Short-Chain Fatty Acid-Producing Gut Microbiota Is Decreased in Parkinson’s Disease but Not in Rapid-Eye-Movement Sleep Behavior Disorder

Twenty studies on gut microbiota in PD have been reported, whereas only one study has been reported on iRBD from Germany. iRBD has the highest likelihood ratio to develop PD. Our meta-analysis of iRBD in Japan and Germany revealed increased mucin-layer-degrading genus Akkermansia in iRBD.

German data set (32,43). We here performed 16S rRNA sequencing (16S rRNA-seq) analysis of 26 iRBD patients and 137 controls. We also meta-analyzed our data set with the German data set using a nonparametric meta-analysis method that we developed previously to identify shared taxonomic changes between the two countries and compared iRBD-associated taxonomic changes in two countries with PD-associated changes in five countries.

RESULTS
Differences in demographic and clinical features, diet, and medications between controls and iRBD in our data set. All iRBD patients were diagnosed according to the International Classification of Sleep Disorders Criteria-Third Edition (57). To search for possible confounding factors, we compared four features (age, sex, BMI, and constipation) between iRBD and controls in our data set (Table 1). Compared to controls, iRBD patients had higher ages and higher BMI and included more males. Similarly, the ratio of constipation was higher in iRBD patients. The Unified Parkinson's Disease Rating Scale (UPDRS) and Mini-Mental Status Examination (MMSE) scores indicated lack of Parkinsonian symptoms and lack of cognitive deficits ( Table 2). iRBD patients had similar autonomic dysfunctions as PD patients in our cohort (32) (Table 2). iRBD and PD patients took proton pump inhibitors (PPI) more frequently than controls (Table 3). PD patients drank coffee less frequently than controls (Table 3), as has been previously reported (58)(59)(60).
PERMANOVA to evaluate the differences in the overall composition of gut microbiota between controls and iRBD, as well as between iRBD and Hoehn and Yahr 1 scale of PD. We performed 16S rRNA-seq analysis of gut microbiota in 26 patients with iRBD and 137 healthy controls. Permutational multivariate analysis of variance (PERMANOVA) compares the overall bacterial compositions between two groups by either taking into account or not the effects of possible confounding factors (61). PERMANOVA manages the effect of controls versus iRBD, and the effects of possible confounding factors, at an equal level. We first compared the overall bacterial compositions between controls and iRBD without considering possible confounding factors and found that the overall bacterial compositions were statistically different by all three distance metrics (Table 4A). PERMANOVA including the possible confounding factors showed that the difference was not accounted for by sex, BMI, constipation, or PPI (Table 4B). Age had an equivocal effect with one significant and two nonsignificant distance metrics (Table 4B). We next compared the overall bacterial compositions between iRBD and Hoehn and Yahr 1 of PD without considering possible confounding factors and found that the overall bacterial compositions were statistically different by two distance metrics (Table 4C). PERMANOVA including the possible confounding factors showed that the difference was not accounted for by age, sex, BMI, constipation, or PPI (Table 4D). Although these possible confounding factors had no essential effect on the overall bacterial composition, we took into account the effects of these possible confounding factors on individual genera and families in the following analysis.
PCoA plot to analyze the overall composition of gut microbiota in controls, iRBD, and PD. We conducted principal-coordinate analysis (PCoA) of gut microbiota in controls and iRBD in our data set, as well as gut microbiota in our previously reported PD subjects (Hoehn and Yahr scales 1 to 5) (32). The centers of gravity moved from the upper left to the lower right with disease progression from controls, to iRBD, to Hoehn and Yahr scales 1 to 5 (Fig. 1A). iRBD was positioned close to the mildest form of PD with Hoehn and Yahr scale 1.
