Decrypting Variants of Uncertain Significance in Long-QT Syndrome

Clinical genetic testing has become standard-of-care for many diseases including hundreds of inherited conditions. However, interpreting genetic test results is often confounded by the discovery of ‘variants of unknown significance’ (VUS) for which there is insufficient data or inadequate predictive tools to establish whether or not a particular variant predisposes to a disease. This problem is particularly vexing for genetic disorders with strong allelic heterogeneity and a preponderance of ‘private’ mutations such as the congenital long-QT syndrome (LQTS), which predisposes young adults and children to sudden death from cardiac arrhythmias. With the anticipated incorporation of personal exome or genome data into routine clinical care, interpreting VUS will become an even greater challenge especially when variants in genes associated with human disorders are incidentally discovered. Unfortunately, there are no reliable methods to predict a priori whether a given variant predisposes an individual to a particular disorder or whether the change is merely a benign rare variant. We propose to develop a novel paradigm for distinguishing disease-causing mutations from benign variants in LQTS as a model for other inherited arrhythmia syndromes and channelopathies. We will focus on variants in KCNQ1, the most commonly mutated gene in LQTS. The central hypothesis of this proposal is that a holistic predictive model that relates experimentally determined protein structure and dynamics to function and disease is highly accurate even for novel variants. Our ultimate objectives are to develop a data-trained, web-accessible algorithm that classifies VUS discovered in KCNQ1 based on reliable predictions of the structure and dynamics of the affected protein, and to achieve prediction accuracy to levels needed to inform medical decisions. The medical importance of correctly classifying KCNQ1 variants provides strong justification for having a dedicated and highly-tailored gene-specific prediction model. The ability to distinguish deleterious from neutral variants would help avoid unnecessary and potentially harmful interventions in carriers of benign alleles, and save the lives of those with true mutations. We propose to collect extensive electrophysiological, biochemical and structural data on a large set of KCNQ1 variants discovered in LQTS subjects as well as several suspected benign or neutral variants (Aims 1-2), then use these data to iteratively train and validate a machine learning based algorithm that can differentiate benign from deleterious KCNQ1 alleles among a set of new VUS (Aim 3). Our proposal is innovative in the use of a multidisciplinary approach to functionally and structurally annotate genomic variant data for a medically important gene at an unprecedented scale, and then to use these experimental findings to train/test a novel computational system to achieve clinical-grade predictions. Targeting KCNQ1 will also validate an approach for parallel work that can be utilized to predict the medical significance of variants in closely related potassium channels associated with heritable epilepsy (KCNQ2, KCNQ3) or deafness (KCNQ4).