1. Introduction
Fibromyalgia (FM) is one of the most common forms of chronic widespread pain, with an estimated prevalence of 2% to 4% of the adult population.14,30,71 Characterised by pain, physical exhaustion, sleep disturbance, and cognitive symptoms, FM leads to a significant impairment in quality of life of affected individuals.30,52,71,79 The pathophysiology of FM is not well understood, with multiple hypotheses being suggested including impaired central nervous system nociceptive processing, altered peripheral nociception, and systemic inflammation.79
The diagnosis of FM is based on the recognition of a typical cluster of symptoms, while excluding other potential sources of pain.30 The diagnostic criteria for FM rely on self-reported symptoms.93 The lack of objective diagnostic criteria is a source of frustration among patients and clinicians, and adds to the possible reasons for inaccurate diagnoses. In 2 retrospective cohorts of patients diagnosed with FM, the rate of false-positive diagnoses was estimated at 66% to 73%.22,89 Even when provided with the best available treatment, many patients continue to suffer from significant symptoms.30
In recent years, evidence is mounting on the critical role of the gut microbiota in a variety of pathologies, including, but not limited to, metabolic, cardiovascular, oncologic, neurologic, and psychiatric disorders.33,49 The scientific literature is rich in studies on different aspects of the interactions between the host and the microbiome; however, data on the possible role of the gut microbiota in the pathophysiology of chronic pain outside of the gastrointestinal tract is still scarce. Our increasing understanding of the interactions between the gut microbiota and the central nervous system, also known as the “gut–brain axis,” makes reasonable the hypothesis that it may also affect pain processing and perception.6,7,15,21,24,53 Recently, this hypothesis has been supported by several animal studies, which have shown that gut microbiota play an important role in the development of visceral pain,48,64,65 of chemotherapy-induced neuropathic pain,76 and of opioid tolerance.35 Human studies have thus far focused on alterations of the gut microbiota in several visceral pain disorders, showing consistent alterations in individuals with irritable bowel syndrome (IBS)16,19,70,83,96 and in patients suffering from chronic dysfunctional pelvic pain.5,9,77 Similarly, patients with chronic fatigue syndrome (CFS), which shares some symptomatic features with FM, were shown to have altered gut microbiome and metabolomic profiles.25,27,60,61 Finally, in several rheumatologic diseases, including rheumatoid arthritis and spondyloarthropathies, microbiome alterations have been reported.32,92,94,95
Indirect evidence hints that the gut microbiome may be altered in FM patients: altered small intestinal permeability was reported in a cohort of FM and complex regional pain syndrome patients28; in a small cross-sectional study of FM patients, a distinct urine metabolomic signature was demonstrated, which could be attributed to gut microbiome alterations.51
Despite recent reports of gut microbiome alterations in certain inflammatory rheumatic diseases, IBS, and interstitial cystitis (IC), there is still no evidence of such alterations in nonvisceral pain. Here, variation in the composition of the gut microbiome between FM patients and control participants is investigated for the first time.
2. Methods
2.1. Study design and oversight
The study took place at the Alan Edwards Pain Management Unit of the McGill University Health Centre, Montreal, and at the West Island Rheumatology Clinic, in Montreal, Quebec, Canada. The study was approved by the McGill University Health Centre, Montreal institutional review board. All participants were given a detailed explanation of the study and signed an informed consent form.
2.2. Patient recruitment and clinical evaluation
Individuals with FM and control participants were recruited as follows: patients were recruited at the Alan Edwards Pain Management Unit and at the West Island Rheumatology Clinic. Patients were contacted either by their treating physician, or by advertisements in the local media and a dedicated web site. Inclusion criteria for patients were: female sex, age 30 to 60 years, widespread pain index and symptom severity scores compatible with the 2016 diagnostic criteria for FM, and ability to provide informed consent in French or in English. Three groups of control participants were recruited: (1) first-degree female relatives of patients participating in the study were recruited as genetic controls. This group included adult women; (2) household members of patients participating in the study were recruited as environment controls. This group included women and men aged 30 to 70 years; and (3) unrelated healthy women aged 30 to 60 years.
Exclusion criteria were: major chronic illness (eg, malignancy, active inflammatory disease, and metabolic disease), antibiotic treatment in the preceding 2 months, any acute illness in the preceding month, change in regularly taken medication in the preceding month, and substantial dietary alterations in the preceding month.
All patients with chronic pain were interviewed by a specialized pain physician for a thorough assessment. Only individuals whose diagnosis of FM was confirmed were deemed eligible to participate in the study as patients. All patient and control participants were interviewed by a specialized pain physician, and data were collected regarding their demographics, anthropometrics, comorbidities (including a specific evaluation for IBS, based on ROME IV criteria44,55), medications, dietary intake, smoking, and alcohol consumption. Comorbidities were recorded based on a systematic clinical interview by a pain medicine physician and a review of participants' regular medications. Participants then filled-in the following questionnaires: the FM Survey Diagnostic Criteria and Severity Scale questionnaire, assessing symptom severity, pain location, fatigue, sleep quality, and cognitive and somatic complaints in FM patients, based on the 2016 criteria for the diagnosis of FM93; the FM Impact Questionnaire, a 10-item questionnaire evaluating physical functioning, work difficulty, pain, fatigue, morning tiredness, stiffness, anxiety, and depression11; physical activity assessment using the Physical Activity self-Administered Questionnaire, a 22-item questionnaire evaluating the average individual physical activity level88; and Sleep Scale from the Medical Outcomes Study, a 12-item sleep quality evaluation questionnaire.58 All questionnaires were offered in English or in French and have been validated in both languages.13,23,68,88 Participants were then interviewed by a dietitian, who supervised them as they filled-in the web-based NIH ASA24-Canada dietary recall, validated in English and in French.86
2.3. Dietary intake assessment
Dietary intake was assessed using the NIH Automated Self-Administered 24-hour recall (ASA24-Canada version 2016, English or French). This tool has been extensively validated.66,86 During the on-site visit, participants were trained and supervised by a dietitian as they filled the first day of recall online. They were asked to fill 2 additional separate days during the following week, allowing for a total of 3-day dietary assessment, including 2 weekdays and one weekend day. Data sets with at least 2 complete days were kept (n = 146) and daily nutrient averages were calculated. Nutrient intakes were compared among groups as absolute intakes and relative to energy or body weight. Underreporting was investigated by applying a ratio of reported energy intake over resting energy expenditure estimated with the Mifflin-St. Jeor equation56; participants with a ratio below 0.9 were deemed as underreporters and excluded from analysis. The Healthy Eating Index (HEI-2015) was calculated to assess diet quality.39
2.4. Sample acquisition and handling
Patients were asked to collect a stool sample at home, using the Omnigen Gut OM-200 kit (DNA Genotek, Ottawa, Ontario, Canada). Stool was fresh-frozen at −20C, and delivered by the participants to the study facility within 10 days of its collection. Proper delivery conditions were assured using a dedicated thermal bag (Thermos, Schaumburg, IL) containing an ice pack, and monitored using an adhesive thermal indicator (Warm Mark; MesaLabs, Lakewood, CO), which was attached to the sample bag. On arrival at the study facility, thermal indicators were read to assure proper delivery conditions, and the stool samples were frozen at −80C.
2.5. DNA extraction, 16S ribosomal RNA gene amplification, and Illumina sequencing
DNA was extracted from all 156 stool samples using the QIAamp PowerFecal DNA kit (Qiagen, Venlo, the Netherlands) following the manufacturer's instructions. The V5-V6 region (based on Escherichia coli) of the 16S ribosomal RNA (rRNA) was targeted for amplification by PCR using the forward primer: S-D-Bact-0785-a-S-18, GGMTTAGATACCCBDGTA and reverse primer: S-*-Univ-1100-a-A-15, GGGTYKCGCTCGTTR.38 The CS1 (ACACTGACGACATGGTTCTACA) and CS2 (TACGGTAGCAGAGACTTGGTCT) tags were used to add a barcode and Illumina adapters. Amplification was performed using Q5 High-Fidelity DNA polymerase (New England BioLabs, Ipswich, MA) with PCR cycles as follows: initial denaturation step of 98°C, for 30 seconds, before 23 cycles of 98°C, for 10 seconds, 58°C, for 15 seconds, and 72°C, for 30 seconds, with the final extension at 72°C, for 2 minutes. The MiSeq250 platform was used for 2 × 250 nucleotides (nt) paired-end sequencing of the resulting PCR products. The ANCHOR pipeline was used to process and annotate sequence reads (https://github.com/gonzalezem/ANCHOR). Briefly, sequences were aligned and dereplicated using Mothur74 before selection of operational taxonomic units (OTUs) using a count threshold of 48 across all samples. Annotation used 4 sequence repositories with strict BLASTn criteria (>99% identity and coverage): NCBI-curated bacterial and Archaea RefSeq, NCBI nr/nt, SILVA, Ribosomal Database Project (NCBI-curated bacterial and Archaea RefSeq is given a priority when at 100% identity and coverage). When the highest identity/coverage is shared among different putative annotation, all are retained and reported (borrowed from the idea of secondary annotation in metatranscriptomics29). Amplicons with low counts (<48) are binned to high-count sequences in a second BLASTn, using a lower threshold of >98% identity/coverage.
