Circulating microbe-derived metabolites are associated with protection from cGVHD. (A) Metabolomic analysis of SCFAs in peri-day 100 plasma samples from MSKCC (n = 10 cGVHD, 43 control, 5 healthy volunteers). P values were determined on the basis of analysis of variance followed by Mann-Whitney U testing. One sample was included per patient; where multiple plasma samples were available, the closest sample to day 100 was chosen. Three cGVHD patients and 13 control patients had both stool and plasma available in the peri-day 100 window. (B) Log₁₀ relative abundance of 14 candidate microbial genera in cGVHD patients and controls. (C) Logistic regression coefficients of the genus abundances associated with cGVHD or control status (9 and 36 patients, respectively, all recipients of unmodified grafts). Bars represent 95% highest posterior density intervals (HDIs), and intervals highlighted in blue and red are entirely >0 or <0. (D-E) Recall and accuracy of our model compared with 100 000 draws from a binomial null model with P = .2 as per the case-control design. (F-G) Predicted probability of cGVHD at different abundances of Lachnoclostridium or Faecalibacterium in 1000 simulated microbiota communities per abundance level (minimum observed [min. obs.], lowest observed, nonzero family abundance). (H) Duke plasma samples from a cross-sectional cohort of patients who did or did not go on to develop cGVHD (n = 11 cGVHD, 17 no cGVHD).