Epigenome-wide profiling identifies distinct DNA methylation architecture underlying ME/CFS and fibromyalgia symptom burden

Abstract

Background

Myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) and fibromyalgia are overlapping chronic disorders characterized by fatigue, pain, cognitive dysfunction, sleep disturbance, and multisystem symptoms. Whether peripheral blood DNA methylation reflects diagnostic categories, quantitative symptom burden, or both remains unclear. We aimed to identify DNA methylation axes associated with diagnosis, symptom dimensions, and clinical differences between these conditions.

Methods

This cross-sectional study included 188 women: 73 healthy controls, 71 with ME/CFS, and 44 with fibromyalgia. DNA methylation was profiled in peripheral blood mononuclear cells using the Illumina MethylationEPIC v2 array. Principal component analysis identified latent methylation axes. Linear models adjusted for age, body mass index, and estimated immune-cell composition tested associations with diagnostic group and clinical measures. Region-level methylation analyses, functional enrichment, bootstrap resampling, permutation testing, leave-one-out analyses, and medication/comorbidity sensitivity analyses were performed.

Results

Both patient groups had greater symptom burden than healthy controls but differed clinically. Fibromyalgia showed greater widespread pain, pain catastrophizing, central sensitization inventory scores, and temporal summation, whereas ME/CFS showed greater post-exertional malaise, cognitive symptoms, and lower physical activity. Two methylation axes showed clinically relevant associations. PC5 differentiated ME/CFS from fibromyalgia and healthy controls and was associated mainly with post-exertional malaise and cognitive symptoms. PC6 separated both patient groups from healthy controls but not from each other, and was associated with broader symptom burden, including widespread pain, pain impact, sleep disturbance, visceral symptoms, and temporal summation. The temporal summation association, together with higher widespread pain and temporal summation in fibromyalgia, suggests that PC6 includes a pain-related component extending to experimentally assessed nociceptive summation. Region-level analyses identified 591 PC5-associated and 54 PC6-associated high-confidence differentially methylated regions. Enrichment implicated neuroimmune, metabolic, cytokine, NF-κB, JAK-STAT, TGF-β, and immune-regulatory pathways. Sensitivity analyses supported the stability of the main associations.

Conclusions

Peripheral blood DNA methylation profiles identified partly distinct but overlapping DNA methylation axes in ME/CFS and fibromyalgia. PC5 was aligned with post-exertional malaise and cognitive symptoms, whereas PC6 was aligned with broader pain-related multisystem burden. Independent replication and longitudinal studies are needed to establish clinical utility.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12967-026-08824-5.

Introduction

Many chronic disorders are defined primarily by clinical criteria based on patterns of reported symptoms rather than by established biological mechanisms. Although these classifications are essential for clinical practice, they are pragmatic constructs and may group together individuals with heterogeneous underlying pathophysiology. This is particularly problematic for complex and heterogeneous conditions in which symptom burden, disease severity, immune state, environmental exposures, and neuroendocrine regulation may interact. As a result, symptom-based diagnoses can hinder biologically meaningful variation and limit progress towards mechanistic understanding and more objective, biologically informed, classifications.

Myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) [1] and fibromyalgia (FM) [2] exemplify this challenge. Both are disabling conditions characterized by chronic fatigue, pain, cognitive dysfunction, sleep disturbance, and multisystem symptoms, but they differ in the relative dominance of post-exertional malaise (PEM), widespread pain, and sensory amplification [3]. Their pathophysiology remains incompletely understood, although converging evidence implicates immune dysregulation, altered stress-response systems, metabolic dysfunction, neuroendocrine signalling, and gene-regulatory mechanisms [3–5]. Molecular studies in these disorders should therefore not only test diagnostic group differences, but also determine whether biological variation tracks the symptom dimensions that define clinical burden.

DNA methylation provides a plausible molecular framework for studying these conditions [6, 7]. Methylation at cytosine–guanine dinucleotides (CpGs) contributes to cell identity, developmental regulation, and gene-expression dynamics, and can be shaped by environmental exposures, inflammation, stress, and disease state [8, 9]. Previous epigenome-wide association studies (EWAS) in patients with ME/CFS and FM have identified differentially methylated positions and regions associated with immune, metabolic, and neurological pathways [9–17]. However, individual loci have rarely been replicated suggesting that biologically relevant methylation differences in complex chronic disorders may be distributed across many CpGs with modest individual effects rather than concentrated at a small number of highly significant loci [18].

