Setting and Participants

The methods for this study were conducted in accordance with the approved guidelines for human experimental research. Ethics approval was obtained from Victoria University Human Research Ethics Committee in May 2013 (HRE13-109). As a retrospective sample, there was no direct contact with participants. Patients obtaining faecal assessment through Bioscreen (Aust.) signed informed consent to allow their microbial results and accompanying self-reported symptoms to be used for research purposes.

The dataset included 274 patients who had signed consent to participate in research during faecal microbial assessment (FMA) through the NATA (National Association of Testing Authorities) accredited laboratory, Bioscreen. All patients received a diagnosis of CFS in accordance with the Canadian Definition Criteria33 or Fukuda criteria34 during treatment from one of the co-authors (DPL) between January 2011 and April 2013. Only the earliest available data were included when multiple FMA results were available for the same patient.

Sex representation within this study was equivalent to prevalence ratios amongst clinical ME/CFS populations10 with 86 male (31.4%) and 188 female (68.6%) participants. The age range of 6 to 81 years (M = 39.25, SD = 14.81) is consistently representative of the occurrence of ME/CFS across developmental stages10. Age was not provided for two participants. Additional demographic information or information about comorbid diagnoses were not available. Therefore, no additional exclusion criteria were applied.

Data sources/measurement

Faecal Microbial Analysis. Sample collection: Prior to faecal sample collection, patients were instructed to cease antibiotic and/or probiotic treatment for four and two weeks, respectively. Patients collected a sample of their first morning bowel motion in a faecal container (anaerobic pouch system) with a perforated lid to aid anaerobiasis (achieved by activating Anaero Gen Compact (Oxoid, Thermo Fisher Scientific, Australia)). Samples were immediately transported to the laboratory in cold conditions (<12 °C) for analysis within 48 hours after collection. Laboratory protocol rejects samples subjected to inaccurate collection, transportation, anaerobiasis or refrigeration procedures. Internal quality assurance investigations validated the anaerobic transport and culture methods (see35).

Microbial identification and quantification: After removal from the anaerobic pouch system, all faecal samples were processed within 10–15 minutes. Between 0.5–1.0 g was transferred to 10 mL of 1% glucose-saline buffer36. Dilution factor was determined by the difference in the weight of the glucose-saline buffer with and without the sample. One hundred and one thousand fold dilutions (beginning from 10−1 to 10−7) of homogenised faecal samples were prepared37. Dilutions (10 and/or 1 μL amounts) were transferred onto dried Columbia horse blood agar (Oxoid), chromogenic medium (Oxoid), colistin and nalidixic acid blood selective agar (Oxoid), and chloramphenicol-gentamicin selective Sabouraud agar for aerobic incubation. Anaerobic incubation (4 day duration) in anaerobic jars (Oxoid) utilised pre-reduced Columbia horse blood haemin agar and Raka Ray medium. Aerobic media were incubated at 35 °C for 48 hours. A stereomicroscope was used to examine both aerobic and anaerobic culture plates for a minimum of 20 min/plate before bacterial identification. Each colony from each medium was microscopically examined and the colony/viable count were quantified for each plate. To assess purity prior to identification, similar morphotypes were sub-cultured onto horse blood agar.

Identification using MALDI-TOF MS analysis: Following overnight purity checks, index bacterial colonies were transferred to a target polished steel plate (MSP 96, Bruker Daltonics Inc.) for drying under exhaust ventilation in a Class II Biohazard Hood (Gelman Sciences Australia) at room temperature. Air-dried samples were subjected to protein extraction with 1 μL 70% formic acid (Sigma). After repeat air-drying under exhaust ventilation, samples were overlaid with 1 μL of matrix solution (saturated solution of α-cyano-4-hydroxycinnamic acid (HCCA) in a mixture of 47.5% ultra-pure water, 2.5% trifluoroacetic acid, and 50% acetonitrile). Dried samples were analysed using Microflex MALDI-TOF mass spectrometer (Bruker Daltonik GmbH, Leipzig, Germany) equipped with a 60 Hz nitrogen laser. Spectra were recorded in the positive linear mode for the mass range of 2,000–20,000 Da at maximum laser frequency. The MALDI Biotyper 3.0 software package (default settings; Daltonik GmbH, Bremen, Germany) was used to automatically analyse and measure raw spectra without user intervention. This technology can detect approximately 5000 species. The most prevalent microorganisms are quantified (viable count detection limits include anaerobes >108 CFU/g, facultative anaerobes >105).