LIGER analysis to reveal unsupervised enterotypes in a combined data set of controls, iRBD, and PD. We applied LIGER that was developed for topic model-based single-cell RNA sequencing (RNA-seq) analysis (62) to make unsupervised clustering of gut microbiota in controls, iRBD, and PD (Hoehn and Yahr scales 1 to 5). Each cluster should represent an enterotype. LIGER revealed four enterotypes A, B, C, and D (Fig. 1B). Examination of the proportion of controls, iRBD, and Hoehn and Yahr scales 1 to 5 in each enterotype revealed that the proportion of controls was decreased in the order of enterotypes A to D, while the proportions of Hoehn and Yahr scales 3 to 5 were increased in the same order (Fig. 1C). The proportion of iRBD was the highest in  The first, second, and third items indicate the number of cases (ratio) who take the indicated food zero times a week, once or twice a week, and three or more times a week, respectively. c The first, second, and third items indicate the number of cases (ratio) who take the indicated alcohol zero times a week, once to three times a week, and four or more times a week, respectively. d P values are calculated by Fisher's exact test. *, P value , 0.05. enterotype A. In factorization by LIGER, the first factor contributes most to differentiate enterotypes A to D, and genera with high loadings in the first factor are major determinants of enterotypes. Color-coding of the first factor in each subject showed a gradual decrease of the first factor from enterotypes A to D (Fig. 1D). The top 10 genera with the highest loadings in the first factor are indicated in Table S1 in the supplemental material. It was interesting that, among the 10 genera, nine produce SCFA and one putatively produces SCFA (Lachnospiraceae ND3007 group). Scatterplots of the 10 genera in each enterotype showed that the abundances of these genera were also decreased in the order of enterotypes A to D (see Fig. S1 in the supplemental material). Among the 10 genera, Faecalibacterium, Roseburia, and Lachnospiraceae ND3007 group were exactly the three genera that were decreased in PD in our previous meta-analysis of five countries (32). To summarize, unsupervised clustering of enterotypes revealed that enterotypes were shifted from A to D with transition from control, to iRBD, to Hoehn and Yahr scales 1 to 5 of PD and that SCFA-producing genera were decreased from enterotypes A to D.
Analysis of each taxon between controls and iRBD in our data set. We examined taxonomic differences between controls and iRBD in our data set using Analysis of Composition of Microbiomes (ANCOM) (63) and Wilcoxon rank sum test (Table S2a at  the genus level and Table S2b at the family level in the supplemental material). ANCOM was developed to reduce false discoveries by exploiting microbial compositional constraints (63,64). The analyses revealed that seven genera were increased in iRBD (Ruminococcus 2, Alistipes, Akkermansia, Ruminococcaceae UCG-005, Ruminococcaceae UCG-004, [Eubacterium] coprostanoligenes group, and Family XIII AD3011 group), two families were increased in iRBD (Rikenellaceae and Akkermansiaceae), and no genera or families were decreased in iRBD. To adjust for the environmental and dietary factors, we next examined taxonomic differences in eight pairs of iRBD patients and their spouses in our data set (Table S3a at the genus level and Table S3b at the family level in the supplemental material). Out of the nine significantly changed taxa in the unpaired analysis above, eight  taxa were similarly changed in the paired analysis, although no statistical significance was observed. We next compared the results of iRBD-ANCOM with those of previously reported PD-ANCOM (32). Nine genera in iRBD and 24 genera in PD were increased with W . 0.5 Â N (see Materials and Methods). Among them, seven genera (Alistipes, Akkermansia, Ruminococcaceae UCG-005, Ruminococcaceae UCG-004, Family XIII AD3011 group, Ruminococcaceae_anonymous, and Oscillibacter) were increased in both iRBD and PD. Similarly, two genera in iRBD and 27 genera in PD were decreased with W . 0.5 Â N. No genera, however, were shared between iRBD and PD.
Possible confounding factors in our data set for nine taxa that were significantly changed in iRBD in our data set. We next asked whether any of the nine taxonomic changes in iRBD were due to confounding factors. We thus performed Generalized Linear Mixed Model (GLMM) analysis with constipation, BMI, sex, age, and PPI. We found that five genera (Ruminococcus 2, Alistipes, Akkermansia, Ruminococcaceae UCG-004, and Family XIII AD3011 group) and two families (Rikenellaceae and Akkermansiaceae) were changed in iRBD after adjusting for constipation, BMI, sex, age, and PPI ( Fig. 2 Table S4 in the supplemental material). In contrast, two genera (Ruminococcaceae UCG-005 and Family XIII AD3011 group) were increased by age ( Fig. 2 and underlines in Table S4 in the supplemental material). Three genera (Akkermansia, Ruminococcaceae UCG-004, and Family XIII AD3011 group) and one family (Akkermansiaceae) were decreased by BMI ( Fig. 2 and underlines in Table S4 in the supplemental material). Two genera (Ruminococcaceae UCG-004 and Family XIII AD3011 group) were increased by constipation ( Fig. 2 and underlines in Table S4 in the supplemental material).