2.6. Whole-genome sequencing, analysis using MetaPhlAn2, and mapping to 16S rRNA gene operational taxonomic units
Genomic DNA was quantified using the Quant-iT PicoGreen dsDNA Assay Kit (Life Technologies, Carlsbad, CA). Libraries were generated using the NEB Ultra II DNA kit (NEB) as per the manufacturer's recommendations. TruSeq adapters with unique dual indices and PCR primers were purchased from IDT (Integrated DNA Technologies, Inc, Coralville, IA). Size selection of libraries at 360 nt was performed using SPRIselect beads (Beckman Coulter, Brea, CA). Libraries were quantified using the Quant-iT PicoGreen dsDNA Assay Kit (Life Technologies) and the Kapa Illumina GA with Revised Primers-SYBR Fast Universal kit (Kapa Biosystems Inc., MA). Average size fragment was determined using a LabChip GX (PerkinElmer, Waltham, MA) instrument. The libraries were normalized, pooled, and denatured in 0.05 N NaOH and then neutralized using HT1 buffer. ExAMP was added to the mix to bring the final concentration to 200 pM following the manufacturer's instructions. A phiX library was used as a control and mixed with libraries at 1% level. The pool was loaded on an Illumina cBot, and the flowcell was ran on a HiSeq 4000 for 2 × 100 cycles (150 nt, paired-end mode). The Illumina control software was HCS HD 3.4.0.38, and the real-time analysis program was RTA v. 2.7.7. Program bcl2fastq2 v2.18 was then used to demultiplex samples and generate fastq reads. Trimmomatic43 was used to trim nucleotides of poor quality and reads <100 nt were removed (parameters: LEADING:25 TRAILING:25 MINLEN:100). Whole-genome sequencing (WGS) reads were analysed in 2 ways: by directly mapping them to the 16S rRNA gene ANCHOR OTU sequences and using MetaPhlAn2.75 Bowtie241 was used to align WGS reads against MetaPhlAn2 v2.2.0 (default parameters). Mapping of WGS reads directly to the ANCHOR OTU table was performed using BLASTn (100% identity and 100% query coverage thresholds).
2.7. Canonical correspondence analysis and prediction
Canonical correspondence analysis was performed on raw OTU counts of FM patients and unrelated controls (PERMANOVA, permutational multivariate analysis of variance P < 0.001). Household and family controls were projected on the axes post hoc (predict.cca function, R Vegan package62).
2.8. Microbiome explained variance analysis
Unconstrained ordination was calculated using Nonmetric Multidimensional Scaling transformation on Bray–Curtis distances (metaMDS function, R Vegan package). Environmental factors were projected onto ordination diagram, and each variable regression was independently tested by Monte Carlo permutation test (envfit function, R Vegan package).62
2.9. DESeq2 differential abundance analysis
Differential abundance analysis was performed using 16S rRNA in accordance with DESeq2,47,84 which has been shown to perform well when applied to the uneven library sizes and sparsity common in 16S rRNA gene marker data.91 Differentially abundant (DA) selection parameter of false discovery rate (FDR; Benjamini–Hochberg procedure) < 0.1 was applied4,46,47 (supplementary file, available at https://links.lww.com/PAIN/A839). Regularized log transformation was applied to raw counts across samples (rlog function, R phyloseq package54). A sparsity and low-count cutoffs were used whereby an OTU count in a single sample is <90% of the count in all samples, and OTU count must be >2 in 40% of the samples.
2.10. Correlation between taxa abundance and clinical indices
Spearman's correlation was calculated between log-transformed OTU abundance and continuous or rank clinical variables. Individual P value was calculated for comparison, and correction for multiple comparisons was done using Benjamini–Hochberg FDR, with a cutoff value of 0.05. A colour-coded correlation matrix representing taxa abundance vs clinical indices was elaborated and sorted based on hierarchical clustering of DA OTUs using MathWorks MATLAB.
2.11. Operational taxonomic unit–operational taxonomic unit correlation
Pearson's correlation was calculated between normalized OTU counts. Correlation matrices were sorted based on hierarchical clustering of OTUs and represented using R package corrplot.90
2.12. Machine-learning instances for fibromyalgia prediction
The L1-penalized regression or LASSO (least absolute shrinkage and selection operator)85 was used to predict FM diagnosis from the ANCHOR DA OTU count table and was run using glmnet R package.26 Several count tables were tested as preselection data sets for the LASSO including full and reduced (sparsity filtered, DESeq2 DA OTUs), each with different count transformation or normalization strategies (raw, log, quantile, square root; supplementary Figure 5, available at https://links.lww.com/PAIN/A839). DESeq2 DA OTUs (72 features) performed better in diagnosis prediction than sparsity-filtered counts (1071 features), and so were considered a good preselection data set in line with previous research.98 Random OTU count table subsets of equivalent size (each with ∼72 OTUs) were also compared against the DESeq2 DA OTU subset, with DA OTUs again showing better prediction power (supplementary Figure 3, available at https://links.lww.com/PAIN/A839). Raw counts outperformed transformed data (log, square root) and normalized counts (quantile) in their prediction and were used for subsequent analysis. LASSO's integrated feature selection (ie, selection of sets of features leading to best prediction) was run on DESeq2 DA count table (72 OTUs) using a 10-fold multinomial cross-validation and repeated 10,000 times. A total of 123 high predictive DA OTU sets, ie, sets of features selected multiple times by LASSO with a high prediction (median area under the curve [AUC] > 0.9) or selected a single time with a perfect prediction (AUC of 100%), were selected for leave-one-out cross-validation multinomial LASSO. Leave-one-out cross-validation was performed on training data sets composed of 90% of the participants and was repeated 1000 times. Prediction was evaluated on the remaining 10% of participants (test data set). Receiver operating characteristic (ROC) curves (pROC R package73) were used to illustrate the discriminative ability of the classifier, whereas prediction performance was evaluated by the AUC metric. The related support vector machine (SVM) technique34 was used based on the best predicting DA OTUs and 90% of the samples, and the diagnosis prediction was evaluated (AUC on 10% on the total number of participant).
2.13. Targeted serum metabolite analysis
Serum levels of butyric acid, isobutyric acid, and propionic acid were measured as previously described.17 Targeted metabolite quantification approach was used to analyze the serum samples using a combination of direct injection mass spectrometry with a reverse-phase liquid chromatography-tandem mass spectrometry (LC-MS/MS custom assay). The method used combines the derivatization and extraction of analytes, and the selective mass-spectrometric detection using multiple reaction monitoring pairs. Isotope-labeled internal standards and other internal standards are used for metabolite quantification. The custom assay contains a 96-deep-well plate with a filter plate attached with sealing tape, and reagents and solvents used to prepare the plate assay. First 14 wells were used for one blank, 3 zero samples, 7 standards, and 3 quality control samples. One hundred fifty microliter of ice-cold methanol and 10 μL of isotope-labeled internal standard mixture were added to 50 μL of serum sample for overnight protein precipitation. Then it was centrifuged at ×13,000g for 20 minutes. Fifty microliter of supernatant was loaded into the center of wells of a 96-deep-well plate, followed by the addition of 3-nitrophenylhydrazine (NPH) reagent. After incubation for 2 hours, BHT stabilizer and water were added before LC-MS injection.
Mass spectrometric analysis was performed on an API4000 Qtrap tandem mass spectrometry instrument (Applied Biosystems/MDS Analytical Technologies, Foster City, CA) equipped with a solvent delivery system. Data analysis was performed using Analyst 1.6.2.
2.14. General statistical considerations
Clinical indices, including demographics, anthropometrics, comorbidities, and medications, in the study groups were compared using IBM SPSS Statistics version 23 and MathWorks MATLAB version 2018b, using analysis of variance (ANOVA) for univariate comparisons and multivariate analysis of variance for multivariate comparisons. Metabolite concentrations were compared similarly.
