henhaobio
Sign in

medRxivGenetic & Genomic MedicinePreprint

Seven replicated genomic associations of myalgic encephalomyelitis/chronic fatigue syndrome: a biobank study

Joshua SlaughteriD, Olivier LabayleiD, Breeshey Roskams-HieteriD, Joshua J DibbleiD, Simon J. McGrath, Sjoerd Viktor BeentjesiD, Ava KhamsehiD, Chris Paul PontingiD

PDFDOICode
In-depth full-text digest available · Jump to digest ↓

Abstract

Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS) is a debilitating female-biased disease with neither diagnostic biomarkers nor effective treatment nor well-understood aetiology. To investigate its biological basis, we used TarGene to perform a genome-wide association study in the UK Biobank with 1,268 ME/CFS cases, using electronic health records and survey responses to affirm ME/CFS status in cases, and non-ME/CFS status in controls. This analysis identified 176 variants as significantly associated with ME/CFS (false discovery rate View larger version (20K): org.highwire.dtl.DTLVardef@12b51d4org.highwire.dtl.DTLVardef@1095197org.highwire.dtl.DTLVardef@cbd0a2org.highwire.dtl.DTLVardef@ee7482_HPS_FORMAT_FIGEXP M_FIG C_FIG No genomic risk loci for myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) have previously replicated across independent cohorts. We find seven variants associated with ME/CFS that replicate in disjoint biobank cohorts. O_LICases and controls are defined from multiple lines of evidence rather than a single diagnostic code. C_LIO_LINo variant replicates across all three cohorts, possibly reflecting differences in phenotype definition and population. C_LIO_LIResults provide candidate loci for follow-up into the biological mechanisms of ME/CFS. C_LI Lay summaryMyalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) is a common and disabling illness with a variety of symptoms. Additionally, little is known about the biological mechanisms that cause ME/CFS. The variety of symptoms and its unknown cause can make it difficult for healthcare professionals to diagnose people with ME/CFS reliably. This poses a significant challenge for ME/CFS research, as misdiagnoses may lead to errors in conclusions drawn from its study. Previously, studies have attempted to find biological mechanisms for ME/CFS by comparing the DNA of people with ME/CFS with the DNA of people without ME/CFS. These studies have identified regions of DNA linked to the illness; however, none of these links have been found in other ME/CFS studies. In our work, we compare the DNA of people with ME/CFS to the DNA of people without ME/CFS, where ME/CFS status is supported by multiple lines of evidence to reduce the likelihood of misdiagnoses in the participants selected for the study. We then use similar selection strategies, using multiple lines of evidence to repeat the study in independent groups of people. By doing so, we found seven regions of DNA linked to ME/CFS status in more than one study. Some of these links are close to or located in regions of DNA called genes. Genes produce molecules known as proteins, which are responsible for many functions in the human body. It is not currently possible to determine exactly whether genes near disease-linked regions of DNA cause disease, so we report nearby genes with the highest likelihood of linkage to ME/CFS. We also investigated whether the linkage of these regions changed when we further compared groups based on sex or socioeconomic status, but no conclusive results were found. Comparison with the larger DecodeME study showed no overlapping results.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (14 Sept 2026) · CC BY 4.0

Targeted-learning GWAS of 1,268 UK Biobank ME/CFS cases found 176 FDR-significant variants; seven replicated in disjoint UK Biobank or All of Us cohorts.

ME/CFS has no diagnostic biomarker or effective treatment, and earlier biobank GWAS used single-source, sometimes self-reported phenotypes with conventional methods too underpowered to detect and replicate associations. Pairing multi-evidence case definitions with doubly-robust targeted estimation and two disjoint replication cohorts yields what the authors call the first ME/CFS genetic associations replicated in independent cohorts.

Key results

  1. ShownA targeted (doubly-robust) GWAS of genotyped SNPs in the UK Biobank discovery cohort identified 176 variants associated with ME/CFS risk at FDR below 5%, with modest genomic inflation.

    1,268 cases (MC1) vs 113,132 controls (NC1); 670,243 QC-passed genotyped autosomal SNPs tested; 176 significant at Benjamini-Hochberg FDR < 5%, of which 15 reached p < 5 x 10^-8; genomic inflation λ = 1.09. · Study populations: cases and controls; Discovery analysis; Discussion; Quality control

  2. ShownSeven discovery variants replicated in one of two smaller disjoint cohorts (two in a second UK Biobank cohort, five in All of Us), but none was FDR-significant in all three cohorts.

    UKB2: 319 cases vs 28,471 controls; AoU: 371 cases vs 33,140 controls; prevalence matched to discovery (q0 ≈ 1.10%). Eight (UKB2) and three (AoU) variants could not be tested because minor allele counts were too low. None of the 15 genome-wide significant UKB1 SNPs reached FDR < 0.05 in UKB2 or AoU. · Study populations: cases and controls; Replication analyses

  3. SuggestedAt the replicated chromosome 13 locus, fine-mapping found a five-variant credible set near CLYBL that colocalised with lower CLYBL expression in putamen. The CLYBL Arg259 loss-of-function variant itself showed no ME/CFS association.

    95% PIP credible set (lead rs9585273, intronic to CLYBL); replicated rs117553493 in moderate LD with it (0.54 ≤ r^2 ≤ 0.60); SuSiE coloc with GTEx v10 putamen eQTL PP.H4 = 0.994; stop-gain rs41281112 p > 0.05, PP.H3 = 0.99. Only one other locus (rs73175505, chromosome 8) gave a credible set, and it did not colocalise. · Fine-mapping and colocalisation; Discussion; Fine-mapping and colocalisation

  4. ShownNeither genetic sex nor area-level socioeconomic deprivation significantly modified the effects of the seven replicated variants. UK Biobank discovery cases (MC1) had higher area-level deprivation than discovery controls (NC1).

