Age-specific alterations of the gut mycobiome in patients with myalgic encephalomyelitis/chronic fatigue syndrome and identification of potential diagnostic biomarkers

Abstract

Background

While bacterial dysbiosis and metabolic disruptions have been widely implicated in the pathogenesis of myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS), the contribution of the gut mycobiome remains largely unresolved, particularly in the context of host age.

Results

Here, high-throughput internal transcribed spacer (ITS) sequencing was applied to fecal samples from 118 individuals (59 ME/CFS patients and 59 healthy matched controls), stratified into three age cohorts: young (18–34 years), middle-aged (35–55 years), and elderly (56–85 years). ME/CFS patients demonstrated significant alterations in gut fungal community composition and diversity compared to controls, with age-specific patterns emerging upon stratified analysis. Reduced fungal amplicon sequence variant (ASV) richness was observed in ME/CFS patients within the young (P < 0.001) and middle-aged (P < 0.01) cohorts, while the elderly ME/CFS group unexpectedly exhibited increased alpha diversity. Principal coordinate analyses based on Bray–Curtis and Jaccard distances robustly differentiated ME/CFS and control groups across all age strata, with the magnitude of separation exceeding that observed in age-unstratified analysis. Age-dependent discriminatory taxa were identified, including Preussia, Endocarpon, Chlorocillium, and Verticillium in the young cohort; Preussia, Romagnesiella, Aspergillus, and Trichothecium in the middle-aged cohort; and Chaetomium, unclassified Ascomycota, and Chlorocillium in the elderly cohort. Classification models based on fungal genera achieved an overall accuracy of 65.2% (AUC = 0.786) without age stratification, but predictive performance was markedly enhanced, yielding accuracies of 87.5% (AUC = 1.00), 100% (AUC = 0.911), and 100% (AUC = 1.00) in the young, middle-aged, and elderly cohorts, respectively. Correlation analyses revealed that taxa positively associated with fatigue severity were consistently depleted in ME/CFS, whereas negatively associated genera were enriched, suggesting that these fungi function primarily as biomarkers of disease burden rather than as causal agents.

Conclusions

These findings underscore the critical importance of age-specific analyses in mycobiome research and highlight gut fungal profiles as promising diagnostic biomarkers for ME/CFS when age is explicitly accounted for.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12866-025-04650-9.

Introduction

Myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) is a debilitating and complex multisystem disorder [1] defined by persistent, unexplained fatigue accompanied by a constellation of symptoms, including cognitive impairment, non-restorative sleep, diffuse musculoskeletal pain (e.g., myalgia, arthralgia, and cephalgia), recurrent pharyngitis, lymphadenopathy, and various systemic manifestations [2]. Despite decades of investigation, the etiology and pathogenesis of ME/CFS remain only partially resolved [3, 4]. Current hypotheses implicate a multifactorial origin, involving post-infectious sequelae [5], immune dysregulation [6], neuroendocrine abnormalities [7], genetic predisposition [8], and dysregulated stress responses [9]. Among these, the gut microbiome has emerged as a critical player in ME/CFS pathophysiology. Mounting evidence indicates that gut microbial dysbiosis may contribute to the disease through disrupted metabolic pathways, immune signaling imbalances, and impaired gut-brain axis communication [10–14]. The gut microbiome, comprising trillions of microorganisms and representing one of the most complex and metabolically active ecosystems in the human body, plays a pivotal role in host immunity, nutrient processing, and neurological function [15]. Dysbiosis within this microbial network has been increasingly implicated in the pathogenesis of various diseases, including gastrointestinal disorders [16], obese [17], cardiovascular pathologies [18], and neuropsychiatric and neurodevelopmental conditions [19]. Several studies have already reported profound disruptions in gut microbial diversity and taxonomic composition between individuals with ME/CFS and healthy controls [2, 10, 12, 20–24], characterized by reduced α-diversity and a marked depletion of commensal Firmicutes [22]. Metabolomic profiling has further revealed an impaired capacity for microbial synthesis of butyrate—a key short-chain fatty acid critical for intestinal barrier function and immune modulation—in ME/CFS patients. Notably, taxa such as Faecalibacterium prausnitzii and Eubacterium rectale, both prominent butyrate-producers, are consistently underrepresented in ME/CFS cohorts [10]. Analogously, individuals with ME/CFS diagnosed within the past four years show decreased abundances of key butyrate-producing taxa, including Roseburia and Faecalibacterium prausnitzii, accompanied by consistently reduced butyrate levels in both fecal and plasma samples [14]. Due to the multifaceted and nonspecific clinical presentation of ME/CFS, diagnosis remains primarily exclusionary, relying on the criteria developed by the Centers for Disease Control and Prevention (CDC) in 1994 and the Institute of Medicine (IOM) in 2015. Given the reproducible gut microbiota alterations observed in ME/CFS, microbial compositional profiling is increasingly recognized as a promising diagnostic biomarker. For instance, a previously reported machine learning model incorporating 16S rRNA sequencing data and inflammatory biomarkers achieved 82.93% cross-validation accuracy in distinguishing ME/CFS patients from healthy controls [22].

Fungi constitute a distinct and functionally significant component of the gut microbiota. Although they comprise a smaller proportion of the microbial community relative to bacteria, gut fungi exert critical regulatory influences on microbial equilibrium and host immune homeostasis [25]. Interkingdom interactions between fungi and bacteria shape the composition, resilience, and functional output of the gut ecosystem, contributing to host physiology and disease susceptibility [26]. Accumulating evidence indicates that fungal communities modulate host metabolic activity, gut barrier integrity, and immunological responses, thereby playing a key role in the pathogenesis of diverse disorders [27]. Inflammatory bowel disease, for instance, is characterized by marked gut mycobiota dysbiosis, which is closely related to disease severity [28]. Higher levels of Candida albicans have been reported in individuals with dyslipidemia, showing positive associations with total plasma cholesterol and low-density lipoprotein cholesterol concentrations [29]. Colorectal cancer (CRC) patients display a distinct intestinal mycobiota profile, marked by elevated abundance of six fungal species, including Aspergillus rambellii, and a concurrent depletion of Aspergillus kawachii compared to both healthy individuals and patients with adenomas. Functional assays have confirmed that Aspergillus rambellii promotes CRC cell proliferation and tumor growth in xenograft models [30]. Gut fungi also influence systemic disease processes through their metabolic byproducts [1]; for example, Aspergillus tubingensis contributes to polycystic ovary syndrome (PCOS) via secretion of its secondary metabolite AT-C1 [1], while Saccharomyces cerevisiae exacerbates colitis and increases gut permeability by enhancing host purine metabolism [31]. Despite this expanding body of research, the role of the gut mycobiome in ME/CFS remains poorly defined. To date, only a single study has investigated fungal profiles in ME/CFS, reporting no significant differences in eukaryotic diversity or taxonomic composition between ME/CFS patients and healthy controls. Furthermore, the limited cohort size of only 34 participants raises concerns regarding the statistical power and generalizability of these findings.

