Taxonomy-free fecal microbiome profiles enable robust prediction of immunotherapy response and toxicity in melanoma

The gut microbiome has been causally linked to the efficacy of immune-checkpoint inhibitor therapy (ICI), prompting numerous clinical trials of microbiome-targeting strategies. Yet, mechanisms by which gut microbiota shape immune responses remain elusive as taxonomic biomarkers have failed to generalize across multiple cohorts. In this study, we develop a taxonomy-agnostic framework to identify microbial biomarkers of ICI response and immune-related adverse event (irAE) occurrence from metagenomic sequencing. Applying this approach to four independent melanoma cohorts from clinical centers across the United States, we uncover gut microbial proteins produced by diverse bacterial taxa that consistently predict ICI response. Notably, we uncover a previously uncharacterized operon involved in cellular redox homeostasis that is encoded by different bacteria and reliably predicts irAE occurrence. We further validated the predictive power of this operon in a prospectively sequenced melanoma cohort. Our results demonstrate that taxa-agnostic microbial protein biomarkers are robust, generalizable, and provide a path towards pretreatment risk stratification for melanoma patients initiating ICI therapy.