Meta-analytic microbiome target discovery for immune checkpoint inhibitor response in advanced melanoma

Immune checkpoint inhibitors have transformed melanoma therapy, yet only a subset of patients achieve durable responses. Gut microbes have been linked to response, but reported biomarkers vary across studies. We aim to identify reproducible microbial features and test their generalizability across cohorts and treatment settings.

We reprocessed stool metagenomic sequencing data from 15 melanoma cohorts (763 samples from 484 individuals), including 12 cohorts treated with immune checkpoint inhibitors alone and 3 trials combining immune checkpoint inhibitors with fecal microbiota transplantation. Using a unified analysis pipeline, we profiled microbial species, metabolic pathways, and biosynthetic gene clusters, and analyzed their associations with treatment response using Tweedie regression, random-effects meta-analysis, and multimodal integration with leave-one-dataset-out validation.

Here, we show that responders in immune checkpoint inhibitor-only cohorts are enriched for several short-chain fatty acid-producing commensals, whereas non-responders show higher abundance of taxa associated with disrupted gut communities. In fecal microbiota transplantation plus immune checkpoint inhibitor trials, response associates with distinct communities and shifts in amino-acid, nucleotide and cofactor metabolism. Across cohorts, multiview prediction models repeatedly select gene clusters linked to antimicrobial peptides and surface polysaccharides, but cross-study discrimination remains modest.

Microbiome signatures of response are treatment-context dependent and are not captured by a single universal species. These harmonized findings prioritize microbial taxa and functions for mechanistic studies and future microbiome-informed interventions.