The human gut microbiota undergoes dynamic and progressive shifts across the lifespan, with compositional changes reflecting age-related physiological, immunological, and metabolic states [32]. Distinct age-associated microbial signatures have been observed, with younger individuals exhibiting higher relative abundances of genera such as Bacteroides, Barnesiella, Odoribacter, Collinsella, Bifidobacterium, Lachnoclostridium, Oscillibacter, Ruminiclostridium 5, and Bilophila, while older adults tend to show greater prevalence of Prevotella, Clostridium sensu stricto 1, Holdemanella, Catenibacterium, Phascolarctobacterium, and Prevotella 9 [33, 34]. These compositional patterns are thought to mirror underlying differences in host biology across age groups and are increasingly recognized as modulators of disease susceptibility and progression [35–38]. Notably, age-dependent gut microbiota compositions have been observed in individuals with type 2 diabetes [39], CRC [40], and inflammatory bowel disease [41], underscoring the complex interplay between host age, microbial ecology, and metabolic pathology. Despite these advances, whether the gut microbiome shows analogous age-specific alterations in the context of ME/CFS remains unexplored.

This study conducted a systematic analysis of gut fungal communities in 59 ME/CFS patients and 59 matched healthy controls using high-throughput internal transcribed spacer (ITS) sequencing. Age-stratified comparisons were performed to identify fungal taxa that significantly differentiated ME/CFS cases from controls within each age group. The analysis explored patterns of mycobiota dysregulation associated with ME/CFS and their relationship to fatigue severity, while also evaluating the diagnostic potential of age-specific fungal signatures using random forest machine classification.

Methods

Study population

All procedures involving human participants were approved by the Ethics Committee of the Sichuan Provincial Hospital of Integrated Traditional and Western Medicine (approval number: KY-2021–030). Patients diagnosed with ME/CFS and healthy controls were recruited between February and October 2021 at the Sichuan Provincial Hospital of Integrated Traditional and Western Medicine. Inclusion criteria were: (i) fulfillment of established ME/CFS diagnostic criteria [42]; (ii) age between 18 and 85 years, with no restrictions on sex; and (iii) absence of severe cognitive impairment and sufficient capacity to understand study procedures. Exclusion criteria were: (i) presence of a primary condition known to induce chronic fatigue; (ii) diagnosis of mood or other psychiatric disorders; (iii) concurrent severe hepatic or renal dysfunction; (iv) current pregnancy (positive test), breastfeeding, or active efforts to conceive; (v) use of medications associated with fatigue, including antibiotics such as amoxicillin or roxithromycin, within the past month; (vi) use of anti-fatigue pharmacologic agents within the previous month; (vii) enrollment in another clinical trial within the previous month; (viii) inability or unwillingness to comply with scheduled follow-ups or intervention protocols; (ix) determination by the investigator of unsuitability for participation.

Fecal samples were self-collected by participants under physician supervision using strict sterile techniques. Fresh samples were deposited directly into sterile containers, immediately placed in pre-chilled transport containers, and delivered to the central laboratory within 2 h of collection. Upon arrival, all samples were stored at − 80 °C in ultra-low temperature freezers pending DNA extraction.

Fatigue assessment

Fatigue was evaluated using the Fatigue Self-Assessment Scale, a validated instrument designed to assess the type, severity, and impact of fatigue in ME/CFS across four dimensions: overall fatigue, physical fatigue, mental fatigue, and fatigue impact [43]. Each domain was scored on a standardized 0–100 scale. Higher scores reflected greater fatigue burden, including increased severity of physical and mental fatigue, higher functional impact, resistance to relief by sleep or rest, and stronger situational manifestation.

ITS gene sequencing

Genomic DNA was extracted from fecal samples using a DNeasy PowerSoil Kit (QIAGEN), incorporating mechanical lysis steps to ensure efficient disruption of fungal cells, according to the manufacturer’s protocols. DNA quality was assessed by 1% agarose gel electrophoresis. The fungal ITS rDNA region was amplified using primers ITS1F (5'-CTTGGTCATTTAGAGGAAGTAA-3') and ITS2R (5'-GCTGCGTTCTTCATCGATGC-3'). Polymerase chain reaction (PCR) was conducted in a 25 µL reaction volume containing Phusion® High-Fidelity PCR Master Mix (15 µL; New England Biolabs), 0.2 µM of each primer, and 10 ng of template DNA. Thermocycling conditions consisted of an initial denaturation at 98 °C for 1 min, followed by 30 cycles at 98 °C for 10 s, 50 °C for 30 s, and 72 °C for 30 s, with a final extension at 72 °C for 5 min. PCR products were purified using a commercial PCR purification kit, and sequencing adapters were ligated to the purified amplicons. Library quality was evaluated by Qubit fluorometry and Agilent Bioanalyzer fragment analysis. Sequencing was performed on the Illumina NovaSeq 6000 platform (Novogene, Beijing, China) using paired-end 2 × 250 bp chemistry. To minimize environmental contamination during laboratory processing, we included three negative controls (sterile water) in each DNA extraction batch. These blanks were processed alongside all samples through DNA extraction and ITS PCR amplification. Only batches in which the negative controls showed no detectable PCR products were advanced to sequencing.

