Enhancing a low-cost, minimally invasive screening test for dihydropyrimidine dehydrogenase mutations linked to 5-fluorouracil sensitivity by integrating computational mutation predictions.

226 Background: 5-Fluorouracil (5-FU), a common chemotherapy for solid tumors, is metabolized primarily by dihydropyrimidine dehydrogenase (DPD), encoded by the DPYD gene. With > 200 known variants, individuals with nonfunctional DPYD alleles exhibit impaired 5-FU metabolism and are at risk of severe toxicity. Approximately 35% of the population has partial DPD deficiency, and 0.2% completely lack enzymatic activity. Despite this, DPYD genotyping is not standard practice. This study evaluates the use of non-invasive saliva samples for DPYD genotyping and combines AI-based molecular prediction modeling to search for novel variants within DPYD coding region. Saliva samples were collected from fifty-six healthy individuals and from gastrointestinal (GI) cancer patients including a family cohort of nineteen individuals and three unrelated patients undergoing chemotherapy. Methods: gDNA was sequenced for nine DPYD variants reported to reduced or abrogate DPD activity: c.1905+1G > A, c.1679T > G, c.2846A > T, c.1129-5923C > G, c.1236G > A, c.299_302delTCAT, c.703C > T, c.2983G > T, and c.557A > G. cBioPortal was utilized to search the TCGA database for studies involving colon cancer patients with novel DPYD mutations exhibiting a total FIS score of at least 2 or identified through 3D modeling of DPD using PyMol based on their proximity to or the presence of polar contacts within the active site. Putative pathogenic mutations were analyzed using AlphaMissense and ChimeraX to assign a RSMD score, assessing their potential negative impact on DPD function. Results: Computational studies identified three mutations in DPYD from TCGA colon cancer patients with an unknown impact on DPD and a FIS greater than 4 (c.198G > C/T, c.2161G > A, c.2185G > A), suggesting potential disruption of its function. Four mutations (c.424T > G, c.427T > G, c.2460G > C, c.2909C > A) were identified through protein modeling using PyMol, but only two had a FIS greater than 2. Of these seven identified mutations, four (c.198G > C/T, c.2161G > A, c.2185G > A, c.2909C > A) also scored highly using AlphaMissense and ChimeraX. In silico prediction models from VarSome listed c.2185G > A as Pathogenic Moderate and c.2909C > A as Pathogenic Very Strong. Sequencing of saliva samples indicated mutations c.1129-5923C > G and c.1236G > A were present in four and three volunteers, respectively, and c.1905+1G > A in one case. Mutations identified in GI cancer patients were c.1129-5923C > G and c.1236G > A. Conclusions: By combining computational modeling with the analysis of naturally occurring mutations in colon cancer patients, we have successfully identified potentially pathogenic mutations in the DPYD gene. This information can enhance existing clinical diagnostic tests, providing a more comprehensive assessment of DPYD mutations, including novel variants.