and bold letters in
Meta-analysis of the Japanese and German data sets. Meta-analysis of gut microbiota was performed using the Japanese and German data sets (43). The effect size and relative abundance of 132 genera and 39 families are collated in Table S5a and b, respectively, in the supplemental material. Our putative criteria (I 2 , 25% and P values of both fixed-effects model [FEM] and random-effects model [REM] after Bonferroni correction of ,0.05) showed that four genera (Ruminococcaceae UCG-004, Alistipes, Family XIII AD3011 group, and Akkermansia) and two families (Rikenellaceae and Akkermansiaceae) were increased in iRBD ( Fig. 3 and Table S6 in the supplemental material). These six taxa were a subset of the seven taxa that were significantly changed in iRBD after adjusting for confounding factors in our data set ( Fig. 2 Table S4 in the supplemental material). Among the seven taxa, genus Ruminococcus 2 was increased in Japan but not in Germany and was excluded from forest plots (Fig. 3A). Forest plots of the six taxa in iRBD in two countries along with those in PD in five countries showed that all taxa tended to be increased in PD, and the most homogenous and significant increases were observed in genus Akkermansia and family Akkermansiaceae (Fig. 3A).

and bold letters in
We previously reported that two SCFA-producing genera (Faecalibacterium and Roseburia) and one putative SCFA-producing genus (Lachnospiraceae ND3007 group) were decreased in PD across countries (32). We assumed that genus Lachnospiraceae ND3007 group is a putative SCFA producer, because most genera in family Lachnospiraceae produce SCFA. None of the three recognized or putative SCFA-producing genera were decreased in iRBD in our meta-analysis. However, forest plots of the three genera showed that genus Faecalibacterium tended to be decreased in iRBD, but genera Roseburia and Lachnospiraceae ND3007 group were not (Fig. 3B).
Relative abundances of four genera in progression of a-synucleinopathy. Plots of relative abundances of genera Akkermansia, Faecalibacterium, Roseburia, and Lachnospiraceae ND3007 group in controls, iRBD, and Hoehn and Yahr scales 1 to 5 showed that genus Akkermansia gradually increased and genera Faecalibacterium, Roseburia, and Lachnospiraceae ND3007 group gradually decreased with progression of a-synucleinopathy (Fig. 4). Comparison of controls and iRBD showed that genus Akkermansia was significantly increased in iRBD. In contrast, genus Faecalibacterium, but not Roseburia or Lachnospiraceae ND3007 group, tended to be decreased in iRBD (Fig. 4).

DISCUSSION
We analyzed gut microbiota in iRBD in our data set and meta-analyzed the Japanese and German data sets (43). We first analyzed the overall compositions of gut microbiota in controls and iRBD by PERMANOVA. Evaluation of the effects of possible confounding factors by PERMANOVA showed that the overall compositions of gut microbiota were statistically different between controls and iRBD by all three distance metrics, and the difference may or may not be affected by age but not by sex, BMI, constipation, or PPI (Table 4). PERMANOVA similarly showed that the overall compositions of gut microbiota were statistically different between iRBD and Hoehn and Yahr scale 1 by Chao and weighted UniFrac but not by unweighted UniFrac (Table 4). Again, the difference was not due to confounding factors. Weighted UniFrac takes read counts into consideration to calculate the distance so that the effects of low-abundance taxa become small, whereas low-abundance taxa have more effects on unweighted UniFrac (65). Thus, iRBD and Hoehn and Yahr scale 1 may have large differences in major taxa but not in minor taxa.
We additionally observed by PCoA that the overall compositions of gut microbiota were gradually changing in controls, iRBD, and Hoehn and Yahr scales 1 to 5 in this order (Fig. 1A). We next applied LIGER (62) to the 16S rRNA-seq analysis for the first time. LIGER, which was developed for single-cell RNA-seq analysis, enables integrative nonnegative matrix factorization (iNMF) by exploiting a topic model. Topic modeling that has been developed for text mining generally fits well to the analysis of gut microbiota (66,67). LIGER revealed four enterotypes in controls, iRBD, and Hoehn and Yahr scales 1 to 5 in an unsupervised manner (Fig. 1B). Enterotypes were shifted with transition from control, to iRBD, to Hoehn and Yahr scales 1 to 5 (Fig. 1C). SCFA-producing genera were similarly decreased with the shift in enterotypes (see Fig. S1 in the supplemental material). We showed that genus Akkermansia was increased in iRBD in this communication, as well as in our previous meta-analysis of PD (32). The increase of genus Akkermansia in PD was also addressed in previous reports (34,35,38,39,45,(47)(48)(49) that could not be included in our previous meta-analysis of PD (32). Some reports, however, did not address the changes in Akkermansia in PD (31, 40-42, 44, 46). Genus Akkermansia, however, was not detected in factorization by LIGER. Genus Akkermansia was likely to be underestimated by LIGER, because multiple SCFA-producing genera were coordinately decreased in PD, whereas genus Akkermansia was increased alone without any accompanying genera, which reduced the chance of detecting genus Akkermansia by topic modeling by LIGER. Both PCoA and LIGER indicate that gut dysbiosis advances with progression of a-synucleinopathy. Alternatively, patients with a-synucleinopathy with marked gut dysbiosis may progress faster than those with mild gut dysbiosis.
