Unraveling Difficult Answers: From Genotype to Phenotype in Coronary Artery Disease

Abstract

Genome-wide association studies (GWASs) have robustly found a correlation between coronary artery disease (CAD) and an intergenic region at locus 9p21.3. However, the mechanistic implication of this association is unknown. Recently in Cell, Lo Sardo et al. used hiPSC genome editing to demonstrate how this locus contributes to CAD predisposition.

GWASs have proven to be powerful tools to discover the associations between genomic variants and patient phenotypes, providing novel insights into the mechanisms involved. GWAS tools have moved from assessing just a small proportion of the genome, commonly 700,000 base pairs on basic SNP-chips, to trending toward the 3.2 billion base pairs of the whole genome. Additionally, the number of participants in these studies has massively increased, to now over 1 million participants (Evangelou et al., 2018). It has become clear that the discovered SNPs are actually less commonly non-synonymous (i.e. they change the amino acid sequence) and in genes with a rational role in the phenotype. Instead, they are in fact synonymous (i.e. they do not change the amino acid sequence) in genomic regions that are not thought to be expressed, or merely coinherited with functional SNPs (in linkage disequilibrium). For example, a large study on cardiac QT intervals completed in >75,000 participants discovered 22 new loci, none of which changed the amino acid sequence (Arking et al., 2014).

CAD is one of the major causes of death worldwide, responsible alone for ~17% of deaths in the USA (Go et al., 2014), and it has been a major target of GWASs, resulting in the identification of 91 risk loci (Howson et al., 2017; Nikpay et al., 2015). The most robustly associated locus is a cluster of about 59–100 SNPs in perfect linkage disequilibrium across an ~60 kb intergenic region located on chromosome 9p21.3 (Figure 1). The closest genes to this location are cyclin-dependent kinase inhibitor 2A (CDKN2A) and 2B (CDKN2B), as well as the long non-coding RNA (lncRNA) (CDKN2B-AS1/ANRIL). The population-attributable risk that predis-poses CAD for this haplotype ranges between 11% and 17% across the population in an age-dependent manner (Gränsbo et al., 2013), making this the most highly associated correlation. Despite this disease importance and robust, well-replicated genetic association, the mechanistic role of this locus has not been elucidated. This is due to the location of the SNP cluster in a non-coding region, which therefore provides no genes of interest to track, as well as the difficulty of isolating the effector cell type for expression analysis that could yield expression quantitative trait loci (eQTL) data. In the case of CAD, vascular smooth muscle cells (VSMCs) are the main contributing cell type. Well-functioning VSMCs are contractile and responsible for the vascular tone in the arteries. When these cells are triggered to de-differentiate to synthetic VSMCs, they become less contractile, less adhesive, and more proliferative, promoting plaque formation, a key facet of atherosclerosis (Alexander and Owens, 2012).

Figure 1. Linkage Disequilibrium (LD) Map at the chr9p21.3 Locus.

Pairwise linkage disequilibrium (r2) is shown in different shades of gray (r² = 0, white, and r² = 1, black) for all SNPs spread over 59.85 kb on chr9p21.3, encompassing risk loci associated with CAD. The VCF file containing genotype calls for 99 CEU individuals was downloaded from the 1,000 genomes data-base for the 9p21.3 locus (chr9:22065657–22125503) using GRCh37.p13 build. Using VCFtools, VCF was converted to PED and MAP files. Plink (Purcell et al., 2007) was used to calculate linkage disequilibrium between SNPs, and finally the haplotype LD map was generated using Haploview V4.2.

In a recent issue of Cell, Lo Sardo et al. reveal the mechanism by which this interesting locus increases susceptibility to CAD using patient-specific human induced pluripotent stem cell (hiPSC)-derived VSMCs and a genome-editing-based approach they term “haplotype editing” (Lo Sardo et al., 2018). Lo Sardo et al. show that the 9p21.3 locus prompts a predisposition to CAD via altering VSMC state and functionality through global transcriptome regulation triggered by overexpression of lncRNA ANRIL. In this work, the authors used hiPSC-derived VSMCs to study CAD. In total, the authors generated 37 hiPSC clones from 4 patients harboring the risk (RR) and 3 patients harboring non-risk (NN) haplotypes (~60 kb). The authors then used TALENs to delete the ~60 kb haplotype in both RR and NN hiPSC clones, resulting in homozygous KO hiPSC clones.

Whole-transcriptome analysis during VSMC development showed that deleting the risk haplotype in RR resulted in an increased significant transcriptome alteration from D3 to D17 of VSMC development when compared to RR-KO, whereas haplotype deletion in NN showed only a minimal effect on VSMC transcriptome profiles. Interestingly, deleting the risk haplotype in CAD patient-derived cells (RR-KO) results in transcriptional profiles that resemble that of NN and NN-KO. Remarkably, transcriptional alteration caused by the RR risk haplotype is translated into a de-differentiated cell state showing more proliferative and less contractile VSMCs. Deleting the risk haplotype in RR (RR-KO) resolves this state back to a normal mature VSMC state (NN and NN-KO). The fact that the risk haplotype is composed of ~59–100 non-coding SNPs, some of which are located in terminal exons of ANRIL, raised a question about the mechanistic implications of ANRIL in transcriptomic changes and VSMC de-differentiation leading to CAD. RNA-seq revealed that a long ANRIL isoform (19 exons) is explicitly expressed in RR and NN in comparable values, whereas the short ANRIL isoform (13 exons) is expressed in NN, NN-KO, RR, and RR-KO. Interestingly, the short ANRIL isoform expression is very low in hiPSCs, but it increases through VSMC differentiation and is significantly higher in RR versus RR-KO, NN, and NN-KO.

This finding demonstrates that the presence of the risk haplotype is associated with higher expression of the shorter ANRIL isoform, which results in a switch in the VSMC phenotype into a functionally impaired state. Exogenous overexpression of the short ANRIL isoform in NN-KO VSMCs results in partial transcriptional, as well as complete phenotypical and functional, changes resembling RR VSMCs. The incomplete transcriptome recapitulation of the short ANRIL isoform overexpressing NN-KO in relation to RR suggests that there are other risk haplotype SNPs that contribute to changes in gene expression and subsequently to CAD predisposition in an ANRIL-independent manner.

Genomics research has identified thousands of loci associated with cardiovascular diseases. Further use of this data has been moderated by the lack of any functional explanation for the implication of variants in disease predisposition. The enormous improvements in hiPSC reprogramming, differentiation protocols toward different cell types involved in heart disease, genome editing, high-throughput phenotypic assays, and sequencing strategies have enhanced the bridge from genotype to phenotype (Magdy et al., 2016). Using an hiPSC-derived VSMC-based platform to study cardiovascular disease offers a future opportunity to investigate real-time transcriptome dynamicity during the VSMC development process, a privilege not possible without this cell model. Using hiPSC-derived VSMCs eliminates the impact of interpatient variability in disease status, age, administered drugs, and environmental factors that might affect VSMC de-differentiation phenotype while maintaining the clear association of patient genetic background with CAD.