Microbiome analysis

Raw sequence reads were processed using QIIME 2 (v.2020.6) [44]. Quality filtering and amplicon sequence variant (ASV) inference were performed using DADA2, which included: (i) trimming of low-quality 3′ ends and removal of primer/adapters; (ii) denoising to distinguish true biological sequences from errors; (iii) merging of paired-end reads; and (iv) filtering out bases with Phred scores < 20. Chimeric sequences were further eliminated using USEARCH 8 [45], yielding high-quality clean reads. ASVs were generated using the DADA2 denoising algorithm and taxonomically classified by BLAST alignment against the UNITE 9.0 fungal database [46]. To minimize bias from sequencing depth, ASV tables were rarefied to the minimum read count per sample prior to performing alpha and beta diversity analyses. Alpha diversity metrics included observed ASV count and Shannon index; beta diversity was assessed using principal coordinates analysis (PCoA) based on Bray–Curtis and Jaccard distances to evaluate inter-group dissimilarity. To address the compositional nature of microbiome relative abundance data, we employed centered log-ratio (CLR) transformation followed by Aitchison distance-based analysis. Fungal composition was also profiled at the phylum and genus levels.

Statistical analysis

Differences in mean values between two groups were compared using the Mann–Whitney U test and among three or more groups were analyzed using Kruskal–Wallis H test. Post-hoc pairwise comparisons were conducted using Dunn’s test with Benjamini–Hochberg correction applied to control the false discovery rate. Statistical significance was defined as a Benjamini–Hochberg–adjusted P-value < 0.05. A random forest classifier (randomForest package; proximity = TRUE, ntree = 1000, mtry = 22) was used to identify key genera and ASVs that best distinguished ME/CFS patients from healthy individuals, including CY vs. HY, CM vs. HM, and CE vs. HE. The top 50 discriminatory fungal genera (or ASVs) were selected to construct a diagnostic model. The dataset was partitioned into training (60%) and validation (40%) subsets using stratified random sampling to preserve class distribution. Model performance was evaluated using receiver operating characteristic (ROC) curve analysis on the validation subset, with the area under the ROC curve (AUC) used as the primary measure of classification accuracy. To ensure robust performance evaluation and mitigate overfitting, we implemented: (1) five rounds of repeated fivefold cross-validation (25 total folds), and (2) a comprehensive comparison of performance across the training, cross-validation, and independent test sets. Pearson correlation analysis was performed to assess associations between the relative abundance of the top 50 fungal genera (ASVs) and individual fatigue factors. Data visualization was carried out using the “boxplot”, “barplot”, “ggplot2”, and “plot” functions in the R base package (v4.5) [47].

Results

Study population characteristics

A total of 118 participants were enrolled, comprising 59 individuals diagnosed with ME/CFS (CFS group) and 59 healthy controls. All ME/CFS patients completed the Fatigue Self-Assessment Scale (Table 1). To evaluate age-dependent effects, participants were further stratified into six subgroups based on age [48, 49]: young ME/CFS group (CY group, 18–34 years old; n = 20), young healthy control group (HY group, 18–34 years old; n = 20), middle-aged ME/CFS group (CM group, 35–55 years old; n = 19), middle-aged healthy control group (HM group, 35–55 years old; n = 20), elderly ME/CFS group (CE group, 56–85 years old; n = 20), and elderly healthy control group (HE group, 56–85 years old; n = 19) (Table 2).

Table 1.

Characteristics of the study population

		CFS group (n = 59)	Health group (n = 59)
Gender	Female	43	41
	Male	16	18
Age	Mean ± SD	46.1 ± 18.7	46.1 ± 19.1
	Median (range)	46(19 ~ 74)	43(21 ~ 85)
The fatigue self-assessment scale	Overall fatigue	49 ± 17.3	NA
	Physical fatigue	51 ± 21.8
	Mental fatigue	53 ± 20.4
	Consequences of fatigue	46 ± 18.7

Table 2.

Study population characteristics by age group

		CY (n = 20)	CM (n = 19)	CE (n = 20)	HY (n = 20)	HM (n = 20)	HE (n = 19)
Gender	Female	13	15	15	17	13	11
	Male	7	4	5	3	7	8
Age	Mean ± SD	21.80 ± 0.31	44.89 ± 1.35	66.45 ± 1.13	25.7 ± 3.4	43.8 ± 6.3	70 ± 7.1
	Median (range)	22(19 ~ 24)	45(36 ~ 55)	67(57 ~ 74)	25.5(21 ~ 31)	44(35 ~ 53)	72(60 ~ 85)
The fatigue self-assessment scale	Overall fatigue	53.60 ± 4.46	44.84 ± 3.33	49.85 ± 3.74	NA	NA	NA
	Physical fatigue	52.65 ± 4.71	51.21 ± 5.50	49.45 ± 4.79
	Mental fatigue	56.80 ± 5.01	48.05 ± 4.69	53.60 ± 4.11
	Consequences of fatigue	52.15 ± 4.77	38.89 ± 3.12	47.80 ± 4.11

Sequencing depth adequacy assessed by rarefaction curves

To evaluate whether sequencing depth was sufficient for robust diversity analysis, rarefaction curves were constructed using a standardized subsampling depth of 8 000 reads per sample. With increasing sequencing depth, both the Shannon diversity index (Fig. S1A) and observed ASV richness (Fig. S1B) approached saturation across all groups, indicating that sequencing depth was sufficient for downstream analyses and captured the majority of fungal diversity within samples.

