different lead variants. That said, some brain-related traits have much stronger genetic correlation with fibromyalgia than others: for instance, pain traits and post-traumatic stress disorder have higher correlations than depression and schizophrenia, which in turn have higher genetic correlation than insomnia. Our strongest association ~9% increased risk of fibromyalgia was with a common coding variant in HTT , the causal gene for HD, although this variant is distinct from the rare repeat expansion that causes HD. The variant results in the deletion of a single glutamic acid residue in the HTT protein. We also observe an association near GPR52 , a regulator of HTT . HTT could contribute to fibromyalgia etiology via both central and peripheral nervous system mechanisms. Centrally, HTT is widely expressed, especially in neurons, and HD neuronal dysfunction is concentrated in pain-related brain regions such as the striatum 53, where we observe enriched fibromyalgia heritability. Peripherally, HTT may affect nociception and spinal nociceptive relay: mouse Htt is expressed in dorsal root ganglion DRG neurons, where it aids peripheral axon regeneration after sciatic nerve injury , as well as in the superficial dorsal horn laminae of the CNS that receive nociceptive inputs from the DRG. While HTT has limited direct evidence of a functional role in peripheral nociception, its interactor Huntingtin-associated protein 1 HAP1 has stronger evidence. Mouse 54 /articles/s41591-026-04492-6 ref-CR54 Hap1 is expressed in nociceptive DRG subpopulations and spinal dorsal horn neurons and upregulated in these regions in chronic pain conditions 55 /articles/s41591-026-04492-6 ref-CR55 . 56 /articles/s41591-026-04492-6 ref-CR56 Hap1 heterozygous knockout mice have reduced nociceptive neuron excitability and attenuated spinal glial activation after nerve injury . 56 /articles/s41591-026-04492-6 ref-CR56 The genetic underpinnings of fibromyalgia are shared with a remarkably wide variety of other disease types. Genetic correlations implicate digestive, genitourinary and respiratory systems. One explanation is that fibromyalgia genetics captures a transdiagnostic central nervous system vulnerability that predisposes individuals to sensory and affective dysregulation. This vulnerability might manifest clinically as fibromyalgia, irritable bowel syndrome, post-traumatic stress disorder or a constellation of other disorders, depending on other genetic and environmental factors. On the other hand, this pleiotropy could merely reflect genetic influence on comorbid conditions that secondarily increase fibromyalgia risk, such as obesity, which also has a strongly neural genetic architecture. Our results inform the controversy over whether fibromyalgia is an autoimmune disorder 4. We did not find a significant GWAS signal in the major histocompatibility complex MHC region, nor heritability enrichment in either peripheral immune cell types or glia, but did observe modest genetic correlations between fibromyalgia and autoimmune disorders including Sjögren’s syndrome and RA. These associations should be interpreted with caution in light of the potential for diagnostic misclassification of fibromyalgia as Sjögren’s syndrome or RA . Notably, recent work using a broader phenotype that included both ICD-10 code-based and self-reported fibromyalgia reported autoimmune-related genetic signal, although differences in case definition may partly explain the apparent discrepancy with our findings 57 /articles/s41591-026-04492-6 ref-CR57 , 58 /articles/s41591-026-04492-6 ref-CR58 . Further supporting divergence between fibromyalgia and classical autoimmune disorders, genetic correlation was stronger with the more atypical seronegative subtype of RA than the seropositive subtype. Overall, our results suggest that fibromyalgia is not primarily an autoimmune disorder, although it may nonetheless have a peripheral immune and/or neuroimmune component. Power to detect this may have been limited by the predominantly European composition of our sample and by healthy participant bias in biobank cohorts. 59 /articles/s41591-026-04492-6 ref-CR59 Despite several-fold-higher prevalence in females, we found no sex difference in the genetic architecture of fibromyalgia. The observed difference in prevalence is therefore unlikely to be driven by sex-specific genetic risk variants, and may instead reflect nongenetic biological factors, environmental exposures or diagnostic bias 60. This study identifies two potential therapeutic targets. The linked associations of HTT and its regulator GPR52 highlight a clear repurposing opportunity, as GPR52 is already an investigational drug target for HD. Our association with CELF4 supports developing CELF4 -based gene therapies—already under investigation for chronic pain—for fibromyalgia in particular. These loci represent genetically supported molecular targets for fibromyalgia. This study has important limitations. Our cohort was mostly European, limiting the generalizability of findings and transferability of our PRSs to non-European ancestries. Our reliance on ICD codes may result in the inclusion of individuals not meeting full syndromic criteria for fibromyalgia, or the exclusion of undiagnosed cases. Incorporating phenotype risk scores 61 into fibromyalgia phenotyping might mitigate diagnostic heterogeneity between cohorts, or exclude controls with conditions such as depression or back pain. Our study maps the genetic architecture of fibromyalgia, identifying 26 risk loci and providing robust genetic validation of the notion that fibromyalgia is primarily a central nervous system disorder. Identifying specific risk loci provides the field with concrete molecular starting points, enabling hypothesis-driven studies of pathophysiology and shared etiology with comorbid conditions. Methods Ethics approval All of Us All of Us Institutional Review Board , UK Biobank North West Multi-centre Research Ethics Committee/North West–Haydock Research Ethics Committee , FinnGen Coordinating Ethics Committee of the Hospital District of Helsinki and Uusimaa , Estonian Biobank Estonian Committee on Bioethics and Human Research, Estonian Ministry of Social Affairs , Genes & Health London–South East Research Ethics Committee/NRES Committee London–South East , Mass General Brigham Biobank Mass General Brigham Institutional Review Board/Mass General Brigham Human Research Committee , Michigan Genomics Initiative University of Michigan Medical School Institutional Review Board , deCODE Genetics/Amgen Iceland National Bioethics Committee of Iceland and Icelandic Data Protection Authority , Copenhagen Hospital Biobank CHB; Danish National Committee on Health Research Ethics and Capital Region Data Protection Agency , Danish Blood Donor Study DBDS; Danish National Committee on Health Research Ethics and Danish Data Protection Agency , Intermountain Health Intermountain Healthcare Institutional Review Board , and Nashville Biosciences/BioVU Vanderbilt University Medical Center Institutional Review Board gave ethical approval for this work. Participants were not compensated for this study. Phenotype definition Cases were individuals with at least one inpatient or primary care record containing ICD-10 code M79.7 fibromyalgia; Supplementary Table 11 /articles/s41591-026-04492-6 MOESM3 . No exclusion criteria were applied: controls were all genotyped participants without code M79.7. Phenotype ascertainment was performed independently within each cohort before genome-wide association analysis. Genotyping, imputation and quality control We included autosomal and chromosome X data for all cohorts. All of Us We obtained whole-genome sequencing WGS , hg38 data from All of Us’s Curated Data Repository version 8 release, which involved variant calling, initial quality control and genetic ancestry assignment. We used the allele count/allele frequency ACAF threshold callset, which contains variants with a minor allele count 100 or minor allele frequency 1% in any genetic ancestry group, as the basis for a second round of ancestry-specific quality control with version 2.0.0 of the plink genetic analysis toolkit 62, in which variants with minor allele frequency MAF < 0.1%, missingness 10% or Hardy–Weinberg equilibrium HWE P < 1 × 10 −15 with mid- P correction were removed. Participants flagged by All of Us in the first round of quality control, those with sex-chromosome aneuploidy or sex at birth not equal to male or female, were excluded. We performed association analysis on each of the 3 largest genetic ancestry groups—EUR, AFR and AMR—with version 3.3 of the regenie GWAS toolkit applying Firth approximation for variants with association value < 0.01, and using as covariates age, sex, age 63 /articles/s41591-026-04492-6 ref-CR63 2, age × sex, age 2× sex and the first 10 genotype principal components PCs . For step 1 of regenie, we used an LD-pruned subset of the full genotypes, calculated with plink version 2.0.0 via the option ‘–indep-pairwise 500 kb 1 0.2’. We gratefully acknowledge All of Us participants for their contributions, without whom this research would not have been possible. We also thank the National Institutes of Health’s All of Us Research Program for making available the participant data examined in this study. UK Biobank We obtained imputed genotypes from the UK Biobank’s Data-Field 22828 https://biobank.ndph.ox.ac.uk/ukb/field.cgi?id=22828 . The UK Biobank imputed genotypes using a combination of two GRCh37 reference panels, the Haplotype Reference Consortium and a combined UK10K and 1000 Genomes Phase 3 panel, using the Haplotype Reference Consortium imputation if a variant was present in both panels 64. We excluded samples with sex-chromosome aneuploidy Data-Field 22019 https://biobank.ctsu.ox.ac.uk/crystal/field.cgi?id=22019 , that were outliers for heterozygosity or missingness rate Data-Field 22027 https://biobank.ctsu.ox.ac.uk/crystal/field.cgi?id=22027 , or that had discordant genetic sex Data-Field 22001 https://biobank.ctsu.ox.ac.uk/crystal/field.cgi?id=22001 versus self-reported sex Data-Field 31 https://biobank.ctsu.ox.ac.uk/crystal/field.cgi?id=31 . We excluded variants with MAF < 0.1%, imputation INFO score <0.8, missingness 10% or HWE P < 1 × 10 −15 with mid- P correction with plink version 