PTPRN2 hypomethylation and PHB2-associated miR-153-3p maturation define dual epigenetic features linked to symptom variability in Myalgic encephalomyelitis

Abstract

Background

Myalgic encephalomyelitis (ME) is a chronic, debilitating condition increasingly linked to epigenetic changes. With its unclear pathophysiology and no validated diagnostic biomarkers, DNA methylation becomes of interest. Specifically, DNA methylation patterns in saliva, to study ME-related epigenetic changes.

Methods

Saliva samples from 54 ME patients and 21 sedentary healthy controls were analyzed by DNA methylation array. Symptom assessment was conducted using validated questionnaires (SF-36, MFI-20, and DSQ).

Results

A significant DNA hypomethylation at the CpG site cg19803194 (Bonferroni-corrected and adjusted for saliva composition, p = 2.14 × 10^− 7) within the PTPRN2 gene body was identified. This hypomethylation was associated with cognitive impairments in both sexes, such as difficulties expressing thoughts and comprehension, commonly known as “brain fog,” and respiratory symptoms in male patients. The hypomethylation also corresponded with reduced circulating levels of miR-153-3p, an intronic microRNA of PTPRN2, which was associated with impaired memory recall in both sexes. Interestingly, the mitochondrial protein Prohibitin 2 (PHB2) was associated with reduced miR-153-3p activity. This association was consistent with a predominant cytoplasmic localization of PHB2 and selective reduction of mature miR-153-3p without changes in its immature forms, suggesting involvement at a post-transcriptional stage that may be attenuated in female patients due to increased extracellular export of PHB2.

Conclusions

These findings suggest a potential epigenetic relationship in ME involving PTPRN2 body hypomethylation and PHB2-associated variation in miR-153-3p levels. While the directionality between gene-body methylation and expression remains a biological hypothesis, these results shed light on potential molecular pathways associated with symptom variability and sex differences in ME severity.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-026-08162-6.

Background

Myalgic encephalomyelitis (ME) is a complex, multi-system disease [1] that presents a significant challenge in clinical practice. It is characterized by profound, debilitating fatigue that does not improve with rest and persists for more than six months [2]. The hallmark symptom of ME is post-exertional malaise (PEM), a worsening of symptoms following minimal physical or mental exertion. Alongside PEM, patients often experience persistent symptoms, including unrefreshing sleep, cognitive impairment, and dysfunctions across the immune, endocrine, and autonomic nervous systems [3–5]. Multiple bodily systems are affected by ME, notably the central nervous system (CNS), as well as the immune, gastrointestinal, and neuroendocrine systems [6, 7]. This clinical heterogeneity makes understanding the pathophysiological mechanisms underlying ME especially challenging [8]. Furthermore, the lack of reliable biomarkers or approved laboratory tests complicates ME diagnosis. Overlapping symptoms with other conditions often lead to diagnostic delays and prolonged medical investigations [9, 10].

The precise causes of ME onset remain unknown, although viral infections are implicated in many cases, with up to 75% of patients reporting an infection episode, often viral, prior to the onset of symptoms [11]. Environmental factors, such as exposure to mold toxins and heavy metals, may also contribute to the onset of the disorder [12]. Additionally, there is a growing interest in epigenetic modifications as potential contributors to ME pathology, though their precise role remains understudied. Although research on DNA methylation changes in peripheral blood mononuclear cells (PBMCs) has provided some insights, findings across studies are often inconsistent. This inconsistency is likely due to limitations in sample size, methodological issues, symptom variability (e.g., “good days vs. bad days”), and differences in analytical approaches [13–17]. Most previous studies have used PBMCs to investigate epigenetic changes in ME, with some employing the Illumina HumanMethylation450 BeadChip to analyze methylation across approximately 450,00 CpG sites [12, 13]. In contrast, others, such as Trivedi et al., 2018 [15], used the expanded MethylationEPIC array, which assesses over 850,000 CpG sites. These studies have primarily focused on pathways related to immune dysfunction and metabolism, yet they provide limited insight into neurocognitive and muscle-related impairments, two key features of ME. Interestingly, epigenetic analyses of alternative sources, such as saliva, are not widely explored but could yield new insights into the epigenetic landscape of ME. Saliva-derived DNA includes contributions from both epithelial and immune cells, offering a broader systemic perspective than PBMC-based studies, which are primarily immune-focused. Given the tissue-specific nature of DNA methylation, examining non-blood sources may help uncover epigenetic changes that would otherwise go undetected.

In this context, saliva sampling presents a practical advantage over blood collection, particularly for patients with severe ME, who are often housebound or bedridden and may face significant barriers to participating in clinical studies. The ease and non-invasiveness of saliva collection make it especially suitable for vulnerable populations [18]. Additionally, prior research indicates that DNA methylation patterns in saliva are often more closely aligned with those in brain and muscle tissues than those in blood [19], making saliva an appealing proxy for studying ME-related epigenetic changes [19]. In this study, we combine our deep phenotyping protocol with complementary methods to examine genome-wide DNA methylation alterations in saliva samples from ME patients, aiming to identify distinct epigenetic markers associated with ME and its diverse symptoms. This strategy holds promise for developing new diagnostic biomarkers and advancing our understanding of sex differences in ME pathophysiology, potentially opening the door to targeted therapeutic strategies.

Materials and methods

Experimental design and study participants

This study was approved by the Institutional Review Board of CHU Sainte-Justine (protocol # 4047) and was conducted in accordance with the guidelines of human ethics regulations. Signed informed consent was obtained from all participants. Figure 1 summarizes our experimental study design. This is a prospective pilot study in which 54 consecutive ME patients diagnosed according to the Canadian Consensus Criteria were enrolled. We also recruited 21 sedentary healthy controls free from chronic pain conditions, mental disorders, recent infections (within 3 months), pregnancy, cohabitation or family history of ME, or use of medications affecting study outcomes. Matched saliva and plasma samples were collected from the same participants, enabling integration of DNA methylation, circulating miRNA, and protein measurements across assays. In this study, we also reported sex-disaggregated analyses to ensure a comprehensive understanding of the potential influences of biological sexes on ME prevalence, symptomatology, and outcomes. To minimize potential confounding effects on DNA methylation profiles, patients and healthy controls receiving medications known or suspected to influence epigenetic regulation were excluded. These included, but were not limited to, systemic corticosteroids, immunomodulatory or immunosuppressive agents, chemotherapeutic drugs, antiepileptics, and other medications with documented effects on DNA methylation. A detailed list of excluded medications is provided in the Supplementary Table S1.

Fig. 1.

Graphical summary of experimental design and workflow. The diagram illustrates the study process, starting with participant recruitment (ME patients and healthy controls), followed by saliva and plasma sample collection. Symptom severity evaluated by health reported questionnaires including 36-Item Short Form Health Survey (SF-36), the Multidimensional Fatigue Inventory-20 (MFI-20) and DePaul Symptom Questionnaire (DSQ). DNA was extracted from the saliva samples, which was then used for Illumina MethylationEPIC array processing, data normalization and quality control, and differential methylation analysis. miRNA was extracted from the plasma samples, followed by cDNA synthesis, qPCR analysis of miR-153-3p expression, and PHB2 protein measurement. Downstream bioinformatic analyses were then performed. ME: myalgic encephalomyelitis; HC: healthy controls