Unsupervised dimensionality-reduction and region-based approaches may therefore be informative. Principal component analysis can summarize methylation variation across the epigenome into latent axes, while differentially methylated region analysis can identify spatially distributed methylation changes that are more interpretable than isolated CpGs [19, 20]. We hypothesized that the molecular heterogeneity and symptom burden of ME/CFS and FM would be associated with distributed DNA methylation architecture. The objective of this study was to identify DNA methylation axes that were directly linked to symptom burden and determine whether they were associated with diagnostic group, clinical phenotypes, DNA methylation differences, and biologically interpretable pathways in women with ME/CFS, FM, and healthy controls.

Methods

Study design and participants

This cross-sectional study used baseline data from 2 clinical studies conducted at the University Hospital Brussels between April 2021 and April 2025. EPIME, a randomized experimental trial including women with ME/CFS and matched healthy controls, and EPISIP, a crossover study including women with fibromyalgia and matched healthy controls. The two studies were separate clinical studies and that there was no participant overlap between cohorts. Both studies were approved by the University Hospital Brussels ethics committee (EC-2020–205 and EC-2022–118) and registered at ClinicalTrials.gov (NCT04378634 and NCT06475859). All participants provided written informed consent. This study followed the STROBE reporting guideline. Women of 18 to 70 years of age were eligible. Patients had physician-confirmed ME/CFS according to Centres for Disease Control and Prevention criteria [21] or fibromyalgia according to American College of Rheumatology criteria [22]. Healthy controls met the same eligibility criteria, were free of chronic pain, and had an inactive lifestyle to approximate the lower activity levels commonly observed in patients [23]. Exclusion criteria included major neurological, psychiatric, cardiopulmonary, endocrine, systemic, immunological, or oncological comorbidities, pregnancy, or breastfeeding within the preceding 12 months. Both studies included only women to reduce sex-related heterogeneity in methylation and neuroimmune measures [24, 25]. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for case-control studies.

Clinical and molecular measures

Clinical data were collected using REDCap before and during the baseline visit [26]. Symptom burden and clinical covariates were assessed using validated questionnaires, including the Widespread Pain Index [27], DePaul Symptom Questionnaire [28], Central Sensitization Inventory [29], CFS Symptom List [30], and Symptom Severity Scale [27]. Covariates were assessed using the International Physical Activity Questionnaire – Short Form [31], Pain Catastrophizing Scale [32], and Beck Anxiety Index [33]. General health status was measured via the SF-36 [34]. DSQ symptom variables were subjected to exploratory factor analysis to identify latent symptom domains and reduce dimensionality. Detailed questionnaire scoring, factor extraction, and quantitative sensory testing procedures are provided in Supplementary Material S1.

Peripheral blood mononuclear cells were isolated from fasting morning blood samples and cryopreserved until analysis. Genomic DNA was extracted using the QIAamp DNA Blood Mini kit, and genome-wide DNA methylation was profiled using the Infinium MethylationEPIC BeadChip Array v2.0 [35]. Raw methylation data were processed in R using the sesame EPICv2 pipeline [36]. After quality control and probe filtering, 881,392 CpG sites with reliable signal across samples were retained. Technical variation was evaluated before downstream analyses. Batch effects were checked and corrected for when necessary, using the ComBat R package (Supplementary Material S2). DNA methylation was summarized as beta values for effect-size interpretation and M values for statistical analyses. Technical effects were assessed and corrected using ComBat. Leukocytes proportions were estimated from methylation data using EpiDISH and included as covariates in downstream models [37]. Detailed laboratory and methylation preprocessing procedures are provided in Supplementary Material S2, including a Figure summarising the analytical workflow of the analyses performed in the study (Figure S2–1).

Statistical analysis

Analyses were conducted in R. Unsupervised principal component analysis was performed on the processed methylation matrix to identify latent methylation axes. Details on the procedure and results can be found in Supplementary Material S4. Associations between methylation principal components, diagnostic group, and clinical variables were tested using linear models adjusted for age, body mass index, and estimated immune-cell composition. Multiple testing was controlled using the Benjamini-Hochberg false discovery rate [38–40]. Differential methylation analyses compared ME/CFS with healthy controls, fibromyalgia with healthy controls, and all patients with healthy controls. Differentially methylated regions were identified using DMRcate on M-value-transformed methylation data after cell-composition adjustment. Candidate regions were retained for main interpretation if they met predefined criteria for regional significance, minimum absolute methylation difference, CpG support, and directional consistency. Functional enrichment analyses were performed using missMethyl to account for probe-number and CpG-density biases. Additional details on regional methylation analyses, enrichment, and filtering thresholds are provided in Supplementary Material S2. Based on one-way ANOVA assumptions and an expected DNA methylation variability of approximately 5% [40], the final sample size provided 80% power to detect a mean between-group methylation difference of 2%. This calculation was intended to estimate power for a clinically meaningful methylation difference under typical variability assumptions and did not represent formal power for all high-dimensional EWAS tests after multiple-testing correction. A methylation difference of this magnitude has been reported to be biologically relevant and associated with downstream differences in gene and protein expression. [40] Accordingly, high-dimensional methylation results were evaluated using false discovery rate correction, region-level filtering, and sensitivity analyses rather than relying on the sample size calculation alone.

