Connecting the Dots from GWAS to Function in Atrial Fibrillation for ZFHX3

Genetics significantly influence susceptibility to atrial fibrillation (AF). Individuals with first-degree relatives under 60 years old with AF have ~5-times higher risk¹ with heritability estimated at 22%.² Genetic burden can be divided into two main categories: rare variants with strong effects, often due to loss of function mutations, and common variants (most often single nucleotide polymorphisms, SNPs) that can additively increase AF susceptibility. Common AF associated variants are identified by genome wide association studies (GWAS). The first published GWAS of AF in 2007 reported SNPs on 4q25 near PITX2¹. In 2010 variants near ZHFX3 were identified.² With increasing cohort size, over 100 AF risk loci have now been identified.³ However, despite advancements in genomics, our understanding of the pathological mechanisms of AF remains nebulous.

Polygenic risk scores for AF, using tens to millions of common variants, estimate that individuals in the top percentile have 4.6-fold increased risk of AF.⁴ Exome sequencing in ~43000 UK Biobank subjects revealed that common variants explain >20-fold more of the AF variance than the rare variants.⁵ Thus, GWAS identify the loci responsible for most of the population-attributable genetic risk for AF. GWAS loci, using Manhattan plots, display variants at each locus with ascending significance by p-value. Most of these variants are a result of linkage disequilibrium (LD), implying they’re often inherited together from ancestral chromosomes. However, discovering these loci is just the first step in the arduous tasks of identifying the responsible gene, variant(s), and mechanism for disease association.

In this issue of Circulation Research, Jameson, Ellinor and colleagues⁶ have performed just these tasks, focusing on the second most significant locus associated with AF, on chromosome 16q22 (Figure). The lead AF-associated SNP at this locus, rs2106261, resides in ZFHX3 gene’s intron, but that is insufficient to conclude ZFHX3 as the causal gene. For example, a common SNP for coronary artery disease, located in an intron of PHACTR1, regulates the expression of a gene, EDN1 encoding endothelin-1, which is approximately 600 kb away.⁷ The prime hypothesis is that most GWAS variants are regulatory variants that alter gene expression, or the use of alternative transcription starts, polyadenylation, and splice sites, as most lead variants are not in LD with protein altering variants.

Figure. Connecting the dots between the AF GWAS locus on chromosome 16 and AF susceptibility.

Top. Some of the steps Jameson et al. used to identify the causal variant, rs12931021, which regulates ZFHX3 expression in iCMs. Top left Manhattan plot modified from Roselli, et al.³ Bottom. Identification of ZFHX3 as an AF susceptibility gene via the use of a tissue specific mouse knockout of Zfxh3.

A key tool in identifying regulatory variants is through transcriptomics (RNA sequencing of the relevant tissue) to determine if the GWAS SNP contributes to the expression of a nearby gene or transcript isoform (also called a cis expression quantitative trait locus, or cis-eQTL).⁸ However, the authors here did not identify any human heart or left atrium (LA) cis-eQTLs for the lead SNP. To identify potential regulatory SNPs, 52 SNPs in LD with rs2106261 were studied, and six regions overlapped with open chromatin in human cardiomyocytes. Each region was tested by reporter gene transfections of the two alleles to determine if they drive expression of an exogenous gene in cardiomyocytes derived from induced human stem cells (iCMs). Only one SNP, rs12931021, had allele-specific enhancer activity. CRISPR-mediated deletion of the region surrounding rs12931921 led to modestly reduced expression of ZFHX3 in iCMs. CRISPR gene-editing was employed on a heterozygous stem cell line for this SNP, generating isogenic iCMs that were homozygous for either the risk or reference alleles. Remarkably, there was an allele dosage specific effect on the expression of ZFHX3. So why was the lead SNP not a cis-eQTL for ZFHX3 expression in human heart tissue? This SNP may only regulate gene expression during development or in a specific, minute tissue area, and eQTL studies typically involve adult bulk tissues. Unlike adult tissues, the iCMs used for allele-specific gene expression resemble fetal cardiomyocytes more closely. It remains possible that there are one or more additional regulatory variants in LD with the lead variant, which were not detectable in the regions of open chromatin in the cells used. In future studies, it may be informative to perform RNA sequencing in the rs12931021 allele-switch isogenic iCMs to see if there were any other genes whose expression are altered in response to the allele switch.