2.0.0. We performed associations on the 3 largest genetic ancestry groups as defined by Pan-UKBB https://pan.ukbb.broadinstitute.org/docs/qc/index.html ancestry-definitions Return 2442 https://biobank.ndph.ox.ac.uk/ukb/dset.cgi?id=2442 —EUR, Central and South Asian CSA and AFR—with version 3.4.1 of regenie, applying Firth approximation for variants with association P value < 0.01 and using as covariates age, sex, age 2, age × sex, age 2× sex and the first 10 genotype PCs. For step 1 of regenie, we used an LD-pruned subset of the full genotypes, calculated with the ‘–indep-pairwise 500 kb 1 0.2’ option from plink version 2.0.0. We thank all participants of the UK Biobank. This research has been conducted using the UK Biobank Resource under application number 116200. FinnGen Genotyping in the FinnGen cohort was performed by using Illumina and Affymetrix arrays Thermo Fisher Scientific and lifted over to GRCh38/hg38. Individuals with high genotype absence 5% , inexplicit sex or excess heterozygosity ±4 standard deviations were excluded from the data. In addition, variants that had high absence 2% , low minor allele count <3 or low HWE P < 1 × 10−6 were removed. More detailed explanations of the genotyping, quality control and genotype imputation are provided elsewhere 65. All individuals in the cohort were Finns and matched against the SiSu v4 reference panel. For the FinnGen cohort Data Freeze 12 , GWAS was conducted using the regenie pipeline https://github.com/FINNGEN/regenie-pipelines https://github.com/FINNGEN/regenie-pipelines . Analysis was adjusted for age at death or end of follow-up, sex, genotyping batches and the first ten genetic PCs. Firth approximation was applied for variants with association P value < 0.01. Study subjects in FinnGen provided informed consent for biobank research, based on the Finnish Biobank Act. Alternatively, separate research cohorts, collected before the Finnish Biobank Act came into effect in September 2013 and start of FinnGen August 2017 , were collected based on study-specific consents and later transferred to the Finnish biobanks after approval by the Finnish Medicines Agency and the National Supervisory Authority for Welfare and Health. Recruitment protocols followed the biobank protocols approved by the Finnish Medicines Agency. The Coordinating Ethics Committee of the Hospital District of Helsinki and Uusimaa HUS statement number for the FinnGen study is Nr HUS/990/2017. The FinnGen study is approved by the Finnish Institute for Health and Welfare permit numbers THL/2031/6.02.00/2017, THL/1101/5.05.00/2017, THL/341/6.02.00/2018, THL/2222/6.02.00/2018, THL/283/6.02.00/2019, THL/1721/5.05.00/2019 and THL/1524/5.05.00/2020 , Digital and Population Data Service Agency permit numbers VRK43431/2017-3, VRK/6909/2018-3, VRK/4415/2019-3 , the Social Insurance Institution permit numbers KELA 58/522/2017, KELA 131/522/2018, KELA 70/522/2019, KELA 98/522/2019, KELA 134/522/2019, KELA 138/522/2019, KELA 2/522/2020 and KELA 16/522/2020 , Findata permit numbers THL/2364/14.02/2020, THL/4055/14.06.00/2020, THL/3433/14.06.00/2020, THL/4432/14.06/2020, THL/5189/14.06/2020, THL/5894/14.06.00/2020, THL/6619/14.06.00/2020, THL/209/14.06.00/2021, THL/688/14.06.00/2021, THL/1284/14.06.00/2021, THL/1965/14.06.00/2021, THL/5546/14.02.00/2020, THL/2658/14.06.00/2021 and THL/4235/14.06.00/2021 , Statistics Finland permit numbers TK-53-1041-17 and TK/143/07.03.00/2020 earlier TK-53-90-20 TK/1735/07.03.00/2021, TK/3112/07.03.00/2021 and Finnish Registry for Kidney Diseases permission and extract from the meeting minutes on 4 July 2019. The Biobank Access Decisions for FinnGen samples and data used in FinnGen Data Freeze 12 include THL Biobank BB2017 55, BB2017 111, BB2018 19, BB 2018 34, BB 2018 67, BB2018 71, BB2019 7, BB2019 8, BB2019 26, BB2020 1 and BB2021 65; Finnish Red Cross Blood Service Biobank 7.12.2017; Helsinki Biobank HUS/359/2017, HUS/248/2020, HUS/430/2021 §28 and §29, HUS/150/2022 §12, §13, §14, §15, §16, §17, §18, §23, §58 and §59, and HUS/128/2023 §18; Auria Biobank AB17-5154 and amendment 1 17 August 2020 and amendments BB 2021-0140, BB 2021-0156 26 August 2021, 2 Feb 2022 , BB 2021-0169, BB 2021-0179, BB 2021-0161, AB20-5926 and amendment 1 23 April 2020 and its modifications 22 September 2021 , BB 2022-0262, BB 2022-0256, Biobank Borealis of Northern Finland 2017 1013, 2021 5010, 2021 5010 Amendment, 2021 5018, 2021 5018 Amendment, 2021 5015, 2021 5015 Amendment, 2021 5015 Amendment 2, 2021 5023, 2021 5023 Amendment, 2021 5023 Amendment 2, 2021 5017, 2021 5017 Amendment, 2022 6001, 2022 6001 Amendment, 2022 6006 Amendment, 2022 6006 Amendment, 2022 6006 Amendment 2, BB22-0067, 2022 0262, 2022 0262 Amendment, Biobank of Eastern Finland 1186/2018 and amendment 22§/2020, 53§/2021, 13§/2022, 14§/2022, 15§/2022, 27§/2022, 28§/2022, 29§/2022, 33§/2022, 35§/2022, 36§/2022, 37§/2022, 39§/2022, 7§/2023, 32§/2023, 33§/2023, 34§/2023, 35§/2023, 36§/2023, 37§/2023, 38§/2023, 39§/2023, 40§/2023 and 41§/2023; Finnish Clinical Biobank Tampere MH0004 and amendments 21.02.2020 and 06.10.2020 , BB2021-0140 8§/2021, 9§/2021, §9/2022, §10/2022, §12/2022, 13§/2022, §20/2022, §21/2022, §22/2022, §23/2022, 28§/2022, 29§/2022, 30§/2022, 31§/2022, 32§/2022, 38§/2022, 40§/2022, 42§/2022 and 1§/2023; Central Finland Biobank 1-2017, BB 2021-0161, BB 2021-0169, BB 2021-0179, BB 2021-0170, BB 2022-0256 and BB 2022-0262, BB22-0067; decision allowing to continue data processing until 31 August 2024 for projects BB 2021-0179, BB22-0067,BB 2022-0262, BB 2021-0170, BB 2021-0164, BB 2021-0161 and BB 2021-0169; Terveystalo Biobank STB 2018001 and amendment 25 August 2020; Finnish Hematological Registry and Clinical Biobank decision 18 June 2021; and Arctic biobank P0844: ARC 2021 1001. Estonian Biobank All the Estonian Biobank participants have been genotyped at the Core Genotyping Lab of the Institute of Genomics, University of Tartu, using Illumina Global Screening Array versions 1, 2 or 3 . Samples were genotyped and PLINK format files were created using Illumina GenomeStudio v2.0.4. Individuals were excluded from the analysis if their call rate was less than 95%, if they were outliers of the absolute value of heterozygosity 3 standard deviations from the mean or if sex defined based on heterozygosity of the X chromosome did not match sex in phenotype data. Before imputation, variants were filtered by call rate <95%, HWE P < 1 × 10−4 autosomal variants only and minor allele frequency <1%. Genotyped variant positions were in build 37 and were lifted over to build 38 using Picard. Phasing was performed using the Beagle v5.4 software. Imputation was performed with Beagle v5.4 software 22 July 2022 release and default settings. The dataset was split into batches of 5,000. A population-specific reference panel consisting of 2,695 WGS samples was used for imputation, and standard Beagle hg38 recombination maps were used. On the basis of the PC analysis, samples that were not of European ancestry were removed. Duplicate and monozygous twin detection was performed with KING 2.2.7, and one sample was removed from the pair of duplicates. Analyses were restricted to individuals of European ancestry. Association analysis in the Estonian Biobank was carried out for all variants with an INFO score 0.4 using the additive model as implemented in regenie v3.0.3 with standard binary trait settings. Logistic regression was carried out with adjustment for current age, age2, sex and the first 10 genetic PCs as covariates, analyzing only variants with a minimum minor allele count of 2. The activities of the Estonian Biobank are regulated by the Human Genes Research Act, which was adopted in 2000 specifically for the operations of the Estonian Biobank. Individual-level data analysis in the Estonian Biobank was carried out under ethical approval 1.1-12/624 from the Estonian Committee on Bioethics and Human Research Estonian Ministry of Social Affairs , using data according to release application 3-10/GI/31689 from the Estonian Biobank. We want to acknowledge the participants of the Estonian Biobank for their contributions. The Estonian Genome Center analyses were partially carried out in the High Performance Computing Center, University of Tartu. The Estonian Biobank Research Team was responsible for data collection, genotyping, quality control and imputation and consisted of A. Metspalu andres.metspalu@ut.ee , M. Metspalu mait.metspalu@ut.ee , L. Milani lili.milani@ut.ee , R. Mägi reedik.magi@ut.ee , M. Nelis mari.nelis@ut.ee , T. Esko tonu.esko@ut.ee and G. Hudjashov georgi.hudjashov@ut.ee . Genes & Health Genotyping was performed on Illumina Infinium Global Screening Array v3 with additional multi-disease variants. Variants with call rates less than 0.99 and/or MAF < 1% were excluded. We excluded individuals unlikely to have genetically inferred Pakistani or Bangladeshi ancestry. Imputation was performed using the TOPMed-r2 panel. We excluded single nucleotide polymorphisms SNPs with low imputation scores INFO < 0.3 . In the Genes & Health cohort, sex was defined on the basis of XX female and XY male chromosomal presence in genotype data. Diagnoses were curated from routine UK NHS EHR data from primary care Systematized Nomenclature of Medicine coded and secondary care ICD-10 coded sources. Data were combined without mapping between coding formats. For each clinical code, the earliest ever measure recorded in a participant’s medical records was used, excluding erroneous code dates preceding the participant’s recorded date of birth. Association analysis was run in regenie, applying Firth approximation for variants with association P value < 0.01 and using as covariates age, sex, age2 and the first 20 genotype PCs. Genes & Health is and has recently been core funded by Wellcome WT102627, WT210561 , the Medical Research Council UK M009017, MR/X009777/1, MR/X009920/1 , Higher Education Funding Council for England Catalyst, Barts Charity 845/1796 , Health Data Research UK for London substantive site and research delivery support from the NHS National Institute for Health Research Clinical Research Network North Thames . We acknowledge the support of the National Institute for Health and Care Research Barts Biomedical Research Centre NIHR203330 ; a delivery partnership of Barts Health NHS