The means and standard deviation for descriptive data describing the samples can be found in Table 1. The average ages for IBS (n = 29, females = 21) and HCs (n = 23, females = 14) were similar, 26.1 years (SD = ±5.72 and 26.0 years (±6.5), respectively. Average symptom duration in IBS subjects was 11.3 (±13.2) years. Even though IBS subjects as a group had significantly greater level of anxiety (p = .03) but not depression symptoms, the anxiety levels for the majority of subjects were within the normal range. However, six IBS subjects (HAD score range 12–17) and one HC (HAD score 13) had scores greater than 11, indicating a probable level of clinical anxiety. Compared to HCs, IBS reported greater levels of perceived stress (p = .02) and higher levels of catastrophizing (p = .02). No group differences were observed in total emotional trauma history scores, but there was a trend for a greater history of emotional trauma in IBS (p = .09). Moderate to high levels of early life trauma (4–5) on the emotional subscale were reported by 36% IBS subjects while no healthy control subjects showed such scores. IBS subjects reported average of overall symptoms in the past week of 9.3 (±4.3) on a 21-point numeric rating scale (0 = no pain, 20 = the most intense symptoms imaginable). Distribution of bowel habit was 11 constipation (37.9%), 10 diarrhea (34.5%), 1 alternating (3.4%), 5 mixed (17.2%), and 2 unspecified (6.9%].
No statistically significant differences in dietary patterns were observed between groups. IBS1 showed slightly higher BMI than HC-like IBS (p = .047) although both groups had an average body weight within the normal range.
IBS subgroup identified based on a microbiome signature
Hierarchical clustering using average linkage and PCoA analysis on unweighted Unifrac distances are depicted in Fig. 1a, b and Additional file 4: Figure S1. Together, these analyses indicated that microbial signatures could be used to group patients and to discriminate among samples. Based on the PCOA and hierarchical clustering, the samples were labeled as IBS1 (IBS patients with microbial profiles distinct from HCs; n = 13), HC-like IBS (IBS patients with similar microbial composition as HCs; n = 16) and HCs (n = 23). One IBS subject clustered with IBS1 in the PCoA but clustered with HCs in the hierarchical clustering. This subject was labeled as HC-like IBS subject. We tested whether this group assignment changed group difference or correlation results but it did not. Adonis analysis  of the unweighted Unifrac distances indicated that a significant proportion of the variance could be explained by IBS1–HC differences F(1,51) = 5.79, p = .004, R
2 = .10. The variance accounted for other group contrasts was not statistically significant (HC-like IBS-IBS1, F(1,51) = 2.2, p = .073, R
2 = .04; HC-like IBS-HCs, F(1,51) = 1.23, p = .25, R
2 = .02).
OTUs contributing to the variation in microbial signatures along the 3rd axis of the PCoA plot (i.e., that which captured the separation of the IBS1 communities from the HC and HC-like IBS communities) were identified using correlation analysis. Those whose relative abundances correlated most strongly with the location along this axis are provided in Additional file 5: Table S3 and included members of the genera Faecalibacterium, Bacteroides, and Blautia.
Random forest analysis provided further support for the differentiation of IBS1 and HC groups. This analysis identified an OTU-based signature which correctly distinguished the majority of IBS1 subjects from HC subjects. Using 10-fold cross-validation, the resulting model had an AUROC of 0.96, sensitivity of 0.95, and specificity of 0.67 (Additional file 6: Figure S2). OTUs contributing to the differentiation of IBS1 and HC gut communities, according to their random forest importance (i.e., mean decrease in accuracy) scores, included members of the genera Blautia, Streptococcus, Faecalibacterium, and Bacteroides (see Additional file 7: Table S4).
Diversity analyses and comparison of bacterial relative abundances among IBS subgroups
Alpha diversity by groups (HC, IBS) and subgroups (IBS1, HC-like IBS, HC) were calculated, and results are depicted in Additional file 8: Figure S3. Comparison of Faith’s phylogenetic diversity indices with nonparametric t tests indicated that as a group, all IBS subjects (n = 29) showed significantly greater diversity than HCs, t(50) = 2.83, p = .007. Examining the IBS subgroups indicated this was largely due to the fact that the IBS1 community was more diverse than the HC-like IBS patients, t(27) = 3.33, p = .012, and HCs, t(36) = 3.80, p = .003. There were no differences observed between the HC-like IBS subjects and HCs, t = .90, p = .99. Similar results were observed for the other diversity measures: chao1 richness estimator, Shannon’s entropy, and number of observed OTUs.