Sensitivity analyses

Robustness of PC-symptom associations was evaluated using bootstrap resampling, permutation testing, leave-one-out analyses, and additional adjustment for estimated immune-cell subsets. Because medication use and comorbidities may influence methylation or index symptom severity, targeted sensitivity analyses additionally adjusted PC-symptom regression models for medication and comorbidity classes. Medication and comorbidity adjustments were not included in the PCA, diagnostic-group comparisons, DMR and functional enrichement analyses, as we did not want to risk correcting for markers of disease burden. Full details are provided in Supplementary Material S7 and S8.

Results

Participants and clinical characteristics

A total of 188 women passed methylation quality control and were included in the analyses: 73 healthy controls, 71 participants with ME/CFS, and 44 participants with fibromyalgia. Participant characteristics are shown in Table 1.

Table 1.

Characteristics of participants included in the present study

	Group
Variable	HCN = 731	FMN = 441	MEN = 711	p-value2	Pairwise (BH adj.)3
Age	40.6 ± 12.9	39.4 ± 11.5	44.2 ± 10.6	0.046	HC vs FM: p = 0.751; HC vs ME: p = 0.044; FM vs ME: p = 0.027
BMI	23.7 ± 3.4	24.8 ± 3.5	24.4 ± 4.1	0.2	HC vs FM: p = 0.059; HC vs ME: p = 0.423; FM vs ME: p = 0.359
Symptoms duration (months)	0 ± 0	92.7 ± 75.5	92.5 ± 85	<0.001	HC vs FM: p = 6.62e-22; HC vs ME: p = 1.03e-25; FM vs ME: p = 0.644
SF-36 Physical component	336.5 ± 48.3	138.7 ± 58.9	128.0 ± 52.5	<0.001	HC vs FM: p = 3.12e-18; HC vs ME: p = 7.51e-24; FM vs ME: p = 0.22
SF-36 Mental component	288.9 ± 74.4	162.6 ± 66.8	204.1 ± 72.1	<0.001	HC vs FM: p = 3.89e-12; HC vs ME: p = 1.68e-09; FM vs ME: p = 0.001
SF-36 Pain	78.9 ± 14.5	33.3 ± 18.5	41.9 ± 18.7	<0.001	HC vs FM: p = 1.12e-17; HC vs ME: p = 2.16e-18; FM vs ME: p = 0.144
WPI	2.0 ± 1.9	11.9 ± 4.1	8.7 ± 4.5	<0.001	HC vs FM: p = 5.27e-19; HC vs ME: p = 5.04e-17; FM vs ME: p = 0.0005
SSS	4.6 ± 1.7	8.5 ± 1.9	9.2 ± 1.5	<0.001	HC vs FM: p = 4.63e-15; HC vs ME: p = 4.13e-22; FM vs ME: p = 0.067
PCS	6.5 ± 7.0	21.6 ± 11.5	15.6 ± 9.8	<0.001	HC vs FM: p = 8.15e-12; HC vs ME: p = 4.42e-09; FM vs ME: p = 0.005
CSI	23.3 ± 11.9	62.7 ± 12.5	57.4 ± 12.3	<0.001	HC vs FM: p = 2.27e-18; HC vs ME: p = 1.78e-22; FM vs ME: p = 0.047
BAI	24.2 ± 3.2	38.6 ± 9.6	38.0 ± 9.0	<0.001	HC vs FM: p = 2.18e-13; HC vs ME: p = 4.2e-19; FM vs ME: p = 0.769
IPAQ	2,505 ± 2,827	3,762 ± 5,545	1,651 ± 1,993	0.002	HC vs FM: p = 0.145; HC vs ME: p = 0.018; FM vs ME: p = 0.0009
CSL	6.3 ± 10.0	48.1 ± 19.3	48.6 ± 16.4	<0.001	HC vs FM: p = 2.15e-17; HC vs ME: p = 4.98e-23; FM vs ME: p = 0.899
DSQ	33.3 ± 28.8	208.5 ± 99.4	223.5 ± 95.0	<0.001	HC vs FM: p = 1.78e-18; HC vs ME: p = 2.38e-24; FM vs ME: p = 0.404
PEM	−0.9 ± 0.3	0.3 ± 0.8	0.7 ± 0.9	<0.001	HC vs FM: p = 3.75e-17; HC vs ME: p = 1.76e-23; FM vs ME: p = 0.007
Cognitive	−0.8 ± 0.2	0.2 ± 0.8	0.6 ± 1.0	<0.001	HC vs FM: p = 3.26e-14; HC vs ME: p = 3.79e-21; FM vs ME: p = 0.031
Visceral	−0.6 ± 0.3	0.5 ± 1.0	0.2 ± 1.1	<0.001	HC vs FM: p = 5.8e-13; HC vs ME: p = 2.85e-08; FM vs ME: p = 0.051
Sleep	−0.6 ± 0.3	0.5 ± 1.1	0.3 ± 0.9	<0.001	HC vs FM: p = 4.23e-12; HC vs ME: p = 1.24e-13; FM vs ME: p = 0.47
Autonomic	−0.6 ± 0.2	0.5 ± 1.0	0.3 ± 1.0	<0.001	HC vs FM: p = 4.89e-15; HC vs ME: p = 1.52e-12; FM vs ME: p = 0.158
Systemic	0.0 ± 0.2	−0.2 ± 1.0	0.2 ± 1.2	0.004	HC vs FM: p = 0.000112; HC vs ME: p = 0.56; FM vs ME: p = 0.034