Altered gut fungal diversity and composition in ME/CFS

Alpha diversity analysis revealed a significant reduction in observed ASVs among ME/CFS patients relative to healthy controls (P < 0.05), while the Shannon index showed no significant difference (Fig. 1A–B). PCoA analysis revealed modest clustering separation between ME/CFS and control groups, with significant but weak overall differences in community structure (Bray–Curtis: R² = 3.39%, P = 0.001; Jaccard: R² = 2.49%, P = 0.001) (Fig. 1C–D). Despite strong statistical support, the small effect sizes suggest that group status accounts for only a small proportion of overall community variance. Homogeneity of variance tests demonstrated significantly greater intra-group variability in the ME/CFS cohort for both Bray–Curtis (adjusted P < 0.001, Fig. 1E) and Jaccard (adjusted P < 0.001, Fig. 1F) distances. Notably, intra-patient fungal community variation in ME/CFS was comparable in magnitude to the overall difference between patient and control groups, suggesting pronounced heterogeneity within the ME/CFS population that may obscure between-group separation at the community level. To address compositional biases inherent in relative abundance data, we employed centered log-ratio (CLR) transformation followed by Aitchison distance-based analysis. This compositionally aware approach corroborated our initial findings: microbial communities showed significant but modest separation between groups (PERMANOVA: R² = 3.14%, P = 0.001, Fig. S2A) and replicated the pattern of increased within-group variation in ME/CFS participants (all pairwise comparisons: adjusted P < 0.001) (Fig. S2B). The convergence of results across both conventional and compositionally robust methods strengthens the conclusion that ME/CFS is associated with heightened microbiome variability.

Fig. 1.

Gut fungal α- and β-diversity in ME/CFS patients and healthy controls. A Shannon index and (B) observed ASV count between healthy controls and ME/CFS patients. Principal coordinates analysis (PCoA) based on (C) Bray–Curtis and (D) Jaccard distances. Community dissimilarity assessed by intra- versus inter-group comparisons using (E) Bray–Curtis and (F) Jaccard distances. Asterisks indicate statistical significance (*P < 0.05, **P < 0.01, ***P < 0.001)

At the phylum level, gut fungal communities in both groups were dominated by Ascomycota and Basidiomycota, although ME/CFS patients showed a relative increase in Basidiomycota abundance (Fig. 2A–B). At the genus level, dominant taxa in healthy individuals included Candida (13.17% ± 22.99%), Simplicillium (7.11% ± 5.02%), Pichia (4.71% ± 19.73%), and Sarocladium (3.91% ± 2.67%). In contrast, ME/CFS patients exhibited a shift toward an increased relative abundance of Aspergillus (13.84% ± 20.13%), Candida (9.14% ± 23.78%), Simplicillium (4.21% ± 5.21%), and Penicillium (3.29% ± 7.12%). Notably, the relative abundance of Aspergillus increased from 2.62% ± 5.52 in healthy controls to 13.84% ± 20.13% in ME/CFS patients, while Candida levels declined from 13.17% ± 22.99% to 9.14% ± 23.78% (Fig. 2C–D). Collectively, these findings demonstrate significant alterations in gut fungal diversity and taxonomic composition in ME/CFS, highlighting increased community variability and distinct taxonomic shifts relative to healthy controls.

Fig. 2.

Taxonomic composition of gut fungi in ME/CFS patients and healthy controls. Relative abundance of taxa at the phylum level in healthy individuals (A) and ME/CFS patients (B) and at the genus-level taxa in healthy individuals (C) and ME/CFS patients (D)

Age-stratified analysis reveals stronger fungal community differences between ME/CFS patients and healthy individuals

Stratification by age uncovered marked heterogeneity in gut fungal alpha diversity between ME/CFS patients and healthy controls. Within both the healthy and ME/CFS groups, Shannon index values did not differ significantly across young, middle-aged, and elderly cohorts (Fig S3A, C). However, healthy elderly individuals exhibited significantly lower observed ASV richness than healthy young individuals, whereas elderly ME/CFS patients showed significantly higher richness than young ME/CFS patients (Fig S3B, D). When comparing ME/CFS patients to controls within specific age groups, no significant differences in the Shannon index were observed between the ME/CFS and control groups within the young and middle-aged cohorts (P > 0.05). However, in the elderly cohort, ME/CFS patients exhibited significantly higher Shannon diversity compared to age-matched controls (P < 0.05), indicating elevated gut fungal diversity in older ME/CFS patients (Fig. 3A). In contrast, the number of observed fungal ASVs was significantly reduced in young (P < 0.001) and middle-aged (P < 0.01) ME/CFS patients relative to their healthy counterparts, consistent with diminished fungal richness in these age groups (Fig. 3B). Interestingly, this trend was reversed in the elderly cohort, where ME/CFS patients showed significantly greater ASV richness than age-matched healthy controls (P < 0.05), suggesting the presence of age-specific mycobiome alterations in ME/CFS.

Fig. 3.

Age-stratified comparison of gut fungal α- and β-diversity between ME/CFS patients and healthy controls within three age cohorts. A Shannon index and (B) observed ASV count in young, middle-aged, and elderly cohorts, comparing healthy controls (blue) and ME/CFS patients (orange). PCoA based on Bray–Curtis distances for young (C), middle-aged (D), and elderly (E) cohorts. Intra‐ vs. inter‐group community dissimilarity (pairwise Bray–Curtis distances) in young (F), middle-aged (G), and elderly (H) cohorts, comparing within-healthy, within-ME/CFS, and between healthy-ME/CFS. CY: young ME/CFS cohort; CM: middle-aged ME/CFS cohort; CE: elderly ME/CFS cohort; HY: young healthy controls; HM: middle-aged healthy controls; HE: elderly healthy controls. Asterisks indicate statistical significance (*P < 0.05, **P < 0.01, ***P < 0.001)

Age-stratified PCoA demonstrates enhanced disease-associated divergence

PCoA based on Bray–Curtis and Jaccard distances revealed significantly enhanced separation between ME/CFS patients and controls within each age cohort: (i) Young cohort: CY vs. HY (R² = 13.24%, P = 0.001, Fig. 3C; R² = 6.67%, P = 0.002, Fig. S3E); (ii) Middle-aged cohort: CM vs. HM (R² = 14.57%, P = 0.001, Fig. 3D; R² = 10.74%, P = 0.001, Fig. S3F); (iii) Elderly cohort: CE vs. HE (R² = 8.97%, P = 0.002, Fig. 3E; R² = 9.28%, P = 0.001, Fig. S3G). These values far exceeded those observed in the unstratified analysis (Bray–Curtis: R² = 3.39%, P = 0.001; Jaccard: R² = 2.49%, P = 0.001), reinforcing age as a critical factor modulating variation in the ME/CFS-associated mycobiome.