Dietary analysis was performed as follows: normality of distribution was explored using Kolmogorov–Smirnov and Shapiro–Wilk tests. Most dietary data were assessed using one-way ANOVA and post hoc Games-Howell test (no assumption of equal variance and appropriate for unequal sample size). Some nutrients were not normally distributed and analysed using nonparametric Kruskal–Wallis test. Analyses were conducted using IBM SPSS Statistics version 23.
Formal power analysis for sample-size calculation was not performed, but sample size was not smaller than that of similar previous studies of microbiome characterization in other medical conditions.94
2.15. Data availability
All data generated as part of this study are included in this published article and its supplementary file or are available from the corresponding authors on reasonable request. Supplementary file is available at https://github.com/gonzalezem/Fibromyalgia. Genome sequences can be found at https://www.ncbi.nlm.nih.gov/sra/PRJNA521398 (16S rRNA sequences) and https://www.ncbi.nlm.nih.gov/sra/PRJNA521587 (WGS sequences).
3. Results
3.1. Participant characteristics
Two hundred potential participants were screened for this study. Twenty-nine participants were excluded during the screening phase and 15 participants, who did not meet the diagnostic criteria for FM, were excluded following a medical interview with 2 specialized pain physicians. Of the 156 participants, 77 were FM patients and 79 were control participants. Control participants included 11 first-degree relatives of participating FM patients, 20 household members of participating patients, and 48 unrelated controls. Other than sex, marital status, and occupational status, differences in demographic and anthropometric characteristics among groups were not statistically significant (supplementary table 1, available at https://links.lww.com/PAIN/A839). One patient in the first-degree relative control group was diagnosed with well-controlled rheumatoid arthritis, with no evidence of active arthritis or elevated inflammatory markers. All participants were recruited and all samples collected between October 2017 and June 2018. No seasonal variation was observed between groups when comparing sample collection season (ANOVA P = 0.30, F = 1.24).
Fibromyalgia patients were women (mean age 46 ± 8 years), who on average had been diagnosed with FM 12 (SD 7.7) years earlier. The majority of participants were of Caucasian ethnicity. Mean widespread pain index in the patients' group was 11 (SD 3.2), and their mean symptom severity score was 9.3 (SD 1.8). The 2016 FM diagnostic criteria scores and FM Impact Questionnaire scores were significantly higher in patients compared with all control groups (Pillai's trace P < 0.0001, F = 4.53; supplementary table 2, available at https://links.lww.com/PAIN/A839). Participants' sleep quality was evaluated using the Insomnia Severity Index questionnaire. Fibromyalgia patients reported more difficulty falling asleep, maintaining sleep, and early awakening (Pillai's trace P < 0.0001, F = 7.75, see univariate analysis in supplementary table 3, available at https://links.lww.com/PAIN/A839). The frequency of depression and anxiety scores of 5/10 or higher among FM patients was 59% and 68%, respectively, in line with previous reports.2,31
Participants' physical activity was evaluated using the Physical Activity Self-administered Questionnaire. No consistent trend towards higher or lower level of activity was observed for any group (supplementary table 3, available at https://links.lww.com/PAIN/A839).
Dietary assessment was performed for all participants. Among participants with complete dietary reports, 16 patients and 3 unrelated controls were excluded for underreporting. Energy and macronutrient intake (total and relative) were not different between patients and controls (supplementary table 4, available at https://links.lww.com/PAIN/A839). In addition, vitamin, mineral, different fatty acids, alcohol, caffeine, sugar, and fiber intakes were not different among groups, nor was overall diet quality score (data not shown).
3.2. Overall gut microbiome composition
To explore the composition of the gut microbiome of FM patients and controls, we studied a cohort of 156 adults. Stool samples (77 FM patients and 79 controls) were collected and sequenced using both 16S rRNA and metagenome methods, and analyzed at multiple taxonomic levels. 16S rRNA sequencing produced 12,137,453 reads, with an average of 77,804 ± 6832 reads per sample.
A total of 1620 OTUs were identified capturing 8,474,421 reads. These could be classified at various taxonomic levels: 8 at class, 25 at order, 179 at family, 716 at genus, and 349 as species, with 343 OTUs remaining as unclassified or unknown (Fig. 1). Predominant phyla across all participants included bacteroidetes (48% of total raw counts), firmicutes (40%), proteobacteria (4%), and actinobacteria (2%). From the 349 OTUs annotated at species taxonomic level, 312 were unique whereas 37 shared a sequence common to multiple species (additional 16S and WGS data are provided in the supplementary file). The 10 most abundant OTUs annotated at species resolution across all samples were from the Bacteroides genera, including Bacteroides dorei, Bacteroides uniformis, Bacteroides stercoris, Bacteroides ovatus, as well as Prevotella copri, Alistipes putredinis, and Faecalibacterium prausnitzii.
;)
Figure 1.: Overall
microbiome composition of 156 study participants. (A) Flower diagram of 1620 OTUs colour coded by phyla. An interactive version is available on
http://www.computationalgenomics.ca/pub/fibromyalgia_microbiome/flower_diagram/. (B) Phyla-level taxonomic composition of the gut
microbiome among FM patients (n = 77), relative controls (RC, n = 11), household members (HC, n = 20), and unrelated controls (UC, n = 48). (C) Distribution of OTUs by taxonomic level of identification. (D) Distribution of relative abundance by the level of taxonomic identification. (E) Comparison of Bray–Curtis dissimilarity between all pairs of 156 participants, demonstrating that pairwise distances between FM patients (FM–FM, 1482 pairs) are smaller as compared to distances between pairs of patients–controls (FM–C, 6083 pairs) or controls–controls (C–C, 1560). (F) Comparison of alpha diversity indices for each participant group. (G) Study design flowchart. (H) Comparison of species identification using 16S rRNA genes (ANCHOR; yellow) and WGS (Metaphlan2; blue) technologies. FM,
fibromyalgia; WGS, whole-genome sequencing.
To validate the taxa identified using 16S rRNA gene amplification, WGS data were directly mapped to the 16S rRNA gene OTU table as well as analysed using MetaPhlAn2. Whole-genome sequencing generated 1,289,867,794 reads with an average of 8,268,383 reads per sample. All 1620 OTUs were successfully recapitulated by WGS reads at 100% identity. The median coverage was 99.1% (SD 4.2) with 1015 OTUs mapped uniquely and 605 sharing reads. Independent analysis using MetaPhlAn2 identified 196 species, including all but 3 species found using 16S rRNA gene amplification with the exception of those not yet present in the MetaPhlAn2 marker database (80 species, Fig. 1H).
3.3. Gut microbiome composition is altered in fibromyalgia patients
Shannon and inverse Simpson alpha-diversity indices showed nonsignificant differences in sample diversity between FM patients and household, relatives, and unrelated control groups (ANOVA P > 0.05; Fig. 1F). Bray–Curtis dissimilarity between pairs of participants showed a small but significant difference in mean dissimilarity metrics among FM–FM pairs as compared to FM–control pairs (P < 0.0001, F = 48.44), or control–control pairs (P < 0.0001, F = 56.71, Fig. 1E).
Canonical correspondence analysis suggests that samples segregate by FM patients and unrelated controls in nonoverlapping clusters (Fig. 2A). Multivariate analysis showed between-group variance to be significantly greater than within-group variance (P < 0.01) between FM patients and unrelated controls. Household members and first-degree relative controls clustered between FM patients and unrelated controls (Fig. 2B).
Figure 2.: (A) Canonical correlation analysis (x axis) vs principal component analysis (y axis) of taxa abundance (raw counts) in FM patients vs unrelated controls (UC). Amount of explained variance is provided (P < 0.001, ANOVA). (B) Relatives (RC) and household controls (HC) cluster midway between FM patients and unrelated controls. ANOVA, analysis of variance; FM, fibromyalgia.
Differential abundance analysis was performed to compare FM patients (77) with unrelated control participants (48) using DESeq2. Seventy-two OTUs were identified as significantly DA, 53 higher in FM and 18 higher in unrelated controls. These included 47 that could be annotated at either genera or species taxonomic levels and included taxa from firmicutes, bacteroidetes, and proteobacteria. The largest positive fold-change difference in DA putative species observed in FM patients were from the OTUs Parabacteroides_merdae_4 and Clostridium_scindens_1, whereas the largest negative fold-change difference in FM patient were in Prevotella_copri_1 and Bacteroides_uniformis_3 (Fig. 3).