ME symptoms evaluation and participant health status

All participants underwent a comprehensive health status assessment using three standardized questionnaires, including the 36-Item Short Form Health Survey (SF-36), the Multidimensional Fatigue Inventory-20 (MFI-20) and the DePaul Symptom Questionnaire (DSQ). These questionnaires evaluated the participant’s overall health and ME-associated symptoms. The SF-36 questionnaire provided scaled scores to assess physical and mental health, with higher scores indicating better health status [20]. The MFI-20 scores were categorized into various dimensions, including General Fatigue, Physical Fatigue, Reduced Activity, Reduced Motivation, and Mental Fatigue higher scores indicated more severe symptoms [21]. The DSQ questionnaire measured ME symptomatology across five factors: neuroendocrine, autonomic, and immune dysfunction; cognitive impairment; post-exertional malaise (PEM); and sleep disturbances [22]. For the DSQ, higher scores indicated more severe symptoms. Additionally, the severity of cognitive impairments, including difficulties expressing thoughts and understanding concepts, and breathing difficulties, was assessed using the DSQ, through questions 38 (Difficulty expressing thoughts), 39 (Difficulty understanding things), and 49 (Breathing problems). Participants rated the severity and frequency of each symptom on a scale from 0 to 4, where 0 indicates no symptom, and 4 indicates a severe symptom. The reported scores were converted to percentages, and the severity and frequency scores were averaged for each symptom. The scores for questions 38 and 39 were used to assess the severity of cognitive impairments related to expressing thoughts and understanding concepts. The score for question 49 was used to evaluate the severity of breathing difficulties, with higher scores indicating greater symptom severity. Twelve patients participating in the study volunteered to undergo a digital cognitive assessment using the BrainCheck test [23]. The BrainCheck platform is a clinically validated, FDA-cleared digital cognitive assessment tool [24, 25]. This 15-minute test consists of six puzzle-like tasks to assess different cognitive domains: attention and mental flexibility (Trail Making Tests A and B), executive functions (digit symbol substitution and Stroop test), and memory (immediate vs. delayed recall). For each task, the participant receives a score and percentile rank, where higher scores and percentiles indicate better cognitive function.

Saliva samples collection and DNA extraction

Saliva samples were collected using DNA Genotek Oragene kits (Ottawa, ON, Canada) according to the manufacturer’s instructions. Genomic DNA was extracted from the saliva samples using the PureLink Genomic DNA Mini Kit (Thermo Fisher Scientific, Waltham, MA, USA) following the provided protocol. This ensured efficient genomic DNA isolation for subsequent analyses.

Illumina DNA methylation array data analyses

We employed a robust data processing and quality control pipeline for the Illumina Methylation EPIC microarray dataset (Illumina, San Diego, CA, USA), encompassing 75 saliva samples using the R Package Champ (Version 2.28.0) within R (Version 4.1.2) to ensure accuracy and reliability [26]. First, the raw signal intensity data for each of the 75 samples were imported into Champ [27, 28]. Subsequently, these raw intensities were transformed to generate a beta value matrix. The beta values, which range from 0 to 1, reflect the methylation ratios of individual probes. To ensure data integrity, low-quality probes and samples were filtered out. Probes were removed based on the following criteria [29], including non-CpG probes, multi-hit probes, probes with Beadcount < 3, low-quality or failed probes with signal-background detection, P values exceeding 0.05 and probes residing on single-nucleotide polymorphism (SNP) sites or chromosomes X and Y. Simultaneously, samples containing more than 10% failed or low-quality probes were excluded from the analysis, in accordance with standard data quality practices.

Beta value normalization and batch effect removals

To ensure the robustness of our analyses, we performed essential preprocessing steps on the beta values obtained from the Illumina Methylation EPIC microarray data. First, beta values were normalized using the beta-mixture quantile dilation method to correct for potential type-2 probe bias, which can arise from differences in the dynamic range between type-1 and type-2 probes. This normalization step enhances the comparability of methylation measurements across our dataset [30]. We applied a robust empirical Bayesian method to refine the data and mitigate any batch effects originating from microarray slides. This approach allowed us to identify and subsequently remove batch-related variations, ensuring that downstream analyses were not confounded by technical biases [31, 32]. The resulting processed beta values after these normalizations and batch effect removals served as the foundation for our subsequent differential methylation analyses.

Cell-type deconvolution and adjustment

To account for the cellular heterogeneity inherent in saliva samples, we estimated cell-type proportions using the EpiDISH R package (Version 2.10.0). We employed the Robust Partial Correlation (RPC) method, utilizing a saliva-specific reference matrix to determine the relative fractions of epithelial cells, fibroblasts, and immune cells. The resulting cell-type proportions were incorporated as covariates in our downstream linear models to ensure that the observed differential methylation was not driven by shifts in cell composition.

Differential methylation analyses

We conducted a comprehensive analysis to identify and quantify differential methylation patterns between cases and controls. Initially, we performed a probe-level differential methylation analysis using the linear model in the Limma package [33, 34]. This approach allowed us to identify individual probes with statistically significant differences in methylation between the case and control two groups. P values were calculated to assess the significance of these observed differences. To control for potential confounders, the model was adjusted for age, sex, BMI, and the estimated cell-type fractions (epithelial, fibroblast, and immune cells). To further address genomic inflation and ensure well-calibrated null distributions, we applied the Bayesian Adjustment for Confounds (BACON) method. P-values were recalculated using the BACON-corrected test statistics to provide more conservative and reliable estimates of significance. Next, we identified differential methylation regions (i.e., extended genomic regions showing differential methylation patterns) using the Bumphunter method [35]. DNA methylation levels were quantified as beta (β) values and differential methylation was defined using a threshold of Δβ ≥ 0.13, representing the difference in methylation levels between groups. This threshold was chosen based on a comprehensive review of established epigenetic research practices, in which thresholds within the 0.1–0.2 range are widely accepted for defining significant methylation differences [36–38].

Blood samples collection

Blood was collected using 6.0 mL BD vacutainer tubes treated with K₂-EDTA (BD, Franklin Lakes, NJ, USA). The collected blood samples were centrifuged at 216 x g for 10 min to separate the plasma, which was then aliquoted and stored at -80 °C until further analysis.

MicroRNA detection and quantification

Circulating miRNAs were isolated from plasma samples using the mirVana PARIS RNA and Native Protein Purification Kit (Thermo Fisher Scientific) according to the manufacturer’s instructions. To ensure accurate normalization, 10 fmol of C. elegans miRNA cel-miR-39-3p 5’-phosphorylated (custom oligonucleotide 20nmol, Thermo Fisher Scientific) was added to each sample as an exogenous control during RNA extraction. Next, cDNA synthesis was performed using the TaqMan Advanced miRNA cDNA Synthesis Kit (Thermo Fisher Scientific) according to the manufacturer’s instructions in a T3000 Thermocycler (Biometra, Gottingen, Germany). The resulting cDNA was stored at -20 °C for further use. Quantification of miRNAs was performed via qPCR using the TaqMan Advanced miRNA Assays and probes (probe ID: 477918_mir, Thermo Fisher Scientific) on the QuantStudio 3 instrument (Thermo Fisher Scientific). To ensure robustness, qPCR reactions were duplicated for each sample and target, and the mean of cycle thresholds (CT) values was used for further calculations. miRNA 153-3p expression was normalized to the exogenous control cel-mir-39a-3p. This normalization was achieved by calculating ΔCTsample, where ΔCTsample = CTmiR - CTcel-miR-39a-3p. Next, to determine the fold change (ΔΔCT) in expression for each sample, the difference between the expression of hsa-miR-153-3p and the mean expression of the controls (mean ΔCTctrls) was calculated. The formula for ΔΔCT is as follows: ΔΔCT = ΔCTsample - mean ΔCTctrls.

Saliva DNA methylome analysis

Genome-wide DNA methylation profiling was conducted at the Genome Quebec Innovation Center (Montreal, QC, Canada) using the Illumina Infinium Methylation EPIC BeadChip (850 K). This state-of-the-art technology enables the detection of DNA methylation changes across CpG sites in the genome, providing high coverage and accuracy for analysis of the methylation patterns in the study participants.

