Fecal Microbial Community Profiling Allows Discrimination of Phenotype and Treatment Response in Pediatric Crohn's Disease and Ulcerative Colitis-An International Meta-Analysis

Abstract

Background and aims: The pathophysiology of pediatric inflammatory bowel disease (PIBD), encompassing Crohn's disease (CD) and ulcerative colitis (UC), is not entirely understood. Dysregulation of the intestinal microbiome is recognized as both a disease-driving and a potential therapeutic target. This study aimed to systematically analyze gut microbiome compositions and its applicability as a biomarker for disease progress and treatment response.

Methods: Bibliographic and nucleotide databases were searched. Raw 16S-rRNA sequencing reads were subjected to a uniform downstream dada2/phyloseq pipeline to extract taxonomy, community structure, and abundance information. Patient metadata were extracted from publications, and study authors were contacted for further details if required.

Results: Twenty-six studies comprising 3956 stool samples (CD 41%, UC 36%, 23% healthy) were included in the analyses. Median age of individuals was 12 (interquartile range 4). Sex distribution was comparable. Alpha diversity was reduced between the healthy and both UC and CD treatment-naïve groups (P < .001) and further reduced with increasing clinical disease activity. Beta diversity revealed altered community structure in treatment-naïve children with PIBD (P < .001). This alteration remained in patients in clinical remission (P < .001). Machine learning models discriminated between treatment-naïve patients with CD or UC with an area under the receiver operating characteristics curve (AUROC) of 98%. Microbial communities differed between patient responders versus nonresponders to treatment (P < .001). Further, microbial community profiling distinguished treatment response (eg, steroid, nutrition, or TNFα) with AUROCs of 82%-90%.

Conclusions: Gut microbial community structure is substantially altered in active and inactive PIBD and may be utilized as a biomarker for differentiating PIBD subtype and predicting treatment response.

Keywords: 16S metagenomics; Crohn’s disease; microbiome; treatment response; ulcerative colitis.

Figure 1.

Study design, cohort composition, and diversity metric: (A) World map indicating countries of origin of patient cohort. (B) Schematic overview of the study cohort. (C) Study level comparison of all data pre (left) and post (right) MMUPHin transformation by using a principal coordinate analysis. (D, E) Alpha diversity, expressed as both observed richness and Shannon diversity, of study corrected fecal gut microbiome data in treatment-naïve patients with PIBD compared to healthy individuals; n.s. not significant, *P < .05, **P < .01, ***P < .001. (F, G) Beta diversity shown as multidimensional scaling (MDS) using a principal coordinate analysis with Bray-Curtis distance measures. (F) Treatment-naïve patients with UC or CD compared to healthy individuals. (G) PIBD patients in clinical remission with healthy individuals; ***P < .001.CD, Crohn’s disease; MMUPHin, Meta-Analysis Methods with a Uniform Pipeline for Heterogeneity in microbiome studies; PIBD, pediatric inflammatory bowel disease; UC, ulcerative colitis.

Figure 2.

Relative abundance, disease association with differential abundance of taxa: (A) Dendrogram depicting top 10 most abundant species (inner circle/box plots) and associated conditions on the 3 outer circles. (B) Relative abundance of 10 most abundant phyla in healthy individuals and patients with PIBD over different clinical disease activities. (C) Relative abundance of 14 most abundant genera in healthy individuals and patients with PIBD over different clinical disease activities. (D-G) Relative abundance of Bifidobacterium, Subdoligranulum, Escherichia-Shigella, Haemophilus genera regression to calprotectin levels. PIBD, pediatric inflammatory bowel disease.

Figure 3.

Discrimination of CD and UC treatment response: (A) Beta diversity shown as multidimensional scaling (MDS) using a principal coordinate analysis with Bray-Curtis distance measures comparing treatment-naïve patients with CD (Paris L2) or UC (Paris E3/E4). (B) Area under the receiver operating characteristics curve of XGB and RF models discriminating steroid response treatment-naïve patients with CD (Paris L2) UC (Paris E3/E4) patients. (C) Dotted box-plot of relative genus abundance of top 10 discriminative genus features identified by supervised machine learning with feature ranking and log2 fold change from differential abundance analysis comparing CD (Paris L2) (yellow) with UC (Paris E3/E4) (red). CD, Crohn’s disease; RF, random forest; UC, ulcerative colitis; XGB, extreme gradient boosting.

Figure 4.

Community diversity and discrimination of treatment response: (A) Alpha diversity of gut microbiome in patients with CD with or without response to treatment (observed); ***P < .001, n.s. not significant. (B) Alpha diversity of gut microbiome in patients with UC with or without response to treatment; *P < .05, n.s. not significant. (C-D) CD and UC respective beta diversity shown as multidimensional scaling (MDS) using a principal coordinate analysis with Bray-Curtis distance measures; ***P < .001. (E) Horizontal bar plot showing the top 10 differential abundant genera comparing patients with CD without (brown yellow) and with (yellow) treatment response. (F) Horizontal bar plot showing the top 10 differential abundant genera comparing UC patients without (dark red) and with (red) treatment response. (G) XGB area under the receiver operating characteristics curve of a supervised machine learning model discriminating treatment response in pooled patients with CD. (H) Dotted box-plot of log10-transformed genus abundance of top 10 discriminative genus features identified by both supervised machine learning with feature ranking and log2 fold change from differential abundance analysis comparing patients with CD without (brown yellow) and with (yellow) treatment response. CD, Crohn’s disease; UC, ulcerative colitis; XGB, extreme gradient boosting.

Figure 5.

Community diversity and discrimination of treatment response: (A) Area under the receiver operating characteristics curve (AUROC) of XGB model discriminating steroid response in patients with CD. (B) Dotted box-plot of log10-transformed genus abundance of top 10 discriminative genus features identified by supervised machine learning with ranked feature importance of the XGB and RF model, and log2 fold change from differential abundance analysis comparing patients with CD without (gray) and with (black) steroid response. (C) AUROC of XGB model discriminating EEN response in patients with CD. (D) Dotted box-plot of log10-transformed genus abundance of top 10 discriminative genus features identified by supervised machine learning with ranked feature importance of the XGB and RF model, and log2 fold change from differential abundance analysis comparing patients with CD without (gray) and with (black) EEN response. (E) AUROC of RF model discriminating TNF response in patients with CD. (F) Dotted box-plot of log10-transformed genus abundance of top 10 discriminative genus features identified by supervised machine learning with ranked feature importance of the XGB and RF model, and log2 fold change from differential abundance analysis comparing patients with CD without (gray) and with (black) TNF response. CD, Crohn’s disease; EEN, exclusive enteral nutrition; RF, random forest; TNF, tumor necrosis factor; XGB, extreme gradient boosting.