Figure 3.: (A) Differentially abundant OTUs between the FM and unrelated control groups. Fold change (FC log2) denotes relative differences in relative abundance between groups (DESeq2). Species are grouped by phylum per comparison. (B) A diagram of species-level differentially abundant bacteria in FM patients (left) and in healthy controls (right). FM, fibromyalgia; OTU, operational taxonomic unit.
When relative abundance of OTUs were correlated, DA species generally clustered according to their abundance in participant groups, whereby species found in higher abundance in FM patients clustered together, whereas those found in higher abundance in controls clustered separately (supplementary Figure 1A, available at https://links.lww.com/PAIN/A839). Among DA species, the majority of correlations were statistically significant (P < 0.1), showing positive correlations between a majority of DA species within FM group and within unrelated control group (Pearson correlation > 0) and significant inverse correlations of DA species between FM and unrelated control groups (Pearson correlation < 0, supplementary Figure 1B, available at https://links.lww.com/PAIN/A839). This observation serves as an independent support that DA taxa represent a valid difference between FM patients and controls.
To control for possible confounding effects of host-related and environmental variables, the amount of variance (r2) explained by various possible confounders was evaluated using a general linear model. The highest amounts of microbiome variance were explained by the diagnosis of FM and disease-specific measures, whereas all other covariates, including diet, medications, comorbidities and anthropometric variables, had a smaller effect on the observed variance (see supplementary Figure 2 Tables 5 and 6, available at https://links.lww.com/PAIN/A839). This suggests that at least some of the observed differences in microbiome composition are likely to be attributed to FM.
3.4. Altered serum levels of fermentation end products in fibromyalgia patients
Consistent alterations in the abundance of butyrate metabolism–related bacteria were observed: F. prausnitzii and B. uniformis were found in lower relative abundance in FM patients, whereas higher relative abundance was observed for Intestinimonas butyriciproducens, Flavonifractor plautii, Butyricoccus desmolans, Eisenbergiella tayi, and Eisenbergiella massiliensis. To explore the possible metabolic effect of these alterations, a targeted metabolite approach was used to measure the serum concentrations of butyric acid, isobutyric acid, propionic acid, and lactic acid. Serum levels of butyric acid in FM patients (n = 73) were higher compared to unrelated controls (n = 46, P = 0.005), whereas levels of propionic acid were lower (P = 0.006) and a trend towards lower levels of isobutyric acid was also observed (P = 0.056). No significant differences in the serum levels of lactic acid were observed (Fig. 4B). Multivariate analysis showed a significant between-group difference (Pillai's Trace, F = 8.97, P < 0.0001).
Figure 4.: (A) Serum concentrations of butyric acid, isobutyric acid, propionic acid, and lactic acid (ANOVA **P < 0.01, ˙P < 0.1). (B) Abundance of some DA OTUs is associated with clinical indices: Heat map of a univariate Spearman correlation matrix between DA taxa abundance (log2; x-axis) vs covariates (y-axis): demographics and anthropometrics; disease severity metrics (FMDC: 2016 FM diagnostic criteria); quality of life scores (FIQ); physical activity measures (PA); and sleep quality (sleep) scores. Heat map is sorted base on a hierarchical clustering of DA OTUs. Blue shades indicate positive correlations, whereas red shades indicate negative correlations (−0.5 < rho < 0.5). Statistically significant correlations are marked by a black dot (Benjamini–Hochberg FDR < 0.05). Note the selective significant correlation of taxa abundance with symptom severity but not with other independent variables. ANOVA, analysis of variance; DA, differentially abundant; FDR, false discovery rate; FIQ, FM Impact Questionnaire; FM, fibromyalgia; OTU, operational taxonomic unit.
3.5. Gut microbiome is associated with clinical indices of fibromyalgia
To explore the relationship between DA taxa abundance and symptom severity, the relative abundance of DA OTUs was correlated against disease-specific and independent variables. Spearman's rank correlation coefficient demonstrated a statistically significant association between the abundance of several taxa and disease severity measures, including pain intensity, widespread pain index, dyscognition, and fatigue (Benjamini–Hochberg FDR < 0.05, Fig. 4A). The abundance of these taxa showed no consistent relationship with patients' demographic, anthropometric, or physical activity variables. Thus, abundance of some DA taxa could be correlated with FM symptom severity.
3.6. Microbiome-based machine-learning identification of fibromyalgia patients
To explore the utility of gut microbiome composition in identifying FM patients, we used machine-learning using LASSO (least absolute shrinkage and selection operator) and SVM algorithms to define the OTUs most predictive of the diagnosis of FM (Fig. 5A). The full and sparsity-filtered count tables, as well as OTU subsets (DA and random), were tested to determine the most useful data set for classification. The DA OTU data set yielded the highest diagnostic accuracy. A 10-fold cross-validation (10,000 iterations) yielded 123 DA OTU sets with the highest diagnostic accuracy. LOOCV (leave-one-out cross-validation) identified a set of OTUs that provided the best overall prediction accuracy (AUC-ROC of 87.2%; Figs. 5B and C). Finally, this best predicting OTU set was used in a SVM trained on 90% of participants. The SVM yielded a prediction accuracy of ROC-AUC 87.8% (Fig. 5D) using cohorts of 10% of participants (100 iterations).
Figure 5.: Machine-learning classifier algorithm: (A) Two-step LASSO
95 workflow adapted from Bollepalli et al.
9 Operational taxonomic units that resulted in the best prediction accuracy in LASSO were used in a support vector machine (SVM). (B) Species-level annotated OTUs most often selected by LASSO. (C) Receiver operating characteristic curve for 1000 iterations of LASSO-LOOCV feature prioritisation using 123 previously selected OTU sets. (D) Receiver operating characteristic curve for 100 iterations of SVM validation test using most important group of OTUs (group_103). The model was trained on 90% of the initial number of samples and tested on the remaining. AUC, area under the curve; CV, cross-validation; LOOCV, leave-one-out cross-validation; OTU, operational taxonomic unit; ROC, receiver operating characteristic.
The superior utility of the DA OTU data set reiterates the notion that DA taxa represent a true difference between FM patients and controls. Furthermore, these results suggest that the composition of the microbiome could be indicative of the diagnosis of FM.
4. Discussion
The overall population structure and diversity of the gut microbiome in patients with FM and a matched cohort of healthy controls was relatively similar. When explored at a higher resolution, however, gut microbiome composition showed significant alterations in FM patients. These differences, independently evident using multiple analytic approaches, revealed a distinct pattern of the fecal microbiome in FM. The diagnosis of FM and its clinical features (pain, fatigue, and cognitive symptoms) explained more microbiome variance than did any other covariate, including demographic, anthropometric, and dietary variables, as well as comorbidities and medication consumption.
We observed a quantitative association between the abundance of several taxa and the severity of FM-related symptoms, including pain intensity, pain distribution, fatigue, sleep disturbances, and cognitive symptoms. The abundance of these taxa correlated selectively with disease-related symptoms, but not with disease-independent variables.
Having established that gut microbiome is altered in FM patients, LASSO machine-learning algorithm showed high prediction accuracy of patients from controls, based only on individual microbiome features.67 Further study, including larger cohorts, greater ethnic sampling, and cross-cohort assessment, is required before a confident contribution can be made to improved FM diagnostics; however, the ability of our model to accurately identify FM patients in this study indicates that gut microbiome may have the potential to contribute to the clinically challenging diagnosis of FM.
When considering the 19 specific species identified as significantly DA between FM patients and unrelated control participants, there was a broad range in how well characterised these species were. Those species putatively depleted (lower in relative abundance) in FM were relatively well characterised and included F. prausnitzii, B. uniformis, P. copri, and Blautia faecis. F. prausnitzii is one of the most abundant and well-studied butyrate-producing bacteria in the human gut.45 This species has been reported to be depleted in multiple intestinal diseases and was therefore suggested as a potential marker for gut disorders. Within the gut, F. prausnitzii has been reported to exert antinociceptive as well as anti-inflammatory effects57,78 and to enhance the intestinal barrier function.45 Similar to our observations, F. prausnitzii was also reported to be depleted in patients with CFS.60
B. uniformis is one of several species that have recently been reported as having altered relative abundance in patients with inflammatory arthritis, along with Haemophilus parainfluenzae, P. copri, and others.95B. uniformis and H. parainfluenzae were detected in synovial tissues of osteoarthritic joints, whereas P. copri and H. parainfluenzae were detected in rheumatoid arthritis synovial fluid. P. copri is believed to mediate inflammatory response through Th17 activation,36,42 and was also shown to induce arthritis in an animal model of arthritis-prone mice.50 In this study, these species were found in lower abundance in FM patients. Although FM is often considered to be a rheumatologic disease, it seems that at least some species previously found at higher abundance in inflammatory rheumatic diseases are depleted in FM.