Data Used for Statistical Analysis. Genera investigated: Anaerobic (Bacteroides, Bifidobacterium, Clostridium, Eubacterium, and Lactobacillus) and aerobic genera (Escherichia, Streptococcus, Enterococcus) were investigated. Species identified during FMA were classified according to genera (data provided is the combined total of species identified within each genus). Species-level analyses were not included due to the heterogeneity of species identified during MALDI-TOF MS assessment and insufficient power to correlate less common species. Whilst genera-level investigations lack specificity, some evidence suggests similar metabolic and functional capacity within taxa and genera38.

Justifications for selected genera: A priori selection of genera was grounded in the literature. Some of the most abundant strains of enteric microbiota within healthy human samples fall within Bacteroides, Clostridium, Eubacterium, and Prevotella as the anaerobic genera and Escherichia and Streptococcus as the aerobic genera39. Within infants, some of the dominant enteric microbiota include Lactobacillus, Bifidobacterium and Streptococcus species40. Whilst the abundance of specific microbiota does not necessarily equate to their purpose, function or importance39, they provide an initial direction for examining specific genera.

The D-lactate hypothesis and relationship between increased D-lactate levels and neurocognitive impairment26 further guided selection of genera investigated in this research. An association between D-lactic acidosis and an overgrowth of enteric lactic acid bacteria (including some species of Lactobacillus, Bifidobacterium, Enterococcus and Streptococcus) has been shown7. An Australian sample of patients with ME/CFS showed significantly higher levels of Enterococcus and Streptococcus genera viable count compared with healthy controls7. This study also showed variable levels of Escherichia coli amongst ME/CFS samples compared with controls, hence the Escherichia genus was also investigated.

A possible cause of D-lactic acidosis is from abnormal metabolism of carbohydrate by enteric microbiota41. Although not a primary byproduct, Eubacterium species can also produce lactic acid42. Evidence of dietary influences on microbial composition supported the rationale for including examination of Eubacterium (associated with dietary fibre and starch43); and Clostridium (associated with increased refined sugar inake26). Additionally, some strains of Clostridium have been associated with health44 and others with pathology45.

Some strains of Lactobacillus and Bifidobacteria are frequently associated with optimal health and used for probiotic supplementation28,29,30,31,32,46. Health-promoting functions of these microbiota contrast the D-lactic hypothesis and provided further justification for examining these genera.

The abundance of Prevotella as well as evidence of an association between colonic overgrowth and neurological symptoms47 suggests the need to further investigate this genus. Unfortunately, Prevotella species were excluded from the analysis due to variable microbial identification and quantification methods during the data collection period.

Selection of the eight genera was supported by post-hoc analyses of the current dataset showing that they accounted for large proportions of detectable microbiota in the majority of stool samples. To assess the level of representation of selected genera within this ME/CFS sample, the Total RA was calculated as the combined proportion of the eight genera investigated within the total detectable bacteria (i.e., including all genera and not specifically limited to those analyzed in this study). From the 270 samples that were assessed for both aerobic and anaerobic bacteria, the eight genera represented between 5–100% of detectable microbiota (M = 92.60%, SD = 16.80%). The most common Total RA score was 100% with 90% of the sample showing a Total RA of equal to or above 72%. Sex comparisons of the Total RA indicated similarity in representation of the eight genera investigated (Mdn males = 99.67%, Mdn females = 99.77%, U = 8529.0, P = 0.263, r = 0.068).

Count: Microbial frequency of each genus was measured in colony-forming units per gram (CFU/g). Genera exponent values were used as a measure of each microbial count per patient.

Total Bacteria: Exponent values for the microbial frequency of all detectable bacteria as measured in CFU/g.

Aerobic:Anaerobic Ratio: Total detectable aerobic bacteria divided by total detectable anaerobic bacteria multiplied by 1000. This includes all genera and not specifically limited to those selected for data analysis.

Relative abundance (RA): Percentages were calculated by dividing the viable count of each genus by the total detectable bacteria count (methods akin to39,48). The expanded whole numbers for both counts were used in this calculation.

Total RA: The sum of RA percentages for the eight selected genera.

