Exploration of Potential Gut Microbiota-Derived Biomarkers to Predict the Success of Fecal Microbiota Transplantation in Ulcerative Colitis: A Prospective Cohort in Korea

Abstract

Background/aims: Although fecal microbiota transplantation (FMT) has been proven as one of the promising treatments for patients with ulcerative colitis (UC), potential prognostic markers regarding the clinical outcomes of FMT remain elusive.

Methods: We collected fecal samples of 10 participants undergoing FMT to treat UC and those from the corresponding donors. We categorized them into two groups: responders and nonresponders. Sequencing of the bacterial 16S rRNA gene was conducted on the samples to explore bacterial composition.

Results: Analyzing the gut microbiota of patients who showed different outcomes in FMT presented a distinct microbial niche. Source tracking analysis showed the nonresponder group had a higher rate of preservation of donor microbiota, underscoring that engraftment degrees are not one of the major drivers for the success of FMT. At the phylum level, Bacteroidetes bacteria were significantly depleted (p<0.003), and three genera, including Enterococcus, Rothia, and Pediococcus, were enriched in the responder group before FMT (p=0.003, p=0.025, and p=0.048, respectively). Furthermore, we applied a machine learning algorithm to build a prediction model that might allow the prediction of FMT outcomes, which yielded an area under the receiver operating characteristic (ROC) curve of 0.844. Notably, the microbiota-based model was much better at predicting outcomes than the clinical features model (area under the ROC curve=0.531).

Conclusions: This study is the first to suggest the significance of indigenous microbiota of recipients as a critical factor. The result highlights that bacterial composition should be evaluated before FMT to select suitable patients and achieve better efficiency.

Keywords: Fecal microbiota transplantation; Fecal microbiota transplantation outcome; Machine learning; Ulcerative colitis.

Fig. 1

Decrease in the ulcerative colitis (UC) activity after fecal microbiota transplantation (FMT). (A) Schematic diagram of the FMT schedule, assessment of UC severity, and sampling at designated time points (T). (B) At T58 (58 days after the first FMT), patients with UC were classified into RP (decrease in both pMayo and CRP, n=6) and NRP (no change or an increase, n=4) groups. PR, responder; NRP, nonresponder; CRP, C-reactive protein; pMayo, partial Mayo.

Fig. 2

Variation in gut microbial composition after fecal microbiota transplantation (FMT). (A) Projection displayed by principal coordinates analysis (PCoA) represented significantly differential gut microbiota composition (p=0.001). The compositional variances explained by each axis in PCoA dimensions were shown on the axes. Box plots next to the PCoA indicate a comparison of Bray-Curtis based distance across groups. The distance was calculated in comparison with donor microbiota. (B) Bar chart representation of BugBase analysis shows predicted phenotype. Samples denoted from each grouping were displayed and colored accordingly. Adonis and ANOVA test (Tukey honestly significant difference for multiple comparison) were used where applicable. ANOVA, analysis of variance; RP, responder; NRP, nonresponder. *p<0.05, ^†p<0.01, ^‡p<0.001.

Fig. 3

Identification of indigenous taxa that shapes fecal microbiota transplantation (FMT)-mediated clinical outcomes. (A) Box plots of top five bacterial phyla abundances of each group before FMT (responder [RP] vs nonresponder [NRP]). (B) Phylogenetic tree generated on Amplicon sequence variant (ASV) level to grant overview of the microbial community. Heatmap of outer circle presents log-transformed abundances of individual ASVs. Box plots indicate bacterial genera with significant differences between the RP and NRP groups before FMT (pre-RP and pre-NRP). NS, not significant. Wilcoxon rank-sum test was used to assess statistical significance, *p<0.05 and ^†p<0.01.

Fig. 4

Construction of a gut microbiota-derived diagnostic model to predict fecal microbiota transplantation (FMT) outcome. A machine learning model was trained to diagnose subjects into responder and nonresponder groups before FMT (pre-RP and pre-NRP). (A) The heatmap plot shows the z scores of selected features across samples, and the bar plots on the left indicate feature weight, including robustness, ordered by effect size. (B, C) The values of area under the curve (AUC) are given under the curve, with the 95% confidence interval shaded to indicate accuracy. Values of 0.844, 0.531, and 0.781 for cross-validation performance were obtained using patients’ indigenous microbiota, clinical features, and donor microbiota, respectively. RP, responder; NRP, nonresponder.