In contrast to the depletion of butyrate producers F. prausnitzii and B. uniformis in FM patients, we observed significant higher relative abundance of a number of other known intestinal butyrate producers: I. butyriciproducens, F. plautii, B. desmolans, E. tayi, and the recently identified E. massiliensis.1,3,10,12,37,81,87 Alterations in butyrate- and propionate-metabolizing species were further supported by alterations in serum levels of these short-chain fatty acids. Coherent with this putative shift in the butyrate-producing community of FM patients, Parabacteroides merdae was also significantly higher in relative abundance in FM patients. Recently, P. merdae has been reported by Olson et al. to be one of 2 key mediators of the antiepileptic effect of the ketogenic diet63: in a mouse model, ketogenic diet can drive an increase in the abundance of P. merdae, which in turn, by regulation of amino acid γ-glutamylation, leads to an increase in hippocampal γ-aminobutyric acid (GABA)/glutamate ratio. The increase in the inhibitory to excitatory neurotransmitter ratio in the brain is believed to protect against seizures. The second key mediator species of the ketogenic diet effect on seizures reported by Olson et al. was Akkermansia muciniphila, which in our cohort was also found at higher abundance in FM patients, although this observation did not reach statistical significance (P = 0.042, Benjamini–Hochberg FDR = 0.27).
Clostridium scindens and B. desmolans, 2 bacterial species capable of converting cortisol to androgens by 20α-hydroxysteroid dehydrogenase activity,8,18,59,72,80 were found in higher abundance in FM patients. Interestingly, abnormal regulation of hypothalamic-pituitary-adrenal axis has been reported in FM patients, although the direction of dysregulation is controversial.20,79
Fibromyalgia shares clinical features with several syndromes, including IBS, CFS, and IC. Specific alterations in gut microbiota were reported in all these syndromes (IBS,69,70,82,97 CFS,60 and IC9). Although several taxa, found here to be DA in FM, were also reported to be DA in other overlapping syndromes, other taxa seem to be unique to FM. As an example, C. scindens was found at higher abundance both in FM and in CFS. H. parainfluenzae and F. prausnitzii were found at decreased abundance in FM and in CFS, and the latter was also reported to be depleted in IC patients.9 Several genera, including Bacteroides, Parabacteroides, and Clostridium were found DA both in IBS and in FM (independent of IBS).70,82,97 Nevertheless, a few taxa showed an inverse differential abundance pattern: eg, P. merdae, significantly higher in FM, was reported to be depleted in CFS.60 Furthermore, FM is associated with a high frequency of mental health comorbidity, mainly anxiety and depression affecting 40% to 80% of individuals.2,31 Anxiety, depression, and emotional stress have been associated with alterations in gut microbiome both in the general population40 and in the context of specific other diagnoses. Recent research by Peter et al. (2018) has revealed the potential importance of mental health as a factor in IBS, suggesting that further cross-disciplinary research is needed to fully understand the microbiome differences identified here.69 It thus seems that FM shares some microbiome characteristics with several overlapping syndromes, while retaining some unique features.
In this study, effort was made to validate the diagnosis of each patient by a thorough medical evaluation by experienced pain specialists, thus excluding over 15% of interviewed patients. We consider the meticulous characterization of patient phenotypes, an important strength of this study. A full dietary intake evaluation was completed for nearly 80% of participants and analyzed by a team of nutritionists. To the best of our knowledge, this type of in-depth evaluation is not routinely performed in similar studies. Finally, each participant underwent a full medical, including documentation of their comorbidities (with a specific evaluation for IBS), medication consumption, physical activity, and sleep. These measures resulted in a well-balanced case-control cohort, almost identical in demographics, anthropometrics, comorbidities, nonpain medication intake, and dietary intake. Furthermore, the extent of collected data allowed us to evaluate the effect of each variable on the observed variance in microbiome composition and to detect potential confounding factors. The cohort of 156 participants provided sound statistical power. We have used 2 sequencing methods—16S rRNA gene amplification and WGS metagenomics analyses—each with its own advantages and drawbacks, to allow for an independent validation of the microbiome composition. The combination of large cohort size with high depth of coverage allowed for the identification of low-prevalence DA species, such as I. butyriciproducens. Targeted serum metabolomics allowed for an independent support of some of the alterations observed in the gut microbiome. Results of this study should be considered in light of several limitations: first, the symptoms of FM include chronic pain, fatigue, and other somatic symptoms, precluding the attribution of specific microbiome alteration to a specific clinical presentation. Second, over 90% of participants self-reported as Caucasians, possibly not reflecting microbiome patterns in other ethnic groups. Third, participants' diagnoses were recorded based on a systematic clinical interview, which could potentially lead to underdiagnosis. Mental health conditions may be underrepresented in FM patients (supplementary Table 5, available at https://links.lww.com/PAIN/A839) as compared to self-reported symptoms (supplementary Table 2, participants were not evaluated by a mental health care specialist, available at https://links.lww.com/PAIN/A839). Finally, this study was designed to find an association between microbiome composition and FM, not allowing for direct conclusions regarding causality or mechanisms of action to be drawn.
The results reported here are, to the best of our knowledge, the first to demonstrate gastrointestinal tract microbiota alteration in patients with somatic (nonvisceral) pain in general, and FM in particular. Some of the DA taxa reported here are known to be involved in metabolic pathways whose effect on symptoms of FM is biologically plausible. Moreover, there seems to be a quantitative correlation between the abundance of several taxa and symptom severity. Finally, DA taxa in FM patients share some commonalities with overlapping dysfunctional syndromes—IBS, CFS, IC—while retaining some unique DA taxa.
Results of this study provide evidence for alterations of gut microbiome alterations in FM. Further studies are needed to measure possible microbiome changes in other chronic pain conditions, and to explore potential causal correlations between the gut microbiome and FM. Our findings may offer important opportunities to improve our understanding of FM, aid in its diagnosis, and perhaps outline future therapeutic modalities. First, exploring the mechanisms by which the gut microbiota may affect nociception could offer new insights into the pathogenesis and pathophysiology of FM, and possibly suggest leads to explore the possible effects of trigger events, personal susceptibility, and individual prognosis. Second, the accuracy of microbiome-based machine learning classification suggests that the microbiome holds the potential to aid in the diagnosis of FM in the future. Finally, if a causal relationship between gut microbiome alterations and FM is established, the way may be paved for the development of new treatment modalities targeting this fascinating community of intestinal microorganisms.
Disclosures
The authors have no conflicts of interest to declare.
Appendix A. Supplemental digital content
Supplemental digital content associated with this article can be found online at https://links.lww.com/PAIN/A839 and https://links.lww.com/PAIN/A840.
Supplemental video content
Video content associated with this article can be found online at https://links.lww.com/PAIN/A840.
Acknowledgments
The authors acknowledge the generous financial support of the Louise and Alan Edwards Foundation. The authors thank the Israeli Society for Musculoskeletal Medicine for their financial support. The authors would like to thank Mrs Sylvie Toupin, Mrs Linda Ferguson, Dr Pierre Lepage (Génome Québec), Dr Rupa Mandal (TMIC), Ms Audrey Moyen, and Mr Stéphan Attara for their invaluable help in the realization of this project. The authors are grateful to L'association de la Fibromyalgie de Montréal and to all participants for their partnership. The authors thank David Anderson for advising on figure design. DNA sequencing was performed by Génome Québec, Montreal. Metabolite analysis was performed by the metabolomic innovation centre (TMIC), Edmonton, Alberta. The Canadian Center for Computational Genomics (C3G) is a Genomics Technology Platform (GTP) supported by the Canadian Government through Genome Canada.