PHB2 determination

Circulating PHB2 levels were assessed using a sandwich enzyme immunoassay, following the manufacturer’s instructions (MyBiosource, San Diego, CA, US).

Vectors and cells for cell-based assays

The reporter plasmid pmirGLO-153 was constructed by inserting the complementary sequence of miR-153-3p between the PmeI/XbaI restriction sites downstream of the Firefly luciferase gene in the pmirGLO vector (Promega, Madison, WI, USA). For the pFLAG-PHB2 (or pFLAG-Ago2) plasmid, the human PHB2 (or Ago2) gene was inserted between the BamHI/EcoRI restriction sites in a modified pcDNA3.0 vector containing a 3XFLAG sequence between the HindIII/BamHI restriction sites. The plasmid expressing human pri-miR-153-2 was cloned by inserting the sequence corresponding to pre-miR-153-2 flanked by 50 nucleotides on each side into the HindIII/EcoRI restriction sites of the pcDNA3.1 vector. The integrity of each plasmid was verified by Sanger sequencing. FLAG-tagged Ago2 and PHB2 expression in HEK-293T cells was confirmed by Western blot using the Anti-FLAG M2 antibody (Sigma-Aldrich, Burlington, MA, USA). HEK-293T cells were maintained in DMEM (Wisent, Saint-Jean-Baptist, QC, Canada) supplemented with 10% (v/v) FBS (Wisent) at 37 °C in 5% CO₂.

Dual-luciferase reporter assay

For the luciferase assay, 10⁵ cells were plated 24 h before transfection and then co-transfected with 200 ng of either pcDNA3.1 or pcDNA3.1-pri-miR-153-2 vector, 200 ng of either pmirGLO, pmirGLO-153, or pmirGLO-mismatch153, and 200 ng of either pcDNA3.0-3xFLAG or pFLAG-PHB2 (or pFLAG-Ago2) vectors using polyethylenimine (PEI; Sigma-Aldrich) at 4:1 PEI: DNA ratio.Forty-eight hours post-transfection, the cell lysate was prepared according to manufacturer instructions (Promega). Luciferase activity was measured using the Dual-Luciferase Reporter Assay (Promega) with a Triathler Multilabel Tester (Hidex, Turku, Finland). miR-153-3p activity was first calculated as an F_luc/R_luc luminescence signal ratio. The global F_luc/R_luc ratio was then calculated as [S (pmirGlo-miR153T + pri-miR-153)/ S (pmirGlo-miR153T)] / [S (pmirGlo + pri-miR-153) / S (pmirGlo)], with F_luc/R_luc signals (S) measured under four different conditions. Finally, in the reported results, the effect of proteins (PHB2, Ago2) on the global F_luc/R_luc ratio are averaged from biological triplicates and normalized with respect to the empty pcDNA3.0-3xFLAG vector. Statistical significance was assessed using basic one-way ANOVA.

Cell-based assay to examine the effect of PHB2 overexpression on miR-153-3p levels

For this assay, cells were plated 24 h before transfection and then co-transfected with 500 ng of pcDNA3.1-pri-miR-153-2 vector and 500 ng of either pcDNA3.0-3xFLAG or pFLAG-PHB2 vector using a 4:1 PEI: DNA ratio. Forty-eight hours post-transfection, the cells were washed with 1X PBS, 90 µL of transzol (Transgene Biotech Co, Illkirch-Graffenstaden, France) was added, the cells were collected, and the total RNA was extracted according to the manufacturer’s instructions. The GeneJET RNA Cleanup and Concentration Microkit (Thermo Fisher Scientific) was then used to purify the RNA in 22 µL of RNase-free water. The genomic DNA was then removed by incubating with 15 U of recombinant DNase I (Roche, Basel, Switzerland) in its reaction buffer for 25 min at 37 °C, followed by enzyme inactivation in 8 mM EDTA at 75 °C for 10 min. The integrity of the purified RNA was verified by agarose gel. Detection of pri-miR-153 and pre-miR-153 was performed using a standard SYBR Green RT-qPCR protocol, whereas detection of miR-153-3p was performed using the polyA-based RT-qPCR protocol [39], which includes a polyadenylation step followed by a universal reverse transcription. Quantitative analysis of changes in expression levels was derived from real-time PCR data obtained with the LightCycler 96 system (Roche). Briefly, pri-miR-153-2 and pre-miR-135-3-2 were detected using specific forward primers (pri153-2-F: AGC GGT GGC CAG TGT CAT TTT and pre153-2-F: GTG TCA TTT TTG TGA TGT TGC AGC TAG) and a common reverse primer (pre/pri153-2-R: TCA CTT TTG TGA CTA TGC AAC TG), and their levels were normalized to GAPDH and HMBS reference genes (GAPDH-F: TGG TCT CCT CTG ACT TCA ACA, GAPDH-R: AGC CAA ATT CGT TGT CAT ACC, HMBS-F: GGC AAT GCG GCT GCA A, and HMBS-R: GGG TAC CCA CGC GAA TCA C), whereas miR-153-3p was detected using a specific forward primer (TTG CAT AGT CAC AAA AGT GA), a universal reverse primer (CCA GTC TCA GGG TCC GAG GTA TTC), and a universal TaqMan probe (5’FAM-TCT CCT CGG TAT CGA GTC GCA CT-3’TAMRA). Using specific forward primers, the miR-153-3p levels were normalized to 5 S rRNA (GCG TAG AGG AAC CAC ACC AAT C), U6 small nuclear RNA (CAC GCA AAT TCG TGA AGC GTT CCA T) and miR-103-3p (TCG GAG CAG CAT TGT ACA GGG). These experiments were performed in four biological replicates. Relative gene expression was quantified using the 2-^ΔΔCt method [40]. Statistical significance was assessed using Student’s unpaired t-test.

Subcellular fractionation

A 10-cm plate of confluent HEK 293T cells transfected with FLAG-PHB2, pmiRGLO and pcDNA3.1 was lysed with EBKL buffer (25 mM HEPES pH 7.6, 5 mM MgCl2, 1.5 mM KCl, 1 mM DTT, 0.5% NP-40, supplemented with a Mini-Complete protease inhibitor tablet (Sigma)) by incubation on ice for 5 min, followed by Dounce homogenizing (total fraction). The lysate was centrifuged 3 min at 600 × g and the supernatant was further centrifuged twice again 10 min at 18 800 × g to obtain the cytoplasmic fraction. The pellet was washed with EBMK buffer (25 mM HEPES pH 7.6, 5 mM MgCl2, 1.5 mM KCl, 75 mM NaCl, 6% sucrose, 1 mM DTT, supplemented with a Mini-Complete protease inhibitor tablet (Sigma)), passed through a syringe and incubated on ice for 15 min. Following two centrifugation steps (3 min at 600 × g), the resulting pellet was resuspended in EBMK to obtain the nuclear fraction. FLAG-tagged proteins (Sigma F1804), β-actin (Themo Fisher MA5-15739) and histone H3 (Abcam ab1791) were detected by immunoblotting.