Analysis of individual taxa by ANCOM and the Wilcoxon rank sum test revealed that seven genera (see Table S2a in the supplemental material) and two families (see Table S2b in the supplemental material) were increased in iRBD. Adjustment for possible confounding factors for the seven genera and two families by GLMM showed that  Table S1 in the supplemental material), individuals colored in blue carry a high proportion of SCFA-producing bacteria.
increases of five genera (Ruminococcus 2, Alistipes, Akkermansia, Ruminococcaceae UCG-004, and Family XIII AD3011 group) and two families (Rikenellaceae and Akkermansiaceae) were indeed accounted for by iRBD (orange arrows in Fig. 2), although age, BMI, and constipation had additional confounding effects on three genera (Akkermansia, Ruminococcaceae UCG-004, and Family XIII AD3011 group) and one family (Akkermansiaceae). Among the five genera and two families that were increased in iRBD, only genus Akkermansia and family Akkermansiaceae were also increased in PD in our meta-analysis of five countries (32).
Meta-analysis of the Japanese and German data sets revealed that four genera (Ruminococcaceae UCG-004, Alistipes, Family XIII AD3011 group, and Akkermansia) and two families (Rikenellaceae and Akkermansiaceae) were increased in iRBD (Fig. 3A and Table S6 in the supplemental material). Among these six taxa, we previously reported  Families increased in iRBD Genera increased in iRBD FIG 2 Generalized linear mixed model (GLMM) analysis to evaluate confounding factors of seven genera and two families that were significantly changed in iRBD compared to controls in our data set. The effects of iRBD, age, sex, body mass index (BMI), constipation, and PPI were individually analyzed by mutually adjusting for confounding factors by GLMM. Arrows indicate taxa that were significantly changed by iRBD (orange arrows), age (red arrows), BMI (green arrows), and constipation (black arrows) after adjusting for the other confounding factors. Upward and downward arrows indicate increased and decreased taxa, respectively. Exact P values are indicated in Table S4 (A and B) Forest plots of four genera (A) and two families (B) that were significantly and homogenously changed in iRBD (Bonferroni-corrected P value , 0.05 and the homogeneity index I 2 , 25%) in the Japanese and German data sets. Forest plots of PD in five data sets are also indicated in parallel.
(C) Forest plots of two recognized and one putative SCFA-producing genera that were significantly and homogenously decreased in PD in five countries in (Continued on next page) that genus Akkermansia and family Akkermansiaceae were consistently increased in PD across countries (32). We found that relative abundances of genus Akkermansia gradually increased from iRBD to Hoehn and Yahr scales 1 to 5 (Fig. 4). Akkermansia muciniphila degrades the mucus layer of the gut (68) and erodes the mucus layer in the lack of dietary fibers (69). Indeed, intestinal permeability is increased in PD (70), and the serum lipopolysaccharide-binding protein levels are decreased in PD (31,70). Reduced expression of a tight junction protein, occludin, in colonic biopsy specimens in PD is similarly in accordance with the reduced mucus layer (71). Increased intestinal permeability may expose the intestinal neural plexus to oxidative stress and pesticide/herbicide (72). which subsequently allows the formation of abnormal a-synuclein aggregates in the intestine. Moreover, in the presence of other gut microbiota, Akkermansia muciniphila in mouse intestine enhances differentiation of follicular T cells, which  Table S2a in the supplemental material) are indicated on the left. Read counts of genera Akkermansia, Faecalibacterium, and Roseburia in Hoehn and Yahr scales 1 to 5 were previously reported, but iRBD was not included (32).

FIG 3 Legend (Continued)
our previous report (32). Forest plots of iRBD in the Japanese and German data sets are also indicated in parallel. An effect size of each data set, as well as the overall effect sizes by the fixed-effects model (FEM) and the random-effects model (REM), is indicated. Both lines and diamonds indicate 95% confidence intervals. Orange and blue symbols represent iRBD and PD, respectively. Exact statistical measures are indicated in Table S6 in the supplemental material.
