Global evaluation of congenital heart disease-associated non-coding variants.

Abstract

Genome-wide association studies have revealed that over 90% of disease- and trait-associated variants have been mapped within the non-coding genome. Non-coding variants can impact cellular and organismal phenotypes by altering biochemical interactions between TFs and CREs, which can disrupt gene expression. Advancements in human sequencing have identified thousands of CHD-associated gene variants, with over 97% occurring within the non-coding genome. However, the biochemical and gene mechanisms behind the causality of CHD-associated variants remain mostly unknown, as previously discussed in detail (see Chapter 1). In this work, we implement a combined in silico, <em>in vitro</em>, and cellular approach to prioritize and evaluate the impact of CHD-associated SNPs on TF-DNA binding and regulatory activity.<br /> <br /> In the 1st aim of this project (see Chapter 2), we identified and prioritized CHD- and cardiovascular disease (CVD)-associated SNPs through a combination of population-specific linkage disequilibrium analyses, regulatory element mapping, cardiac-specific eQTL data integration, and predictive modeling. The LS-GKM SVM model, specifically trained with ChIP-seq data for the cardiac TFs NKX2-5, GATA4, and TBX5 accurately predicted changes in TF binding affinity for selected variants. Biochemical validation through EMSA and luciferase assays confirmed the model's predictions, establishing the reliability of this integrated computational pipeline for functional variant prioritization. These findings highlight the utility of machine learning approaches over traditional PWM-based methods in identifying biologically relevant variants affecting cardiac TF binding.<br /> <br /> In the 2nd aim of this project (see Chapter 3), we performed a high-throughput TF-DNA binding assay on all 3,232 CHD-associated SNPs for NKX2-5, GATA4, and TBX5. Towards this goal, we developed SNP Bind-n-Seq, a gel shift assay-based high-throughput approach that measures the allele-specific binding of non-coding variants to a TF. Using SNP Bind-n-Seq, we constructed binding curves for 3,232 CHD-associated variants to evaluate allele-specific binding of the cardiac TFs NKX2-5, GATA4, and TBX5. We generated 9,600 binding curves for every possible allele of CHD-associated SNPs, resulting in ~38,400 enrichment measurements (~12,928 per TF). In doing so, we identified 256 variants with allele-specific binding for at least one of the cardiac TFs used in this work. This approach uncovered significant allele-specific binding differences for NKX2-5,GATA4, and TBX5 across hundreds of SNPs, revealing both motif-dependent and more-independent regulatory mechanisms. Importantly, approximately 40% of the identified SNPs impacted TF binding without disrupting core motifs, suggesting that regulatory variant effects are more complex and context-dependent than previously appreciated. This systematic evaluation provides a robust dataset valuable for training predictive models capable of capturing the complexities of TF binding dynamics.<br /> <br /> In the 3rd aim of this project (see Chapter 4), we expanded on the 256 SNPs with allele-Specific binding to understand their potential to alter gene expression. SNPs with allele-specific binding for NKX2-5, GATA4, or TBX5 showed genotype-dependent gene expression when evaluated through luciferase reporter assays. Additionally, we identified 65 SNPs in cardiac eQTL with 29 genes in the heart atrial appendage and left ventricle. We constructed a complex TF-SNP-Gene interaction networks linked to cardiovascular diseases and traits. Finally, we performed an in vivo reporter assay on transgenic zebrafish larvae and observed altered gene expression with allele-specific binding SNPs. In short, we provided a comprehensive evaluation of CHD-associated SNPs with allele-specific binding and regulatory activity in cellular (reporter assays), tissue (eQTL analysis), and in vivo (zebrafish transgenic system) context. Taken together, the three aims presented in this thesis substantially advance our understanding of how non-coding gene variants contribute to CHDs through complex regulatory mechanisms. Our integrated approach combining computational predictions, extensive biochemical validations, and robust cellular and organismal models provide a comprehensive framework for dissecting regulatory gene mechanisms underlying complex diseases.<br /> <br /> Although there is still much to be explored regarding the link between disease-associated SNPs and CHD development, we provide a comprehensive and systematic evaluation of CHD-associated variants on cardiac TF-DNA binding. First, we provide in silico CHD- and CVD-associated SNP predictions on cardiac TF-DNA binding and the trained models for public use. Second, we performed in vitro validation on all CHD-associated SNPs that can be used to train better predictive models to identify potential disease-causal variants. Finally, we complemented biochemical affinity validations with reporter assays and eQTL analysis to identify potential causal SNPs and genes in CHD development. This work provides the latest and most extensive characterization of CHD-associated SNPs in an in silico, <em>in vitro</em>, and cellular context.