Trust, Queen Mary University of London, St George’s University Hospitals NHS Foundation Trust and St George’s University of London. Genes & Health is and has recently been funded by Alnylam Pharmaceuticals, Genomics and a Life Sciences Industry Consortium of AstraZeneca, Bristol-Myers Squibb, GlaxoSmithKline Research and Development, Maze Therapeutics, Merck Sharp & Dohme, Novo Nordisk A/S, Pfizer and Takeda Development Centre Americas. We thank Social Action for Health, Centre of The Cell, members of our Community Advisory Group and staff who have recruited and collected data from volunteers. We thank the NIHR National Biosample Centre UK Biocentre , the Social Genetic & Developmental Psychiatry Centre King’s College London , Wellcome Sanger Institute and Broad Institute for sample processing, genotyping, sequencing and variant annotation. This study uses data provided by patients and collected by the NHS as part of their care and support. This research used Queen Mary University of London’s Apocrita HPC facility, supported by QMUL Research-IT, https://doi.org/10.5281/zenodo.438045 https://doi.org/10.5281/zenodo.438045 . We thank Barts Health NHS Trust, NHS Clinical Commissioning Groups City and Hackney, Waltham Forest, Tower Hamlets, Newham, Redbridge, Havering, Barking and Dagenham , East London NHS Foundation Trust, Bradford Teaching Hospitals NHS Foundation Trust, Public Health England especially D. Wyllie , Discovery Data Service and Endeavour Health Charitable Trust especially D. Stables , Voror Health Technologies especially S. Don , NHS England for what was NHS Digital for GDPR-compliant data sharing backed by individual written informed consent. Most of all, we thank all of the volunteers participating in Genes & Health. A favorable ethical opinion for the main Genes & Health research study was granted by NRES Committee London - South East reference 14/LO/1240 on 16 September 2014. Queen Mary University of London is the sponsor and data controller. Mass General Brigham Biobank The Mass General Brigham Biobank genotyped 53,297 participants on the Illumina Global Screening Array GSA and 11,864 on Illumina Multi-Ethnic Global Array MEG . The GSA arrays captured approximately 652,000 SNPs and short insertions–deletions indels , while the MEG arrays captured approximately 1.38 M SNPs and short indels. These genotypes were filtered for high missingness 2% and variants out of HWE P < 1 × 10−12 , as well as variants with an allele frequency AF discordant P < 1 × 10−150 from a synthesized AF calculated from gnomAD subpopulation frequencies and a genome-wide GnomAD model fit of the entire cohort. This resulted in approximately 620,000 variants for the GSA and 1.15 M for MEG. The two sets of genotypes were then separately phased and imputed on the TOPMed imputation server Minimac4 algorithm using the TOPMed-r2 reference panel. The resultant imputation sets were both filtered at an r 2 0.4 and an MAF 0.001, and then the two sets were merged and intersected resulting in approximately 19.5 M hg38 autosomal variants. The sample set for analysis was then restricted to just those classified as EUR according to a random-forest classifier trained with the Human Genome Diversity Project 66 as the reference panel, with the minimum probability for assignment to an ancestral group of 0.5, in 19 of 20 iterations of the model. To correct for population stratification, PCs were computed in genetically European participants. Association analysis for the full cohort N = 51,053 was performed with variants using regenie v3.2.8 with adjustment for age, sex, genotype-chip, tranche and the first ten genetic PCs. The summary statistics were filtered with a minimum minor allele count of 50. These summary statistics were then lifted over to human genome version 19 hg19 . 63 /articles/s41591-026-04492-6 ref-CR63 In the MGB Biobank, sex-specific association analyses were performed in the hg38 build and carried out using regenie v3.2.8 with covariates of age, tranche, genotype-chip and the first 10 PCs of ancestry calculated by performing the within-EUR principal component analysis PCA . The summary statistics were filtered with a minimum minor allele count of 25. Michigan Genomics Initiative We included autosomal and chromosome X data from the Michigan Genomics Initiative MGI , a health system-based biobank of patients recruited primarily during surgical encounters at Michigan Medicine. As of Freeze 6, MGI comprised 80,529 participants with linked genotype and electronic health record data. Detailed descriptions of the MGI cohort, recruitment protocols and overall design are available in a previous study 67. DNA samples were genotyped at the University of Michigan Advanced Genomics Core on customized versions of the Illumina Infinium CoreExome-24 v1.0, v1.1, v1.3; ~570,000 markers or Illumina Infinium GSA v1.3; ~682,000 markers . Array content included standard backbones with additional custom content to capture GWAS candidate variants, predicted loss-of-function alleles, ancestry-informative markers and pharmacogenomic variants. Genotype calling was performed in GenomeStudio, supplemented by zCall for recovery of rare variants. Sample-level quality control QC excluded individuals for consent withdrawal, genotype-inferred sex mismatch, sex-chromosome aneuploidy, call rate <99%, contamination 2.5%, unresolved technical duplicates or batch-level DNA extraction issues. Relatedness was estimated with KING v2.1.3, and contamination with VICES. Variant-level QC removed probes that did not uniquely map to hg38, sites with call rate <98%, HWE P < 1 × 10−4 in unrelated European-ancestry samples or poor clustering metrics GenTrain <0.15, cluster separation <0.3 . Additional harmonization excluded variants with large allele frequency deviations compared with 1000 Genomes reference populations. Phasing was performed with Eagle v2.4 using the TOPMed reference panel, followed by imputation against TOPMed haplotypes via the Michigan Imputation Server. Post-imputation, variants with r 2 < 0.3 or MAF < 0.01% were removed, yielding ~52 million high-quality variants. PCs were calculated using FlashPCA2 after pruning rare and correlated variants. For GWAS, we analyzed participants of genetically inferred European ancestry as defined by PCA projection and ADMIXTURE K = 7 reference populations from the Human Genome Diversity Project . Association testing was performed with SAIGE v0.35, a generalized mixed-model framework that accounts for relatedness and case–control imbalance. Covariates included age, sex, genotyping array and the first ten genetic PCs. Variants with MAF ≥ 0.01% and imputation r 2 ≥ 0.3 were included in association testing. We acknowledge the Michigan Genomics Initiative participants, AI & Digital Health Innovation at the University of Michigan, the University of Michigan Medical School Central Biorepository and the University of Michigan Advanced Genomics Core for providing data and specimen storage, management, processing and distribution services, and the Center for Statistical Genetics in the Department of Biostatistics at the School of Public Health for genotype data curation, imputation and management in support of the research reported in this Article. deCODE Genetics/Amgen Iceland At the time of analysis, 63,118 samples from Icelandic participants were whole-genome sequenced at deCODE using Illumina standard TruSeq methods to a mean depth of 38× ref. 68 . Only samples with a genome-wide average coverage of 20× and higher were included. Genotypes of SNPs and indels were identified and called jointly by Graphtyper v.2.7.1 . In all, 173,025 samples from Icelandic participants had been chip-genotyped using various Illumina SNP arrays 69 /articles/s41591-026-04492-6 ref-CR69 . The chip-typed individuals were long-range phased 68 /articles/s41591-026-04492-6 ref-CR68 , and the variants identified in the whole-genome sequences of Icelanders imputed into the chip-typed individuals. Using extensive and encrypted Icelandic genealogy data, familial imputation of genotypes in first- and second-degree relatives was used to increase sample size 70 /articles/s41591-026-04492-6 ref-CR70 . The final dataset used included 19,657,761 variants with imputation information over 0.8 and MAF over 0.1%. 68 /articles/s41591-026-04492-6 ref-CR68 , 70 /articles/s41591-026-04492-6 ref-CR70 The Icelandic samples and diagnostic data were obtained from Icelandic medical record data repositories and analyzed under approval from the National Bioethics Committee 17-035-V11-S2, previously 12-162 following review by the Icelandic Data Protection Authority. Data were anonymized and encrypted by a third-party system, approved and monitored by the Icelandic Data Protection Authority. CHB and the DBDS study We obtained genotypes from the CHB 71 study on pain and degenerative musculoskeletal diseases CHB-PDS; approval: NVK-1803812, P-2019-51 and the DBDS approval: NVK-1700407, P-2019-99 . All samples were genotyped on Illumina’s Infinium Global Screening Array versions 1.0 and 3.0 and imputed to build hg38. Genotyping and imputation were performed by deCODE Genetics. Before imputation, duplicate samples and those with genotype call rates <98% were removed. Phasing was carried out with SHAPEIT4 ref. 72 /articles/s41591-026-04492-6 ref-CR72 , and imputation was performed using deCODE’s in-house workflow, with a joint Graphtyper-based reference panel comprising ~50,000 individuals, including ~10,800 Danes, including danmac5.dk ref. 73 /articles/s41591-026-04492-6 ref-CR73 . Initial quality control of genotypes included removal of samples with sex discrepancies, missingness 5% and variants with missingness 10%, or HWE 74 /articles/s41591-026-04492-6 ref-CR74 P < 1 × 10 −5. For genome-wide association analyses, we removed variants with MAF < 0.1%, imputation INFO < 0.8, missingness 10% or HWE P < 1 × 10 −15 with mid- P correction , using plink version 2.0.0. Intermountain Health Under research collaboration between Intermountain Health and deCODE Genetics/Amgen, samples were obtained from consenting participants of European descent in two ongoing studies: the Intermountain Inspire Registry and the HerediGene Population Study 75. The Intermountain Healthcare Institutional Review Board approved both studies, and all participants provided written informed consent before enrollment. Samples were genotyped at deCODE Genetics/Amgen using Illumina Global Screening Array chips. In all, 138,006 individuals of European origin were chip typed. Intermountain and Danish imputation was based on a multi-ethnic reference panel of 50,179 whole-genome-sequenced individuals of mostly European descent, including 23,288 individuals from Intermountain and 11,722 individuals from Denmark. Samples were filtered on 98% variant yield and duplicates removed. The WGS protocol was the same as described above for the Icelandic and Danish data. Sequence variants were imputed into 138,006 chip-typed individuals. Over 245 million high-quality sequence variants and indels, sequenced to a mean depth of 20×, were identified using Graphtyper v.2.7.1 69. Quality-controlled chip genotype data were phased using SHAPEIT4 ref. . A phased haplotype reference panel was prepared from the sequence variants using the long-range phased chip-genotyped samples 73 /articles/s41591-026-04492-6 ref-CR73 . In all, 21,316,504 variants imputation info 0.8 and MAF 0.1% were tested. 