Ranking and classifying the VUS for family counseling: Using a model - PowerPoint PPT Presentation

Ranking and classifying the VUS for family counseling: Using a model from cancer researchers Martin Tristani-Firouzi, MD Division of Pediatric Cardiology Nora Eccles Harrison CVRTI University of Utah School of Medicine

Disclosure The speaker has no commercial or financial relationships to disclose.

Variants / genome (million) 150 protein truncating variants 10,000 amino acid changing variants 500,000 variants in regulatory regions

The dreaded result: a variant of unknown significance (VUS)

How do we interpret the significance of VUS? Segregation of variant w/ disease in families • Occurrence in multiple unrelated individuals • Functional assays • In silico prediction algorithms •

Most VUS are rare and “private” Segregation of variant w/ disease in families • Occurrence in multiple unrelated individuals • Functional assays • In silico prediction algorithms •

SIFT: Sorting Tolerant from Intolerant tool Kumar et al, Nature Protocols, 2009

Physico-chemical properties of amino acid substitution: predicted mutation effect charged polar hydrophobic Livingstone & Barton, CABIOS , 9 , 745-756, 1993

Other in silico tools for pathogenicity prediction

Meta-SVM: combining multiple prediction tools improves accuracy Dong et al, Hum Mol Genetics, 2015

LQTS and BrS SCN5A variants Receiver operator curve for when >4 in silico tool are in agreement True positive rate False positive rate Kapplinger et al, Circ Cardio Genet, 2015

The addition of topology to pathogenicity prediction Whicher and MacKinnon, Science, 2016

The addition of topology to pathogenicity prediction Kapplinger et al, Circ Cardio Genet, 2015

The Cancer Field approach to VUS

Variant classification scheme Class 5- Pathogenic, > 99% probability of pathogenicity Class 4- Likely Pathogenic, 95-99% probability of pathogenicity Class 3- Uncertain, 5-95% probability of pathogenicity Class 2- Likely Neutral, 0.1-5% probability of pathogenicity Class 1- Neutral, <0.1% probability of pathogenicity

Bayesian multi-factorial model of pathogenicity Posterior OR = Prior OR x OR for Pathogenicity from Data (D i ) predictive value for each D i is the probability of Class 5 pathogenicity, P(D i |C5) divided by probability of Class 1 benign, P(D i |C1) D 1 = in silico D 2 = functional assay D 3 = family history Prior OR x P(D 1 |C5)/P(D 1 |C1) x P(D 2 |C5)/P(D 2 |C1) x P(D 3 |C5)/P(D 3 |C1)

Variant classification scheme and clinical recommendations

Applying the Bayesian multi-factorial model of pathogenicity for LQTS Posterior OR = Prior OR x OR for Pathogenicity from Data (D i ) predictive value for each D i is the probability of Class 5 pathogenicity, P(D i |C5) divided by probability of Class 1 benign, P(D i |C1) D 1 = in silico D 2 = functional assay D 3 = family history Prior OR x P(D 1 |C5)/P(D 1 |C1) x P(D 2 |C5)/P(D 2 |C1) x P(D 3 |C5)/P(D 3 |C1)

Case-control comparison for LQTS variants (prior odds) Ruklisa et al, Genome Med, 2015

Applying the Bayesian multi-factorial model of pathogenicity for LQTS Posterior OR = Prior OR x OR for Pathogenicity from Data (D i ) predictive value for each D i is the probability of Class 5 pathogenicity, P(D i |C5) divided by probability of Class 1 benign, P(D i |C1) D 1 = in silico D 2 = functional assay D 3 = family history Prior OR x P(D 1 |C5)/P(D 1 |C1) x P(D 2 |C5)/P(D 2 |C1) x P(D 3 |C5)/P(D 3 |C1)

Functional characterization of candidate variants zebrafish Human iPSC-CMs

Why zebrafish?: repolarization properties similar to human

High throughput screening platform for phenotype-based repolarization screen

Functional effects of putative LQT2 mutations and polymorphisms as determined by zebrafish cardiac assay LQT-2 mutants polymorphisms +/- 95% CI

Comparison of in vivo zebrafish cardiac assay with in vitro mammalian cell assay

Applying the Bayesian multi-factorial model of pathogenicity for LQTS Posterior OR = Prior OR x OR for Pathogenicity from Data (D i ) predictive value for each D i is the probability of Class 5 pathogenicity, P(D i |C5) divided by probability of Class 1 benign, P(D i |C1) D 1 = in silico D 2 = functional assay D 3 = family history Prior OR x P(D 1 |C5)/P(D 1 |C1) x P(D 2 |C5)/P(D 2 |C1) x P(D 3 |C5)/P(D 3 |C1)

Findmyvariant.org The FindMyVariant team is affiliated with the University of Washington, Department of Laboratory testing.

Steps involved in investigating the family disease Talking with Your Immediate • Family About Your Variant Talking with Living Relatives • to Find Your Ancestors Using Online Social • Networking Sites to Find Descendants of Your Ancestors

Final family history and pedigree

Bayesian multi-factorial model of pathogenicity Posterior OR = Prior OR x OR for Pathogenicity from Data (D i ) predictive value for each D i is the probability of Class 5 pathogenicity, P(D i |C5) divided by probability of Class 1 benign, P(D i |C1) D 1 = in silico D 2 = functional assay D 3 = family history Prior OR x P(D 1 |C5)/P(D 1 |C1) x P(D 2 |C5)/P(D 2 |C1) x P(D 3 |C5)/P(D 3 |C1)

Establishing a consortium of LQTS experts to adjudicate VUS Identifying a funding mechanism Partnering with commercial and research genetic testing Partnering with SADS and findmyvariant.org Ideally review QTc, symptoms, functional data, family history

Thank you…and questions?