¹ Mean ± SD; n. ² Kruskal-Wallis rank sum test; Fisher’s exact test. ³ Benjamini–Hochberg correction for post-hoc tests. Abbreviations: BAI, Beck Anxiety Index; BH, Benjamini–Hochberg; BMI, Body Mass Index; Central Sensitisation Inventory; CSL, Chronic fatigue syndrome Symptoms List; DSQ, DePaul Symptom Questionnaire; FM, Fibromyalgia; HC, Healthy Controls; IPAQ, International Physical Activity Questionnaire; ME, Myalgic Encephalomyelitis; PCS, Pain Catastrophising Scale; PEM, post-Exertional Malaise; SSS, Symptom Severity Scale; WPI, Widespread Pain Index

Although both patient groups showed substantially greater symptom burden than healthy controls, ME/CFS and fibromyalgia differed in core clinical features. Participants with fibromyalgia had higher widespread pain, pain catastrophizing, central sensitization inventory scores, and temporal summation than participants with ME/CFS. In contrast, participants with ME/CFS had higher PEM and cognitive symptom scores, and showed lower physical activity levels. The two patient groups thus showed overlapping clinical presentation but with a different clinical weighting: a more PEM-Cognitive and activity-limitation profile in ME/CFS, and a more pain-dominant profile in fibromyalgia (Supplementary Material S1). Estimated immune-cell proportions differed by diagnostic group and symptom burden, particularly for B cells and CD8+ T cells. B-cell proportions were higher in ME/CFS than in healthy controls and were associated with post-exertional malaise, cognitive symptoms, and widespread pain, although only the association with post-exertional malaise remained significant after multiple-testing correction. CD8+ T-cell proportions showed a group-level difference and nominal inverse associations with post-exertional malaise, visceral symptoms, and widespread pain, but these associations did not survive false discovery rate correction (Supplementary Material S3).

Latent methylation axes associated with diagnosis and symptoms

An unsupervised principal component analysis of PBMC DNA methylation profiles was performed. The first ten principal components were evaluated for associations with diagnostic group and clinical variables. PC5 and PC6 were identified as two clinically relevant methylation axes and prioritized because they showed the most consistent and clinically interpretable associations with both diagnostic status and symptom burden after false discovery rate correction. Among the other components, PC2, PC3, and PC7 were found significantly different between groups in the 3-group model, but showed no associations with clinical symptoms (Supplementary Material S4, Table S4–3). PC5 showed the clearest separation of ME/CFS from both fibromyalgia and healthy controls. In adjusted models, PC5 differed between ME/CFS and healthy controls and between ME/CFS and fibromyalgia, whereas the difference between fibromyalgia and healthy controls was smaller. PC6 separated both patient groups from healthy controls, but did not distinguish ME/CFS from fibromyalgia (Supplementary material S4, Table S4–4). PC5 and PC6 were also associated with symptom dimensions (Fig. 1). PC5 was most strongly associated with post-exertional malaise and cognitive symptoms. Higher PC5 scores were associated with greater post-exertional malaise (r = 0.29; β = 8.82; FDR-adjusted p = 5.6 × 10^-5) and cognitive symptoms (r = 0.28; β = 8.00; FDR-adjusted p = 2.9 × 10^-4). PC6 had a broader clinical profile. It was associated with post-exertional malaise, cognitive symptoms, sleep disturbance, visceral symptoms, widespread pain, pain impact, and temporal summation. Among pain-related measures, PC6 was associated with widespread pain (r = 0.24; β = 1.22; FDR-adjusted p = 0.002), SF-36 pain impact (r = −0.27; β = −0.29; FDR-adjusted p = 0.0004), and temporal summation (r = 0.32; β = 6.16; FDR-adjusted p = 4.0 × 10^-5). These patterns suggested that PC5 was more closely aligned with ME/CFS and its post-exertional malaise/cognitive phenotype, whereas PC6 was associated with broader multisystem symptom burden across patient groups.