68 /articles/s41591-026-04492-6 ref-CR68 Nashville Biosciences Nashville Biosciences NashBio is a data and analytics provider owned by Vanderbilt University Medical Center VUMC . NashBio uses the biobank collection BioVU of VUMC that includes a collection of de-identified DNA samples linked to de-identified data from the electronic health records of VUMC referred to as the Synthetic Derivative SD database. All patients of VUMC consented to their residual samples from routine clinical testing and data being contributed to BioVU. The NashBio dataset included in this study is a subset of BioVU, consisting of, in total, 80,965 whole-genome-sequenced individuals of genetically defined European descent and 31,025 individuals of genetically defined African descent. All samples were whole genome sequenced at deCODE Genetics/Amgen using Illumina NovaSeq in accordance with deCODE/Amgen protocols deCODE Genetics/Amgen Iceland /articles/s41591-026-04492-6 Sec23 . Using the same quality criteria as for other analyzed deCODE/Amgen datasets, in total, 22,316,589 variants were tested in the European cohort and 33,108,534 in the African cohort. BioVU extracts and banks germline DNA samples that are de-identified and only linked to the SD through a randomly assigned unique identifier, not back to the patient or their underlying medical record. The use of BioVU is classified as non-human subject research by the Institutional Review Board IRB of VUMC, and NashBio is not required to seek study-specific consent for use of these datasets. The overall biobanking program is reviewed annually by the IRB to maintain this determination and make decisions about patient protections, privacy and ethical issues. Each individual study seeking to use the SD database and BioVU biobank is filed with the IRB of VUMC to validate its non-human subject classification and appropriate use of data. Genetic ancestry analysis in deCODE datasets For the non-Icelandic datasets analyzed at deCODE Genetics/Amgen Copenhagen Hospital Biobank and the DBDS Intermountain Health and Nashville Biosciences , genetic ancestry analysis was performed to identify a subset of individuals with similar ancestry. For the Danish samples, we used ADMIXTURE v1.23 76 run in supervised mode using the 1000 Genomes populations CEU Utah residents with northern and western European ancestry , CHB Han Chinese in Beijing, China , ITU Indian Telugu in the UK , PEL Peruvian in Lima, Peru and YRI Yoruba in Ibadan, Nigeria as training samples. These training samples had themselves been filtered for ancestry outliers using PCA and unsupervised ADMIXTURE. Samples assigned <0.93 CEU were excluded, resulting in 358,483 samples included in the analysis. For the Intermountain samples, we used the same methods as for the Danish samples to identify a subset of 108,149 individuals of European descent that were included in the analysis. To identify ancestry-homogeneous subsets in the Nashville Biosciences cohort, sample ancestry was estimated using similar methods as in the Danish cohort. Samples with admixed African and European ancestry defined as YRI + CEU 0.9, YRI 0.3 and CEU 0.02 or predominantly African ancestry YRI 0.9 were placed in the AFR subcohort. Genotypes were prepared for PCA with PLINK v1.9 ref. 62 using–maf 0.01–thin-count 1000000–indep-pairwise 60000 6000 0.3 to minimize effects of local LD and very recent population structure. Pairwise relatedness was calculated using KING v2.3.0–ibdseg–degree 3 ref. to identify relatives of third degree or closer, with one sample from each pair of relatives removed before PCA and projected onto the resulting PCs using a deCODE in-house script that adjusts for shrinkage. PCA was performed with PCAone v0.3.4 ref. 77 /articles/s41591-026-04492-6 ref-CR77 using default parameters. 78 /articles/s41591-026-04492-6 ref-CR78 Association testing in deCODE datasets The four fibromyalgia case–control GWASs were performed at deCODE Genetics/Amgen Iceland, Denmark, US Intermountain Health and US NashBio with software developed at deCODE Genetics, using logistic regression assuming an additive model 68. For the Icelandic data, the model included sex, county of birth, current age or age at death first- and second-order terms included , blood sample availability for the individual, sequencing status and an indicator function for the overlap of the lifetime of the individual with the time span of phenotype collection. To include imputed but ungenotyped individuals in Iceland, we used county of birth as a proxy covariate for the first PC because county of birth has been shown to be in concordance with the first PC in Iceland . For the Danish data, 12 PCs were used, in addition to sex, year of birth and sequencing status as covariates, whereas for the Intermountain Health data, we included 4 PCs, year of birth, sex and sequencing status as covariates. For the NashBio data, associations were adjusted for the top 20 PCs in addition to sex, year of birth and sequencing batch. The number of PCs used to adjust for population stratification was determined by identifying the point at which further PCs appeared to capture local LD rather than population structure as reflected by sharp peaks in PC loadings and in plateauing of eigenvalues 79 /articles/s41591-026-04492-6 ref-CR79 . All statistical tests were two sided unless otherwise indicated. 80 /articles/s41591-026-04492-6 ref-CR80 Harmonization and meta-analysis Each cohort’s summary statistics were harmonized via an in-house analysis pipeline. For each summary statistics file that is, for each cohort and ancestry , we removed variants with missing data in any column, non-ACGT alleles, P values outside 0, 1 , allele frequencies outside 0, 1 , imputation INFO scores outside 0, 1 , or non-positive odds ratios, standard errors, or sample sizes. We normalized indels to their minimal representation by iteratively removing common trailing then leading base pairs from the reference and alternate alleles—subject to the constraint that each allele retains at least one base pair—and advancing the genomic position by the number of leading base pairs removed. We inferred Reference SNP cluster ID rs numbers based on chromosome and base-pair columns using dbSNP build 156, requiring an exact match to both the ref and the alt allele, allowing allele flips for single-nucleotide variants but not indels and converting the dbSNP variants to minimal representation as well before doing the matching. We removed variants without rs numbers in dbSNP, or that mapped to multiple genomic positions this happens very rarely . To account for the tendency of dbSNP to merge rs numbers over time, if a variant mapped to multiple rs numbers, we took the lowest-numbered rs number. We calculated each cohort and ancestry’s effective sample size per variant via the formula N eff = 4/ 2 × AAF × 1 − AAF × INFO − BETA2 /SE2, skipping the ‘× INFO’ if INFO was not available, where INFO denotes the imputation information score, BETA the estimated log odds ratio for the variant, and SE its standard error. We then summed these N eff across cohorts and ancestries. To account for P value inflation due to residual confounding such as cryptic relatedness, we applied stratified LD score regression 81 to each cohort and ancestry’s summary statistics before meta-analysis, using the effective sample size N effrather than the raw N to avoid bias . We used the standard 52-annotation baseline model precomputed by the developers of stratified LD score regression. We manually calculated LD scores based on ancestry-specific reference panels from the corresponding 1000 Genomes Phase 3 superpopulation, where 1000 Genomes EUR was used for European cohorts, SAS South Asian for Central/South Asian cohorts, AFR for African cohorts, and AMR for admixed American cohorts. However, we subset to variants present in Europeans in HapMap 3, as the standard 52 annotations are available only for those variants. For cohort–ancestry pairs in which the LD score regression intercept was greater than 1, we divided each variant’s 82 /articles/s41591-026-04492-6 ref-CR82 χ 2statistic by the LD score regression intercept, then adjusted the standard errors and P values accordingly. This has the effect of reducing the significance of every variant in summary statistics with evidence of inflation, while leaving non-inflated summary statistics alone. After harmonization and LD score regression correction, we performed standard fixed-effects inverse-variance weighted meta-analyses via the ‘–meta-analysis’ option from plink version 1.9.0. Variants were matched across cohorts based on the combination of rs number, reference allele and alternate allele. In total, 54,629 cases and 2,509,126 controls went into the multi-ancestry meta-analysis, and 49,000 cases and 2,264,287 controls went into the European genetic ancestry-only analysis. We also performed sex-stratified meta-analyses restricted to males and females, and leave-one-cohort-out analyses for each of the 11 constituent cohorts, with both multi-ancestry leaving out all ancestries from that cohort, if the cohort had multiple and European-only versions. For females, we performed both European-only and multi-ancestry meta-analyses, but for males, we performed only the European-only meta-analysis owing to the lack of male summary statistics in Genes & Health and the low number of male cases in non-European ancestries in the multi-ancestry cohorts. After meta-analysis, we filtered to variants with an overall minor allele frequency of 1% across all cohorts that went into that particular meta-analysis. As two cohorts the UK Biobank and Mass General Brigham Biobank reported variants in hg19 coordinates, we avoided having to perform an error-prone remapping of variant positions from hg19 to hg38 by filtering to variants present in at least 5 cohorts in the multi-ancestry meta-analyses and at least 3 studies in the European meta-analyses, thereby ensuring that every variant would be present in at least one hg38 study by the pigeonhole principle. Causal gene prioritization We prioritized causal genes for each risk locus in the multi-ancestry meta-analysis via a combination of approaches. First, we performed manual literature search for each lead variant, focusing heavily on the nearest gene to each lead variant as nearest genes are expected to be causal about two-thirds of the time 83. We supplemented this search with two orthogonal bioinformatic approaches: the state-of-the-art machine learning method FLAMES and an in-house quantitative trait locus QTL -based pipeline developed by deCODE Genetics. 