Figures 2 and 3 depict the microbial composition of each group and individual at the phylum and class levels. The ratio of Firmicutes to Bacteroidetes (F-B ratio) was significantly higher in the IBS1 subgroup compared to HCs (q = .02). The F-B ratio in the HC-like IBS group did not differ significantly from either group (Fig. 4a).
The relative abundances of OTUs and the mean relative abundance for identifiable taxa demonstrating overall group differences at each taxonomic level (phylum, class, order, family, and genus) are depicted in Fig. 4. The means and standard deviations as well as Kruskall Wallis tests of significance can be found for identifiable and unidentified taxa and OTUs in Additional file 9: Table S5.
At the phylum level (Fig. 4c), the relative abundance of Firmicutes was significantly greater in the IBS1 subjects compared to HCs (q = .002) but not the HC-like IBS subjects (q = .075). On the other hand, the relative abundance of Bacteroidetes was significantly lower in IBS1 group compared to HCs (q = .003) but not in the HC-like IBS subjects (q = .12). For all comparisons, the HC-like IBS group did not differ from the HCs with respect to the relative abundance of either Firmicutes or Bacteroidetes. Furthermore, no differences were observed at the phylum level for the relative abundance of Actinobacteria (IBS1 .1%, HC-like IBS 0%, and HCs 0%) or Proteobacteria (1.3, .8, .4%).
At the class level (Fig. 4d), differences were observed in the Firmicutes classes, Bacilli and Clostridia. The relative abundance of members of Bacilli was greater in IBS1 compared to HC-like IBS (q = .001) and HCs (q = .005). For Clostridia, IBS1 showed greater abundances than HCs (q = .004) but not HC-like IBS (q = .14). IBS1 had lower abundances of Bacteroidetes class Bacteroidia compared to HCs (q = .002) but not HC-like IBS (q = .075).
At the order level (Fig. 4e), differences were observed with respect to the Bacilli order Lactobacillales with greater abundance observed for IBS1 compared to HC-Like IBS (q = .004) and HCs (q = .004). The relative abundance of Bacteroidales order was lower in IBS1 compared to HCs (q = .002) but not HC-like IBS (q = .075).
At the genus level (Fig. 4f), members of the Erysipelotrichaceae genus Holdemania were more abundant among IBS1 compared to HC-like IBS (q < .001) and HCs (q < .001). Members of the Porphyromonadaceae genus Parabacteroides were less abundant in IBS1 compared to HCs (q = .003) but not HC-like IBS (q = .097).
Mean differences in clinical metadata between IBS subgroups based on microbial clusters
Detailed clinical data for the subgroups of IBS based on microbial community profiles, IBS1 (9 females, 5 males) and HC-like IBS (12 females, 3 males), are available in Additional file 10: Table S6. IBS1 subjects reported a longer duration of symptoms than HC-like IBS (t(26) = 2.80, p = .01) as well as higher scores on the ETI emotional subscale, t(25) = 3.14, p = .004. Although overweight status (BMI >25.0) was not significantly different between the two IBS clusters, IBS1 [m(sd), 25.52 (5.68)] showed a trend (t(27) = 1,99, p = .057) toward greater BMI compared to HC-like IBS [20.99 (6.51)], Hedge’s effect size g (Cohen’s d corrected for small sample sizes) = .74. This was associated with a trend for higher dietary plant fat intake in the IBS1 group compared to HC-like IBS (p = .047) and HC (p = .017).
In contrast, microbial community clustering in IBS was not associated with sequencing depth, age, predominant bowel habit, symptoms of anxiety or depression, levels of catastrophizing, perceived stress, or medication usage.
Subgroup analysis in only females indicated that the mean differences in the duration of symptoms, t(19) = 2.72, p = .014, and emotional ETI scores remained significant as did the trend for greater BMI (t(19) 1.80, p = .08 in IBS.
Associations between clinical metadata and microbiota taxa differentiating IBS1 from HC-like IBS and HCs
Order level taxa
In IBS subjects, Lactobacillales, which had significantly higher abundance in IBS1, had a moderately positive correlation with ETI total score (r = .51, p = .008), as well as sexual (r = .59, p = .001), physical (r = .51, p = .007), and emotional (r = .42, p = .028) subscores. No correlations were observed for Bacteroidales. Examining correlations in female IBS only revealed that Bacteroidales positively correlated with overall symptom severity (r = .423, p = .056). Correlations with the ETI total score (r = .56, p = .01), and its physical (r = .53, p = .013), emotional (r = .52, p = .02), and sexual (r = .58, p = .008) subscores persisted.
Genus level taxa
Abundance of Parabacteroides was positively associated with bloating r = .48 p = .03. Because Holdemania was not identified in any group but IBS1, associations with this genus were not examined.