Fig. 1.

Associations between DNA methylation axes PC5 and PC6 and core disease symptoms. Plots represents linear regression analyses to test associations between PC5 (A) or PC6 (B) and core symptoms. Models were adjusted for age, BMI, and cell composition. p-values are FDR-corrected p-values. Abbreviations: BMI, body mass Index; central sensitisation Inventory; FDR, false discovery Rate; FM, Fibromyalgia; HC, healthy Controls; ME, Myalgic Encephalomyelitis; PEM, post-exertional Malaise; TS, temporal Summation; WPI, widespread pain Index

High-confidence differentially methylated regions and functional enrichment

Region-level analyses were performed to characterize methylation patterns associated with PC5 and PC6 and to compare them with diagnostic group-associated regions. After predefined filtering for regional significance, methylation effect size, CpG support, and directional consistency, PC5 was associated with 591 high-confidence differentially methylated regions, and PC6 was associated with 54 high-confidence regions. Diagnostic contrasts identified 104 high-confidence regions for ME/CFS vs healthy controls, 70 for fibromyalgia vs healthy controls, and 25 for all patients vs healthy controls. PC5 showed the strongest overlap with ME/CFS-associated regions, whereas PC6 showed less diagnostic overlap but a broader symptom-association profile (Table 2).

Table 2.

Summary of latent DNA methylation axes associated with diagnosis and symptom burden

Feature	PC5	PC6
Main symptom associations	Post-exertional malaise and cognitive symptoms	Pain impact, widespread pain, temporal summation, sleep disturbance, post-exertional malaise, and cognitive symptoms
Diagnostic association	Differentiated ME/CFS from fibromyalgia and healthy controls	Differentiated both patient groups from healthy controls, but not ME/CFS from fibromyalgia
Main clinical pattern	ME/CFS-aligned axis	Multisystem pain and symptom-burden axis
High-confidence DMRs	591 regions	54 regions
Overlap with group-associated DMRs	Strongest overlap with ME/CFS-associated regions	Smaller diagnostic overlap; broader symptom association profile
Selected loci	OXT, HIF3A, FAM171A2, NAPRT, SPRED3	FURIN, PDCD1, HIF3A, ESR2/SYNE2, FAM171A2, PREX1
Main enriched pathways	Neuroactive ligand signalling, calcium signalling, PI3K-Akt, Wnt, PPAR, arachidonic acid metabolism	Cytokine-cytokine receptor interaction, NF-κB, JAK-STAT, TGF-β, Notch signalling
Robustness	Stable across bootstrap, permutation, leave-one-out, immune-cell, medication, and comorbidity sensitivity analyses	Visceral symptom association was more sensitive in leave-one-out and medication/comorbidity analyses; widespread pain association attenuated after opioid adjustment but remained statistically significant
Interpretation	Methylation axis associated with PEM and cognitive symptoms, and ME/CFS diagnosis	Methylation axis associated with broader symptom burden across ME/CFS and fibromyalgia, with a particular relevance for pain-related symptoms

PC5 and PC6 were identified using a-priori unsupervised principal component analysis of processed PBMC DNA methylation data. Diagnostic and symptom associations were tested using linear models adjusted for age, body mass index, and estimated immune-cell composition. DMRs were retained using predefined region-level filtering criteria. Robustness was assessed using bootstrap resampling, permutation testing, leave-one-out analyses, additional immune-cell adjustment, and medication/comorbidity sensitivity analyses. DMR indicates differentially methylated region; ME/CFS, myalgic encephalomyelitis/chronic fatigue syndrome; PBMC, peripheral blood mononuclear cell; PEM, post-exertional malaise.

Selected PC5-associated regions mapped to loci including OXT, HIF3A, FAM171A2, NAPRT, and SPRED3. Selected PC6-associated regions mapped to loci including FURIN, PDCD1, HIF3A, ESR2/SYNE2, FAM171A2, and PREX1. Detailed DMR results are provided in Supplementary Material S5. Functional enrichment analyses showed partially distinct pathway profiles for PC5 and PC6 (Fig. 2). PC5-associated regions were enriched for developmental and regulatory processes and for pathways including neuroactive ligand signaling, calcium signaling, PI3K-Akt signaling, Wnt signaling, PPAR signaling, and arachidonic acid metabolism. PC6-associated regions showed enrichment for inflammatory and immune-regulatory pathways, including cytokine-cytokine receptor interaction, NF-κB signaling, JAK-STAT signaling, TGF-β signaling, Notch signaling, osteoclast differentiation, and cell-cycle regulation. These enrichment results supported the interpretation that PC5 and PC6 were associated with partly different methylation patterns: one more closely related to ME/CFS and post-exertional malaise/cognitive symptoms, and the other related to broader pain and multisystem symptom burden.