84 /articles/s41591-026-04492-6 ref-CR84 FLAMES FLAMES is a framework that nominates candidate causal genes at risk loci from a GWAS, via an ensemble of two complementary approaches. First, a gradient boosting model scores genes based on locus-specific variant-to-gene evidence, aggregating features such as QTLs, variant effect predictor annotations, chromatin interactions and variant-to-gene distance. The model’s predictive weights were established by training it on external GWAS loci, in which ‘silver-standard’ causal genes had been identified via missense or predicted loss-of-function variants from exome studies. For this analysis, we supplied the model with 99% credible sets derived from the sum of single effects 85 fine-mapping of variants within 500 kb of the lead variant at each locus from our primary meta-analysis. Second, FLAMES incorporates gene-level scores from polygenic priority score PoPS 86, another tool for causal gene prioritization. PoPS prioritizes genes by identifying gene features that are enriched for genetic association across the entire GWAS—such as pathway memberships, protein–protein interaction networks, co-expression data and expression in various tissues and cell types—then scoring each gene based on which of these enriched features it has. PoPS identifies these trait-relevant gene features by leveraging multi-marker analysis of genomic annotation to aggregate variant-level GWAS 87 /articles/s41591-026-04492-6 ref-CR87 P values into gene-level association scores, then using a regression model to identify which gene features are predictive of high multi-marker analysis of genomic annotation scores. FLAMES integrates these two evidence streams by multiplying the gradient boosting scores with scaled PoPS scores, effectively upweighting genes supported by both locus-specific and GWAS-wide evidence. At the recommended confidence threshold scaled FLAMES score 0.248 , FLAMES nominated a candidate causal gene at 19 of the 26 loci from the primary meta-analysis, while abstaining from prediction at the remaining 7 loci owing to a lack of confidence in the causal gene. deCODE pipeline As a second strategy to nominate candidate causal genes, we performed functional annotation and QTL analyses on all 26 lead variants from the primary meta-analysis, as well as all variants in high LD r 2 ≥ 0.8 and within ±1 Mb with these lead variants. We used variant effect predictor 88 to attribute to the studied variants the most severe predicted variant consequences in canonical and non-canonical transcripts. We classified as high-impact variants those predicted as start-lost, stop-gained, stop-lost, splice-donor, splice-acceptor or frameshift, collectively called loss-of-function variants, whereas moderate impact variants are those predicted to otherwise affect coding or splicing of a protein missense . For all lead and correlated variants, we studied their association with 1 mRNA expression top local expression quantitative trait locus eQTL , splicing QTL or alternative polyadenylation QTL in multiple tissues analyzed at deCODE, in addition to data from GTEx 89 and other public datasets, and 2 plasma protein levels top cis -protein QTL identified in large proteomic datasets from Iceland and the UK . RNA sequencing was performed on whole blood from 17,848 Icelanders and on subcutaneous adipose tissue from 769 Icelanders, respectively. Gene expression was computed based on personalized transcript abundances using kallisto 90 /articles/s41591-026-04492-6 ref-CR90 . The association between sequence variants and gene expression 91 /articles/s41591-026-04492-6 ref-CR91 cis -eQTL was tested via linear regression, assuming additive genetic effect and normal quantile gene expression estimates, adjusting for measurements of sequencing artifacts, demographic variables, blood composition and PCs . The gene expression PCs were computed per chromosome using a leave-one-chromosome-out method. All variants within 1 Mb of each gene were tested. 92 /articles/s41591-026-04492-6 ref-CR92 The Icelandic proteomics data were analyzed using the SomaLogic SOMAscan v4 proteomics assay that scans 4,907 aptamers, measuring 4,719 proteins in samples from 35,892 Icelanders with genetic information available at deCODE Genetics 90,93. Plasma protein levels were standardized and adjusted for year of birth, sex and year of sample collection 2000–2019 . The UK proteomics dataset was analyzed using the Olink Explore 3072 proximity extension assay platform with 2,941 immunoassays characterizing 2,925 proteins in 54,265 participants in the UK Biobank 90 /articles/s41591-026-04492-6 ref-CR90 , 93 /articles/s41591-026-04492-6 ref-CR93 . 90 /articles/s41591-026-04492-6 ref-CR90 , 93 /articles/s41591-026-04492-6 ref-CR93 Phenome-wide associations of lead variants For each of our 26 lead variants from the primary meta-analysis, we looked up genome-wide significant P < 5 × 10−8 associations in the GWAS Catalog that overlapped either the variant itself or any proxy variants in high LD r 2 ≥ 0.8 and within ±1 Mb . To account for the GWAS Catalog containing large numbers of closely related phenotypes, we grouped related traits for visualization. Specifically, we removed GWAS results obtained through MTAG analysis or performed on multiple traits combined, identified by ‘and’ or ‘or’ in their trait names. We then standardized trait names by removing words in parentheses and laterality descriptors ‘left’ and ‘right’ , and standardized the spelling of ‘neutrophill’ to ‘neurophil’. Finally, we grouped trait names that became identical after this standardization. We also looked up the P values of our 26 lead variants in preexisting GWAS summary statistics for each of 330 disease endpoints from the November 2024 release of MVP–Finngen–UKBB https://mvp-ukbb.finngen.fi/about https://mvp-ukbb.finngen.fi/about , a GWAS meta-analysis of Million Veteran Program, FinnGen and the UK Biobank. We performed Bonferroni correction across the 330 diseases and 26 variants tested, leading to a significance threshold of P = 0.05/8,580 ≈ 5.83 × 10−6. We applied the same approach to 124 GWAS of drug prescription endpoints from FinnGen’s DF12 release, in which each GWAS was on whether an individual was ever prescribed any drug from a particular category for example, analgesics . We performed Bonferroni correction across the 124 drug categories and 26 variants tested, leading to a significance threshold of P = 0.05/3,224 ≈ 1.55 × 10−5. Heritability For the European-only meta-analysis, we estimated single-nucleotide polymorphism-based heritability—the proportion of phenotypic variance explained by the aggregated effect of single-nucleotide polymorphisms across the autosomal genome—using stratified LD score regression 81 with the standard 52-annotation baseline model mentioned above. We used LD scores computed from the European-ancestry subsample of 1000 Genomes Phase 3 precomputed by LDSC authors , and subset to variants present in Europeans in HapMap 3. The observed-scale heritability was calculated from the LD score regression slope. As with all LD score regression-based analyses in this study, we used the effective sample size N effrather than the raw N to avoid bias due to case–control imbalance. Tissue and cell-type enrichment We used LDSC-SEG 94 to infer tissue and cell-type enrichments for the European-only meta-analysis. As for the global heritability analysis, we used LD scores computed from the European-ancestry subsample of 1000 Genomes Phase 3 precomputed by LDSC authors and the 52-annotation baseline model, subsetting to variants present in Europeans in HapMap 3. LDSC-SEG is a variant of stratified LD score regression that analyzes each tissue or cell type in turn, adding a binary annotation to the 52-annotation baseline model for each one. This annotation flags variants located within or within 100 kb of the 10% of genes most specifically expressed in that tissue or cell type. The heritability enrichment for each tissue or cell type is computed by dividing its per-SNP heritability derived from the LD score regression slope by the average per-SNP heritability across all SNPs in the analysis. To compute tissue enrichments, we used annotations provided by the LDSC-SEG authors that were derived for each of 53 tissues in the GTEx project, in which the top 10% of specifically expressed genes were defined by ranking genes by their differential expression t -statistic for expression in that tissue versus all other tissues. For GTEx brain tissues, the comparison was instead between that tissue and all non-brain tissues. To compute cell-type enrichments, we generated gene sets from a comprehensive single-cell transcriptomic atlas of the mouse, PanSci, which profiles over 20 million cells from 14 organs and tissues 95. From this dataset, we included 119 cell types and 7 broad lineages that were represented by at least 1,000 cells after quality control filtering to cells