Patient Questionnaire: Concurrently to faecal sample collection, patients completed an 88-item Bioscreen Patient Questionnaire (BPQ). The BPQ is used for all referring patients regardless of clinical presentation. Items address diverse symptomatology similar to the Symptom Checklist-90-Revised49 and Beck Depression Inventory-II50. Patients rated symptom severity (past 7 days) and frequency (past 12 months) using a 5-point Likert scale (0–4). Frequency scores ranked from none at all (0) to extreme (4, severity) or constant (4, frequency). The BPQ showed high internal consistency within this ME/CFS population (Cronbach’s α = 0.974).

ME/CFS Symptom Factors: Seventy-six BPQ items were clinically classified into 13 factors reflecting ME/CFS symptoms in accordance with the ICC (F1-F1010) and mood symptoms (F11; Table S1). Seventeen items were omitted that were inconsistent with the ICC, could be misinterpreted as representative of two or more factors, or did not pertain to mood symptoms. Whilst psychological or mood symptoms are not specified under the ICC, high comorbidity with depression and anxiety symptoms in the ME/CFS population provided the rationale for further investigation of mood symptoms (predominantly depressive and anxiety symptoms). An impact score (severity × frequency) was calculated for each item (possible range 0–16) and relevant items were added to form corresponding factors. As measures of combined symptomatology, an ICC Symptoms Score (summation of F1-10) and Total Symptoms Score (summation of F1-F11) were calculated.

Bias: To reduce item selection bias, the factor classification was performed according to face validity as assessed by A.W., D.B. and M.B. and confirmed by consultation with clinician, D.P.L. No changes to the factor structure were made after commencing data analysis.

As a retrospective data sample, FMA results were initially performed for clinical purposes. Hence, no a priori hypotheses influenced data collection methods reducing the potential for investigator bias or falsification of data.

Statistical Methods. Descriptive statistics were performed for all ME/CFS symptom (Table 1 and Table S1) and microbial variables (Table S2) for the total sample, males and females. No outliers were found for microbial variables. The heterogeneity of symptom scores influenced the decision to include any clusters of outliers identified by SPSS on the ME/CFS Symptom Factors. Pairwise exclusion was used for missing data. All variables defied normality, therefore, nonparametric analyses were employed.

Examining Subgroups and Interactions. Sex comparisons on ME/CFS symptom factors: Descriptive statistics confirmed that each symptom factor (total, females and males separately) defied normality. A series of Mann-Whitney tests were used to compare the distribution of female and male symptom scores for each factor.

Sex comparison for microbial levels: Descriptive statistics confirmed that each microbial genus (count and RA) defied normality. A series of Mann-Whitney tests were used to compare the distribution of female and male microbial levels for count and percentage distribution independently. Effect sizes were calculated using equation:

where N was the total sample used in the analysis. Effect sizes were classified as small (0.01), moderate (0.03) and large (0.05) according to Cohen’s (1988) guidelines51.

Associations between microbial level and ME/CFS symptoms: Spearman’s rank order correlations (r s ) were used to investigate sex-interactions between microbial RA and ME/CFS symptom factors (Table S3). Missing data were excluded pairwise from the analyses. Correlations were deemed statistically significant at P < 0.05. Positive correlations indicated an increase in microbial relative abundance was monotonically associated with an increase in symptom scores. The direction of a positive association could also be explained in reverse. Negative correlations indicate an inverse monotonic relationship between the two variables. Correlations were classified as small (0.01), moderate (0.03) and large (0.05) effect sizes51.

Observed z scores were calculated using equation (2)52 to examine whether there was a statistically significant difference between the sexes for the strength of the correlation between symptom factor and microbial RA. Differences were deemed statistically significant when z obs < −1.96 or z obs > 1.9652.

Design Limitations and Advantages

We caution against over-interpretation of these findings considering the limitations of cross-sectional, observational research design (unable to establish causation or consequence, difficulty excluding confounding variables53) and categorical analysis of genera rather than species. Other genera that were not selected during this investigation may also have relevance for ME/CFS symptomatology. Technological advancement enabling 16S amplicon sequencing of viable count will be able to identify and compare a broader range of genera and species. This will then allow comparison with other ME/CFS samples (e.g. 20) and the ability to examine the representative nature of these results whilst considering the impact of ethnic and geographic diversity on microbial composition. Applied human research has clinical relevance54 and can appropriately direct the pursuits in animal investigations where mechanistic studies are needed21. A symbiotic, interdisciplinary approach that integrates sex differences in clinical observational data and mechanistic data will inform therapeutic directions and treatment utility.