Meta-analysis of gut microbiota alterations in patients with irritable bowel syndrome

Abstract

Introduction: Irritable bowel syndrome (IBS) is a common chronic disorder of gastrointestinal function with a high prevalence worldwide. Due to its complex pathogenesis and heterogeneity, there is urrently no consensus in IBS research.

Methods: We collected and uniformly reanalyzed 1167 fecal 16S rRNA gene sequencing samples (623 from IBS patients and 544 from healthy subjects) from 9 studies. Using both a random effects (RE) model and a fixed effects (FE) model, we calculated the odds ratios for metrics including bacterial alpha diversity, beta diversity, common genera and pathways between the IBS and control groups.

Results: Significantly lower alpha-diversity indexes were observed in IBS patients by random effects model. Twenty-six bacterial genera and twelve predicted pathways were identified with significant odds ratios and classification potentials for IBS patients. Based on these feature, we used transfer learning to enhance the predictive capabilities of our model, which improved model performance by approximately 10%. Moreover, through correlation network analysis, we found that Ruminococcaceae and Christensenellaceae were negatively correlated with vitamin B6 metabolism, which was decreased in the patients with IBS. Ruminococcaceae was also negatively correlated with tyrosine metabolism, which was decreased in the patients with IBS.

Discussion: This study revealed the dysbiosis of fecal bacterial diversity, composition, and predicted pathways of patients with IBS by meta-analysis and identified universal biomarkers for IBS prediction and therapeutic targets.

Keywords: gut microbiota; irritable bowel syndrome; meta-analysis; random forest model; transfer learning.

Figure 1

Description of the selection of the included studies following a PRISMA flow diagram.

Figure 2

Principal coordinates analysis (PCoA) based on the Bray–Curtis distance according to the genera. Each point in the diagram represents a sample. The shapes represent the healthy control and IBS groups, respectively. The colors represent the different studies.

Figure 3

Comparison of bacterial alpha diversity and beta diversity between the individuals with IBS and HS. (A) Forest plot of the alpha diversity metrics richness, Shannon Index, and evenness (J) between the individuals with IBS and HS; (B) Forest plot of the Bray–Curtis distance between the individuals with IBS and HS. The error bar depicts the 95% confidence interval. The ORs less than 1.0 (left side of the dashed line) indicate that the metric was higher in the patients with IBS than in the controls. The ORs greater than 1.0 (right side of the dashed lines) indicate that the metric was lower in the patients with IBS than in the controls. No intersection between the dashed line and the error bar indicates a significant difference between the patients with IBS and the controls.

Figure 4

Forest plots of (A) the genera or (B) the KOs with the significant ORs.

Figure 5

Establishment of an IBS discriminative model based on the abundance of the bacterial genera. (A) Diagnostic potential of intestinal bacteria for IBS across all studies. The specific AUC values for the training set and test set are shown in Table 2; (B) The rank of feature importance for the optimal model.

Figure 6

Cross-prediction matrix reporting the prediction performances as the AUC values obtained using a random forest (RF) model on the (A) genus and (B) pathway level-relative abundances. The values boxed in yellow squares on the diagonal are the AUC values obtained by training and validating within the individual cohorts. The non-diagonal values refer to training a classifier on the dataset corresponding to the row and applying it to the dataset corresponding to the column to obtain the AUC values. The row averages are the average values excluding the diagonal values.

Figure 7

Co-occurrence network plot showing the correlations between the 12 pathways and 26 genera that had significant Spearman correlation coefficients (i.e., with |correlation coefficient | > 0.5 and p < 0.05). The size of the nodes varies with the number of edges (red lines, negative correlations; green lines, positive correlations).