with ≤10% mitochondrial reads, ≥100 genes detected and non-zero Malat1 expression . Our analysis was restricted to a universe of 16,404 protein-coding genes with identifiable human orthologs. For each cell type and lineage, we defined its specifically expressed genes as those with a detection rate 10% within that cell type and a fold change 2 compared with all other cells. Genetic correlation We performed genetic correlation between our leave-FinnGen-out European-only fibromyalgia meta-analysis and each of 1,284 clinical endpoints from FinnGen Release 13 with https://www.finngen.fi/en/researchers/clinical-endpoints https://www.finngen.fi/en/researchers/clinical-endpoints , using LD score regression 96. We calculated SNP heritability with LDSC for all 2,691 endpoints and performed genetic correlation only on those endpoints with significant heritability, resulting in a list of 1,284 endpoints. Variants were aligned and filtered to HapMap release 3 SNPs, and the LD reference was obtained from the European population of the 1000 Genomes Project Phase 3 release. We used the 80th percentile of per-variant effective sample sizes N eff to munge the summary statistics. To avoid diluting the results with low-specificity phenotypes, we then excluded endpoints containing any of the case-insensitive keywords ‘other’ for example, ‘Other disorders of ear’ , ‘unspecified’ for example, ‘Disorder of external ear, unspecified’ , classified for example, ‘Other viral diseases, not elsewhere classified’ , ‘any’ for example, ‘Any mental disorder’ or ‘all’ for example, ‘All kidney diseases’ , as well as derived endpoints for which the listed category did not correspond to a chapter of the ICD. We also excluded fibromyalgia itself as an endpoint. These exclusions reduced the number of endpoints tested from 1,284 to 855, resulting in a Bonferroni significance threshold of P = 0.05/855 ≈ 5.85 × 10−5. For visualization purposes, we grouped endpoints by their ICD chapter. We grouped four categories with few significant endpoints—‘III Diseases of the blood and blood-forming organs and certain disorders involving the immune mechanism’, ‘XVIII Symptoms, signs and abnormal clinical and laboratory findings, not elsewhere classified’, ‘XXI Factors influencing health status and contact with health services’ and ‘XXII Codes for special purposes’—into an ‘Other’ category. Because fibromyalgia has been previously proposed to share etiology with autoimmune disorders, we also created a separate ‘Autoimmune’ category, as these disorders would otherwise be scattered across ICD chapters based on the affected organ system. We also estimated the genetic correlation between fibromyalgia and the subtypes of asthma and RA using summary statistics available at deCODE. In line with other LDSC-based analyses, we used per-variant effective sample sizes, filtered variants to HapMap release 3 SNPs and used precomputed LD scores for European populations downloaded from https://data.broadinstitute.org/alkesgroup/LDSCORE/eur w ld chr.tar.bz2 https://data.broadinstitute.org/alkesgroup/LDSCORE/eur w ld chr.tar.bz2 . PRSs We computed PRSs for the leave-UK Biobank-out European and multi-ancestry meta-analyses, using PRS-CS 97. PRS-CS takes GWAS summary statistics and an LD reference panel as input, and applies Bayesian shrinkage to the variants’ effect sizes to infer posterior effect sizes, which are used as the weights of the PRS. These weights can then be scored on any cohort of interest, which should be non-overlapping with the cohorts used to create the PRS to avoid bias. We used the European-ancestry subsample of 1000 Genomes Phase 3 as the reference panel, removing indels from the summary statistics before the analysis. We ran PRS-CS with the ‘–n burnin 5000’ and ‘–n iter 10000’ options to specify 5,000 burn-in iterations and 10,000 total iterations of the Markov Chain Monte Carlo sampling. We then scored the PRS weights on the UK Biobank cohort with the ‘–score’ option from plink version 2.0.0. As PRS-CS requires a single sample size across all variants, the sample size we provided to PRS-CS was the 80th percentile of per-variant effective sample sizes N eff , as recommended by others for PRS analyses 98. To assess the accuracy of the PRS across ancestries, we scored the PRS weights from the multi-ancestry leave-UK Biobank-out meta-analysis on each of the 3 largest genetic ancestries from the UK Biobank European, Central and South Asian, and African . We assessed PRS accuracy via the AUC, as well as by computing odds ratios for each quintile of polygenic risk relative to the middle quintile, and prevalence of fibromyalgia within each quintile. Statistics and reproducibility No statistical method was used to predetermine sample size. No data were excluded from the analyses. The experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary /articles/s41591-026-04492-6 MOESM2 linked to this article. Data availability GWAS summary statistics are available in the GWAS Catalog accession codes GCST90838603 https://www.ebi.ac.uk/gwas/search?query=GCST90838603 to GCST90838628 https://www.ebi.ac.uk/gwas/search?query=GCST90838628 and PRSs are available in the PGS catalog score IDs PGS012560 and PGS012561 . All summary statistics and PRSs are also available at https://paingenomics.org https://paingenomics.org . This study used data from the All of Us Research Program’s Controlled Tier Dataset Curated Data Repository version 8, available to authorized users on the Researcher Workbench. Code availability The code is available via GitHub at https://github.com/i-kerrebijn/fibromyalgia GWAS meta-analysis https://github.com/i-kerrebijn/fibromyalgia GWAS meta-analysis . References Weir, P. T. et al. The incidence of fibromyalgia and its associated comorbidities: a population-based retrospective cohort study based on International Classification of Diseases, 9th Revision codes. J. Clin. Rheumatol. 12 , 124–128 2006 .Fitzcharles, M.-A. et al. Nociplastic pain: towards an understanding of prevalent pain conditions. Lancet 397 , 2098–2110 2021 .Latremoliere, A. & Woolf, C. J. Central sensitization: a generator of pain hypersensitivity by central neural plasticity. J. Pain 10 , 895–926 2009 .Clauw, D., Sarzi-Puttini, P., Pellegrino, G. & Shoenfeld, Y. Is fibromyalgia an autoimmune disorder? Autoimmun. Rev. 23 , 103424 2024 .Arnold, L. M. et al. Family study of fibromyalgia. Arthritis Rheum. 50 , 944–952 2004 .Arnold, L. M. et al. The fibromyalgia family study: a genome-wide linkage scan study. Arthritis Rheum. 65 , 1122–1128 2013 .Docampo, E. et al. Genome-wide analysis of single nucleotide polymorphisms and copy number variants in fibromyalgia suggest a role for the central nervous system. Pain 155 , 1102–1109 2014 .Moscati, A. et al. Life is pain: fibromyalgia as a nexus of multiple liability distributions. Am. J. Med. Genet. B 192 , 171–182 2023 .Zorina-Lichtenwalter, K. et al. Genetic risk shared across 24 chronic pain conditions: identification and characterization with genomic structural equation modeling. Pain 164 , 2239–2252 2023 .Cooksey, R. & Choy, E. Exploring gender differences, medical history, and treatments used in patients with fibromyalgia in the UK using primary-care data: a retrospective, population-based, cohort study. Lancet Rheumatol. 4 , S20 2022 .Saudou, F. & Humbert, S. The biology of huntingtin. Neuron 89 , 910–926 2016 .Kay, C. et al. Huntingtin haplotypes provide prioritized target panels for allele-specific silencing in Huntington disease patients of European ancestry. Mol. Ther. 23 , 1759–1771 2015 .Yao, Y. et al. A striatal-enriched intronic GPCR modulates huntingtin levels and toxicity. eLife 4 , e05449 2015 .Song, H. et al. Targeting Gpr52 lowers mutant HTT levels and rescues Huntington’s disease-associated phenotypes. Brain 141 , 1782–1798 2018 .Salamon, I. et al. Celf4 controls mRNA translation underlying synaptic development in the prenatal mammalian neocortex. Nat. Commun. 14 , 6025 2023 .Mueth, M. G. et al. The RNA-binding protein CELF4 is a negative regulator of sensory neuron excitability and mechanical and heat behavioral sensitivity. Neurobiol. Pain 18 , 100184 2025 .Hirsch, D. & Zukowska, Z. NPY and stress 30 years later: the peripheral view. Cell. Mol. Neurobiol. 32 , 645–659 2012 .Diaz-delcastillo, M., Woldbye, D. P. D. Heegaard, A. M. Neuropeptide Y and its involvement in chronic pain. Neuroscience 387 , 162–169 2018 .Zamboni, V. et al. Disruption of ArhGAP15 results in hyperactive Rac1, affects the architecture and function of hippocampal inhibitory neurons and causes cognitive deficits. Sci. Rep. 6 , 34877 2016 .Liaci, C. et al. Loss of ARHGAP15 affects the directional control of migrating interneurons in the embryonic cortex and increases susceptibility to epilepsy. Front. Cell Dev. Biol. 10 , 875468 2022 .Alberati-Giani, D. et al. Isolation and expression of a cDNA clone encoding human kynureninase. Eur. J. Biochem. 239 , 460–468 1996 .Hazrati, E., Eftekhar, S. P., Mosaed, R., Shiralizadeh Dini, S. & Namazi, M. Understanding the kynurenine pathway: a narrative review on its impact across chronic pain conditions. Mol. Pain 20 , 17448069241275097 2024 .Arnau-Soler, A. et al. Genome-wide by environment interaction studies of depressive symptoms and psychosocial stress in UK Biobank and Generation Scotland. Transl. Psychiatry 9 , 14 2019 .Ford, C. P. The role of D2-autoreceptors in regulating dopamine neuron activity and transmission. Neuroscience 282 , 13–22 2014 .Vukojevic, V. et al. Evolutionary conserved role of neural cell adhesion molecule-1 in memory. Transl. Psychiatry 10 , 217 2020 .Jacobs, R. et al. CD56 brightcells differ in their KIR repertoire and cytotoxic features from CD56dimNK cells. Eur. J. Immunol. 31 , 3121–3127 2001 .Liang, Z. et al. The pseudokinase CaMKv is required for the activity-dependent maintenance of dendritic spines. Nat. Commun. 7 , 13282 2016 .Andersson, E. R., Sandberg, R. & Lendahl, U. Notch signaling: simplicity in design, versatility in function. Development 138 , 3593–3612 2011 .Louvi, A. & Artavanis-Tsakonas, S. Notch signalling in vertebrate neural development. Nat. Rev. Neurosci. 7 , 93–102 2006 .Ables, J. L., Breunig, J. J., Eisch, A. J. & Rakic, P. Not ch just development: Notch signalling in the adult brain. Nat. Rev. Neurosci. 