Base editing for systematic reclassification of variants of unknown significance in NF1

NF1 is an established tumor suppressor, frequently mutated in cancer genomes and responsible for the cancer-predisposing syndrome neurofibromatosis. NF1 is very large (~350 kb) and has one of the highest mutation frequency in the human genome. Due to incompletely understood biology of the coded protein, high sequence homology, allelic heterogeneity and lack of functional assays, the interpretation of NF1 variants is highly challenging in clinical practice. According to the ClinVar database, 13460 NF1 variants are listed (data extracted up to July 2024) with missense variants being the most represented type, of which 84.8% are variant of unknown significance - VUS.We recently identified a novel function in microtubular damage repair for NF1 (Duso et al Biorxiv 2025). NF1-deficient cells become highly sensitive to microtubule-depolymerising maytansinoids like DM1. We designed a high-throughput base editing screen that exploits this differential sensitivity to identify NF1 variants that lead to its loss of function.SpliceR was employed to design control guides (non targeting, intragenic and splice essential and non essential targeting regions) while for gRNAs directing to NF1, the BEstimate tool was utilised. As a model, we generated HER2+ breast cancer cell line, HCC-1954 and to 293T cells to stably express lentiviral cytidine-(CBE) and adenosine-base editors (ABE). Editing efficiency is evaluated through the BEAR-GFP reporter plasmid. We obtained a library covering 8191 variants with 5569 total guides. Of these, the majority are nonsynonymous. We analysed the coverage of known NF1 variants. Based on ClinVar classification, we cover 882 VUS, 68 conflicting interpretation of pathogenicity (CIP), 202 pathogenic/likely pathogenic (P/LP) variants, and 25 benign/likely benign (B/LB) variants. Based on our RENOVO tool for VUS reinterpretation, 4968 covered variants are considered Low-precision Pathogenic or Benign.Key functional domains in NF1 are covered with high density, in particular with 573 variants covering the functional GTPase-activating domain (GAP), 2745 covering the HEAT domain, likely to mediate interactions with microtubules, and 267covering the SecPH domain, essential for membrane localization. 4257 variants in splicing sites and 3 prime untranslated regions (UTR) variants are covered.The results of the functional screen are integrated with public data to develop a novel NF1 predictor based on our RENOVO architecture (Favalli V et al., Am J Hum Genet. 2021 and Bonetti E, et al., Hum Genomics 2025).These tools will be essential to overcome key diagnostic challenges in neurofibromatosis and precision oncology.