Abstracts

Rationale: The multi-institutional, highly collaborative Epilepsy Multiplatform Variant Prediction (EpiMVP) Center Without Walls develops prediction tools (EpiPred) for variants of uncertain significance (VUS) in epilepsy genes. Starting with non-ion channel/receptor epilepsy-associated genes, EpiPred develops precise algorithms to correctly classify VUS with clinical impact as pathogenic or benign. Machine learning and in silico modeling are combined with biological data from in vitro and in vivo platforms. First, STXBP1 properties of pathogenic/likely pathogenic (P/LP) or benign/likely benign (B/LB) variants are characterized using a combination of HEK293 cells and patient derived or engineered human pluripotent stem cell (hPSC) models to establish STXBP1 loss-of-function (LoF) and variant phenotypes. These platforms will be used to assay VUS to inform refinement of the computational prediction tool.

Methods: HEK293T cells were transfected with increasing amounts of constructs encoding STXBP1 full-length protein with or without an introduced P/LP, B/LB or VUS variant to titrate expression. We also reprogrammed iPSC lines from STXBP1 patient and control fibroblasts and made neural progenitor cells (NPCs) to differentiate into neurons in 2-D culture. In addition, CRISPR/Cas9 edited or patient iPSCs transformed with NGN2 were used for forced differentiation into homogeneous cultures of iNeurons (see Wei et al., this meeting). Lysates from cultures were used in Western blot analyses of protein expression, stability/turnover, solubility in 0.1% TritonX-100, and protein-protein interactions with known SNARE complex binding partners. Anti-STXBP1 immunostained neurons from mutant vs WT lines were co-labeled with neuronal markers and the size of puncta within neurites were compared to assess protein aggregation.

Results: STXBP1 baseline and LoF biochemical phenotypes were established for transfected cells or iPSC-derived neurons. Compared to WT protein, B/LB variants showed slight variation in stability, solubility, and SNARE protein binding, while P/LP missense or truncated versions were significantly unstable, insoluble, with lower binding capacity for STX. Using STXBP1 patient variant (p.Gly544Val fs*2) iPSCs and controls, heterozygous mutant neurons displayed larger mean STXBP1-labeled puncta within neurites and decreased spontaneous MEA firing. With these assays we are evaluating, in a blinded manner, patient VUS selected with the first iteration of the EpiMVP prediction tool were used to test the validity and strength of features used in our machine learning approach to in silico modeling.

Conclusions: Our STXBP1 LoF biochemical assays in patient derived and engineered human cells successfully distinguish phenotypes caused by patient variants. Future assays will probe the accuracy of EpiPred to predict pathogenicity of VUS in STXBP1. Classifications based on 2-D in situ data will feed into more detailed investigation in 3-D cortical organoids and animal models.

Funding: Please list any funding that was received in support of this abstract.: U54NS117170 and Leon Levy Foundation.