Nevertheless, now that Jameson, et al, identified the responsible gene and variant, they needed to identify the mechanism by which reduced expression of ZFHX3 leads to AF susceptibility. This required an animal model, and the easiest one to create and study is the mouse; thus, heart specific Zhfx3 heterozygous and homozygous knockout (KO) mice were created by breeding Zfhx3 floxed mice with mice carrying a cardiomyocyte expressed Cre transgene. Although these mice were viable, premature death was higher in Zhfx3 KO and heterozygous (het) mice with mortality rates of 74% and 56% respectively, compared to 10% in wild type (WT) mice.

Cardiac magnetic resonance imaging revealed that the 3-month old KO mice had increased LA size and reduced left ventricular ejection fraction, but no evidence of fibrosis. To explore potential cardiac contributions to early mortality, researchers assessed 9–11 month old mice; the KO mice had significantly larger atria, often with substantial atrial thrombi, dilated hearts with ventricular dysfunction, and increased fibrosis in both the atria and left ventricle. These mice also showed signs of end-organ complications typically associated with advanced clinical heart failure.

Despite lack of baseline electrocardiogram (ECG) changes at 3 months of age, a gene-dose response in atrial arrhythmia inducibility was seen using programmed stimulation, with the KO mice showing the greatest susceptibility to atrial arrhythmia induction. Ex vivo optical mapping performed on the mouse hearts revealed afterdepolarization-associated calcium transients in the LA of the KO hearts, marked by an increase in calcium release and decay time, a milieu that favors enhanced automaticity. Additional studies would be of interest to determine if the older mice have spontaneous AF and if there are any sex differences in the phenotypes.

RNA sequencing analysis revealed differential upregulation in genes tied to cardiac injury, inflammation, and dendritic cells. Intriguingly, these inflammatory markers were significantly more upregulated in the left than in the right atrium, including Lfit1, Mx1, and IL2ra. Genes involved with fatty acid metabolism, oxidative phosphorylation, muscle contraction, and protein translation were downregulated, while genes linked to cardiac fibrosis, endothelial dysfunction, and markers of cell stress and unfolded protein response were upregulated. Additionally, a considerable number of genes involved in the regulation of Wnt signaling showed significant differential expression. These transcriptomic changes align with our recent data on genes co-expressed with AF,⁹ where ZHFX3 and its co-expressed genes were associated with cardiac necrosis/cell death, cardiac fibrosis, cardiac hypertrophy, retinoic acid receptor activation, NRF2-mediated oxidative stress response, and cell cycle G1/S checkpoint regulation. Furthermore, ZHFX3 was grouped in a Weighted Gene Co-expression Network Analysis module that included 23 AF risk genes (MYH6, SGCG, TTN, FBXO32, CUL4A, DMTN, PPP2R3A, XPO7, THAP9-AS1, AKAP6, FRMD4B, FAM13B, MYOCD, NR3C1, IGF1R, CEP68, KCND3, ATXN1, KIAA1841, SORL1, CDK6, ZFHX3, KCNN3). Genes in the module were linked to pathways such as PPARα/RXRα activation, cardiac fibrosis, cardiac hypertrophy, and dysregulated aryl hydrocarbon receptor signaling.

Since ZFHX3 encodes a transcription factor, the authors wanted to determine where it binds in the genome of heart cells. They generated a consensus ZFHX3 binding motif and through the analysis of human cardiac single nucleus ATACseq data (measures open chromatin) from the four heart chambers, they determined that ZFHX3 biding sites are enriched in the open chromatin of atrial cardiomyocytes. Many of these binding sites were near genes that were differentially expressed in the KO mouse atria, including genes also identified via AF GWAS (FDN1, PLN, CAV1, TBX5, SYNE2, CYS1).

Functional genomics, cellular, and mouse studies are all required to go from the GWAS locus to the responsible gene and variant, and the disease mechanism, and Jameson, et al. have elegantly accomplished these tasks. Their compelling findings demonstrate that Zfhx3 is a causal gene for AF in mice, providing a rational explanation for the association between genetic variants near ZFHX3 and AF in humans. In the mouse studies, Zfhx3 not only modulates gene expression and the structural architecture of the myocardium, but also influences susceptibility to arrhythmias through abnormal calcium handling. Despite the number of years since the GWAS results for AF have been published, only a few other studies have connected genetic variants to direct functional mechanisms explaining AF. These include studies on the top locus on chromosome 4q25 near PITX2,^10–12 which is associated with development of pulmonary veins, and on a locus on chromosome 5, associated with functional changes in FAM13B.¹³ However, identification of putative function for these loci has yet to achieve clinical applicability, which will be critical if these genomic findings are to be translated to the bedside. This highlights the need for further in-depth research on additional AF associated genes and insights to translate this knowledge into AF prevention or treatment.