Fig. 2.

Functional enrichment of differentially methylated regions associated with PC5 and PC6. A, gene ontology enrichment analysis of DMRs associated with PC5 and PC6. B, KEGG pathway enrichment analysis of DMRs associated with PC5 and PC6. Bar length represents the number of DMR-associated genes within each term or pathway. Color intensity represents enrichment significance. DMR indicates differentially methylated region; FDR, false discovery rate; GO, gene Ontology; KEGG, kyoto encyclopedia of genes and genomes

Robustness and sensitivity analyses

Robustness analyses supported the stability of the main PC–symptom associations (Table 2; Supplementary Material S7). Bootstrap resampling showed confidence intervals that did not include zero for all tested PC–symptom associations, with direction of association preserved in 98.2% to 100% of resamples. Permutation testing showed that observed associations were stronger than expected under random symptom-label assignment, with empirical p values surviving false discovery rate correction. Leave-one-out analyses indicated that associations were not driven by individual participants, although the PC6–visceral symptom association was more sensitive to removal of individual observations. PC5 and PC6 were not significantly associated with estimated immune-cell proportions after false discovery rate correction, and PC–symptom associations remained significant after additional adjustment for CD8+ T cells, CD4+ T cells, natural killer cells, B cells, and monocytes. Medication and comorbidity sensitivity analyses yielded similar results (Supplementary Material S8). PC5 associations with post-exertional malaise and cognitive symptoms remained significant across all sensitivity models, with no relative change greater than 20% in regression coefficients. PC6 associations also remained significant, except for the PC6–visceral symptom association. The PC6–widespread pain association was attenuated by more than 20% after opioid adjustment but remained statistically supported, suggesting that opioid use may partly index greater pain burden rather than acting solely as a confounder.

Discussion

ME/CFS and fibromyalgia showed substantial clinical overlap, but the dominant symptom profiles differed. Compared with fibromyalgia, ME/CFS was characterized by more severe PEM, cognitive symptoms, and lower physical activity, whereas fibromyalgia showed greater widespread pain, pain symptoms burden, and temporal summation. This clinical structure was paralleled by partly distinct patterns of DNA methylation. Unsupervised DNA methylation analyses identified 2 latent methylation axes associated with such clinical phenotypes. PC5 was associated with core ME/CFS symptoms such as PEM and cognitive symptoms, and was able to separate patients with ME/CFS from both fibromyalgia and healthy controls. PC6 showed a broader clinical profile: it was associated with multisystem symptom burden, including pain impact, widespread pain, temporal summation, sleep disturbance, PEM, and cognitive symptoms. PC6 separated both patient groups from healthy controls but did not distinguish ME/CFS from fibromyalgia. As widespread pain and temporal summation were higher in fibromyalgia than in ME/CFS, PC6 might reflect a pain-related axis. These findings were supported by region-level DNA methylation analyses, pathway enrichment, and sensitivity analyses. Together, these findings suggest that ME/CFS and fibromyalgia are neither fully separate nor molecularly interchangeable, but partially overlapping conditions with distinguishable clinical and DNA methylation dimensions. A central implication is that diagnostic status and symptom burden should not be treated as interchangeable phenotypes in molecular studies of ME/CFS and fibromyalgia. Our findings challenge broad overlap models of ME/CFS, fibromyalgia, and related syndromes and terminology like functional somatic or central sensitivity syndromes [41, 42]. In addition, the prominence of PEM in this axis supports its status as a defining ME/CFS feature, consistent with diagnostic frameworks that place PEM at the core of the illness [43].

The PC5 pattern is consistent with emerging evidence that ME/CFS involves immune-metabolic and neuroendocrine alterations linked to exertion intolerance and cognitive dysfunction [9–17]. PC5-associated regions included loci related to metabolic stress, hypoxia signalling, neuroendocrine regulation, and immune function, including NAPRT, HIF3A, OXT, FAM171A2, and SPRED3. Enrichment analyses implicated neuroactive ligand signalling, calcium signalling, PI3K-Akt, Wnt, PPAR signalling, and arachidonic acid metabolism. These findings align with deep phenotyping studies of post-infectious ME/CFS reporting altered catecholamine and tricarboxylic acid cycle metabolites, tryptophan and polyamine pathways, and and increased PD‑1+ CD8 T-cells associated with effort intolerance and cognitive symptoms [5]. They are also consistent with proteomic and multi-omics studies implicating oxidative phosphorylation, redox regulation, NAD+ and fatty-acid metabolism, and neuroendocrine‑immune regulation in ME/CFS [44, 45]. Though, our study design can not demonstrate altered expression or functional activity of the implicated genes and pathways, the present results are in line with previous multi-omics findings and suggests that the PEM-cognitive phenotype might be associated with DNA methylation variation.

