Machine learning identifies a distinct microbiota signature in immune checkpoint inhibitor colitis compared with...

Immune checkpoint inhibitors (ICIs) have revolutionized oncology by enhancing antitumor immune responses. However, their use is frequently associated with immune-mediated adverse events, including colitis (ICIs-colitis). This condition shares clinical and histological features with inflammatory bowel diseases (IBD), such as Crohn's disease (CD) and ulcerative colitis (UC). Emerging evidence highlights the gut microbiota's role in both ICIs-colitis and IBD pathogenesis. This study aimed to determine whether a distinct microbiota profile, assessed via machine learning, could differentiate ICIs-colitis from IBD and healthy controls (HCs).

A prospective study was conducted with patients diagnosed with ICIs-colitis, alongside historical cohorts of IBD patients (active and inactive UC/CD) and HCs. Stool samples were analyzed using 16S rRNA gene sequencing. Diversity metrics (alpha and beta) and differential abundance at multiple taxonomic levels were evaluated. Machine learning techniques, including supervised and unsupervised algorithms, were employed to identify microbiota patterns and signatures distinguishing the groups.

Nineteen patients with ICIs-colitis, 40 with UC (20 active, 20 inactive), 34 with CD (14 active, 20 inactive), and 36 HCs were analyzed. Alpha diversity differed between ICI-colitis and UC (P = .03) and between ICI-colitis and CD (P = .0002), but not versus healthy controls (P = .94). Beta diversity showed significant disease-associated clustering among ICI-colitis, UC and CD (PERMANOVA P < .001). Differential abundance analyses identified higher Enhydrobacter in ICI-colitis versus IBD and higher Bifidobacterium longum in UC. Machine-learning approaches (sparse partial least squares discriminant analysis, sPLS-DA, and Random Forest) supported group discrimination.

A unique microbiota signature characterizes ICIs-colitis compared to IBD and HCs. These findings underscore the potential of microbiota profiling and machine learning to aid in diagnosing and managing ICIs-colitis. Future studies should validate these findings in larger, multicenter cohorts, and explore therapeutic implications.