Deciphering associations between gut microbiota and clinical factors using microbial modules

Abstract

Motivation: Human gut microbiota plays a vital role in maintaining body health. The dysbiosis of gut microbiota is associated with a variety of diseases. It is critical to uncover the associations between gut microbiota and disease states as well as other intrinsic or environmental factors. However, inferring alterations of individual microbial taxa based on relative abundance data likely leads to false associations and conflicting discoveries in different studies. Moreover, the effects of underlying factors and microbe-microbe interactions could lead to the alteration of larger sets of taxa. It might be more robust to investigate gut microbiota using groups of related taxa instead of the composition of individual taxa.

Results: We proposed a novel method to identify underlying microbial modules, i.e. groups of taxa with similar abundance patterns affected by a common latent factor, from longitudinal gut microbiota and applied it to inflammatory bowel disease (IBD). The identified modules demonstrated closer intragroup relationships, indicating potential microbe-microbe interactions and influences of underlying factors. Associations between the modules and several clinical factors were investigated, especially disease states. The IBD-associated modules performed better in stratifying the subjects compared with the relative abundance of individual taxa. The modules were further validated in external cohorts, demonstrating the efficacy of the proposed method in identifying general and robust microbial modules. The study reveals the benefit of considering the ecological effects in gut microbiota analysis and the great promise of linking clinical factors with underlying microbial modules.

Availability and implementation: https://github.com/rwang-z/microbial_module.git.

Figure 1

Overview of the study. (A) Fecal samples were collected consecutively from IBD patients and healthy controls for 1 year. (B) Number of healthy controls and IBD patients included in the study. (C) Number of control and IBD samples in the “10-visit-set” and “24-visit-set”. (D) Number of samples collected from each subject. E. Pie chart of the distribution of samples collected from healthy controls and IBD patients in the “10-visit-set” and “24-visit-set” additionally. The samples in the “10-visit-set” are also included in the “24-visit-set”. (F) Illustration of microbial module identification and downstream analysis in this study. The longitudinal data are represented as a third-order tensor and factorized into three factor matrices, which are further used for module analysis, subject stratification, and module member detection, respectively. PERMANOVA, permutational multivariate analysis of variance.

Figure 2

Decomposition results of the longitudinal gut microbiota data. (A) The Bayesian tensor factorization model of the proposed method (visit-correlated model). (B) Classification performance (AUROC) of different numbers of top-ranking microbial modules of IBD identified from the results of “24_visit”, “10_visit”, and “10_visit-correlated”, respectively. Dots indicate the performance of different random runs. Average values are demonstrated as grey diamonds. (C) Boxplot of the average AUROC of each run under different settings. The best performing runs of “10_visit-correlated” are marked in the grey dashed rectangle. (D) Comparison of the best performing runs of “10_visit-correlated” in terms of the classification performance and reconstruction error measured by the average AUROC using different numbers of top-ranking modules and RMSE, respectively.

Figure 3

Relationships between the microbial modules and clinical factors. (A) The variation in terms of the microbial module activities and the taxonomic composition explained by several clinical factors, quantified by PERMANOVA. The color represents the proportion of variance explained by each factor. (B) Associated modules of antibiotics, immunosuppressants, disease states, and diarrhea, respectively (Wilcoxon rank-sum test, $\mathrm{FDR}\le 0.25$). Each column lists the associated modules of one factor, and each panel demonstrates the activities of a module in distinct groups (“Yes” and “No”) of the associated factor. Member taxa of the associated modules of antibiotics (C), immunosuppressants (D), and diarrhea (E), respectively. Each column represents a module. The color demonstrates the activities of the taxa in the modules.

Figure 4

Intramodule associations of the member taxa. The intramodule co-occurrence of taxa measured by DI (A) and absolute Spearman correlation (B) calculated individually (average of subjects) and across all samples, compared with that of equal-size random groups formed by the taxa pairs not included in any module. (C) The intragroup functional similarity of the modules calculated across all samples. ***, P-value ≤ .001. The values are multiplied by a scaling factor of 100 and then log-transformed. (D) Pairwise DI of all taxa calculated individually (upper triangle) and across all samples (lower triangle). Taxa pairs in the IBD-associated modules (+) are distributed in the blue rectangle. (E) Examples of the pairwise DI of an IBD-associated module, an equal-size module, and an equal-size random group, respectively.

Figure 5

The IBD-associated modules. (A) The activities of the member taxa in each IBD-associated module. (B) The association support from different sources. Top, the venn plot comparing the taxa in the IBD-associated modules, the IBD-associated taxa detected by DA analysis and the taxa reported associated with IBD in HMDAD. Bottom, the venn plot demonstrating the supports from another two studies. The color of the species indicates the phyla they belong to. (C) The pie chart of the number of taxa supported by each source in each IBD-associated module. (D–F) The contributions of the member taxa to the top pathways in each IBD-associated module. Blue circles, microbial taxa. Brown circles, the top 20 abundant pathways. Size of the circles, the abundance of the pathways (for pathway circles), or the contribution of the taxa to the pathways (for the taxon circles). Lines, contributions of the taxa to the pathways. (G) The classification performance (AUROC) of the IBD-associated modules and taxa relative abundance in the discovery set.

Figure 6

Validation of the microbial modules in the external cohorts. (A) The intramodule co-occurrence in the validation cohorts measured by DI and absolute Spearman correlation, compared with that of equal-size random groups formed by the taxa pairs not included in any module. ***, P-value ≤ .001. NS, not significant. (B) The activities of the IBD-associated modules in the control and IBD groups of the validation set “Val_Hall”. *, FDR ≤ 0.25. (C) The classification performance (AUROC) of the IBD-associated modules and the taxa relative abundance in “Val_Hall”. (D) Comparison of the classification performance (AUROC) in the discovery set (iHMP) and the validation set “Val_Hall”.