Statistical analysis

Outlier detection using the ROUT method (Q = 1%) was applied selectively to specific analyses (e.g., circulating PHB2 measurements), but not to genome-wide DNA methylation analyses. DNA methylation analyses were performed without outlier removal to preserve the integrity of genome-wide signal distributions. Comparisons between two groups were made using independent samples t-tests, and comparisons between three or more groups were made using one-way ANOVA with Tukey’s HSD post-hoc test. Pearson correlation coefficients were calculated for exploratory correlation analyses. A two-tailed p-value of < 0.05 was considered significant. Graphical representations and statistical analyses were performed using GraphPad Prism (version 8, GraphPad Software, Inc., San Diego, CA, United States). To ensure the technical robustness of the findings, technical variance and batch effects originating from microarray slides were mitigated using a robust empirical Bayesian method implemented via BACON. This adjustment was performed using a phenotype-free model to avoid biasing biological signals. The effectiveness of this calibration was verified through diagnostic QQ plots and scatter plot comparisons of log10(P) values (Supplementary Figure S2), demonstrating a significant reduction in genomic inflation and technical noise. The final sample sizes for clinical and molecular modeling (n = 54 whole cohort; n = 42 females; n = 12 males) reflect the dataset after quality control procedures; outlier removal was applied only to specific targeted analyses as described above. Associations between cg19803194 methylation and clinical symptoms were evaluated using multivariable linear regression models adjusted for age, sex (when applicable), BMI, illness duration, batch, and salivary cell-type fractions. Cell-type proportions (Epithelial, Immune, and Fibroblast cells) were estimated for all samples using EpiDISH to account for tissue heterogeneity. Similarly, associations between circulating miR-153-3p levels and delayed recognition memory scores were evaluated using multivariable linear regression adjusted for age, sex, and BMI. Multivariable linear regression models were specified as follows: Outcome ~ Predictor + Age + Sex (when applicable) + BMI + Illness duration + Batch + Cell-type fractions (Epithelial, Immune, Fibroblast). Sex-stratified analyses were conducted as post-hoc exploratory analyses to investigate potential sex-specific effects. Model diagnostics included assessment of residual normality using the Shapiro–Wilk test and leverage analysis to ensure that results were not driven by influential observations. Multiple testing was controlled using the Benjamini–Hochberg false discovery rate (FDR) procedure. To account for multiple comparisons, FDR correction was applied separately to each set of analyses, with a total of 7 tests performed for methylation-based models and 5 tests for miRNA-based models. All analysis scripts (R and Python) and processed data have been deposited in the Gene Expression Omnibus (GEO) repository (GEO accession number GSE327255). Computational analyses were supported by specific R packages, including bacon (v1.24.0) for Bayesian adjustment, EpiDISH (v2.12.0) for cell-type deconvolution, and limma (v3.52.2) for linear modeling.

Results

Demographics and clinical characteristics of participants

A total of 75 participants were enrolled in this cross-sectional prospective study conducted before the COVID-19 pandemic, including 54 individuals with ME (42 women and 12 men) who met the Canadian Consensus Criteria (mean age: 52 ± 1.7 years) and 21 age- and sex-matched sedentary healthy controls (11 women and 10 men; mean age: 53 ± 2.6 years). The clinical and demographic characteristics of the participants are reported in Table 1. All participants were Caucasians (French-Canadians) of European ancestry. No significant difference in body mass index (BMI) was observed between the patients and controls. The scores obtained from three self-reported questionnaires, namely the Short Form 36-Item Health Survey (SF-36), Multidimensional Fatigue Inventory-20 (MFI-20), and DePaul Symptom Questionnaire (DSQ), were consistently indicative of more severe symptoms in the ME group compared to the control group (Table 1).

Table 1.

Clinical and demographic characteristic of participants

Characteristics	All ME	All HC	P value
Number of subjects (Female/Male)	54 (42/12)	21 (11/10)	N/A
BMI (kg/m2)	26 ± 0.8	25 ± 0.7	0.44
Age (years)	52 ± 1.7	53 ± 2.6	0.84
Reported mean age at disease onset (years)	38 ± 1.5	N/A	N/A
Illness duration (years)	14 ± 1.5	N/A	N/A
Questionnaire Scores
36-Item Short Form Health Survey (SF-36) Scores
Physical score	31 ± 1.7****	90 ± 1.9	9.91E-31
Mental score	43 ± 2.3****	85 ± 3.6	6.16E-15
Multidimensional Fatigue Inventory-20 (MFI) Scores
General Fatigue	18 ± 0.3****	7 ± 0.8	4.07E-27
Physical Fatigue	18 ± 0.4****	6 ± 0.7	3.02E-25
Reduced Activity	17 ± 0.4****	7 ± 0.5	1.01E-20
Reduced Motivation	11 ± 0.4***	7 ± 0.8	0.000102
Mental Fatigue	16 ± 0.5****	6 ± 0.7	1.8E-16
DePaul Symptom Questionnaire (DSQ) Scores
Autonomic/Neuroendocrine/Immune	40 ± 1.9****	5.8 ± 1	9E-17
Cognitive	59 ± 2.2****	10 ± 2.9	2.5E-19
Post-exertional malaise	70 ± 2.7****	7 ± 1.6	1.81E-22
Sleep	49 ± 2.4****	12 ± 2.7	3.31E-13

BMI: Body Mass Index; ME: myalgic encephalomyelitis; HC: healthy controls. All data are represented as mean ± standard error of the mean. Two-Tailed Student T-test comparing ME patients and HC group were performed, and changes in specific scores were considered significant with ***P-value < 0.001 and ****P-value < 0.0001. SF-36: Higher scores indicate better physical and mental health status, MFI-20: Higher scores indicate more severe fatigue symptoms across various dimensions (General Fatigue, Physical Fatigue, Reduced Activity, Reduced Motivation, Mental Fatigue) and DSQ: Higher scores indicate more severe symptoms across the four domains assessed: autonomic/neuroendocrine/immune dysfunction, cognitive impairment, post-exertional malaise (PEM), and sleep disturbances

DNA methylation alterations associated with ME and symptom variability

In our DNA methylation analysis of approximately 850,000 CpG sites using the Illumina Infinium MethylationEPIC 850 K BeadChip array, DNA methylation levels were quantified as β values, representing the ratio of methylated signal to total signal. Using a threshold of an Δβ ≥ 0.13, we identified seven differentially methylated CpG sites (Table 2). This threshold aligns with established epigenetic research practices, in which Δβ values within the 0.1–0.2 range are commonly used to define biologically meaningful differences [36–38]. We applied the Bonferroni correction to adjust for multiple comparisons, setting a significance threshold between 10^{− 8} and 10^{− 7} based on adjusted p-values and the correlation between neighbouring CpG sites. After correction, one CpG site, cg19803194, located within the PTPRN2 gene body, showed significant DNA hypomethylation (mean Δβ = -0.22; adjusted p-value = 7.85 10^{− 8}) associated with ME (Table 2; Fig. 2a). The cg19803194 probe (Illumina ID: cg19803194) is located on chromosome 7 (MAPINFO: 158,155,733; hg19/GRCh37) within the gene body of PTPRN2. The locus overlaps regions marked by H3K27ac, consistent with accessible regulatory chromatin. To ensure this signal was not driven by tissue heterogeneity, we estimated salivary cell-type proportions using the EpiDISH algorithm. The samples were primarily composed of immune cells (mean = 0.67 ± 0.13) and epithelial cells (mean = 0.31 ± 0.14), with a minor contribution from fibroblasts (Figure S1, Supplementary Table S3). Upon adjusting our linear models for these cell fractions, the hypomethylation at cg19803194 remained highly significant (adjusted p-value = 2.14 × 10^{− 7}). Furthermore, applying BACON correction to calibrate the null distribution and account for genomic inflation further confirmed the robustness of this association (adjusted p-value = 3.32 × 10^{− 9}) (Supplementary Figures S2–S3, Supplementary Table S4). To address the statistical stability of this finding given our sample size (n = 75), we performed a post-hoc power analysis. Based on the observed effect size at cg19803194 (logFC ≈ -2.55), our study achieved a power of 0.914 at the stringent Bonferroni-corrected alpha level (7.04 × 10^{− 8}), indicating that the discovery is well-powered and statistically stable. While no sex-based differences in DNA methylation levels were observed among ME patients for this specific CpG site, both male and female ME patients exhibited lower methylation levels at cg19803194 compared to their respective healthy controls (Fig. 2b). Nevertheless, ME patients were classified into two subgroups based on their DNA methylation status at cg19803194: a hypomethylation group (β-value < 0.65, n = 32) and a normal methylation group (β-value ≥ 0.65, n = 20). Notably, the hypomethylation ME group exhibited a significantly longer illness duration (five years longer) compared to the normal methylation ME group at the time of testing (Table 3).