PC6 showed a different clinical and biological profile. Its strongest associations extended across pain, temporal summation, sleep, cognitive symptoms, and PEM-fatigue. The association between PC6 with widespread pain and temporal summation is noteworthy because both measures clearly differentiated fibromyalgia from ME/CFS in this cohort. This suggests that PC6 may relate to pain and central nociceptive modulation processes [46]. This profile is relevant to fibromyalgia and to the broader overlap between chronic fatigue and chronic pain syndromes. Previous methylation studies in chronic widespread pain and fibromyalgia have reported suggestive signals involving neurological, inflammatory, immune, and nociceptive pathways, but individual loci have been inconsistent across studies [14, 15, 47]. Candidate gene approaches identified differential methylation in GCSAML, GRM2, COMT, BDNF, and OPRM1 genes, suggesting roles for immune inflammation and nociceptive sensitisation and modulation [48–51]. Though fragmented, previous findings are in line within our PC6 axis enrichment analyses, which showed over‑representation of cytokine–cytokine receptor signalling, NF‑κB, JAK–STAT, TGF‑β, Notch and immune checkpoint pathways, suggesting a regulatory inflammatory/neuroimmune sensitisation axis.

The study has several strengths. We applied a stringent statistical and methodological approach to a well-characterised cohort of women with ME/CFS, FM, and healthy controls. DNA methylation was profiled a-priori (that is, blinded to diagnosis or symptom burden) using the Illumina EPIC v2 array and analysed after stringent quality control, batch correction, and adjustment for estimated immune-cell composition [37]. As cell proportions were estimated and adjusted for, our results reflect cell-intrinsic or regulatory DNA methylation differences rather than simple shifts in leukocyte abundance. Nevertheless, the observed B-cell and CD8+ T-cell differences remain biologically informative, especially because several enrichment results pointed toward immune activation, T-cell regulation, and immune-metabolic pathways. Region-level analyses used predefined filtering criteria for statistical significance, methylation effect size, CpG support, and directional consistency. This filtering strategy was designed to preserve biological coherence while reducing the risk of overinterpreting isolated or unstable CpG signals. Our study aligns with methodological literature emphasising power, cell‑composition correction, batch adjustment and region‑based analyses [18, 36, 38, 52], and offers a template for future EWAS in complex chronic diseases.

Several limitations should also be acknowledged. Our cohort included only women. This reduced sex-related heterogeneity in DNA methylation and neuroimmune measures but limits the generalisability of the findings to men and mixed-sex populations. Replication in cohorts including male participants is therefore required, ideally with sufficient sample sizes to evaluate sex-specific methylation patterns. The cross‑sectional design cannot establish causality or temporal stability, for which longitudinal studies are needed. The findings were not evaluated in an independent external cohort. The bootstrap, permutation, leave-one-out, medication, and comorbidity sensitivity analyses assess the internal robustness of the associations but cannot substitute for independent replication. These findings should therefore be replicated before proposing them as validated biomarkers or established molecular subtypes. Peripheral-blood methylation may not reflect epigenomic changes in brain, muscle or autonomic tissues, though it may capture systemic immune programming or metabolic stress. Gene-level interpretation should remain cautious. The study did not include matched transcriptomic, proteomic, metabolomic, or functional measurements. Differential methylation at a locus does not necessarily imply altered transcription of the annotated gene, and pathway enrichment identifies the over-representation of methylation-associated loci rather than activation or suppression of the corresponding pathways. Functional studies are required to validate the roles of identified loci. PC5 and PC6 each explained only a small proportion of total methylation variance. This should not be interpreted as evidence that they are biologically trivial. In high-dimensional EWAS data, most variance reflects broad cellular, technical, genetic, environmental, and random influences, many of which are unrelated to disease. The relevance of PC5 and PC6 lies in their convergence across independent layers of evidence including group differences, symptom associations, DMR overlap, and pathway enrichment. Robustness analyses further supported the main findings: bootstrap resampling, permutation testing, leave-one-out analyses, additional immune-cell adjustment, medication and comorbidity sensitivity analyses generally preserved the principal PC–symptom associations. The association between PC6 and widespread pain was attenuated after opioid adjustment but remained statistically significant, which may reflect both potential medication confounding and the fact that opioid use likely indicate greater pain burden. Thus, we are confident that these components should be interpreted clinically relevant methylation axes.

