Decrypting Variants of Uncertain Significance in Long-QT Syndrome

Clinical genetic testing has become standard-of-care for many diseases including the congenital long-QT syndrome (LQTS). However, interpreting genetic test results is often confounded by the discovery of ‘variants of uncertain significance’ (VUS) for which there are insufficient data to determine whether a particular variant is pathogenic or benign. The goal of this project is to use a novel paradigm for distinguishing pathogenic from benign variants in LQTS with a focus on KCNQ1, the most common cause of LQTS. During the prior periods of support, we implemented high throughput strategies to determine the functional consequences of ~180 KCNQ1 variants, the functional consequences of all known disease-associated KCNE1 variants, and assessed the stability, structure, and cell surface expression of several dozen KCNQ1 variants. We then integrated data on KCNQ1 structure, function and sequence conservation with machine learning tools to build a gene-specific algorithm in a web-based format to predict the likelihood that specific KCNQ1 variants are deleterious. In the next funding period, we propose to continue this powerful and productive multidisciplinary paradigm to extend our research. We used our machine learning approach incorporating an artificial neural network (ANN) model to predict the functional consequences of 136 KCNQ1 VUS from ClinVar. In Aim 1, we will experimentally evaluate predictions made using our machine learning algorithm of functional consequences of 136 KCNQ1 VUS from ClinVar using automated patch clamp recording. In separate experiments, we will perform deep mutational scanning (DMS) of major regions of KCNQ1 (pore and voltage-sensing domains, C-terminus) to identify all possible single nucleotide variants that cause impaired trafficking of the channel to the plasma membrane. In Aim 2, we will use our quantitative flow cytometry-based method to evaluate cell surface expression of disease-associated KCNQ1 variants in other less well-studied regions of the channel (pore domain, C-terminus), and to determine if dysfunctional KCNE1 variants interfere with KCNQ1 trafficking. We will also employ biophysical methods (NMR spectroscopy, differential scanning fluorimetry, cellular thermal shift assay) to evaluate the stability of trafficking-impaired KCNQ1 variants in the context of purified channel protein consisting of the voltage-sensor and pore domains. In Aim 3, we will evolve our machine learning algorithm as a deep neural network and train with new data from Aims 1 and 2. Further enhancements to algorithm performance will be achieved using structural channel models built with a custom version of AlphaFold2.0, computed free energy, and outputs from molecular dynamics simulations of KCNQ1-KCNE1 channels. Our study will yield a large and unprecedented database of functional, structural, and biochemical properties of hundreds of KCNQ1 and KCNE1 variants, along with an advanced, data-trained computational prediction algorithm capable of accurately discriminating deleterious from benign variants. These results will contribute to improving genetic test interpretation and medical decision-making for LQTS.