Table 2.

Top differentially-methylated CpG sites in ME patients compared to HC group

Probe ID	Chr	Feature	Gene	HC β	ME β	Δβ	HC Methylation%	ME Methylation%	P.Value	adj.P.Val
cg19803194	7	Body	PTPRN2	0.85	0.63	-0.22	85.5%	63.5%	4.00E-13	7.85 10 − 8
cg22351833	19	IGR		0.49	0.36	-0.13	49.5%	36.3%	2.00E-06	0.04
cg23549902	7	IGR		0.52	0.37	-0.15	52.3%	37.4%	4.00E-06	0.06
cg20964957	4	IGR		0.52	0.40	-0.12	52.3%	40.1%	2.00E-05	0.12
cg15572235	7	IGR		0.53	0.39	-0.14	53.7%	39.6%	2.00E-05	0.13
cg15549821	19	Body	PLEKHA4	0.71	0.54	-0.17	71.6%	54.8%	3.00E-05	0.15
cg08506353	6	Body	HLA-DPB2	0.60	0.42	-0.18	60.0%	42.1%	4.00E-05	0.16
cg15991630	18	IGR		0.70	0.49	-0.21	70.0%	49.2%	8.00E-05	0.19

ME: myalgic encephalomyelitis; HC: healthy controls; CpG: Methylation site; CHR: Chromosome; Feature: Genome position; cgi: CpG islands; PTPRN2: Protein Tyrosine Phosphatase Receptor Type N2; PLEKHA4: Pleckstrin Homology Domain Containing A4; HLA-DPB2: Major Histocompatibility Complex, Class II, DP Beta 2 (Pseudogene); IGR: Intergenic region. DNA methylation changes at CpG sites were considered significant at adjusted p-value ˂ 0.05 and Δβ value > 0.12Δβ: Delta Beta values (Difference in HC_AVG and ME_AVG scores). Negative delta indicates hypomethylation in the ME group compared to the HC group. %-percent and adj.P.Val-adjusted p-value. HC β and ME β values were calculated as follows: Raw signal intensity data from the Illumina MethylationEPIC microarray were imported into the Champ R package. Beta values were computed as the ratio of methylated probe intensity to total intensity. Low-quality probes and samples were removed based on quality control criteria. Beta-mixture quantile dilation normalization was performed. Batch effects were removed using an empirical Bayesian method. Differential methylation analysis was conducted using limma. Final HC β and ME β values were calculated as the mean normalized beta values for each group

Fig. 2.

The multifaceted relationship between DNA methylation at CpG site cg19803104 and symptom variability in male and female ME patients. (a) DNA methylation levels at cg19803194 compared among all ME patients (n = 54) and matched healthy controls (HC) (n = 21). The cg19803194 probe (Illumina ID: cg19803194) is located on chromosome 7 (MAPINFO: 158,155,733; hg19/GRCh37) within the gene body of PTPRN2. The locus overlaps regions marked by H3K27ac, consistent with accessible regulatory chromatin. Mean and standard error of the mean (SEM) are indicated. Statistical analysis: independent samples t-test. (b) Variation of DNA methylation levels at cg19803104 between male (ME n = 12, HC n = 11) and female participants (ME n = 42 vs. HC n = 11). Median values are indicated. Statistical analysis: One-way ANOVA with Tukey’s post-hoc test. Box plots showing the impact of DNA hypomethylation levels (β-value < 0.654) at cg19803104 on the severity of expressing thoughts (c) and understanding things (d), assessed using scores from the DSQ questionnaire are shown (hypomethylation n = 34, normal methylation n = 20). Min to max whiskers show all data points. Statistical analysis: independent samples t-test. Scatter plots representing correlations between DNA methylation levels at cg19803104 and breathing problems (DSQ scores) in (e) female (n = 42) and (f) male (n = 12) ME patients are illustrated. A significant negative correlation was observed only in male patients (r = -0.657; p = 0.02). Statistical analysis: Pearson correlation (for male patients) and Pearson correlation (for female patients). (g) Box plots illustrating the relative expression levels of miR-153-3p in plasma samples of ME patients with DNA hypomethylation (n = 34) compared to those with normal methylation levels (n = 20) at cg19803194. Min to max whiskers show all data points. Statistical analysis: independent samples t-test. (h) Scatter plot showing a positive correlation between delayed memory recognition percentile ranks assessed using the BrainCheck test and miR-153-3p expression levels in a subset of ME patients (n = 12, r = 0.584; p = 0.046). Statistical analysis: Pearson correlation. Higher BrainCheck scores indicate better cognitive function, reflecting improved processing speed, executive function, and memory. Statistical significance is indicated by * p < 0.05

Table 3.

Clinical and demographic characteristics of ME patients based on methylation status

Characteristics	Hypo ME	Normal ME
Number	34	20
Female/male	26/8	16/4
Age	53 ± 2	50 ± 3
BMI(Kg/m2)	25 ± 1	26 ± 2
Illness duration	17 ± 2 *	11 ± 2
Physical score (SF-36)	32 ± 2	29 ± 3
Mental score (SF-36)	44 ± 3	42 ± 4
General Fatigue (MFI-20)	18 ± 0.4	19 ± 1
Physical Fatigue (MFI-20)	18 ± 0.4	18 ± 1
Reduced Activity (MFI-20)	17 ± 0.5	17 ± 1
Reduced Motivation (MFI-20)	11 ± 0.5	10 ± 1
Mental Fatigue (MFI-20)	15 ± 0.6	16 ± 1
Autonomic/Neuroendocrine/Immune (DSQ)	42 ± 0.3	38 ± 2
Cognitive (DSQ)	61 ± 3	55 ± 4
Post-exertional malaise (DSQ)	70 ± 3	72 ± 5
Sleep (DSQ)	48 ± 3	52 ± 4
Respiration problem score	48 ± 5	48 ± 6

Hypo ME: DNA methylation β-value at CpG site cg19803194 < 0.653

Normal ME: DNA methylation β-value at CpG site cg19803194 ≥ 0.654

This hypomethylation at cg19803194 was associated in both sexes with cognitive impairments, such as difficulties expressing thoughts and understanding concepts, commonly known as “brain fog” (Fig. 2c-d). The DNA methylation level at cg19803194 was inversely correlated with the severity of breathing difficulties, as measured by the DSQ questionnaire (r = − 0.642, p = 0.02), but this association was observed only in male patients (Fig. 2e–f). To ensure the robustness of these clinical findings, we performed multivariable linear regression adjusted for age, sex, BMI, illness duration, batch, and salivary cell fractions (Epithelial, Immune, and Fibroblast cells). These models confirmed that the inverse association between cg19803194 methylation and respiratory severity remains a significant male-specific phenotype (β= -280.0, 95% CI: -515.1 to -44.86, p = 0.02; Supplementary Table S7), independent of cellular heterogeneity. FDR-adjusted p-values for these associations are reported in Supplementary Tables S7 and S8.