12 , 269–283 2011 .Connor, S. A., Elegheert, J., Xie, Y. & Craig, A. M. Pumping the brakes: suppression of synapse development by MDGA–neuroligin interactions. Curr. Opin. Neurobiol. 57 , 71–80 2019 .Joset, P. et al. Rostral growth of commissural axons requires the cell adhesion molecule MDGA2. Neural Dev. 6 , 22 2011 .Zhao, D. et al. Mdga2 deficiency leads to an aberrant activation of BDNF/TrkB signaling that underlies autism-relevant synaptic and behavioral changes in mice. PLOS Biol 23 , e3003047 2025 .Keino-Masu, K. et al. Deleted in colorectal cancer DCC encodes a netrin receptor. Cell 87 , 175–185 1996 .Bull, S.-J. et al. Progressive disorganization of paranodal junctions and compact myelin due to loss of DCC expression by oligodendrocytes. J. Neurosci. 34 , 9768–9778 2014 .Thigpen, A. E. et al. Molecular genetics of steroid 5 alpha-reductase 2 deficiency. J. Clin. Invest. 90 , 799–809 1992 .Wu-Chen, E. et al. The role of androgens on experimental pain sensitivity: a systemic review and meta-analysis. Pain Rep. 10 , e1263 2025 .Denton, K. R. et al. Loss of spastin function results in disease-specific axonal defects in human pluripotent stem cell-based models of hereditary spastic paraplegia. Stem Cells 32 , 414–423 2014 .Solowska, J. M. & Baas, P. W. Hereditary spastic paraplegia SPG4: what is known and not known about the disease. Brain 138 , 2471–2484 2015 .Peron, A. et al. BCL11A intellectual developmental disorder: defining the clinical spectrum and genotype–phenotype correlations. Eur. J. Hum. Genet. 33 , 312–324 2025 .Glinton, K. E. et al. Phenotypic expansion of the BPTF-related neurodevelopmental disorder with dysmorphic facies and distal limb anomalies. Am. J. Med. Genet. A. 185 , 1366–1378 2021 .Sandal, P. et al. De novo missense variants in the PP2A regulatory subunit PPP2R2B in a neurodevelopmental syndrome: potential links to mitochondrial dynamics and spinocerebellar ataxias. Hum. Mol. Genet. 34 , 193–203 2024 .O’Hearn, E., Holmes, S. E. & Margolis, R. L. Spinocerebellar ataxia type 12. Handb. Clin. Neurol. 103 , 535–547 2012 .Xia, L.-P. et al. GPR151 in nociceptors modulates neuropathic pain via regulating P2X3 function and microglial activation. Brain J. Neurol. 144 , 3405–3420 2021 .Jiang, B.-C. et al. G protein-coupled receptor GPR151 is involved in trigeminal neuropathic pain through the induction of Gβγ/extracellular signal-regulated kinase-mediated neuroinflammation in the trigeminal ganglion. Pain 162 , 1434–1448 2021 .Walkley, S. U. & Suzuki, K. Consequences of NPC1 and NPC2 loss of function in mammalian neurons. Biochim. Biophys. Acta 1685 , 48–62 2004 .Vaites, L. P., Paulo, J. A., Huttlin, E. L. & Harper, J. W. Systematic analysis of human cells lacking ATG8 proteins uncovers roles for GABARAPs and the CCZ1/MON1 regulator C18orf8/RMC1 in macroautophagic and selective autophagic flux. Mol. Cell. Biol. 38 , e00392-17 2018 .Carr, T. F., Zeki, A. A. & Kraft, M. Eosinophilic and noneosinophilic asthma. Am. J. Respir. Crit. Care Med. 197 , 22–37 2018 .Kristjansdottir, K. et al. A partial loss-of-function variant in STAT6 protects against type 2 asthma. J. Allergy Clin. Immunol. 155 , 228–235 2025 .Doss, J., Mo, H., Carroll, R. J., Crofford, L. J. & Denny, J. C. Phenome-wide association study of rheumatoid arthritis subgroups identifies association between seronegative disease and fibromyalgia. Arthritis Rheumatol. 69 , 291–300 2017 .Chaudhary, N. S. et al. Multi-ancestry GWAS of long COVID identifies immune-related loci and etiological links to chronic fatigue syndrome, fibromyalgia and depression. Preprint at medRxiv https://doi.org/10.1101/2024.10.07.24315052 https://doi.org/10.1101/2024.10.07.24315052 2024 .Genetics Delivery Team et al. Initial findings from the DecodeME genome-wide association study of myalgic encephalomyelitis/chronic fatigue syndrome. Preprint at medRxiv https://doi.org/10.1101/2025.08.06.25333109 https://doi.org/10.1101/2025.08.06.25333109 2025 .Li, J. et al. Pain in Huntington’s disease and its potential mechanisms. Front. Aging Neurosci. 15 , 1190563 2023 .Schaeffer, J. et al. Customization of the translational complex regulates mRNA-specific translation to control CNS regeneration. Neuron 111 , 2881–2898.e12 2023 .Islam, M. N. et al. Expression of huntingtin-associated protein 1 in adult mouse dorsal root ganglia and its neurochemical characterization in reference to sensory neuron subpopulations. IBRO Rep. 9 , 258–269 2020 .Pan, J. et al. Huntington-associated protein 1 inhibition contributes to neuropathic pain by suppressing Cav1.2 activity and attenuating inflammation. Pain 164 , e286–e302 2023 .Di Franco, M. et al. Misdiagnosis in fibromyalgia: a multicentre study. Clin. Exp. Rheumatol. 29 , S104–S108 2011 .Choi, B. Y., Oh, H. J., Lee, Y. J. & Song, Y. W. Prevalence and clinical impact of fibromyalgia in patients with primary Sjögren’s syndrome. Clin. Exp. Rheumatol. 34 , S9–S13 2016 .Bright, U. et al. The genetics of fibromyalgia and its relationships to psychiatric and medical traits. Preprint at medRxiv https://doi.org/10.1101/2025.09.18.25335991 https://doi.org/10.1101/2025.09.18.25335991 2025 .Mogil, J. S. Sex differences in pain and pain inhibition: multiple explanations of a controversial phenomenon. Nat. Rev. Neurosci. 13 , 859–866 2012 .Bastarache, L. et al. Phenotype risk scores identify patients with unrecognized Mendelian disease patterns. Science 359 , 1233–1239 2018 .Chang, C. C. et al. Second-generation PLINK: rising to the challenge of larger and richer datasets. Gigascience 4 , s13742-015-0047–8 2015 .Mbatchou, J. et al. Computationally efficient whole-genome regression for quantitative and binary traits. Nat. Genet. 53 , 1097–1103 2021 .Bycroft, C. et al. The UK Biobank resource with deep phenotyping and genomic data. Nature 562 , 203–209 2018 .Kurki, M. I. et al. FinnGen provides genetic insights from a well-phenotyped isolated population. Nature 613 , 508–518 2023 .Dashti, H. S. et al. Interaction of obesity polygenic score with lifestyle risk factors in an electronic health record biobank. BMC Med. 20 , 5 2022 .Zawistowski, M. et al. The Michigan Genomics Initiative: a biobank linking genotypes and electronic clinical records in Michigan Medicine patients. Cell Genom. 3 , 100257 2023 .Gudbjartsson, D. F. et al. Large-scale whole-genome sequencing of the Icelandic population. Nat. Genet. 47 , 435–444 2015 .Eggertsson, H. P. et al. Graphtyper enables population-scale genotyping using pangenome graphs. Nat. Genet. 49 , 1654–1660 2017 .Kong, A. et al. Detection of sharing by descent, long-range phasing and haplotype imputation. Nat. Genet. 40 , 1068–1075 2008 .Sørensen, E. et al. Data Resource Profile: the Copenhagen Hospital Biobank CHB . Int. J. Epidemiol. 50 , 719–720e 2021 .Hansen, T. F. et al. DBDS Genomic Cohort, a prospective and comprehensive resource for integrative and temporal analysis of genetic, environmental and lifestyle factors affecting health of blood donors. BMJ Open 9 , e028401 2019 .Delaneau, O., Zagury, J.-F., Robinson, M. R., Marchini, J. L. & Dermitzakis, E. T. Accurate, scalable and integrative haplotype estimation. Nat. Commun. 10 , 5436 2019 .Banasik, K. et al. DanMAC5: a browser of aggregated sequence variants from 8,671 whole genome sequenced Danish individuals. BMC Genom. Data 24 , 30 2023 .Azriel, E. et al. Utilizing public health frameworks and partnerships to ensure equity in DNA-based population screening. Front. Genet. 13 , 886755 2022 .Alexander, D. H., Novembre, J. & Lange, K. Fast model-based estimation of ancestry in unrelated individuals. Genome Res. 19 , 1655–1664 2009 .Manichaikul, A. et al. Robust relationship inference in genome-wide association studies. Bioinformatics 26 , 2867–2873 2010 .Li, Z., Meisner, J. & Albrechtsen, A. Fast and accurate out-of-core PCA framework for large scale biobank data. Genome Res. 33 , 1599–1608 2023 .Price, A. L. et al. The impact of divergence time on the nature of population structure: an example from Iceland. PLOS Genet. 5 , e1000505 2009 .Young, A. I., Benonisdottir, S., Przeworski, M. & Kong, A. Deconstructing the sources of genotype–phenotype associations in humans. Science 365 , 1396–1400 2019 .Finucane, H. K. et al. Partitioning heritability by functional annotation using genome-wide association summary statistics. Nat. Genet. 47 , 1228–1235 2015 .Grotzinger, A. D., Fuente, J., de la, Privé, F., Nivard, M. G. & Tucker-Drob, E. M. Pervasive downward bias in estimates of liability-scale heritability in genome-wide association study meta-analysis: a simple solution. Biol. Psychiatry 93 , 29–36 2023 .Stacey, D. et al. ProGeM: a framework for the prioritization of candidate causal genes at molecular quantitative trait loci. Nucleic Acids Res. 47 , e3 2019 .Schipper, M. et al. Prioritizing effector genes at trait-associated loci using multimodal evidence. Nat. Genet. 57 , 323–333 2025 .Wang, G., Sarkar, A., Carbonetto, P. & Stephens, M. A simple new approach to variable selection in regression, with application to genetic fine mapping. J. R. Stat. Soc. Ser. B 82 , 1273–1300 2020 .Weeks, E. M. et al. Leveraging polygenic enrichments of gene features to predict genes underlying complex traits and diseases. Nat. Genet. 55 , 1267–1276 2023 .de Leeuw, C. A., Mooij, J. M., Heskes, T. & Posthuma, D. MAGMA: generalized gene-set analysis of GWAS data. PLOS Comput. Biol. 11 , e1004219 2015 .McLaren, W. et al. The Ensembl Variant Effect Predictor. Genome Biol. 17 , 122 2016 .Lonsdale, J. et al. The Genotype-Tissue Expression GTEx project. Nat. Genet. 45 , 580–585 2013 .Ferkingstad, E. et al. Large-scale integration of the plasma proteome with genetics and disease. Nat. Genet. 53 , 1712–1721 2021 .Bray, N. L., Pimentel, H., Melsted, P. & Pachter, L. Near-optimal probabilistic RNA-seq quantification. Nat. Biotechnol. 34 , 525–527 2016 .Stegle, O., Parts, L., Piipari, M., Winn, J. & Durbin, R. Using probabilistic estimation of expression residuals PEER to obtain increased power and interpretability of gene expression analyses. Nat. Protoc. 7 , 500–507 2012 .Eldjarn, G. H. et al. Large-scale plasma proteomics comparisons through genetics and disease associations. Nature 622 , 348–358 2023 .Finucane, H. K. et al. Heritability enrichment of specifically expressed genes identifies disease-relevant tissues and cell types. Nat. Genet. 