Dose-limiting toxicities remain a major barrier to drug development and therapy, revealing the limited predictive power of human genetics. Herein, we demonstrate the utility of a more comprehensive approach to studying drug toxicity through longitudinal study of the human gut microbiome during colorectal cancer (CRC) treatment ( NCT04054908 ) coupled to cell culture and mouse experiments. 16S rRNA gene sequencing revealed significant shifts in gut microbial community structure during oral fluoropyrimidine treatment across multiple patient cohorts, in mouse small and large intestinal contents, and in patient-derived ex vivo communities. Metagenomic sequencing revealed marked shifts in pyrimidine-related gene abundance during oral fluoropyrimidine treatment, including enrichment of the preTA operon, which is sufficient for the inactivation of active metabolite 5-fluorouracil (5-FU). preTA + bacteria depleted 5-FU in gut microbiota grown ex vivo and the mouse distal gut. Germ-free and antibiotic-treated mice experienced increased fluoropyrimidine toxicity, which was rescued by colonization with the mouse gut microbiota, preTA + E. coli , or preTA -high CRC patient stool. Finally, preTA abundance was negatively associated with fluoropyrimidine toxicity in patients. Together, these data support a causal, clinically relevant interaction between a human gut bacterial operon and the dose-limiting side effects of cancer treatment. Our approach is generalizable to other drugs, including cancer immunotherapies, and provides valuable insights into host-microbiome interactions in the context of disease. One Sentence Summary: Gut microbial enzymes can be used to predict and prevent anticancer drug toxicity.