Association of circulating miR-153-3p on ME pathogenesis and symptom variability

PTPRN2 encodes Protein Tyrosine Phosphatase Receptor Type N2, also known as islet cell autoantigen-related protein (ICAAR, IA-2β) or phogrin. It includes the intronic primary microRNA of miR-153-3p (pri-miR-153-2), which is co-expressed with its host gene [41]. Although saliva can reflect brain and muscle methylation patterns, the direct sampling of brain or muscle tissues remains a substantial barrier to accurately evaluating changes in PTPRN2 expression at the tissue level. We have further strengthened the rationale for assessing circulating miR-153-3p levels as a surrogate marker by emphasizing its strong association with PTPRN2 and its relevance to brain function in ME. Interestingly, we observed a correlation between hypomethylation at cg19803194 and reduced plasma levels of miR-153-3p (Fig. 2g), which were associated with delayed memory recognition in both sexes in the 12 participating patients who volunteered for an additional cognitive assessment using the BrainCheck test (Fig. 2h). This 15-minute test consists of six puzzle-like tasks to assess different cognitive domains: attention and mental flexibility (Trail Making Tests A and B), executive functions (digit symbol substitutions and Stroop test) and memory (immediate vs. delayed recall).

To validate this molecular link, we evaluated the cognitive relevance of the PTPRN2 locus using multivariable linear regression adjusted for age, sex, and BMI. This analysis confirmed that circulating levels of miR-153-3p were significantly associated with objective memory performance, specifically delayed recognition population percentiles (β = 11.3, 95% CI: 0.17 to 22.42, p = 0.046; Supplementary Table S8). These findings suggest a potential dual association for the locus, whereby epigenetic status is associated with respiratory distress in males, while the hosted miRNA is associated with cognitive memory outcomes across the cohort.

Association of PHB2 with posttranscriptional regulation of miR-153-3p

Prohibitin 2 (PHB2) is a multifunctional mitochondrial protein that plays critical roles in cellular homeostasis, including maintaining mitochondrial integrity, regulating apoptosis, and facilitating nuclear signaling [42]. Our focus on PHB2 stemmed from the fact that a significant difference in plasma PHB2 levels was observed between male and female ME patients, with females showing a strong correlation (r = 0.48; p = 0.003) with circulating miR-153-3p levels (Fig. 3a-c). Although there is no documented evidence that PHB2 directly regulates miR-153-3p or any other microRNA, or that PTPRN2 modulates PHB2 expression, we explored a potential association through which PHB2 could influence the post-transcriptional regulation of miR-153-3p. PHB2 is a versatile mitochondrial protein that moves dynamically between various cellular compartments, including the nucleus, cytoplasm, mitochondria, and cell membrane. It is also released into the bloodstream through shedding from its membrane-bound form. The ability of PHB2 to shift its localization is associated with various age-related diseases, including Alzheimer’s, Parkinson’s, diabetes, and cancer [43–47]. While circulating PHB2 levels were globally higher in the ME group compared to the healthy controls (Fig. 3d), stratification by biological sex revealed that this elevation was primarily driven by female patients, a difference not observed in healthy controls (Fig. 3e).

Fig. 3.

Sex-specific correlation and elevation of circulating PHB2 in ME patients ((a–c) Scatter plots showing correlations between circulating PHB2 levels and miR-153-3p expression levels in all ME patients, female patients (n = 42, r = 0.48; p = 0.003), and male patients, respectively. Statistical analysis: Pearson correlation. (d) all ME patients (n = 46) and healthy controls (HC) (n = 21) and (e) further stratified by biological sex (ME females n = 36 vs.ME male n = 10 and HC females n = 11, HC males n = 10). Min to max whiskers show all data points. Statistical analysis: Independent samples t-test (for ME vs. HC), and Statistical analysis: One-way ANOVA followed by Tukey’s HSD post-hoc test (for male ME vs. male HC, and female ME vs. female HC, female ME vs. male ME). Statistical significance indicated by *p < 0.05; ***p < 0.01

Our ribonucleoproteomic analysis indicated that intracellular PHB2 physically interacts with the stem-loop of immature forms of miR-153-3p (miR-153-2) (Di Tomasso, G. Miller-Jenkins, LM. and Legault, P., unpublished results). Of note, miRNA maturation is a tightly regulated process in which proteins that bind to immature miRNAs can either promote or inhibit miRNA maturation and activity [48]. Using a dual-luciferase assay in HEK-293T cells, we tested the impact of PHB2 overexpression on miR-153-3p activity (Fig. 4a). This assay incorporates a Firefly luciferase (Fluc) reporter containing the miR-153-3p target site, whereas Renilla luciferase (Rluc) serves as an internal control. The assay enables robust and specific detection of miR-153 activity, as a mismatch target site does not produce a measurable response (Supplementary Fig. S6). By comparing the global Fluc/Rluc ratio, we assessed the effect of PHB2 expression while controlling for nonspecific influences from the reporter construct alone [49]. As anticipated, overexpression of Argonaute 2 (Ago2), a key mediator of miRNA function, reduced the global Fluc/Rluc ratio by 1.7-fold relative to the control (Fig. 3d.60 ± 0.03 for Ago2). In contrast, PHB2 overexpression significantly increased the global Fluc/Rluc ratio by 1.5-fold (Fig. 4a.5 ± 0.1 for PHB2), suggesting that PHB2 may be associated with reduce miR-153-3p maturation through interactions with its immature forms. Subcellular fractionation combined with immunoblotting indicate that PHB2 predominantly localize in the cytoplasmic fraction (Fig. 4b). To further delineate the stage of maturation affected, we quantified immature forms of miR-153-2. While PHB2 overexpression resulted in a 1.5-fold decrease in mature miR-153-3p levels (Fig. 4c), the levels of pri-miR-153-2 and pre-miR-153-2 remained unchanged (Fig. 4d). Considering the maturation process of miR-153-3p (Supplementary Fig. S7), these findings suggest that the PHB2-associated inhibition occurs post-transcriptionally in the cytoplasm, likely at the Dicer-processing or Argonaute-loading steps.

Fig. 4.

In vitro associations between PHB2 expression and miR-153-3p maturation. (a) Histogram illustrating the results from a dual-luciferase reporter assay in HEK-293T cells. Cells were transfected with a pmirGlo vector containing the miR-153-3p target sequence, a vector expressing pri-miR-153-2, and either an empty vector (E.V.) or a vector expressing Ago2 or PHB2. The normalized Fluc/Rluc ratio is shown. Mean and SEM are indicated. Statistical analysis: Independent samples t-test. (b) Transfected Flag-PHB2 is expressed in the cytoplasmic fraction of the cell. Western blot of a subcellular fractionation assay (T: total fraction, C: cytoplasmic fraction, N: nuclear fraction). (c) Histogram showing the expression levels of mature miR-153-3p measured by RT-qPCR in HEK-293T cells transfected with a pri-miR-153-2 expression vector and either an E.V. or a PHB2-expressing vector. Mean and SEM are indicated. Statistical analysis: Independent samples t-test. (d) Histogram showing the expression levels of immature pri-miR-153-2 and pre-miR-153-2 measured by RT-qPCR in HEK-293T cells transfected with a pri-miR-153-2 expression vector and either an E.V. or a PHB2-expressing vector. Mean and SEM are indicated

Analysis according to the DNA methylation status at cg19803194 showed that ME patients in the hypomethylation group showed lower plasma PHB2 levels resulting in a further reduction of circulating miR-153-3p levels compared to the normal methylation ME group (Fig. 5a). However, when ME patients were further stratified by both DNA methylation status at cg19803194 and biological sex, a significant elevation in plasma PHB2 levels remained in female patients within the hypomethylation group compared to their male counterparts (Fig. 5b). This divergence suggests that the sex-specific regulation of PHB2 may reflect a compensatory pattern in females to mitigate the effects of PTPRN2 hypomethylation on miR-153-3p availability.

Fig. 5.