Beta dispersion analysis showed that ME/CFS patients exhibited significantly greater within-group distances than healthy controls in both the young and middle-aged cohorts (adjusted P < 0.001 for both Bray–Curtis and Jaccard distances), with within-patient variability approaching the magnitude of the inter-group differences between ME/CFS patients and healthy controls (Fig. 3F–G; Fig. S3H–I). In contrast, in the elderly cohort, within-group dispersion among ME/CFS patients was significantly lower than that of healthy controls (Bray–Curtis: adjusted P < 0.001; Jaccard: adjusted P < 0.001) and also lower than the inter-group variation (adjusted P < 0.001 for both metrics; Fig. 3H; Fig. S3J). These patterns were further validated using compositionally aware analyses. CLR-based PCA visualization revealed consistent age-stratified clustering (Young cohort: R² = 12.24%, P = 0.001; Middle-aged cohort: R² = 12.81%, P = 0.001; Elderly cohort: R² = 8.1%, P = 0.001; Fig. S3A–C), while Aitchison distance metrics quantitatively confirmed elevated within-ME/CFS dispersion in the young and middle-aged groups (adjusted P < 0.001) and reduced dispersion in the elderly (adjusted P < 0.001; Fig. S3D–F). The concordance across both ordination and distance-based analyses underscores these age-dependent heterogeneity patterns as robust biological features of ME/CFS.

Age-stratified random forest analysis identifies distinct gut fungal signatures in ME/CFS

Random forest modeling identified the top 50 fungal genera and ASVs most effective in distinguishing ME/CFS patients from healthy individuals. Distinct sets of discriminatory taxa were observed across comparisons: all ME/CFS patients vs. controls (Fig. 4A; Fig. S4A), CY vs. HY (Fig. 4C; Fig. S4C), CM vs. HM (Fig. 4E; Fig. S4E), and CE vs. HE (Fig. 4G; Fig. S4G). For genus-level classifiers: Debaryomyces, Aspergillus, Wallemia, Preussia, and Penicillium contributed most to classification in the overall comparison. In age-stratified subsets, Preussia, Endocarpon, Chlorocillium, and Verticillium were dominant in CY vs. HY; Preussia, Romagnesiella, Aspergillus, and Trichothecium were most important in CM vs. HM; Chaetomium, g_unclassified (phylum Ascomycota), and Chlorocillium were dominant in CE vs. HE. At the ASV level, key overall discriminators included Debaryomyces (ASV_822), Aspergillus (ASV_3839), Penicillium (ASV_696), and Fusarium (ASV_1387). In CY vs. HY, important ASVs included Preussia (ASV_4948). For CM vs. HM, discriminatory ASVs included Preussia (ASV_8982), Preussia (ASV_4948), and Sarocladium (ASV_3425). In CE vs. HE, key ASVs included Talaromyces (ASV_946), Cyphellophora (ASV_8261), Chaetomium (ASV_6959), and Preussia (ASV_8982).

Fig. 4.

Discriminatory fungal genera identified by random forest and their abundance profiles in ME/CFS patients and healthy controls. Variable importance scores (mean decrease in accuracy) for the top 50 genera distinguishing ME/CFS from healthy individuals in all samples (A), young cohort (C), middle-aged cohort (E), and elderly cohort (G). Corresponding heatmaps of normalized genus-level abundances in all samples (B), young cohort (D), middle-aged cohort (F), and elderly cohort (H), grouped by health status. CY: young ME/CFS cohort; CM: middle-aged ME/CFS cohort; CE: elderly ME/CFS cohort; HY: young healthy controls; HM: middle-aged healthy controls; HE: elderly healthy controls

Notably, the top 50 genera and ASVs identified from the unstratified comparison failed to clearly separate ME/CFS patients and healthy controls based on relative abundance alone (Fig. 4B; Fig. S4B). In contrast, age-stratified models yielded substantially improved discrimination. Relative abundances of the top 50 genera and ASVs exhibited significant discriminatory power between patients and controls within each age cohort: CY vs. HY (Fig. 4D; Fig. S4D), CM vs. HM (Fig. 4F; Fig. S4F), and CE vs. HE (Fig. 4H; Fig. S4H). For example, Symmetrospora and Debaryomyces were elevated in CY compared to HY, while other genera were depleted in CY, with the exception of four outliers. Penicillium, Aspergillus, and Verticillium were enriched in CM, whereas different taxa were more abundant in HM, excluding four anomalous CM samples. In the elderly group, Chaetomium, Staphylotrichum, g_unclassified (order Glomerales), and Saccharomyces were enriched in a subset of HE individuals (n = 7), contrasting with broader enrichment profiles observed in CE, except for seven discordant HE samples.

Age stratification substantially improves the discriminatory power of gut fungal biomarkers for ME/CFS

Machine learning classification based on the top 50 fungal genera or ASVs identified by random forest analysis enabled effective differentiation between ME/CFS patients and healthy controls (Fig. 5A; Fig. S5A). In the unstratified analysis using genus-level data, ME/CFS and control samples were classified with accuracies of 65.22% (15/23) and 78.26% (18/23), respectively, yielding an AUC of 0.786 (Fig. 5B), indicative of moderate diagnostic performance. ASV-based classification yielded an accuracy of 60.87% (14/23) for ME/CFS patients and 91.30% (21/23) for controls, with a corresponding AUC of 0.807 (Fig. S5B), also reflecting moderate diagnostic performance.

Fig. 5.

