Testing for four dihydropyrimidine dehydrogenase (DPYD) variants (DPYD*2 A, DPYD*13, c.2846 A > T, DPYD-HapB3) is currently implemented in clinical practice to prevent fluoropyrimidines (FLs) related toxicity but with limited sensitivity. This study aimed to identify novel genetic factors in FL-related genes to enhance risk prediction using data from the PREPARE trial (NCT03093818). Two hundred seventy-four patients receiving FL-based chemotherapy with severe toxicity were sequenced for 60 candidate genes. Gene and pathway-level association analyses focusing mainly on rare variants were performed using dedicated statistical tests, including gene-wise variant burden (GVB) analysis. DPYD germline variant burden beyond the four routinely tested markers emerged to contribute to toxicity, indicating that rarer genetic variants could help in refining the optimal FL dosage (p < 0.1). Functional rare variant burden in ABCB5, PARP1, ENOSF1, CYP3A4 and nuclear receptors pathway impacted on toxicity risk (p < 0.05 in at least one statistical test). GVB analysis confirmed ABCB5 as a significant risk gene and highlighted ABCC4, HNF4A, and XRCC3 as additional candidates. A predictive model combining genetic burden scores with clinical variables improved the identification of high-risk patients (sensitivity=0.71, specificity=0.74, accuracy=0.73). This study indicated a paradigm shift from population to individual-level arguing for an extension of testing beyond the four DPYD currently considered variants to predict FL-related toxicity.
