Developing A Prediction Model for SCN5A Variants in BrS and LQT3

SCN5A基因是一個重要的鈉離子通道基因，當此基因上出現突變時，容易造成突發性心臟疾病包括布魯格達症候群 (Brugada Syndrome,BrS)以及長Q-T間期症候群 (Long QT Syndrome type 3 LQT3)。此類型的突發性心臟疾病，臨床上常常被醫生忽略，因此無法作出精確的診斷與治療。自SCN5A基因與心臟疾病相關性被研究以來，至今已有上百個SCN5A的基因序列變異位點被發現，但是其潛在的致病機轉及基因型與表現型之間的關係尚不清楚，需要更深入的研究，才能確定新的變異是否會造成疾病。本篇論文的研究目為分析SCN5A基因變異的特性，並嘗試探究SCN5A造成心臟疾病: 布魯格達症候群或長Q-T間期症候群的可能性，並期待透過統計模型分析每個變異位點具有的風險性。 在本篇研究中，共計使用4種演算法來分析基因變異是否會導致疾病，所使用的演算法包括Sorts Intolerant From Tolerant, Protein Variation Effect Analyzer, Polymorphism Phenotyping v2 and Genomic Evolutionary Rate Profiling++。本篇研究的SCN5A相關變異資料是從過去的文獻及發表研究中所收集，其中在布魯格達症候群上有425個突變位點，而在長Q-T間期症候群上則有136個突變位點。我們針對這些位點使用Estimated Predictive Values （EPV）來檢定所有變異的風險，並依據蛋白質結構及基因表現子的結構情形進行EPV區域計算。實驗結果顯示當基因變異位點被3種以上演算法認定為惡性時，可加強EPV所測定風險的能力。舉例來說， 布魯格達症候群 中PFAM-B3701和Na Transmembrane Associate區域之EPV，經過位點篩選後，前者由原來的56％提高到75％，後者則由60％提升至83％。 整體而言，本方法能夠有效地區分病患和正常人發現的基因變異位點，根據篩選結果，可以利用具有差異性的突變位點進一步建立機率模型來預測變異位點之風險值。此外， SCN5A突變位點的結構位置與病患發展成布魯格達症候群和長Q-T間期症候群具有關聯性，數據顯示座落於Transmembrane Domain II (DII)的變異有較高的機率導致布魯格達症候群，而在C-terminus的變異有較高的可能造成長Q-T間期症候群。總結來說，本論文推導的SCN5A基因變異分類模型，結合了演算法以及EPV，將可用於分析SCN5A基因變異的危險性與區分形成布魯格達症候群和長Q-T間期症候群的機率。SCN5A encodes a cardiac sodium channel. Its mutations are associated with Brugada Syndrome (BrS) and Long QT Syndrome Type 3 (LQT3). Both diseases are often neglected by the clinicians because they are difficult to diagnose. Hundreds of non-synonymous variants have been identified in SCN5A; however, the underlying mechanism and the relationship between the genotype and phenotype remain unclear. A new approach that helps to screen and prioritize identified mutations is beneficial for researchers to identify a novel pathogenic mutation in this high-throughput sequencing era. Therefore, we aim to study and analyze the characteristics of SCN5A variants in order to evaluate the possibility of these mutations developing into BrS or LQT3. In this study, 4 prediction algorithms were used to predict whether a variant is pathogenic or benign. The algorithms includes: Sorts Intolerant From Tolerant (SIFT), Protein Variation Effect Analyzer (PROVEAN), Polymorphism Phenotyping v2 (PolyPhen2) and Genomic Evolutionary Rate Profiling++ (GERP++). Several variants (BrS N=425, LQT3 N=136) were collected from literatures and published reports. Furthermore, Estimated Predictive Values (EPV) is used to evaluate the frequency of one variant in a rare disease, such as BrS or LQT3. Therefore, for each variant, EPV was calculated and all variants were classified into different groups based on the protein structures and exon information. The results demonstrated that higher prediction performances can be obtained when at least 3 prediction algorithms agreed on pathogenicity. For example, the EPVs increased from 56% to 75% and 60% to 83% in the domains of Pfam-B3701 and Na transmembrane in BrS, respectively. In general, the results showed that the proposed approach was able to discriminate case-derived variants and general-population-derived variants. Based on the filtered variants, a prediction model was developed to evaluate potential risk for each variant. In addition, the associations between the SCN5A domains and the two diseases, BrS and LQT3, were evaluated. Intriguingly, the results showed that a variant in domain II (DII) transmembrane has a higher possibility that can develop into BrS. Similarly, a variant in the C-terminal may have a higher chance turning into LQT3. In conclusion, a probability model that integrates EPV and 4 prediction algorithms was developed in this study in order to classify variants identified in SCN5A and evaluate the chance that such variants may lead to BrS or LQT3.誌謝 I 中文摘要 II ABSTRACT IV Table of Contents VI List of Tables VIII List of Figures IX Chapter 1 Introduction 1 1.1 Brugada Syndrome (BrS) 1 1.2 Long QT3 (LQT3) 1 1.3 SCN5A 2 1.4 Prediction Algorithms 3 1.5 Motivation 4 1.6 Specific Aim 6 Chapter 2 Materials and Methods 8 2.1 Data Collection 9 2.2 Prediction Algorithms 10 2.3 Statistical Analysis 12 Chapter 3 Results 14 3.1 EPVs in BrS and LQT3 Variants 16 3.2 Differentiation of Case- Versus General-Population-Derived Variants by Prediction Algorithms 18 3.3 Enhancement of EPV Analysis by In-silico Algorithms 21 3.4 Variants Distribution in Domain and Exon 29 Chapter 4 Discussion 35 4.1 Functional Analysis of SCN5A Variants 35 4.2 Model Reliability 37 4.3 Characteristics of the Prediction Algorithms 39 4.4 Limitations 43 Chapter 5 Conclusion 46 References 47 Appendix 56 Supplementary Table A1 56 The List of all Variants Collected for BrS 56 The List of all Variants Collected for LQT3 7