ROC curve analysis of classification performance of the top 50 discriminatory fungal genera between ME/CFS patients and healthy controls. ROC curves based on the top 50 fungi genera identified in all samples (A), young cohort (B), middle-aged cohort (C), and elderly cohort (D). Each panel presents the AUC for the corresponding random forest classifier, reflecting its diagnostic accuracy in distinguishing healthy individuals from ME/CFS patients within each age group

To determine whether age stratification enhances the discriminatory power of gut fungal biomarkers for ME/CFS, participants were grouped into age-based cohorts, and the machine learning models were re-evaluated within each subgroup. Classification accuracy improved markedly upon age stratification. Using genus-level predictors, age-stratified classification yielded high accuracy across all cohorts: young cohort (CY vs HY) demonstrated 87.5% accuracy for predicting ME/CFS and 100% accuracy for predicting healthy controls (Fig. 5C); middle-aged cohort (CM vs HM) showed 100% accuracy for predicting ME/CFS and 87.5% accuracy for predicting healthy controls (Fig. 5E); elderly cohort (CE vs HE) showed 100% accuracy for predicting both ME/CFS and healthy controls (Fig. 5G). Using ASV-level predictors, similar improvements were observed: young cohort (CY vs HY) showed 87.5% accuracy for predicting ME/CFS and 87.5% accuracy for predicting healthy controls (Fig. S5C); middle-aged cohort (CM vs HM) demonstrated 85.71% accuracy for predicting ME/CFS and 87.5% accuracy for predicting healthy controls (Fig. S5E); elderly cohort (CE vs HE) showed 100% accuracy for predicting both ME/CFS and healthy controls (Fig. S5G). ROC analysis further supported enhanced performance in age-stratified groups: young group, AUC = 1 (genera, Fig. 5D) and AUC = 0.961 (ASVs, Fig. S5D); middle-aged group, AUC = 0.911 (genera, Fig. 5F) and AUC = 0.929 (ASVs, Fig. S5F); elderly group, AUC = 1 (genera, Fig. 5H) and AUC = 1 (ASVs, Fig. S5H). As shown in Table S1 and Table S2, the training-set AUC, cross-validation mean AUC, and independent test-set AUC were highly consistent, with no substantial performance discrepancies. These results indicate that the models for the smaller subgroups did not exhibit clear signs of overfitting.

Specific fungal biomarkers are significantly associated with fatigue severity in ME/CFS and vary across age groups

To explore the potential contribution of gut fungal biomarkers in fatigue in ME/CFS, correlations were assessed between fatigue scores and the top 50 fungal genera previously identified as discriminatory by random forest analysis. In the overall cohort (Fig. 6A), Wickerhamomyces exhibited significant positive correlations with both the consequences of fatigue score (P < 0.01) and overall fatigue score (P < 0.05), while Paraphoma showed a significant negative correlation with the consequences of fatigue score (P < 0.05). In age-stratified analyses, distinct correlation patterns emerged. In the young cohort (Fig. 6B), Fungi_gen_incertae_sedis was positively correlated with the mental fatigue score (P < 0.05), consequences of fatigue score (P < 0.01), Physical fatigue score (P < 0.01), and overall fatigue score (P < 0.01). In the middle-aged cohort (Fig. 6C), Trichothecium, Solicozyma, Setophoma, Orbiliales_gen_incertae_sedis, Ochronectria, Mortierellales_gen_incertae_sedis, Humicola, Curvularia, and Ceratobasidium were significantly positively correlated with the physical fatigue score (P < 0.05), whereas Eukaryota_gen_incertae_sedis was negatively correlated with physical fatigue severity (P < 0.01). In the elderly cohort (Fig. 6D), Psathyrellaceae_gen_incertae_sedis was positively correlated with consequences of fatigue score (P < 0.05), physical fatigue score (P < 0.05), and overall fatigue score (P < 0.05), while Poaceae_gen_incertae_sedis was positively correlated with mental fatigue score (P < 0.05), consequences of fatigue score (P < 0.05), and overall fatigue score (P < 0.05). In contrast, several taxa, such as, Preussia, Olpidiaster, Mortierellales_gen_incertae_sedis, Mortierellaceae_gen_incertae_sedis, Knufia, Erythrobasidium, Curvularia, and Cladophialophora were negatively correlated with fatigue severity.

Fig. 6.

Correlation heatmaps between fatigue dimensions and the top 50 discriminatory genera in ME/CFS patients. A Spearman correlation heatmap showing associations between the top 50 discriminatory genera and four fatigue dimensions (physical fatigue, mental fatigue, consequences of fatigue, and overall fatigue) in the full ME/CFS cohort. Age-specific heatmaps showing correlations between the top 50 discriminatory genera and fatigue scores in the young (B), middle-aged (C), and elderly (D) ME/CFS cohorts. Color intensity represents correlation strength and direction. Asterisks indicate statistical significance (*P < 0.05, **P < 0.01, ***P < 0.001)

A parallel analysis was conducted using the top 50 fungal ASVs identified as discriminatory between ME/CFS patients and healthy controls. Notably, ASV-based associations differed from those observed at the genus level. In both the full dataset (Fig. S6A) and the young cohort (Fig. S6B), none of the top 50 fungal ASVs demonstrated significant association with fatigue scores. However, in the middle-aged cohort (Fig. S6C), several ASVs exhibited positive associations with physical fatigue, including g_unclassified_k_fungi (ASV_8950, ASV_3191), Cladophialophora (ASV_6844), Pleosporales_gen_incertae_sedis (ASV_6834), Ceratobasidium (ASV_4799),and Fusarium (ASV_3086) (P < 0.05). In the elderly cohort (Fig. S6D), most ASVs exhibited negative correlations with fatigue scores. The sole exception was g_unclassified_k_fungi (ASV_1083), which was positively correlated with both consequences of fatigue (P < 0.05) and overall fatigue scores (P < 0.05). ASVs negatively correlated with fatigue scores included Penicillium (ASV_9513), Bionectriaceae (ASV_7803), Cladophialophora (ASV_6844, ASV_3213), Preussia (ASV_6277), Paraphoma (ASV_6184), Fusarium (ASV_1477), Olpidiaster (ASV_1399), Sarocladium (ASV_1442), and Papiliotrema (ASV_1114).

Fatigue-associated fungal biomarkers likely reflect correlative rather than causal relationships

To assess whether fungal taxa associated with fatigue severity may play a causal role in ME/CFS, relative abundances were compared between ME/CFS patients and healthy controls. At the genus level, most fatigue-associated fungi, excluding Wickerhamomyces, were more abundant in healthy individuals in the young and middle-aged cohorts. In contrast, these same genera showed elevated abundance in the elderly ME/CFS cohort, with the exception of Psathyrellaceae_gen_incertae_sedis (Fig. 7A–C). ASV-based analysis yielded consistent trends (Fig. S7A–C). Critically, two paradoxical patterns emerged: genera positively correlated with fatigue scores demonstrated lower abundance in ME/CFS patients, whereas those negatively correlated were more abundant.