Author contributions: A. Minerbi and Y. Shir conceived, designed, and supervised the study. Y. Shir, A. Minerbi, M.A. Fitzcharles,K. Dewar, and S. Chevalier were involved in the design of the study. A. Minerbi and M.A. Fitzcharles interviewed and evaluated FM patients. A. Anjarkouchian performed the nutritional evaluations. A. Minerbi collected all biological samples. Genomic data was analysed by E. Gonzalez, N.J. Brereton, and A. Minerbi. Dietary analysis was performed by S. Chevalier and A. Anjarkouchian. Metabolomic data was analysed by A. Minerbi. K. Dewar, S. Chevalier, E. Gonzalez, N.J. Brereton, and A. Minerbi were involved in data analysis, evaluation, and integration. The manuscript was written by A. Minerbi, Y. Shir, N.J. Brereton,E. Gonzalez, and S. Chevalier.
References
[1]. Ahn S, Jin TE, Chang DH, Rhee MS, Kim HJ, Lee SJ, Park DS, Kim BC. Agathobaculum butyriciproducens gen. nov. sp. nov., a strict anaerobic, butyrate-producing gut bacterium isolated from human faeces and reclassification of Eubacterium desmolans as Agathobaculum desmolans comb. nov. Int J Syst Evol Microbiol 2016;66:3656–61.
[2]. Alciati A, Sgiarovello P, Atzeni F, Sarzi-Puttini P. Psychiatric problems in
fibromyalgia: clinical and neurobiological links between mood disorders and
fibromyalgia. Reumatismo 2012;64:268–74.
[3]. Amir I, Bouvet P, Legeay C, Gophna U, Weinberger A. Eisenbergiella tayi gen. nov., sp. nov., isolated from human blood. Int J Syst Evol Microbiol 2014;64:907–14.
[4]. Anders S, McCarthy DJ, Chen Y, Okoniewski M, Smyth GK, Huber W, Robinson MD. Count-based differential expression analysis of RNA sequencing data using R and Bioconductor. Nat Protoc 2013;8:1765–86.
[5]. Arora HC, Eng C, Shoskes DA. Gut
microbiome and chronic prostatitis/chronic pelvic pain syndrome. Ann Transl Med 2017;5:30.
[6]. Attar N.
Microbiome: good for the gut, good for the brain. Nat Rev Microbiol 2016;14:269.
[7]. Bauer KC, Huus KE, Finlay BB. Microbes and the mind: emerging hallmarks of the gut microbiota-brain axis. Cell Microbiol 2016;18:632–44.
[8]. Bloem LM, Storbeck KH, Schloms L, Swart AC. 11β-hydroxyandrostenedione returns to the steroid arena: biosynthesis, metabolism and function. Molecules 2013;18:13228–44.
[9]. Braundmeier-Fleming A, Russell NT, Yang W, Nas MY, Yaggie RE, Berry M, Bachrach L, Flury SC, Marko DS, Bushell CB, Welge ME, White BA, Schaeffer AJ, Klumpp DJ. Stool-based biomarkers of interstitial cystitis/bladder pain syndrome. Sci Rep 2016;6:26083.
[10]. Bui TPN, Shetty SA, Lagkouvardos I, Ritari J, Chamlagain B, Douillard FP, Paulin L, Piironen V, Clavel T, Plugge CM, de Vos WM. Comparative genomics and physiology of the butyrate-producing bacterium Intestinimonas butyriciproducens. Environ Microbiol Rep 2016;8:1024–37.
[11]. Burckhardt CS, Clark SR, Bennett RM. The
fibromyalgia impact questionnaire: development and validation. J Rheumatol 1991;18:728–33.
[12]. Carlier JP, Bedora-Faure M, K’ouas G, Alauzet C, Mory F. Proposal to unify Clostridium orbiscindens Winter et al. 1991 and Eubacterium plautii (Séguin 1928) Hofstad and Aasjord 1982, with description of Flavonifractor plautii gen. nov., comb. nov., and reassignment of Bacteroides capillosus to Pseudoflavonifractor capillosus gen. nov., comb. nov. Int J Syst Evol Microbiol 2010;60:585–90.
[13]. Chahoud M, Chahine R, Salameh P, Sauleau EA. Reliability, factor analysis and internal consistency calculation of the Insomnia Severity Index (ISI) in French and in English among Lebanese adolescents. eNeurologicalSci 2017;7:9–14.
[14]. Clauw DJ.
Fibromyalgia: a clinical review. JAMA 2014;311:1547–55.
[15]. Cryan JF, Dinan TG. Mind-altering microorganisms: the impact of the gut microbiota on brain and behaviour. Nat Rev Neurosci 2012;13:701–12.
[16]. De Palma G, Lynch MDJ, Lu J, Dang VT, Deng Y, Jury J, Umeh G, Miranda PM, Pigrau Pastor M, Sidani S, Pinto-Sanchez MI, Philip V, McLean PG, Hagelsieb MG, Surette MG, Bergonzelli GE, Verdu EF, Britz-McKibbin P, Neufeld JD, Collins SM, Bercik P. Transplantation of fecal microbiota from patients with irritable bowel syndrome alters gut function and behavior in recipient mice. Sci Transl Med 2017;9.
[17]. Deng L, Chang D, Foshaug RR, Eisner R, Tso VK, Wishart DS, Fedorak RN. Development and validation of a high-throughput mass spectrometry based urine metabolomic test for the detection of colonic adenomatous polyps. Metabolites 2017;7.
[18]. Devendran S, Méndez-García C, Ridlon JM. Identification and characterization of a 20β-HSDH from the anaerobic gut bacterium Butyricicoccus desmolans ATCC 43058. J Lipid Res 2017;58:916–25.
[19]. Distrutti E, Monaldi L, Ricci P, Fiorucci S. Gut microbiota role in irritable bowel syndrome: new therapeutic strategies. World J Gastroenterol 2016;22:2219–41.
[20]. Eller-Smith OC, Nicol AL, Christianson JA. Potential mechanisms underlying centralized pain and emerging therapeutic interventions. Front Cell Neurosci 2018;12:35.
[21]. Felice VD, Quigley EM, Sullivan AM, O'Keeffe GW, O'Mahony SM. Microbiota-gut-brain signalling in Parkinson's disease: implications for non-motor symptoms. Parkinsonism Relat Disord 2016.
[22]. Fitzcharles MA, Boulos P. Inaccuracy in the diagnosis of
fibromyalgia syndrome: analysis of referrals. Rheumatol (Oxford) 2003;42:263–7.
[23]. Fitzcharles MA, Ste-Marie PA, Panopalis P, Ménard H, Shir Y, Wolfe F. The 2010 American College of Rheumatology
fibromyalgia survey diagnostic criteria and symptom severity scale is a valid and reliable tool in a French speaking
fibromyalgia cohort. BMC Musculoskelet Disord 2012;13:179.
[24]. Forsythe P, Kunze W, Bienenstock J. Moody microbes or fecal phrenology: what do we know about the microbiota-gut-brain axis? BMC Med 2016;14:58.
[25]. Frémont M, Coomans D, Massart S, De Meirleir K. High-throughput 16S rRNA gene sequencing reveals alterations of intestinal microbiota in myalgic encephalomyelitis/chronic fatigue syndrome patients. Anaerobe 2013;22:50–6.
[26]. Friedman J, Hastie T, Tibshirani R. Regularization paths for generalized linear models via coordinate descent. J Stat Softw 2010;33:1–22.
[27]. Giloteaux L, Goodrich JK, Walters WA, Levine SM, Ley RE, Hanson MR. Reduced diversity and altered composition of the gut
microbiome in individuals with myalgic encephalomyelitis/chronic fatigue syndrome.
Microbiome 2016;4:30.
[28]. Goebel A, Buhner S, Schedel R, Lochs H, Sprotte G. Altered intestinal permeability in patients with primary
fibromyalgia and in patients with complex regional pain syndrome. Rheumatol (Oxford) 2008;47:1223–7.
[29]. Gonzalez E, Pitre FE, Pagé AP, Marleau J, Guidi Nissim W, St-Arnaud M, Labrecque M, Joly S, Yergeau E, Brereton NJB. Trees, fungi and bacteria: tripartite metatranscriptomics of a root
microbiome responding to soil contamination.
Microbiome 2018;6:53.
[30]. Häuser W, Ablin J, Fitzcharles MA, Littlejohn G, Luciano JV, Usui C, Walitt B.
Fibromyalgia. Nat Rev Dis Primer 2015;1:15022.
[31]. Häuser W, Fitzcharles MA. Facts and myths pertaining to
fibromyalgia. Dialogues Clin Neurosci 2018;20:53–62.
[32]. Hernandez CJ. The
microbiome and bone and joint disease. Curr Rheumatol Rep 2017;19:77.
[33]. Human
Microbiome Project Consortium. Structure, function and diversity of the healthy human
microbiome. Nature 2012;486:207–14.