    No gene-by-sex or gene-by-deprivation interaction at FDR < 0.05. Nominal sex-differential effects at rs261902 (p = 0.013) and rs76847656 (p = 0.029); nominal deprivation effect at rs73175505 (p = 0.024). MC1 vs NC1: Townsend p = 5.3 × 10^-25, English IMD p = 3.2 × 10^-27, Scottish IMD p = 5.4 × 10^-7 (two-sided t-tests). · Interaction analyses; Defining deprivation

  5. ShownIn a look-up using community-recruited, clinically diagnosed DecodeME cases, none of the testable replicated variants was associated with ME/CFS despite high power.

    2,143 DecodeME cases vs 191,336 UKB controls at matched prevalence (1.1%). Only four of seven replicated variants passed DecodeME QC. All four were non-significant, with estimates near zero, despite 83-100% power for discovery-sized effects. Shared UKB controls make this a look-up rather than formal replication. · DecodeME validation; Discussion; Validation cohort: DecodeME

Methods, data and reproducibility

Case-control GWAS with TarGene targeted learning (weighted TMLE, XGBoost outcome and propensity models, 6 leave-one-chromosome-out PCs, non-additive genotype coding tested via Hotelling's T^2) on multi-evidence phenotypes, then disjoint-cohort replication, IVW meta-analysis, SuSiE fine-mapping, GTEx eQTL colocalisation, and gene-by-sex/deprivation tests.

  • UK Biobank discovery cohort UKB1 (White British, genotyped SNPs): cases by self-reported CFS plus poor/fair health; controls by pain questionnaire and good/excellent health · human · n = 1,268 cases (MC1); 113,132 controls (NC1) · UK Biobank Application Number 76173
  • UK Biobank replication cohort UKB2, disjoint from UKB1: cases via pain-questionnaire fatigue items, ICD10 G93.3 or a primary-care ME/CFS record · human · n = 319 cases (MC2); 28,471 controls (NC2) · UK Biobank Application Number 76173
  • All of Us replication cohort (predicted European ancestry): ME/CFS diagnostic codes in EHRs plus Overall Health Survey ratings · human · n = 371 cases (MC3); 33,140 controls (NC3) · All of Us Controlled Tier Dataset v8
  • DecodeME validation look-up: community-recruited European-ancestry cases meeting Canadian Consensus/IoM 2015 criteria, against UKB controls meeting NC2 criteria · human · n = 2,143 cases; 191,336 UK Biobank controls
  • GTEx v10 eQTL data (including putamen) for fine-mapping and colocalisation · human
  • 1000 Genomes European Phase 3 genotypes as LD reference for the eQTL population · human (1000 Genomes)
  • FUMA 'UKB release2b 10k White British' LD reference panel for gene mapping · human
Methods & tools
TarGene (Targeted Genomic Estimation) pipeline, TMLE.jl (Julia), Weighted targeted minimum loss-based estimation (wTMLE), XGBoost (cross-validated tree depth and L2 regularisation), Hotelling's T-squared joint test, PLINK2, FlashPCA2 (leave-one-chromosome-out PCA), SuSiE fine-mapping via coloc R package (coloc.susie), FUMA SNP2GENE, ANNOVAR with Ensembl Genes v113, GTEx v10 eQTL database, Fixed-effect inverse-variance-weighted meta-analysis (Cochran's Q, I^2)
Software & environment
Julia TMLE.jl within TarGene; PLINK2; FlashPCA2; coloc R package (coloc.susie); FUMA with Ensembl Genes v113; GTEx v10 eQTLs. Package version numbers and containers are not given.
Reproducibility gaps
Individual-level UK Biobank (Application 76173) and All of Us Controlled Tier v8 data are open only to authorised users. Supplementary files hold summary statistics for the 176 FDR-significant discovery SNPs; DecodeME data access is not described.
Limitations (6)
  • The replication cohorts (319 and 371 cases) are much smaller than the discovery cohort, and effects selected in discovery are expected to shrink (winner's curse), so failure to replicate does not imply absence of effect. · stated by the authors
  • UK Biobank and All of Us do not record explicit post-exertional malaise (PEM), so cases were not defined by PEM-requiring clinical criteria. Phenotyping, healthcare systems, ages and ancestry also differ across cohorts, so each may capture different parts of a heterogeneous phenotype. · stated by the authors
  • Analyses were limited to White British (UK Biobank) or predicted European (All of Us) ancestry; the authors note that more diverse populations are needed for broader relevance. · stated by the authors
  • LD prevents variant-level causal identification. Colocalisation was possible only where credible sets overlapped known eQTLs, so tissue specificity is unverified, and the putamen CLYBL signal is only indirectly tied to the replicated variant rs117553493. · stated by the authors
  • The CLYBL credible-set lead variant is given as rs9585273 in the Results but as rs79477241 in the Discussion, so check the supplementary tables before prioritising a causal SNP at this locus. · noted in this summary
  • Interaction tests covered only the 7 replicated variants and split continuous deprivation indices at the median (TarGene needs categorical exposures), which likely limits power to detect modest gene-by-environment effects. · noted in this summary

Read this if you study ME/CFS genetics, build biobank case definitions from health records and surveys, or use doubly-robust targeted-learning GWAS and want to see which associations survive replication.

Citations
0

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISTeam '25Herrera '18Zhang '25xiong '21Sardell '26Gimenez-Ore '23
79% semantic matchGenetic & Genomic Medicine
Genetics Delivery Team, Thibaud Boutin, Andrew D. Bretherick+58 more·medRxiv·Aug 6, 2025· 6 citations