Meta-analysis of the Japanese and German data sets also showed that no SCFA-producing genera were decreased in iRBD ( Fig. 3B and Table S6 in the supplemental material). We previously reported that three recognized and putative SCFA-producing genera (Faecalibacterium, Roseburia, and Lachnospiraceae ND3007 group) were consistently decreased in PD across countries (32). Although genus Faecalibacterium tended to be decreased in iRBD, no significance was observed (Fig. 4). In contrast, genera Roseburia and Lachnospiraceae ND3007 group were not decreased in iRBD (Fig. 4). Preservation of most of SCFA-producing bacteria in iRBD was also implicated in the LIGER analysis, which showed that both controls and iRBD were enriched in enterotype A (Fig. 1C), in which SCFA-producing bacteria were high (Fig. 1D). Major constituents of gut SCFAs, butyrate and propionate, induce anti-inflammatory regulatory T (Treg) cells by inhibiting histone deacetylase (77,78) and by binding to G protein-coupled receptors of GPR41, GPR43, and GPR109A (79,80). Indeed, in mouse models of PD, SCFAs may (81)(82)(83) or may not (84) have beneficial effects on PD symptoms. In addition, in another German cohort, fecal SCFA concentrations were decreased in PD (35). Our analysis suggests that reduced fecal SCFA concentrations may be a prerequisite for the development of PD but not of iRBD. Reduction of SCFA-producing bacteria culminating in the development of PD may start from genus Faecalibacterium. A decrease of genus Faecalibacterium may thus be a hallmark to predict transition from iRBD to PD. We expect that administration of SCFA and probiotics/prebiotics to increase the intestinal SCFA possibly retards the progression of a-synucleinopathy at the stage of iRBD.

MATERIALS AND METHODS
Patients in our data set. All studies were approved by the Ethical Review Committees of the Nagoya University Graduate School of Medicine (approval no. 2016-0151), Iwate Medical College Hospital (H28-123), Okayama Kyokuto Hospital (kyoIR-2016002), and Fukuoka University School of Medicine (2016M027). We obtained written informed consent from all patients and controls.
We recruited 26 patients with iRBD and 137 healthy controls from four hospitals to participate in this study from September 2015 to February 2018. Among the 137 healthy controls, 8 were spouses of iRBD patients. All iRBD patients were diagnosed by International Classification of Sleep Disorders Criteria-Third Edition (57). The severity of Parkinson's disease was determined according to Hoehn and Yahr scales 1 to 5 (85). Briefly, scale 1 indicates that a patient has only unilateral movement disability. Scale 2 indicates that a patient has bilateral movement disability but no impairment of balance. Scale 3 indicates that a patient has bilateral movement disability and impairment of balance but that his/her daily life is independent. Scale 4 indicates that a patient is severely disabled but manages to walk or stand without assistance. Scale 5 indicates that a patient is confined to bed or a wheelchair unless assisted. Subjects with diabetes mellitus, heart failure, liver cirrhosis, any malignancy, hematological diseases, and autoimmune diseases were excluded from our study. Subjects who had taken any antibiotics in the past 1 month were similarly excluded.
DNA isolation and 16S rRNA V3-V4 sequencing in our data set. The detailed procedures for transportation of a fecal sample from the participant's home to the Nagoya University, freeze-drying of the fecal sample (86), and DNA isolation were described previously (32). The V3-V4 hypervariable region of the bacterial 16S rRNA gene was amplified by primer 341F, 59-CCTACGGGNGGCWGCAG-39 and primer 805R, 59-GACTACHVGGGTATCTAATCC-39. Paired-end sequencing of 300-nucleotide fragments was performed using the MiSeq reagent kit V3 on a MiSeq system (Illumina). Taxonomic analysis was performed with QIIME2 (87). Operational taxonomic units (OTUs) were generated using DADA2, and the SILVA taxonomy database release 132 (88) was used for taxonomic identification.
Differences in demographic and clinical features, diet, and medications between controls and iRBD in our data set. Four demographic and clinical features (age, sex, body mass index [BMI], and constipation), five medications, and 17 kinds of foods were compared between iRBD and controls in our data set using either Student's t test or Fisher's exact test. Subjects with the stool frequency of twice a week or less were defined to be constipated (89).
We examined lack of multicollinearity between iRBD, constipation, BMI, sex, age, and PPI by calculating the variance inflation factor (VIF) using the R package HH version 3.1-40. We verified that the VIFs were all less than 2, indicating that there was no multicollinearity between iRBD, constipation, BMI, sex, age, and PPI.