[34]. Jaggi M. An Equivalence between the Lasso and Support Vector Machines. ArXiv13031152 Cs Stat, 2013. Available:
http://arxiv.org/abs/1303.1152. Accessed February 20, 2019.
[35]. Kang M, Mischel RA, Bhave S, Komla E, Cho A, Huang C, Dewey WL, Akbarali HI. The effect of gut
microbiome on tolerance to morphine mediated antinociception in mice. Sci Rep 2017;7:42658.
[36]. Kim D, Kim WU. Editorial: can prevotella copri Be a causative pathobiont in rheumatoid arthritis? Arthritis Rheumatol 2016;68:2565–7.
[37]. Kläring K, Hanske L, Bui N, Charrier C, Blaut M, Haller D, Plugge CM, Clavel T. Intestinimonas butyriciproducens gen. nov., sp. nov., a butyrate-producing bacterium from the mouse intestine. Int J Syst Evol Microbiol 2013;63:4606–12.
[38]. Klindworth A, Pruesse E, Schweer T, Peplies J, Quast C, Horn M, Glöckner FO. Evaluation of general 16S ribosomal RNA gene PCR primers for classical and next-generation sequencing-based diversity studies. Nucleic Acids Res 2013;41:e1.
[39]. Krebs-Smith SM, Pannucci TE, Subar AF, Kirkpatrick SI, Lerman JL, Tooze JA, Wilson MM, Reedy J. Update of the healthy eating index: HEI-2015. J Acad Nutr Diet 2018;118:1591–602.
[40]. Lach G, Schellekens H, Dinan TG, Cryan JF. Anxiety, depression, and the
microbiome: a role for gut peptides. Neurotherapeutics 2018;15:36–59.
[41]. Langmead B, Salzberg SL. Fast gapped-read alignment with Bowtie 2. Nat Methods 2012;9:357–9.
[42]. Larsen JM. The immune response to Prevotella bacteria in chronic inflammatory disease. Immunology 2017;151:363–74.
[43]. Lohse M, Bolger AM, Nagel A, Fernie AR, Lunn JE, Stitt M, Usadel B. RobiNA: a user-friendly, integrated software solution for RNA-Seq-based transcriptomics. Nucleic Acids Res 2012;40:W622–627.
[44]. Longstreth GF, Thompson WG, Chey WD, Houghton LA, Mearin F, Spiller RC. Functional bowel disorders. Gastroenterology 2006;130:1480–91.
[45]. Lopez-Siles M, Duncan SH, Garcia-Gil LJ, Martinez-Medina M. Faecalibacterium prausnitzii: from microbiology to diagnostics and prognostics. ISME J 2017;11:841–52.
[46]. Love MI, Anders S, Kim V, Huber W. RNA-Seq workflow: gene-level exploratory analysis and differential expression. F1000Res 2015;4:1070.
[47]. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol 2014;15:550.
[48]. Luczynski P, Tramullas M, Viola M, Shanahan F, Clarke G, O'Mahony S, Dinan TG, Cryan JF. Microbiota regulates visceral pain in the mouse. Elife 2017;6.
[49]. Lynch SV, Pedersen O. The human intestinal
microbiome in health and disease. N Engl J Med 2016;375:2369–79.
[50]. Maeda Y, Kurakawa T, Umemoto E, Motooka D, Ito Y, Gotoh K, Hirota K, Matsushita M, Furuta Y, Narazaki M, Sakaguchi N, Kayama H, Nakamura S, Iida T, Saeki Y, Kumanogoh A, Sakaguchi S, Takeda K. Dysbiosis contributes to arthritis development via activation of autoreactive T cells in the intestine. Arthritis Rheumatol 2016;68:2646–61.
[51]. Malatji BG, Meyer H, Mason S, Engelke UFH, Wevers RA, van Reenen M, Reinecke CJ. A diagnostic biomarker profile for
fibromyalgia syndrome based on an NMR metabolomics study of selected patients and controls. BMC Neurol 2017;17:88.
[52]. Mansfield KE, Sim J, Croft P, Jordan KP. Identifying patients with chronic
widespread pain in primary care. PAIN 2017;158:110–19.
[53]. Mayer EA, Tillisch K, Gupta A. Gut/brain axis and the microbiota. J Clin Invest 2015;125:926–38.
[54]. McMurdie PJ, Holmes S. phyloseq: an R package for reproducible interactive analysis and graphics of
microbiome census data. PLoS One 2013;8:e61217.
[55]. Mearin F, Lacy BE, Chang L, Chey WD, Lembo AJ, Simren M, Spiller R. Bowel disorders. Gastroenterology 2016.
[56]. Mifflin MD, St Jeor ST, Hill LA, Scott BJ, Daugherty SA, Koh YO. A new predictive equation for resting energy expenditure in healthy individuals. Am J Clin Nutr 1990;51:241–7.
[57]. Miquel S, Martín R, Lashermes A, Gillet M, Meleine M, Gelot A, Eschalier A, Ardid D, Bermúdez-Humarán LG, Sokol H, Thomas M, Theodorou V, Langella P, Carvalho FA. Anti-nociceptive effect of Faecalibacterium prausnitzii in non-inflammatory IBS-like models. Sci Rep 2016;6:19399.
[58]. Morin CM. Insomnia: psychological assessment and management. New York: Guilford Press, 1993.
[59]. Morris DJ, Ridlon JM. Glucocorticoids and gut bacteria: “The GALF Hypothesis” in the metagenomic era. Steroids 2017;125:1–13.
[60]. Nagy-Szakal D, Williams BL, Mishra N, Che X, Lee B, Bateman L, Klimas NG, Komaroff AL, Levine S, Montoya JG, Peterson DL, Ramanan D, Jain K, Eddy ML, Hornig M, Lipkin WI. Fecal metagenomic profiles in subgroups of patients with myalgic encephalomyelitis/chronic fatigue syndrome.
Microbiome 2017;5:44.
[61]. Newberry F, Hsieh SY, Wileman T, Carding SR. Does the
microbiome and virome contribute to myalgic encephalomyelitis/chronic fatigue syndrome? Clin Sci (Lond) 2018;132:523–42.
[62]. Oksanen J, Blanchet FG, Friendly M, Kindt R, Legendre P, McGlinn D, Minchin PR, O'Hara RB, Simpson GL, Solymos P, Stevens MHH, Szoecs E, Wagner H. Vegan: community ecology package. 2019. Available at:
https://CRAN.R-project.org/package=vegan. Accessed February 7, 2019.
[63]. Olson CA, Vuong HE, Yano JM, Liang QY, Nusbaum DJ, Hsiao EY. The gut microbiota mediates the anti-seizure effects of the ketogenic diet. Cell 2018;173:1728–41.e13.
[64]. O'Mahony SM, Felice VD, Nally K, Savignac HM, Claesson MJ, Scully P, Woznicki J, Hyland NP, Shanahan F, Quigley EM, Marchesi JR, O'Toole PW, Dinan TG, Cryan JF. Disturbance of the gut microbiota in early-life selectively affects visceral pain in adulthood without impacting cognitive or anxiety-related behaviors in male rats. Neuroscience 2014;277:885–901.
[65]. O'Mahony SM, Dinan TG, Cryan JF. The gut microbiota as a key regulator of visceral pain. PAIN 2017;158(suppl 1):S19–28.
[66]. Park Y, Dodd KW, Kipnis V, Thompson FE, Potischman N, Schoeller DA, Baer DJ, Midthune D, Troiano RP, Bowles H, Subar AF. Comparison of self-reported dietary intakes from the Automated Self-Administered 24-h recall, 4-d food records, and food-frequency questionnaires against recovery biomarkers. Am J Clin Nutr 2018;107:80–93.
[67]. Pasolli E, Truong DT, Malik F, Waldron L, Segata N. Machine learning meta-analysis of large metagenomic datasets: tools and biological insights. PLoS Comput Biol 2016;12:e1004977.
[68]. Perrot S, Dumont D, Guillemin F, Pouchot J, Coste J; French Group for Quality of Life Research. Quality of life in women with
fibromyalgia syndrome: validation of the QIF, the French version of the
fibromyalgia impact questionnaire. J Rheumatol 2003;30:1054–9.
[69]. Peter J, Fournier C, Durdevic M, Knoblich L, Keip B, Dejaco C, Trauner M, Moser G. A microbial signature of psychological distress in irritable bowel syndrome. Psychosom Med 2018;80:698–709.
