Predicting Gut Microbiota Dynamics and Allo-HCT Survival By Global Microbiota Community

Intestinal microbiota composition is strongly associated with HCT patient outcomes. We have reported associations between antibiotic exposures and microbiota diversity and GVHD-related mortality. However, understanding microbiota-composition dynamics in response to specific perturbations is challenging due to the high-dimensional nature of microbiota data. This study identified clusters of intestinal microbiota compositions and investigated their dynamics using transition probabilities in a large dataset of allo-HCT fecal specimens. The bacterial compositions of 7,930 samples from 1,076 allo-HCT patients were determined by 16S rRNA deep-sequencing and visualized by tSNE ( Fig. a ). Samples were clustered into 10 distinct microbiota configurations by k-means clustering of a b-diversity matrix ( Fig. b ). Visually, clusters reflected monodominant taxonomic groups: cluster 5 overlaps with the Enterococcus group, while cluster 10 overlaps with the Streptococcus group. These clusters also captured variations in diversity, as clusters 1-2 and 5-10 represented high- and low-diversity states, respectively ( Fig. c ). Clusters also exhibited dynamic behaviors: high-diversity clusters 1-2 were common in pre-HCT samples, while most post-HCT samples belonged to low-diversity clusters 5-10 ( Fig. d ). The temporal behaviors of microbiota composition per patient could be modeled by cluster transition probabilities from 1 week pre- to 1 week post-HCT. Without piperacillin-tazobactam (pip-tazo) administration, patients who had a diverse composition pre-HCT were most likely to maintain their diverse state (probability P = 20%), and had a low chance of transitioning to the Enterococcus cluster (cluster 5, P = 5%) and to the Streptococcus cluster (cluster 10, P = 0%) post-HCT. However, pip-tazo exposure was associated with increased transition to the Enterococcus cluster (P = 15%) and to the Streptococcus cluster (P = 9%), and decreased diversity maintenance (P = 14%) ( Fig. e ). The occurrence of each cluster at peri-engraftment time (days 7-21) was used to estimate its association with patient survival. The Streptococcus cluster was associated with increased mortality, highlighting a disadvantageous cluster transition under pip-tazo exposure ( Fig. f ). While prior studies have associated bacterial taxa or diversity indices with biomarkers of clinical outcomes, here we considered the entire intestinal communities and demonstrated that post-HCT mortality risk can be predicted by the global microbiota composition at days 7-21. This computational framework can be used to predict cluster transitions in response to various clinical variables such as specific drug exposures and clinical events by means of a high-resolution transition matrix, ultimately informing strategies to optimize treatment plans for HCT patients to maximize a healthy gut microbiota state and clinical outcomes.