Analysis of the overall gut microbiota in controls, iRBD, and Hoehn and Yahr 1 scale of PD using PERMANOVA in our data set. Next, we analyzed the effects on the overall composition of gut microbiota of (i) iRBD versus controls and (ii) iRBD versus controls, age, sex, BMI, constipation, and PPI in our data set that was comprised of controls and iRBD using PERMANOVA (61). We similarly analyzed the effects on the overall composition of gut microbiota of (iii) iRBD versus Hoehn and Yahr scale 1 and (iv) iRBD versus Hoehn and Yahr scale 1, age, sex, BMI, constipation, and PPI in a combined data set that was comprised of our current iRBD subjects and PD subjects with Hoehn and Yahr scale 1 in our previous report (32) using PERMANOVA (61). All genera were included in this analysis. The effects were evaluated by three distance metrics of Chao (90), unweighted UniFrac (91), and weighted UniFrac (91). Chao and unweighted/weighted UniFrac distances were calculated using the R package vegan and QIIME2, respectively.
Analysis of the overall gut microbiota in controls, iRBD, and PD using PCoA and LIGER in our data set. For the overall analysis of gut microbiota, PD samples in our previous report (32) were included. We first performed principal-coordinate analysis (PCoA) of each subject, and the centers of gravity and standard errors in seven categories of controls, iRBD, and Hoehn and Yahr scales 1 to 5 were plotted.
We next employed the Linked Inference of Genomic Experimental Relationships (LIGER) (62), which uses integrative nonnegative matrix factorization (iNMF) for single-cell RNA-seq analysis, for unsupervised clustering of gut microbiota of controls, iRBD, and Hoehn and Yahr scales 1 to 5. LIGER enabled us to identify four enterotypes, each of which was comprised of a set of bacteria that were synchronously changed in each subject.
Analysis of each taxon between controls and iRBD in our data set. Taxa were filtered at the genus and family levels using the following conditions. For each taxon, we counted the number of samples in which the relative abundance of the taxon of interest was greater than 1E24. The number of such samples should be 17 or more (more than ;10% of all samples). We thereby chose 50 families and 168 genera.
The difference in the abundance of each taxon between iRBD and controls was analyzed by Analysis of Composition of Microbiomes (ANCOM) (63), as well as by the Wilcoxon rank sum test. ANCOM was performed on R (https://github.com/antagomir/scripts/tree/master/R/ancom). The Wilcoxon rank sum test was performed with the mannwhitneyu functionality of scipy.stat on Python 3.6.5. The threshold of W calculated in ANCOM was set to more than 0.6 Â N, where N is the number of taxa. The difference in the abundance of each taxon between iRBD and controls was also analyzed by the Wilcoxon rank sum test followed by calculation of the false-discovery rate (FDR) using the Benjamini-Hochberg procedure. The FDR threshold was set to 0.05. Bacterial taxa filtered for both W and FDR were assumed to be significant.
We also analyzed the sub-data set of eight pairs of iRBD patients and their spouses by ANCOM and the Wilcoxon signed-rank sum test to adjust for the effects of diet and lifestyle.
Possible confounding factors in our data set for nine taxa that were significantly changed in our data set. Seven genera and two families that were identified in our data set were subjected to GLMM (Generalized Linear Mixed Model) analysis using the function "glmer.nb" of the R package lme4 by setting an option to accept taxonomic variations from subject to subject.
Meta-analysis of the Japanese and German data sets. Our Japanese data set was comprised of 26 iRBD patients and 137 healthy controls, whereas the German data set was comprised of 20 iRBD patients and 38 healthy controls (43). We first collated the experimental methods and demographic features (see Table S7 in the supplemental material), as well as statistical measures of sequencing depths (see Table S8 in the supplemental material) of the two data sets. The read count of each sample was all more than 10,000 in the two data sets, and no sample was excluded from our meta-analysis. For each taxon, we counted the number of samples in which the relative abundance of the taxon was more than 1E24. We then filtered 39 families and 132 genera, in which the number of such samples was more than 10% (17/163 and 6/58) in both data sets.
In the meta-analysis, we applied two criteria that we used in our previous report (32) to identify homogeneously and significantly changed taxa in the Japanese and German data sets. The two criteria were that I 2 , representing heterogeneity in meta-analysis, was below 25% (92) and that the P values after Bonferroni correction for FEM and REM were both less than 0.05.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. FIG S1, EPS file, 2.8 MB.