Across all subjects (n = 48), the relative abundance of the Firmicutes-associated class Bacilli was positively correlated with ETI total score (r = .34, p = .018) as well as with sexual (r = .43, p = .002) and emotional (r = .38, p = .007) subscores. No correlations with EALs were found for the Firmicutes-associated class Clostridia or the Bacteroidetes-associated class Bacteroidia. IBS1 had greater scores on the emotional scale of the ETI than HC-like IBS (p = .004) and HCs (p = .001), while no differences were observed between IBS and HCs as a group for this score.
Association between brain structure and discriminative microbiota in IBS subjects
Table 2 shows the results of the correlations between the relative abundance of classes of bacteria associated with Firmicutes and Bacteroidetes that discriminated the microbial subgroups (Fig. 4d) and the parcellated 165 regional brain volumes for all IBS (n = 29). Moderately sized correlations were observed for the Clostridia belonging to phylum Firmicutes (higher in IBS1), and for the Bacteroidia belonging to phylum Bacteroidetes (lower in IBS1) with several sensory integration regions including the thalamus, basal ganglia (caudate nucleus, putamen, pallidum, nucleus accumbens), and the superior part of the precentral gyrus (motor cortex). Similar correlations were found for the anterior insula and ventral prefrontal regions.
Based on the relative abundance of these classes of bacteria, volumes in the sensory brain regions were increased, while volumes of insula and prefrontal cortices were decreased in IBS1 compared to HC-like IBS. Very few correlations were observed for the parietal, occipital, and temporal regions. The abundance of the Bacilli belonging to phylum Firmicutes showed correlations with the volume of fewer regions including the nucleus accumbens, prefrontal cortices, and ventral posterior cingulate cortex.
As can be seen in Table 3, compared to HC-like IBS and HCs, IBS1 had smaller cortical thickness and larger surface area in the anterior insula. Compared to HC-like IBS, IBS1 also showed greater volume and surface area in the posterior insula, right globus pallidum, and lower cortical thickness and surface area of motor cortex. Finally, compared to HCs, IBS has greater volume in the mid/posterior insula.
Compared to HCs, HC-like IBS had greater cortical thickness in the sensory brain regions, including the left post central and central sulci and the right posterior insula.
Metagenes enriched in the IBS1 subgroup are associated with posterior insula morphometry
We investigated on an exploratory basis whether the distinct microbial composition of IBS1 compared to HC-IBS and HC was associated with a shift in the functional profile of the microbiome. This was motivated by the possibility that altered bacterial production of neuroactive metabolites could contribute to the association of microbial clusters with brain morphometric parameters. Counts of bacterial genes (metagenes) were predicted from the 16S rRNA gene sequence data using PICRUSt, which imputes gene content for each OTU based upon the phylogenetically closest reference genome. The resulting count data were fitted to negative binomial models using DESeq2 with microbial cluster as a predictor. In pairwise comparisons, 606 metagenes (i.e., bacterial genes predicted by PICRUSt) differed in abundance between the IBS1 subgroup and HC-IBS with adjusted p value <.05 and 1223 metagenes differed between IBS1 and HC. In contrast, there were no differential metagenes between HC-IBS and HC. In parallel analyses, the same methodology was used to identify metagenes associated with the 10 brain morphometric parameters (each analyzed separately) that differed among the three microbial clusters at p < .01 (Table 3). Twenty-seven metagenes were found that were significantly associated with both the IBS1 subgroup and with brain morphometric parameters, including 20 that were associated with the surface area of the right inferior segment of the inferior segment of the circular sulcus (a region of the posterior insula) (Additional file 11: Table S7). Of these, we focused on two metagenes enriched in IBS1 that are involved in neurotransmitter metabolism—4-hydroxybutyrate dehydrogenase, which degrades γ-hydroxybutyric acid (GHB), and glutamate dehydrogenase, which mediates glutamate synthesis/breakdown—and on two metagenes enriched in IBS1 that are involved in short chain fatty acid metabolism—acetate co-A transferase subunit B, which mediates the final step of butyrate synthesis, and propionate catabolism operon regulatory protein, which regulates an operon of propionate catabolism genes (Fig. 5a). The individual taxa that made the greatest contribution to the association of these metagenes with IBS1 were identified using FishTaco. This highlighted a prominent role for one particular taxon, an unclassified Peptostreptococcaceae, which had a trend toward significant correlation (p = .058) with the right inferior segment of the circular sulcus though it fell short of significance as compared to the metagene encoding 4-hydroxybutyrate dehydrogenase (p = .008) (Fig. 5b).