Integration of PTPRN2 methylation status and sex-specific PHB2 dynamics in ME patients. Box plots representing the variation in circulating PHB2 levels based on the DNA methylation status at cg19803194 among (a) ME patients (ME hypomethylation n = 29, ME normal methylation n = 19) and (b) following stratification by biological sex (ME hypomethylation females n = 22, ME hypomethylation males n = 6, ME females normal methylation n = 15, ME males normal methylation n = 4). Min to max whiskers show all data points. Statistical analysis: Independent samples t-test (for ME hypomethylation, ME normal methylation), and Statistical analysis: One-way ANOVA followed by Tukey’s HSD post-hoc test (for male ME hypomethylation vs. male normal methylation, and female ME hypomethylation vs. female normal methylation, female ME hypomethylation vs. male ME hypomethylation). Statistical significance indicated by *p < 0.05

Discussion

Our study provides compelling evidence linking PTPRN2 and its intronic-embedded microRNA, miR-153-3p [41], to cognitive decline and other symptom variability associated with ME. The detection of DNA hypomethylation at cg19803194 in the PTPRN2 gene body from saliva samples strongly supports this connection, as salivary DNA methylation profiles closely mirror those observed in brain and muscle tissues [18, 19]. Importantly, adjustment for saliva cellular heterogeneity and the absence of group-level differences in cell-type proportions support the interpretation that this signal reflects a biologically meaningful and systemic epigenetic alteration rather than a tissue composition artifact. The relationship between gene-body methylation and gene expression is complex and context-dependent, and the observed association between PTPRN2 hypomethylation and miR-153-3p levels should be interpreted as a hypothesis rather than a direct measure of transcriptional activity. Gene body methylation is increasingly recognized as a positive regulator of transcription, ensuring accurate splicing, suppressing spurious transcription, regulating alternative splicing, and promoting transcriptional stability [50, 51].

Conversely, hypomethylation in the gene body has been reported to be associated with altered gene expression, although this relationship remains context-dependent. In our study decreased circulating miR-153-3p levels were observed in association with PTPRN2 hypomethylation; however, miR-153-3p should be considered an indirect surrogate rather than a direct measure of PTPRN2 transcriptional activity. The hypomethylation at cg19803194 is particularly significant given the role of PTPRN2 in the brain [52–57] and its established functions in neurotransmitter regulation, insulin secretion [58], and epigenetic modulation in metabolic and neurological disorders. While DNA methylation changes in PTPRN2 have been implicated in other diseases [59–61], their link to cognitive dysfunction and ME-related respiratory issues is novel. Our data highlight that PTPRN2 gene body hypomethylation correlates with reduced circulating miR-153-3p levels and impaired cognitive performance, suggesting that intragenic PTPRN2 methylation may be associated with miR-153-3p expression. This aligns with prior evidence showing the co-regulation of miRNAs and their host genes in various tissues, particularly in neuroendocrine and neuro-immune contexts. Although sex-specific methylation patterns have been observed in diseases such as Alzheimer’s (AD) [56] and Parkinson’s [62], no such differences were detected at the cg19803194 site in ME. Impaired neurotransmitter regulation, linked to reduced PTPRN2 expression, could also be associated with changes in the brainstem, potentially contributing to the respiratory dysfunctions observed in male ME patients. Notably, recent studies highlight the role of PTPRN2 hypomethylation in the pathogenesis and progression of respiratory disorders [63, 64], further underscoring its relevance to ME. Although we observed a correlation between hypomethylation at cg19803194 and reduced plasma levels of miR-153-3p, we acknowledge that the interpretation of this finding is complex. The DNA methylation paradox illustrates the intricate relationship between gene body methylation and transcription, and PTPRN2 undergoes alternative splicing, generating transcripts that may or may not include the miR-153-3p-encoded intron. Furthermore, given that PTPRN2 is not expressed in PBMCs, the primary source of cells in blood, plasma miR-153-3p levels may not directly reflect intracellular PTPRN2 expression or PTPRN2-derived miR-153-3p production. Future studies should investigate miR-153-3p expression in tissues where PTPRN2 is expressed to further clarify this relationship. These sex-stratified findings should be interpreted with caution, as they were derived from post-hoc exploratory analyses and require validation in larger independent cohorts. In addition, future studies in larger independent cohorts will further validate the stability and reproducibility of the cg19803194 signal using resampling-based approaches such as bootstrap or split-half validation.

Our findings also suggest a potential involvement of miR-153-3p in ME pathophysiology. Indeed, its known roles in regulating key genes involved in inflammation, mitochondrial function, and neuroprotection strengthen its relevance to ME [65–69]. Moreover, its ability to modulate α-synuclein and amyloid precursor protein, both implicated in neurodegeneration, raises intriguing possibilities about shared pathways between ME and other chronic neurological diseases [70–74]. Reduced miR-153-3p levels, as seen in our ME cohort, may be associated with neuro-inflammation and impair synaptic plasticity, contributing to the hallmark symptom of brain fog.

The discovery of PHB2 as an associated factor in miR-153-3p maturation adds a new layer to this regulatory network. Our ribonucleoproteomic data indicate that PHB2 interacts with immature forms of miR-153-2, and thus may be associated with its maturation into functional miR-153-3p. A negative regulatory effect was validated by dual-luciferase assays and RT-qPCR, supporting an association between PHB2 expression and reduced miR-153-3p levels. To further contextualize the relationship between PHB2 and miR-153-3p, we performed additional in vitro analyses to examine the subcellular localization and stage of miRNA regulation. Our results indicate that PHB2 is predominantly localized in the cytoplasmic compartment under the experimental conditions used. Notably, PHB2 overexpression was associated with reduced levels of mature miR-153-3p, while levels of primary and precursor forms remained unchanged. This pattern suggests that PHB2 may be associated with the regulation of miRNA levels at a cytoplasmic stage rather than at the level of transcription or early nuclear processing. Interestingly, sex-specific differences in PHB2 localization and circulating levels suggest a compensatory pattern in females, where elevated extracellular PHB2 may be associated with reduced intracellular inhibitory effects of miR-153-3p maturation. This pattern may contribute to the observed resilience of female ME patients to severe respiratory dysfunction, as miR-153-3p has been shown to protect against hypoxia-induced pulmonary vascular remodeling [75].

Notably, in line with the significant symptom burden captured by the SF-36, MFI-20, and DSQ questionnaires, ME patients exhibited markedly lower physical and mental health scores, alongside significantly higher fatigue levels and symptom severity across all DSQ domains compared to healthy controls. However, no significant correlation was found between illness duration and DNA hypomethylation levels (Supplementary Fig. S5). This finding suggests that while ME profoundly impacts symptom severity, the observed epigenetic changes may not be directly proportional to disease duration. Instead, other factors such as the initial triggering event, genetic predisposition, or ongoing environmental exposures may play a more influential role in establishing or maintaining these methylation patterns. Furthermore, the cross-sectional nature of our study limits our ability to assess temporal changes in DNA methylation. Future longitudinal studies are necessary to elucidate the dynamic interplay between illness duration, epigenetic modifications, and symptom trajectories in ME.

From a translational perspective, the identification of a common epigenetic signal at the PTPRN2 locus across patients with different illness durations, clinical trajectories, and likely different triggering events suggests the presence of a stable molecular feature associated with disease state or symptom domains rather than disease onset alone. This observation is particularly relevant in ME, a clinically heterogeneous disease currently diagnosed based on symptoms rather than biological markers. The integration of DNA methylation, circulating miRNA, and protein measurements in this study suggests the existence of a multi-layer molecular axis that may contribute to biologically informed patient stratification. Such molecular stratification approaches could eventually help clinicians better characterize disease heterogeneity, particularly in relation to cognitive impairment, respiratory symptoms, and sex-specific differences. Although these findings do not yet translate into immediate diagnostic or therapeutic applications, they provide a biologically grounded framework that may support future biomarker development, patient subgroup identification, and precision medicine approaches in ME.