[70]. Pozuelo M, Panda S, Santiago A, Mendez S, Accarino A, Santos J, Guarner F, Azpiroz F, Manichanh C. Reduction of butyrate- and methane-producing microorganisms in patients with Irritable Bowel Syndrome. Sci Rep 2015;5:12693.
[71]. Rahman A, Underwood M, Carnes D.
Fibromyalgia. BMJ 2014;348:g1224.
[72]. Ridlon JM, Ikegawa S, Alves JMP, Zhou B, Kobayashi A, Iida T, Mitamura K, Tanabe G, Serrano M, De Guzman A, Cooper P, Buck GA, Hylemon PB. Clostridium scindens: a human gut microbe with a high potential to convert glucocorticoids into androgens. J Lipid Res 2013;54:2437–49.
[73]. Robin X, Turck N, Hainard A, Tiberti N, Lisacek F, Sanchez JC, Müller M. pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics 2011;12:77.
[74]. Schloss PD, Westcott SL, Ryabin T, Hall JR, Hartmann M, Hollister EB, Lesniewski RA, Oakley BB, Parks DH, Robinson CJ, Sahl JW, Stres B, Thallinger GG, Van Horn DJ, Weber CF. Introducing mothur: open-source, platform-independent, community-supported software for describing and comparing microbial communities. Appl Environ Microbiol 2009;75:7537–41.
[75]. Segata N, Waldron L, Ballarini A, Narasimhan V, Jousson O, Huttenhower C. Metagenomic microbial community profiling using unique clade-specific marker genes. Nat Methods 2012;9:811–14.
[76]. Shen S, Lim G, You Z, Ding W, Huang P, Ran C, Doheny J, Caravan P, Tate S, Hu K, Kim H, McCabe M, Huang B, Xie Z, Kwon D, Chen L, Mao J. Gut microbiota is critical for the induction of chemotherapy-induced pain. Nat Neurosci 2017;20:1213–6.
[77]. Shoskes DA, Wang H, Polackwich AS, Tucky B, Altemus J, Eng C. Analysis of gut
microbiome reveals significant differences between men with chronic prostatitis/chronic pelvic pain syndrome and controls. J Urol 2016;196:435–41.
[78]. Sitkin S, Pokrotnieks J. Clinical potential of anti-inflammatory effects of faecalibacterium prausnitzii and butyrate in inflammatory bowel disease. Inflamm Bowel Dis 2018;14:e40–e41.
[79]. Sluka KA, Clauw DJ. Neurobiology of
fibromyalgia and chronic
widespread pain. Neuroscience 2016;338:114–129.
[80]. Swart AC, Storbeck KH. 11β-Hydroxyandrostenedione: downstream metabolism by 11βHSD, 17βHSD and SRD5A produces novel substrates in familiar pathways. Mol Cell Endocrinol 2015;408:114–23.
[81]. Takada T, Watanabe K, Makino H, Kushiro A. Reclassification of Eubacterium desmolans as Butyricicoccus desmolans comb. nov., and description of Butyricicoccus faecihominis sp. nov., a butyrate-producing bacterium from human faeces. Int J Syst Evol Microbiol 2016;66:4125–31.
[82]. Tap J, Derrien M, Törnblom H, Brazeilles R, Cools-Portier S, Doré J, Störsrud S, Le Nevé B, Öhman L, Simrén M. Identification of an intestinal microbiota signature associated with severity of irritable bowel syndrome. Gastroenterology 2017;152:111–23.e8.
[83]. Theodorou V, Ait Belgnaoui A, Agostini S, Eutamene H. Effect of commensals and probiotics on visceral sensitivity and pain in irritable bowel syndrome. Gut Microbes 2014;5:430–6.
[84]. Thorsen J, Brejnrod A, Mortensen M, Rasmussen MA, Stokholm J, Al-Soud WA, Sørensen S, Bisgaard H, Waage J. Large-scale benchmarking reveals false discoveries and count transformation sensitivity in 16S rRNA gene amplicon data analysis methods used in
microbiome studies.
Microbiome 2016;4:62.
[85]. Tibshirani R. Regression shrinkage and selection via the lasso. J R Stat Soc Ser B Methodol 1996;58:267–88.
[86]. Timon CM, van den Barg R, Blain RJ, Kehoe L, Evans K, Walton J, Flynn A, Gibney ER. A review of the design and validation of web- and computer-based 24-h dietary recall tools. Nutr Res Rev 2016;29:268–80.
[87]. Togo AH, Khelaifia S, Bittar F, Maraninchi M, Raoult D, Million M. “Eisenbergiella massiliensis”, a new species isolated from human stool collected after bariatric surgery. New Microbes New Infect 2016;13:15–16.
[88]. Vol S, Bedouet M, Gusto G, Leglu C, Beslin E, Decou P, Nègre E, Planage B, Chazelle E, Mercier F, Lantieri O, Tichet J. Evaluating physical activity: the AQAP questionnaire and its interpretation software. Ann Phys Rehabil Med 2011;54:478–95.
[89]. Walitt B, Katz RS, Bergman MJ, Wolfe F. Three-quarters of persons in the US population reporting a clinical diagnosis of
fibromyalgia do not satisfy
fibromyalgia criteria: the 2012 National Health Interview Survey. PLoS One 2016;11:e0157235.
[90]. Wei T, Simko V, Levy M, Xie Y, Jin Y, Zemla J. corrplot: Visualization of a Correlation Matrix. 2017. Available at:
https://CRAN.R-project.org/package=corrplot. Accessed February 7, 2019.
[91]. Weiss S, Xu ZZ, Peddada S, Amir A, Bittinger K, Gonzalez A, Lozupone C, Zaneveld JR, Vázquez-Baeza Y, Birmingham A, Hyde ER, Knight R. Normalization and microbial differential abundance strategies depend upon data characteristics.
Microbiome 2017;5:27.
[92]. Wen C, Zheng Z, Shao T, Liu L, Xie Z, Le Chatelier E, He Z, Zhong W, Fan Y, Zhang L, Li H, Wu C, Hu C, Xu Q, Zhou J, Cai S, Wang D, Huang Y, Breban M, Qin N, Ehrlich SD. Quantitative metagenomics reveals unique gut
microbiome biomarkers in ankylosing spondylitis. Genome Biol 2017;18:142.
[93]. Wolfe F, Clauw DJ, Fitzcharles MA, Goldenberg DL, Häuser W, Katz RL, Mease PJ, Russell AS, Russell IJ, Walitt B. 2016 revisions to the 2010/2011
fibromyalgia diagnostic criteria. Semin Arthritis Rheum 2016;46:319–29.
[94]. Zhang X, Zhang D, Jia H, Feng Q, Wang D, Liang D, Wu X, Li J, Tang L, Li Y, Lan Z, Chen B, Li Y, Zhong H, Xie H, Jie Z, Chen W, Tang S, Xu X, Wang X, Cai X, Liu S, Xia Y, Li J, Qiao X, Al-Aama JY, Chen H, Wang L, Wu QJ, Zhang F, Zheng W, Li Y, Zhang M, Luo G, Xue W, Xiao L, Li J, Chen W, Xu X, Yin Y, Yang H, Wang J, Kristiansen K, Liu L, Li T, Huang Q, Li Y, Wang J. The oral and gut microbiomes are perturbed in rheumatoid arthritis and partly normalized after treatment. Nat Med 2015;21:895–905.
[95]. Zhao Y, Chen B, Li S, Yang L, Zhu D, Wang Y, Wang H, Wang T, Shi B, Gai Z, Yang J, Heng X, Yang J, Zhang L. Detection and characterization of bacterial nucleic acids in culture-negative synovial tissue and fluid samples from rheumatoid arthritis or osteoarthritis patients. Sci Rep 2018;8:14305.
[96]. Zhou XY, Li M, Li X, Long X, Zuo XL, Hou XH, Cong YZ, Li YQ. Visceral hypersensitive rats share common dysbiosis features with irritable bowel syndrome patients. World J Gastroenterol 2016;22:5211–27.
[97]. Zhuang X, Xiong L, Li L, Li M, Chen M. Alterations of gut microbiota in patients with irritable bowel syndrome: a systematic review and meta-analysis. J Gastroenterol Hepatol 2017;32:28–38.
[98]. Zuo Y, Cui Y, Yu G, Li R, Ressom HW. Incorporating prior biological knowledge for network-based differential gene expression analysis using differentially weighted graphical LASSO. BMC Bioinformatics 2017;18:99.
