Development and Validation of a Novel Microbiome-Based Biomarker of Post-antibiotic Dysbiosis and Subsequent Restoration

Abstract

Background: The human gut microbiota are important to health and wellness, and disrupted microbiota homeostasis, or "dysbiosis," can cause or contribute to many gastrointestinal disease states. Dysbiosis can be caused by many factors, most notably antibiotic treatment. To correct dysbiosis and restore healthier microbiota, several investigational microbiota-based live biotherapeutic products (LBPs) are in formal clinical development. To better guide and refine LBP development and to better understand and manage the risks of antibiotic administration, biomarkers that distinguish post-antibiotic dysbiosis from healthy microbiota are needed. Here we report the development of a prototype Microbiome Health Index for post-Antibiotic dysbiosis (MHI-A). Methods: MHI-A was developed and validated using longitudinal gut microbiome data from participants in clinical trials of RBX2660 and RBX7455 - investigational LBPs in development for reducing recurrent Clostridioides difficile infections (rCDI). The MHI-A algorithm relates the relative abundances of microbiome taxonomic classes that changed the most after RBX2660 or RBX7455 treatment, that strongly correlated with clinical response, and that reflect biological mechanisms believed important to rCDI. The diagnostic utility of MHI-A was reinforced using publicly available microbiome data from healthy or antibiotic-treated populations. Results: MHI-A has high accuracy to distinguish post-antibiotic dysbiosis from healthy microbiota. MHI-A values were consistent across multiple healthy populations and were significantly shifted by antibiotic treatments known to alter microbiota compositions, shifted less by microbiota-sparing antibiotics. Clinical response to RBX2660 and RBX7455 correlated with a shift of MHI-A from dysbiotic to healthy values. Conclusion: MHI-A is a promising biomarker of post-antibiotic dysbiosis and subsequent restoration. MHI-A may be useful for rank-ordering the microbiota-disrupting effects of antibiotics and as a pharmacodynamic measure of microbiota restoration.

Keywords: Clostridioides difficile infection; antibiotics; biomarker; dysbiosis; intestinal microbiota.

Figure 1

Development of MHI-A. (A) Mean relative abundance (π) at the class level for samples from the PUNCH CD2 trial, denoted as before treatment (BL), 7, 30, and 60 days (7 D, 30 D, and 60 D) after treatment, or the administered doses of RBX2660 investigational product. Means (π) with upper and lower confidence intervals were calculated based on maximum likelihood estimate fit to a Dirichlet multinomial distribution, with classes comprising less than 3% relative abundance at all time points combined as “Other.” (B) Dirichlet-multinomial Recursive Partitioning (DM-RPart) was fit to regress the N = 96 microbiome composition data from the PUNCH CD2 trial onto BL, 7 D, 30 D, and 60-day timepoints. Ten-fold cross-validation was used to fit the optimal tree which resulted in 3 terminal nodes: BL (n = 22 samples), 7 D and 30 D (n = 52 samples), and 60 D (n = 22 samples). The taxa abundances for the three terminal nodes are shown in the color bar chart. This indicates that at BL Bacilli and Gammaproteobacteria were the top two dominant classes, and in post-treatment timepoints Bacteroidia and Clostridia became the dominant classes. This suggests these taxa can be used to distinguish representative post-antibiotic dysbiosis (BL) from healthy (RBX2660) populations. (C) MHI-A values for PUNCH CD2 samples, shown on a logarithmic scale as median and interquartile ranges with individual samples overlaid. Timepoints shown are baseline (BL), 7 days (7 D), 30 days (30 D), and 60 days (60 D) after treatment with RBX2660 or placebo (PBO). MHI-A values for the administered doses of RBX2660 investigational product are also shown (RBX2660). The dotted line shows the MHI-A = 7.2 threshold, above which MHI-A values correspond to a healthy based on ROC analysis.

Figure 2

MHI-A values for participants in the PUNCH Open Label trial of RBX2660 and a Phase 1 trial of RBX7455 (A) MHI-A values for all PUNCH Open Label samples, shown as individual sample values with median and interquartile ranges. Timepoints shown are baseline (BL), 7 days (7 D), 30 days (30 D), 60 days (60 D), 6 months (6 M), 12 months (12 M), and 24 months (24 M) after treatment with RBX2660. MHI-A values for the administered doses of RBX2660 investigational product are also shown (RBX2660). The dotted line shows the MHI-A = 7.2 threshold, above which MHI-A values correspond to a healthy based on ROC analysis. (B) Longitudinal within-participant MHI-A values for the subset of RBX2660-treated responders from whom all three displayed timepoints were received. (C) MHI-A values for all RBX7455 responder samples received and for administered RBX7455 drug product.

Figure 3

MHI-A values for RBX2660 investigational product administered in PUNCH CD2 (RBX2660) and three published healthy cohort studies, including the Human Microbiome Project (HMP), healthy Scandanavian adults (PopCol), and FMT donors from three published studies. The dotted line shows the MHI-A = 7.2 threshold, above which MHI-A values correspond to healthy based on ROC analysis.

Figure 4

MHI-A values for patients in three published studies of the effect of antibiotics on the microbiome, shown with individual sample values, median, and interquartile range. (A) MHI-A values before (BL) and at 4 days (4 D), 8 days (8 D), 6 weeks (6 W), or 6 months (6 M) after the start of a 4-day course of a cocktail of three last-resort antibiotics: meropenem, gentamicin, and vancomycin. The 4-day timepoint coincided with the end of treatment. The dotted line shows the MHI-A = 7.2 threshold, above which MHI-A values correspond to healthy based on ROC analysis. (B) MHI-A values before (BL), at the end of treatment (EOT), and 1 month (1 M), 2 months (2 M), 4 months (4 M), or 12 months (12 M) after treatment with one of four antibiotics (ciprofloxacin, Cip; clindamycin, Clin; minocycline, Mino; or amoxicillin, Amox) or placebo (Pbo). The treatment course for all antibiotics was 1 week. (C) MHI-A values before (BL), and 5 days (5 D), 10 days (10 D), 25 days (25 D), and 40 days (40 D) after the start of treatment with a 10-day course of ridinilazole (Rid) or vancomycin (Van) for CDI.