While our findings offer insights into ME pathophysiology, several limitations warrant attention. Due to the cross-sectional design of this study, causal relationships cannot be established, and all observed findings should be interpreted as associative rather than mechanistic. First, our reliance on saliva as a surrogate tissue, while practical, may not fully capture methylation patterns in brain or muscle tissues. While saliva is a robust proxy for neuro-epigenetic studies, future work utilizing single-cell methylation sequencing would further refine our understanding of which specific cell types within the saliva milieu contribute most significantly to the cg19803194 signal. Future studies should validate these findings in larger cohorts and incorporate multi-tissue analyses to better delineate tissue-specific epigenetic signatures. Secondly, we observed correlations between DNA methylation changes and clinical features of ME; however, confirming the functional impact of the identified cg19803104 hypomethylation will require further investigation. Due to limited tissue availability and the fact that PTPRN2 is not expressed in PBMCs, the only cell type collected from our participants, we were unable to perform luciferase reporter assays, a key method for assessing regulatory function. Future studies should prioritize conducting investigations in relevant cell types and tissues to better understand the biological processes by which these methylation changes may contribute to the ME pathophysiology. Additionally, while plasma miR-153-3p levels provided valuable insights, measuring miR-153-3p in saliva, the same biological compartment used for detecting PTPRN2, represents a promising direction for future research. Lastly, the interplay between PTPRN2 expression, PHB2 dynamics, and miR-153-3p targets merits deeper exploration through integrative transcriptomic and proteomic approaches to fully understand their role in the pathogenesis of ME. Moreover, the fact that the DNA hypomethylation mark we identified in PTPRN2 is located within the intron harboring miR-153 raises important mechanistic questions, as it may impact transcription and co-transcriptional splicing of PTPRN2, both of which are likely to be tightly coupled to miR-153 mirtron maturation.

Conclusion

This study identifies a potential dual regulatory relationship in ME involving PTPRN2 hypomethylation located within the intron hosting miR-153 and PHB2-associated variation in miR-153-3p levels, together linked to symptom variability. The intronic hypomethylation may reflect coordinated processes related to PTPRN2 transcription, splicing, and miR-153 processing, while the sex-specific patterns of PHB2 and observed associations of miR-153-3p highlight potential molecular features relevant to patient stratification. Collectively, these findings suggest a link between DNA methylation and miRNA dynamics, providing insight into the molecular landscape of ME pathogenesis and the basis of sex-specific differences in symptom expression. These observations remain exploratory and will require validation in larger, longitudinal, and functional studies to assess their potential diagnostic, prognostic, and therapeutic relevance.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AD

Alzheimer

Ago2

Argonaute 2

ANOVA

Analysis of variance

AQEM

Association Québécoise de l’Encéphalomyélite Myalgique

BamHI

Restriction endonuclease genus Bacillus, species amyloliquefaciens, strain H, first enzyme isolated

BMI

Body mass index

cDNA

Complementary DNA

CFS

Chronic fatigue syndrome

cgi

CpG islands

CHR

chromosome

CHU Sainte-Justine

Centre hospitalier universitaire Sainte-Justine

CIHR

Canadian Institutes of Health Research

CNS

Central nervous system

CO₂

Carbon dioxide

COVID-19

Coronavirus disease 2019

CpG

Cytosine-phosphate-guanine

CT

Cycle thresholds

Ctrls

Controls

DMEM

Dulbecco’s Modified Eagle Medium

DNA

Deoxyribonucleic acid

DNase

Deoxyribonuclease

DSQ

DePaul Symptom Questionnaire

EcoRI

Restriction endonuclease genus Esherichia, species coli, strain RY13, first enzyme isolated

EDTA

Ethylenediaminetetraacetic acid

EV

Empty vector

FBS

Fetal bovine serum

FDA

Food and Drug Administration

Fluc

Firefly luciferase

fmol

Femtomole

g

Gravitational force

GEO

Gene Expression Omnibus

HC

Healthy controls

HEK-293T

Human embryonic kidney 293T

HindIII

Restriction endonuclease Haemophilus influenzae strain Rd, the third enzyme isolated

HLA-DPB2

Major Histocompatibility Complex, Class II, DP Beta 2

IA-2β

Islet cell autoantigen-related protein

ICAAR

Islet cell autoantigen-related protein

ICanCME

Interdisciplinary Canadian Collaborative Myalgic Encephalomyelitis Research Network

IGR

Ntergenic region

K₂-EDTA

Dipotassium ethylenediaminetetraacetic acid

Max

Maximum

ME

Myalgic encephalomyelitis

ME

Myalgic encephalomyelitis

MFI-20

Multidimensional Fatigue Inventory-20

Min

Minimum

mM

Millimolar

ng

Nanogramme

nmol

Nanomole

PBMC

Peripheral blood mononuclear cells

pcDNA

Plasmid cloning DNA

PEI

Polyethylenimine

PEM

Post-exertional malaise

pFLAG

Expression vector containing the FLAG-tag

PHB2

Prohibitin 2

PLEKHA4

Pleckstrin Homology Domain Containing A4

PmeI

Restriction enzyme from Pseudomonas mendocina

pmirGLO

PmirGLO dual-luciferase miRNA target expression vector

PTPRN2

Protein tyrosine phosphatase receptor type N polypeptide 2

p-value

Probability value

QC

Quebec

qPCR

Quantitative polymerase chain reaction

Rluc

Renilla luciferase

RNA

Ribonucleic acid

RNase

Ribonuclease

ROUT

Robust regression and outlier removal

RT-qPCR

Reverse transcription-quantitative polymerase chain reaction

SEM

Standard error of the mean

SF-36

36-Item Short Form Health Survey

SNP

Single-nucleotide polymorphism

Tukey's HSD

Tukey’s Honestly Significant Difference

XbaI

Restriction enzyme from Xanthomonas badrii

Δβ

Delta beta

Author contributions

LC, WE, GDT, PL, CG, and AM designed research; LC, WE, DL, YF, IC, CL, GDT, PL and AM performed the analysis; LC, WE, ME, GDT, PL and AM wrote the first draft of the manuscript; CL, EN, BR-A, and MG performed the qPCR and LC, WE, AF, YF, BR-A, and ME helped run experiments. GDT performed the cell-based assays, including the luciferase assays and the miR-153 RT-qPCR assay. All co-authors reviewed and approved the final version of the manuscript.

Funding

This work was supported by grants to AM from the Sibylla-Hesse Foundation and The Open Medicine Foundation Canada, as well as grants to PL from Parkinson Canada, the Laparé and Legault Funds in Biochemistry for Research in RNA Biology and Neurodegenerative Diseases, and the Canadian Institutes of Health Research (CIHR PJT 186036). PL holds the Bristol-Myers-Squibb Chair in Molecular Biology at the Université de Montréal.

Data availability

The datasets generated and analyzed during the current study, including normalized DNA methylation β-values, sample metadata (including estimated saliva cell-type fractions), and raw and processed miR-153-3p and PHB2 measurements, are deposited in the Gene Expression Omnibus (GEO) repository (GEO accession number GSE327255).

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the Declaration of Helsinki, and approved by the Institutional Review Board of CHU Sainte-Justine (protocol #4047). Informed consent was obtained from all participants involved in the study.

Consent for publication

Not applicable.

Competing interests

AM is the Director of the Interdisciplinary Canadian Collaborative Myalgic Encephalomyelitis (ICanCME) Research Network, a national research network funded by The Canadian Institutes of Health Research (grant MNC–196095 and MNC–196095 to A.M.) and a member of the Scientific Advisory Board of the Open Medicine Foundation (USA).