👤 Sonja Loges

Variants of uncertain significance represent the biggest challenge for genomics-based precision oncology. Activated fibroblast growth factor receptors (FGFRs) frequently drive tumorigenesis by different genetic aberrations. However, it remains unknown which of the many point mutations affecting FGFR1, FGFR2, FGFR3 or FGFR4 in cancer are druggable, that is, activating signaling while not mediating FGFR inhibitor resistance. Here we implemented a saturation mutational scanning platform to screen all 11,520 possible point mutations covering the kinase domains of FGFR1-4. Pooled positive selection screens identified 474 activating and 738 mutations mediating resistance to the FGFR inhibitors pemigatinib and futibatinib, together revealing 301 druggable FGFR mutations analogous to a strong PS3/BS3 evidence level. The screens also identified loss-of-function mutations and an association of gain-of-function mutations with hydrophobic changes. The functional screens identified 97% of acquired resistance mutations in clinical trials. Our comprehensive catalog of every druggable mutation in the FGFR kinase domains is readily available for clinical decision support. Show less

Fibroblast growth factor receptors (FGFRs; FGFR1, FGFR2, FGFR3, FGFR4) are frequently mutated oncogenes in solid cancers. The oncogenic potential of FGFR rearrangements and few hotspot point mutations is well established, but the majority of variants resulting from point mutations especially outside of the tyrosine kinase domain are currently considered variants of unknown significance (VUS). Recurrent nonkinase domain FGFR VUS variants were collected from the Catalog of Somatic Mutations in Cancer and their oncogenic potential was assessed in vitro by different functional assays. We compiled published clinical and preclinical data on FGFR variants and compared the data with results from our functional assays and pathogenicity predictions of state-of-the-art artificial intelligence (AI) models. We identified 12 novel FGFR extracellular small variants with potential driver function. Comparison of clinical and preclinical data on FGFR variants with pathogenicity predictions of state-of-the-art AI models showed limited usefulness of the AI predictions. Sensitivity profiles of activating FGFR variants to targeted inhibitors were recorded and showed good targetability of FGFR nonkinase domain variants. The collected results extend the spectrum of suitable FGFR variants for potential treatment with FGFR inhibitors in the context of clinical trials and beyond. Current AI models for variant pathogenicity prediction require further refinement for use in oncogenic decision making. Show less