50 , 621–629 2018 .Zhang, Z. et al. A panoramic view of cell population dynamics in mammalian aging. Science 387 , eadn3949 2024 .Bulik-Sullivan, B. et al. An atlas of genetic correlations across human diseases and traits. Nat. Genet. 47 , 1236–1241 2015 .Ge, T., Chen, C.-Y., Ni, Y., Feng, Y.-C. A. & Smoller, J. W. Polygenic prediction via Bayesian regression and continuous shrinkage priors. Nat. Commun. 10 , 1776 2019 .Privé, F., Arbel, J., Aschard, H. & Vilhjálmsson, B. J. Identifying and correcting for misspecifications in GWAS summary statistics and polygenic scores. HGG Adv. 3 , 100136 2022 . Acknowledgements We thank the All of Us, Estonian Biobank, FinnGen, Genes & Health, deCODE Iceland, UK Biobank, Nashville Biosciences, Copenhagen Hospital and Danish Blood Donor Study, Intermountain Health, Michigan Genomics Initiative and Mass General Brigham Biobank participants. Funding We acknowledge funding and institutional support from multiple sources. Canadian contributions were supported by the Canadian Institutes of Health Research CIHR FBD-199459; CIHR MHP-192163 and by the Natural Sciences and Engineering Research Council of Canada NSERC Undergraduate Student Research Award URSA . In the United States, support was provided by the Michigan Genomics Initiative and AI & Digital Innovation at the University of Michigan, Ann Arbor, MI. S.F. is funded by an NIH K08 Clinical Investigator Award K08AR082454 . Further support from the United States included NIH funding to R.S. and H.M.O NIH R01AI170850 , and NIH funding NIH RM1HG010461 and a generous donation to N.S.-A. In Denmark, the Copenhagen Hospital Biobank CHB was supported by the Department of Clinical Immunology, Rigshospitalet, Copenhagen University Hospital; by grants from the Novo Nordisk Foundation NNF23OC0082015, NNF17OC0027594 ; and by the Rigshospitalet Research Council Framework grant . The Danish Blood Donor Study DBDS is funded by an annual grant from Bio- and Genome Bank Denmark, with additional support from the Danish Administrative Regions 02/2611 and the Danish Council for Independent Research 09–069412 , as well as the Novo Nordisk Foundation NNF23OC0082015, NNF17OC0027864, NNF17OC0027594 . The Parker Institute is supported by a core grant from the Oak Foundation OFIL-24-074 . Further support in Denmark was received from the Novo Nordisk Foundation NNF17OC0027594, NNF14CC0001 . In the United Kingdom, support was provided to F.M.K.W. by Versus Arthritis. In Iceland, support from the European Commission to the painFACT project T.E.T. H2020-2020-848099 is acknowledged. In Estonia, the work of E.A. was funded by the European Union through Horizon Europe research and innovation programs grants 894987 and 101137201 and by Estonian Research Council grants PRG1291 and Roadmap II project number TT17. In Finland, H.M.O was supported by Research Council of Finland 350181 . Author information Authors and Affiliations Consortia Contributions I.K. performed data curation, formal analysis, funding acquisition, investigation, methodology, project administration, visualization and original draft preparation. G.B. contributed to phenotype definition, data acquisition, writing, critical review and final approval. K.A. and C.J. contributed to formal analysis and original draft preparation. L.U., H.H. and J.V. contributed to data analysis. G.T., L.S. and T.E.T. contributed to data acquisition, writing, critical review and final approval. S.F., L.K. and C.M.B. contributed to data acquisition, critical review and final approval. E.A. contributed to data curation, formal analysis, funding acquisition, investigation, methodology and project administration. M.K., S.F. and D.H. contributed to data curation and formal analysis. B.A., H.B., S.B., M.T.B., M.D., C.E., A.J.G., D.F.G., T.F.H., I.J., S.K., K.U.K., C.M., L.D.N., T.A.O., S.R.O., O.B.V.P., S.S., A.T.S., E.S., H.S., P.S., O.A.S., G.E.T., U.T., H.U., A.V. and T.M.W. contributed to data acquisition, phenotype definitions, critical review and final approval. R.S. contributed to conceptualization and supervision. K.S. provided overall supervision. B.G. contributed to critical review and final approval. D.J.C. contributed to conceptualization, critical review and final approval. F.M.K.W. contributed to conceptualization, writing and final approval. N.S.-A., H.M.O. and M.W. contributed to conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, supervision and both original draft writing and review and editing. Corresponding authors Ethics declarations Competing interests C.M.B. is a consultant for Vertex Pharmaceuticals and Merck Pharmaceuticals, providing expert medicolegal testimony. S.F. is a consultant for Vertex Pharmaceuticals. B.G. has received research grants paid to institution from Sandoz, AbbVie, AlfaSigma and Eli Lilly. S.B. has ownership interests in Hoba Therapeutics Aps, Novo Nordisk A/S, Lundbeck A/S and Eli Lilly and Co. C.E. has received unrestricted research grants from Novo Nordisk and Abbott Diagnostics administered by Aarhus University Hospital; no personal fees received . G.B., T.E.T., G.E.T., T.A.O., S.S., H.S., I.J., A.T.S., G.T., L.S., U.T., P.S. and D.F.G. are employees of Amgen deCODE Genetics. The other authors declare no competing interests. Peer review Peer review information Nature Medicine thanks Victoria Jackson, Ru-Rong Ji and Weihua Meng for their contribution to the peer review of this work. Primary Handling Editor: Anna Ranzoni in collaboration with the Nature Medicine team. Additional information Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Extended data Extended Data Fig. 1 Manhattan plot of the European-only meta-analysis. /articles/s41591-026-04492-6/figures/6 Association statistics were obtained from REGENIE genome-wide association analyses, with variant-level significance assessed using two-sided tests of association. Genome-wide significance was defined as p < 5 × 10−8. Gold-highlighted variants have linkage disequilibrium r2 0.001 and are within 5 megabases of the lead variants diamonds . The nearest gene to each lead variant is labeled. Extended Data Fig. 2 Miami plot of the multi-ancestry female-only and male-only meta-analyses. /articles/s41591-026-04492-6/figures/7 Association statistics were obtained from REGENIE genome-wide association analyses, with variant-level significance assessed using two-sided tests of association. Genome-wide significance was defined as p < 5 × 10−8. Gold-highlighted variants have linkage disequilibrium r2 0.001 and are within 5 megabases of the lead variants diamonds . Note that males are European only, due to low male case counts in non-European ancestries see Methods /articles/s41591-026-04492-6 Sec12 . Extended Data Fig. 3 Manhattan plot of the European-only, female-only meta-analysis. /articles/s41591-026-04492-6/figures/8 Association statistics were obtained from REGENIE genome-wide association analyses, with variant-level significance assessed using two-sided tests of association. Genome-wide significance was defined as p < 5 × 10−8. Gold-highlighted variants have linkage disequilibrium r2 0.001 and are within 5 megabases of the lead variants diamonds . The nearest gene to each lead variant is labeled. The male version of this analysis is not shown here because it would be the same as the bottom half of Extended Data Fig. 2 /articles/s41591-026-04492-6 Fig7 : the male-only meta-analysis is restricted to European males due to low male case counts in non-European ancestries see Methods /articles/s41591-026-04492-6 Sec12 . Extended Data Fig. 4 Sex differences in fibromyalgia lead variant associations. /articles/s41591-026-04492-6/figures/9 Male, female, and full-cohort odds ratios with 95% confidence intervals for the 26 lead variants from the primary meta-analysis. Extended Data Fig. 5 Polygenic risk for fibromyalgia in the UK Biobank. /articles/s41591-026-04492-6/figures/10 Genetic ancestry-stratified a areas under the receiver-operating curve, b odds ratios with 95% confidence intervals for each quintile of polygenic risk relative to the middle quintile, and c prevalences of fibromyalgia with 95% confidence intervals within each quintile, using polygenic risk scores computed from the multi-ancestry leave-UK Biobank-out meta-analysis. Supplementary information Supplementary Information download PDF https://media.springernature.com/original/springer-static/esm/art%3A10.1038%2Fs41591-026-04492-6/MediaObjects/41591 2026 4492 MOESM1 ESM.pdf Supplementary Figs. 1–3. Supplementary Tables download XLSX https://media.springernature.com/original/springer-static/esm/art%3A10.1038%2Fs41591-026-04492-6/MediaObjects/41591 2026 4492 MOESM3 ESM.xlsx Supplementary Table 1: Case and control sample sizes in full-cohort analysis stratified by cohort, ancestry and sex. Supplementary Table 2: Lead variants and candidate causal genes from sex-specific meta-analyses. Supplementary Table 3: Candidate causal genes prioritized through functional annotations and identification of coding variants in high linkage disequilibrium with lead variants. Supplementary Table 4: GWAS Catalog associations for lead variants and variants in high LD with lead variants. Supplementary Table 5: Disease PheWAS associations of lead variants queried from MVP + FinnGen + UKBB meta-analysis summary statistics. Supplementary Table 6: Drug PheWAS associations of lead variants queried from FinnGen summary statistics. Supplementary Table 7: Tissue enrichment results from LDSC-SEG analysis. Supplementary Table 8: Cell-type enrichment results from LDSC-SEG analysis of main mouse cell types. Supplementary Table 9: Cell-type enrichment results from LDSC-SEG analysis of mouse cell lineages. Supplementary Table 10: Genetic correlation results between fibromyalgia and FinnGen disease endpoints. Supplementary Table 11: Overview of QC parameters and phenotype definition per cohort. Rights and permissions Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author s and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ http://creativecommons.org/licenses/by/4.0/ . About this article Cite this article Kerrebijn, I., Bjornsdottir, G., Arbabi, K. et al. The genetic architecture of fibromyalgia across 2.5 million individuals. Nat Med 2026 . https://doi.org/10.1038/s41591-026-04492-6 Received: Accepted: Published: Version of record: DOI: https://doi.org/10.1038/s41591-026-04492-6