Fig. 7.

Relative abundance of fatigue-associated fungal genera in ME/CFS patients and healthy controls across age cohorts. Boxplots showing genus-level relative abundance of fatigue-associated fungi in the gut of young (A), middle-aged (B), and elderly (C) cohorts, comparing ME/CFS patients and healthy controls

Discussion

Extensive research has established that individuals with ME/CFS exhibit notable dysbiosis in the gut bacterial community [2, 12, 21, 22, 24], including reduced microbial diversity and depletion of butyrate-producing taxa such as Faecalibacterium prausnitzii and Eubacterium rectale. In contrast, the gut mycobiome in ME/CFS remains largely unexplored. To date, only one published study has examined intestinal fungal communities in ME/CFS, reporting no significant differences between patients and healthy controls [50]. Diverging from these findings, the present study revealed pronounced alterations in gut fungal composition in ME/CFS. Specifically, alpha diversity based on observed ASVs was significantly reduced in ME/CFS patients, and PCoA demonstrated a modest yet statistically significant separation between groups. The lack of similar findings in the earlier report may be attributable to its limited sequencing depth (approximately 100 reads per sample) and small sample size (17 ME/CFS vs. 17 healthy controls), which may have restricted detection sensitivity. Previous investigations have consistently identified Ascomycota and Basidiomycota as the predominant fungal phyla in the human gut, including both ME/CFS patients and healthy individuals [50]. The present findings corroborate earlier reports indicating an elevated relative abundance of Basidiomycota in ME/CFS. Within this phylum, a striking expansion of Aspergillus was observed among ME/CFS patients. Members of this genus are widespread in the natural environment, and certain species, such as Aspergillus fumigatus, are opportunistic pathogens capable of causing invasive aspergillosis [51]. Moreover, Aspergillus tubingensis has been shown to aggravate PCOS via active secondary metabolites [1], suggesting a broader immunometabolic impact. The substantial enrichment of Aspergillus in ME/CFS may reflect a multifactorial process involving environmental exposure, host immune modulation, and disruption of microbial homeostasis, potentially contributing to disease pathogenesis. Random forest analysis revealed a significant difference in the abundance of Aspergillus between individuals with ME/CFS and healthy controls. Furthermore, age-stratified machine learning models confirmed that Aspergillus serves as a robust discriminative feature, particularly distinguishing middle-aged ME/CFS patients from middle-aged healthy individuals, underscoring its potential as a key biomarker in this subpopulation. Of course, we must consider that the expansion of Aspergillus abundance observed in the ME/CFS patient group may be related to inhaled spores or contamination. This conclusion should be further validated using qPCR in future studies. Conversely, Candida—a major component of the healthy gut mycobiota—seems to deplete in ME/CFS patients. Under normal physiological conditions, Candida species help maintain balance within the gut microbial ecosystem. Elevated Candida abundance has been consistently reported in inflammatory bowel disease [52–54], highlighting its association with mucosal inflammation and barrier disruption. The observed trend suggesting depletion of Candida in ME/CFS requires careful interpretation due to high inter-individual variability (SD ± 23%), which undermines the reliability of this difference. Random forest analysis further indicated that the abundance of Candida did not significantly differ between ME/CFS patients and healthy controls. Similarly, stratified machine-learning models confirmed that Candida does not serve as a discriminatory biomarker between patients and healthy individuals across any age subgroup. These results highlight substantial fungal heterogeneity among individuals and emphasize the necessity of age-stratified approaches for identifying robust mycobiome-based biomarkers that can reliably distinguish ME/CFS from healthy states.

The gut mycobiome exhibits marked inter-individual variability, yet its composition and developmental trajectory are strongly shaped by host age and behavioral factors [32, 55, 56]. Gut fungal communities evolve with age, displaying age-specific taxonomic profiles and ecological structures [57]. In this context, stratified analysis revealed divergent patterns of fungal diversity in ME/CFS patients across different age groups. Both young and middle-aged ME/CFS cohorts demonstrated significantly reduced gut fungal diversity relative to their age-matched controls. In contrast, elderly ME/CFS patients exhibited significantly greater fungal diversity than healthy individuals of similar age. Normally, comparing to non-elderly populations, the gut microbiota of elderly populations is characterized by reduced diversity, shifts in dominant species, decreased numbers of probiotics, and increased levels of facultative anaerobes [58]. However, ME/CFS patients of the elderly group exhibited increased gut fungal diversity, suggesting that increased microbial diversity may not always signify a healthy state but could also indicate dysbiotic expansion. These findings emphasize the importance of age as a factor in studying ME/CFS and confirm that intestinal fungal community diversity and composition exhibit significant differences across age groups Notably, a recent study in elderly Chinese individuals also reported higher bacterial diversity in patients with cardiovascular disease compared to age-matched healthy controls [59]. Collectively, these findings suggest that ME/CFS is associated with distinct, age-specific patterns of gut fungal dysregulation, and that the trajectory of mycobiome alterations may diverge between elderly and non-elderly patients.

PCoA revealed distinct clustering of ME/CFS and control samples within each age group, with more pronounced separation than in the unstratified analysis, indicating age-specific divergence in gut fungal communities. Age stratification also markedly enhanced the performance of fungal biomarkers in distinguishing ME/CFS patients and healthy controls. Classification accuracy and AUC improved substantially across age-stratified models: pooled data yielded an AUC of 0.786 with 65.22% accuracy, whereas the young group achieved an AUC of 1.00 (87.5% accuracy), the middle-aged group an AUC of 0.911 (100% accuracy), and the elderly group an AUC of 1.00 (100% accuracy). These findings confirm the diagnostic potential of fungal biomarkers and highlight age as a key determinant of classification performance. Consistent with studies in other diseases, predictive models trained within age-homogeneous cohorts often achieve superior discriminatory power compared to those developed from mixed-age populations [60]. This phenomenon reflects age-related heterogeneity in disease pathophysiology and microbiome-host interactions [61, 62]. Thus, age-stratified random forest models based on gut fungal signatures may serve as a valuable tool for accurately identifying ME/CFS patients across the lifespan. However, the random forest classifier achieved exceptionally high performance in the elderly subgroup (100% accuracy, AUC = 1.0). Given the limited sample size (approximately 20 individuals), this raises legitimate concerns regarding potential overfitting. The use of k-fold cross-validation, together with the comparative evaluation of training-set AUC, cross-validation mean AUC, and independent test-set AUC, provides preliminary evidence that the perfect AUC values observed in the small subgroups are unlikely to be driven by overfitting. Nevertheless, a highly significant increase in fungal diversity was observed in elderly ME/CFS patients compared to healthy elderly controls. This shift, which contrasts with trends seen in younger groups, suggests a profound and unique dysbiosis specific to elderly ME/CFS patients. Although the model was trained on a small cohort, this distinctive ecological alteration may provide a plausible biological explanation for the model’s strong discriminatory performance. That said, these preliminary yet highly promising findings require further validation in a larger, independent cohort before any definitive conclusions can be drawn regarding the model’s generalizability.