Conclusion

Our study aimed to identify DNA methylation axes that were directly linked to symptom burden and determine whether they were associated with diagnostic group, clinical phenotypes, DNA methylation differences, and biologically interpretable pathways in women with ME/CFS, FM, and healthy controls. We wanted to go beyond clinical classification –we first identified latent methylation structures through a-priori unsupervised analyses, that were thus independent from diagnostic groups or symptom burden. This revealed that PC5 was most strongly aligned with ME/CFS and its core PEM phenotype, whereas PC6 captured a broader pain-burden axis shared across ME/CFS and fibromyalgia. Therefore, the main contribution is not simply the identification of disease-associated methylation differences, but the demonstration that methylation architecture relates to clinically meaningful symptom variation beyond diagnostic labels. These findings should not be interpreted as validated diagnostic biomarkers or molecular subtypes. Rather, they identify methylation axes associated with clinical phenotypes and symptom burden in ME/CFS and fibromyalgia. If replicated, such axes could help future studies investigate symptom heterogeneity, clinical trajectories, and treatment-response patterns in overlapping chronic fatigue and pain disorders. Longitudinal studies and replication in independent cohorts including male participants, ideally integrating methylation with transcriptomic, proteomic, metabolomic, and immune phenotyping, are needed to determine whether these methylation axes are stable over time, generalisable across sexes, and clinically informative.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

Andrea Polli contributed to study conception and design, data collection, data analysis, interpretation of results, and drafting of the manuscript. Jolien Hendrix had a leading role in participant recruitment, data collection, and study coordination, and contributed to manuscript revision. Arne Wyns and Jente Van Campenhout contributed to participant recruitment, data collection, and manuscript revision. Joeri L. Aerts and Thessa Laeremans contributed to study organization, laboratory coordination, sample handling and storage infrastructure, and manuscript revision. Bernard Thienpont contributed to methylation data analysis, methodological supervision, interpretation of epigenomic results, and manuscript revision. Sabine Allard, Huanyu Xiong, Yanthe Buntinx, Joni Michiels, Jinane Ben Amar, and Lode Godderis contributed to data acquisition, technical or material support, and critical revision of the manuscript. Jo Nijs supervised the study, contributed to study conception, interpretation of results, funding acquisition, and manuscript revision.

Funding

Jolien Hendrix and Huanyu Xiong are PhD fellows funded by the Research Foundation – Flanders (FWO). Arne Wyns is a PhD fellow funded by the Berekuyl Academy, the Netherlands. Jente Van Campenhout and Yanthe Buntinx are PhD fellows supported by the Scottish Charity ME Research UK. Andrea Polli is a senior postdoctoral research fellows funded by FWO. Thessa Laeremans is a postdoctoral fellow funded by the Strategic Research Program (SRP86) of the VUB (awarded to Sabine Allard and Joeri L. Aerts). Jo Nijs holds the Berekuyl Academy Chair, funded by the European College for Lymphatic Therapy, the Netherlands, awarded to the Vrije Universiteit Brussel. The funders had no role in the design and conduct of the study, data collection, management, analysis, or interpretation; manuscript preparation, review, or approval; or to the decision to submit the manuscript for publication.

Data availability

Prof. Polli had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Prof. Polli conducted and were responsible for the statistical and epigenomic analyses. Deidentified participant-level data underlying the findings of this study are not publicly available because of participant privacy, ethical restrictions, and the sensitive nature of clinical and epigenomic data. Deidentified data and analytic code may be made available from the corresponding author upon reasonable request, subject to approval by the relevant ethics committee and completion of an appropriate data use agreement.

Declarations

Ethical approval and consent to participate

The study was performed in accordance with the Declaration of Helsinki and approved by the Medical Ethics Committee of the University Hospital Brussels (ref. 2016/134). All participants received a detailed explanation of the procedure, methods, and objectives of the study and the research and had to agree via signing an informed consent form.

Consent for publication

No identifiable details are present in the manuscript and related material.

Competing interests

All authors have read the journal’s policy on disclosure of potential conflicts of interest. The Vrije Universiteit Brussel and Jo Nijs received teaching fees from various professional associations and educational organizations. Jo Nijs authored books on pain science education and pain management, but the royalties are collected by the Vrije Universiteit Brussel, Brussels, Belgium. Joeri L. Aerts is a member of the Commission for Medicines for Human Use of the Federal Agency for Medicine and Health Products (FAMHP). The remaining authors declare no conflict of interest.