Distinct fungal genera exhibited age-specific alterations in ME/CFS, suggesting complex and potentially non-causal associations with disease phenotype. In the young cohort, Symmetrospora and Debaryomyces were markedly elevated in ME/CFS patients relative to age-matched healthy controls. While Symmetrospora has been reported in the gut of non-human mammals such as horses [63] and macaques [64], its relevance in the human gut remains poorly characterized. Debaryomyces, particularly Debaryomyces hansenii, is a well-established commensal in the human gut and has been shown to accumulate in inflamed intestinal tissue in Crohn’s disease [65], implicating this genus in context-dependent pro-inflammatory activity. Whether Debaryomyces contributes to intestinal inflammation in ME/CFS requires further experimental validation. Penicillium was significantly enriched in the middle-aged ME/CFS cohort. Although Penicillium chrysogenum is the strain used to produce penicillin for the treatment of bacterial infections [66], numerous Penicillium species can trigger allergic reactions or produce mycotoxins under specific environmental conditions, posing potential health risks independent of their therapeutic applications [66]. Enrichment of Penicillium has also been observed in the gut microbiota of individuals with metabolic disturbances, such as obesity [67], suggesting potential relevance to suboptimal health states. In contrast, Chaetomium was more abundant in elderly healthy controls than in their ME/CFS counterparts. Chaetomium is a genus commonly enriched in the healthy human gut [68], and its representative species, Chaetomium globosum, produces antimicrobial polysaccharides that modulate gut microbial composition and enhance short-chain fatty acid synthesis in murine models [69]. The presence of Chaetomium in the human gut may play an important role in maintaining the gut microbiome, its depletion in elderly ME/CFS patients may reflect impaired microbial regulation or host-microbe interactions. These findings appear to support that significant alterations in fungal genus abundance are associated with ME/CFS. However, by comparing the relative abundance of fatigue-related fungal biomarkers between ME/CFS patients and healthy controls, we observed some interesting results. Several fungal taxa identified as potential biomarkers in this study, including Endocarpon, Verticillium, Preussia, and Chaetomium, are predominantly known as environmental fungi commonly associated with soil, plants, and decaying organic matter. Their repeated detection in gut mycobiome studies, including ours, raises the possibility that their presence may not necessarily reflect true gut colonization. Instead, it could originate from dietary intake, environmental contamination, or even sequencing artifacts related to database bias or amplification efficiency. While these taxa showed differential abundance between groups, their biological relevance as gut commensals or pathogens remains uncertain. Therefore, we caution against overinterpreting their role as reliable biomarkers without further validation through culture-dependent methods or targeted replication in independently sampled cohorts. Future studies should aim to distinguish true gut residents from transient or technically introduced fungi to improve the specificity of mycobiome-based biomarkers.

Although fungal taxa demonstrated significant variation in abundance across age-stratified cohorts, an unexpected inverse pattern emerged when aligned with fatigue severity. Genera that showed strong positive correlations with fatigue scores were frequently depleted in ME/CFS patients, while those negatively correlated were often enriched. This pattern contradicts the hypothesis that fatigue-associated fungi would be elevated in individuals with greater symptom burden. These contradictory findings suggest that rather than acting as direct effectors of disease, these taxa more likely represent secondary, correlative features—mirroring shifts driven by upstream physiological perturbations such as immune dysregulation, microbial overgrowth, compromised intestinal barrier integrity, altered metabolic signaling, or environmental influences including medication and diet. Within the ME/CFS population, the abundance of these fungi may scale with the intensity of these underlying drivers, functioning as microbial proxies or accomplices that reflect, rather than cause, fatigue severity.

Conclusion

This study revealed significant differences in the gut fungal community composition between ME/CFS patients and healthy controls, with distinct age-dependent patterns emerging upon stratified analysis. Integration of random forest and predictive modeling demonstrated that age stratification enhanced the identification of gut fungal biomarkers with stronger discriminatory capacity, substantially improving classification accuracy. These findings establish age as a critical factor influencing gut mycobiota profiles in ME/CFS and highlight its importance in biomarker discovery and disease stratification.

Authors’ contributions

W.G. designed this study. Y.G., R.N., and W.G. wrote this manuscript. W.G., Y.X. and S.Z. analyzed the data. Y.G., R.N., Y.X., W.Z. J.X., and M.D. performed the laboratory experiments. Y.X., Z.Z., and R.N. collected the samples. All authors revised the manuscript and approved the submission of this article.

Funding

This work was supported by the Sichuan Science and Technology Program (No. 2024NSFSC1179), the National Natural Science Foundation of China (No. 82472328), CMC Excellent-talent Program (No.2024yxGzn05), the Open Project of Collaborative Innovation Center of Sichuan for Elderly Care and Health (YLKYZD2203), the School-level fund of Chengdu medical college (CYZYB23-06), and the Scientific Research Development Fund Project of Chengdu Medical College (CYYZZ25-15).

Data availability

The raw data of metagenome sequences in this study have been deposited into Sequence Read Archive (SRA) in NCBI with the accession BioProject number PRJNA1294412.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethics Committee of the Sichuan Provincial Hospital of Integrated Traditional and Western Medicine (approval number: KY-2021–030). All procedures involving human participants were performed in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.