Long-range Pitx2c enhancer–promoter interactions prevent predisposition to atrial fibrillation

Significance

Noncoding single-nucleotide polymorphisms (SNPs) associated with atrial fibrillation (AF) remain poorly understood. Significant AF-associated risk variants are located upstream of PITX2. No functional evidence directly links this noncoding region to PITX2 expression and AF. We identified a murine Pitx2 enhancer, the deletion of which results in AF predisposition providing insight into how noncoding variants at 4q25 promote AF.

Abstract

Genome-wide association studies found that increased risk for atrial fibrillation (AF), the most common human heart arrhythmia, is associated with noncoding sequence variants located in proximity to PITX2. Cardiomyocyte-specific epigenomic and comparative genomics uncovered 2 AF-associated enhancers neighboring PITX2 with varying conservation in mice. Chromosome conformation capture experiments in mice revealed that the Pitx2c promoter directly contacted the AF-associated enhancer regions. CRISPR/Cas9-mediated deletion of a 20-kb topologically engaged enhancer led to reduced Pitx2c transcription and AF predisposition. Allele-specific chromatin immunoprecipitation sequencing on hybrid heterozygous enhancer knockout mice revealed that long-range interaction of an AF-associated region with the Pitx2c promoter was required for maintenance of the Pitx2c promoter chromatin state. Long-range looping was mediated by CCCTC-binding factor (CTCF), since genetic disruption of the intronic CTCF-binding site caused reduced Pitx2c expression, AF predisposition, and diminished active chromatin marks on Pitx2. AF risk variants located at 4q25 reside in genomic regions possessing long-range transcriptional regulatory functions directed at PITX2.

Atrial fibrillation (AF) increases the risk of death, stroke, heart failure, and dementia. AF incidence increases with age, and with life expectancy rising, AF prevalence will increase over the next decade. AF treatments, including stroke prevention and rate and rhythm control, have limited efficacy with side effects (1). Catheter ablation of ectopic foci, for drug refractory AF, carries the risk of procedural complications and recurrence. Given the challenges in treating AF, it is critical to develop a deeper understanding of molecular pathways leading to AF.

Population-based genome-wide association studies (GWAS) identified many AF-associated loci (2–8). A recent meta-analysis of GWAS for AF identified over 100 AF risk loci (7). The region most significantly associated with AF in European, Japanese, and African American populations is located on chromosome 4q25 (7). The gene in closest proximity to these 4q25 variants is the paired-like homeodomain transcription factor 2 (PITX2). PITX2 encodes a homeobox transcription factor that is required for mammalian development. There are 3 PITX2 isoforms (PITX2a, PITX2b, and PITX2c), and the major cardiac isoform is PITX2c. Previous work carried out in mice indicated that Pitx2 haploinsufficiency and conditional postnatal ablation promotes an atrial arrhythmogenic phenotype (9–11). Postnatally, Pitx2 is expressed in the left atrium (LA) and pulmonary vein, and its expression level decreases with age commensurate with onset of AF susceptibility in humans (11). AF often arises from ectopic electrical activity originating from the pulmonary veins and the LA, and Pitx2 suppresses a pacemaker gene expression signature (11, 12). However, a functional connection between the upstream AF-associated gene desert, PITX2 expression, and predisposition to AF has not been shown.

Insight into how intergenic noncoding GWAS single-nucleotide polymorphisms (SNPs) contribute to disease risk remain a challenge (13). It is thought that noncoding variants reside within or near regulatory elements controlling expression of distal target genes through topological interactions. However, the chromatin state of regulatory elements is variable across individuals, and the disruption of single transcription factor binding sites may not have any measurable phenotypic effect (14). Modeling risk variants in animal models is further complicated by the differences in regulatory sequence composition that have accumulated over millions of years of evolution. Even deletion of the most “ultra-conserved” enhancers may not produce detectable phenotypes (15, 16).

Chromosome conformation capture (3C) discovered a topological connection between potential enhancers within the 4q25 gene desert and the Pitx2c and Enpep promoters (17). Other studies on the most proximal association signal found an AF-associated SNP (rs2595104) that was correlated with altered PITX2c expression in human embryonic stem-cell–derived cardiomyocytes (CMs) (13). A Tbx5-binding site, located ∼150 kb upstream of Pitx2, had enhancer activity in vitro (18). However, the interaction between AF risk variants proximal to PITX2 and the PITX2c promoter remains incompletely understood (19).

We generated a mouse model for studying the molecular and phenotypic features associated with 4q25 regulatory region disruption. Through LA CM-specific epigenomics, we determined the chromatin accessibility of the PITX2 locus in humans and used these data to design CRISPR/Cas9 deletions in mice. We showed that deletion of a noncoding region upstream of Pitx2 predisposes mice to AF. The disruption of the upstream Pitx2 regulatory element led to a shift of chromatin marks on the Pitx2 gene body. Analysis of mice with deletion of an intronic CCCTC-binding factor (CTCF) site in the Pitx2 gene led to reduced Pitx2 expression and AF predisposition. Our findings reveal direct functional topological interactions between PITX2 and upstream genetic AF risk variants and provide insight into noncoding AF variants identified through GWAS.

Results

Epigenetic Characterization of Human LA Cardiomyocytes.

Distal regulatory elements are placed into close proximity to the promoters they regulate via chromatin looping. We hypothesized that AF-associated SNPs, located at 4q25, may disrupt the transcription factor (TF) binding, enhancer activity, and enhancer–promoter looping responsible for maintaining physiologic levels of PITX2 (Fig. 1A). First, to investigate the epigenetic landscape specific to LA CMs, the most relevant cell type for AF, we performed Assay for Transposase Accessible Chromatin using Sequencing (ATAC-seq) (20) on PCM1-enriched nuclei (21) derived from frozen human LA and left ventricle (LV) tissue (Fig. 1B). We obtained CM-enriched ATAC-seq results from a total of 11 LA samples and 3 LV samples. Importantly, the LA CMs possessed high accessibility at known LA-specific loci, like NPPA, compared to LV CMs (Fig. 1C). Principal component analysis revealed that the LA samples clustered separately from LV samples (Fig. 1D). Differential accessibility analysis uncovered 4,962 LA-enriched peaks (Fig. 1E and Dataset S1). Of the annotated genes from these peaks, 219 overlapped with atrial-enriched human proteome (22) (Dataset S2). To look at differential TF activity between LA and LV CMs, we deployed the HINT-ATAC (Hmm-based IdeNtification of Transcription factor footprints for ATAC-seq) algorithm (23), which corrects for Tn5 transposase cleavage bias when searching for TF footprints (Fig. 1F). Among LA significant footprints, we identified the SP1-related motifs SHOX, ZNF740, and RREB1. Notably, RREB1 was also enriched in atrial proteomics analysis (22). LV CMs were enriched for JUN and FOS motifs, including the JDP2 (JUN Dimerization Protein 2) motif (Fig. 1F). De novo motif enrichment on LA-specific peaks combined with expression data on the same human tissue (7) uncovered similar TF motif enrichment, including SP1/4 (Fig. 1G).

Fig. 1.

LA human cardiomyocyte-specific ATAC-seq. (A) Graphical depiction of the putative topological association of the AF-associated region with PITX2. (B) Experimental strategy. (C) Genome browser track for NPPA and NPPB. (D) Principal component analysis of LA and LV ATAC-seq (adjusted P value < 0.01). (E) Heatmap of differentially accessible region-specific CM ATAC-seq peaks. (F) HINT-ATAC analysis of TF activity. Red TFs are those with absolute activity greater than 0.3. Dot size represents normalized expression. (G) De novo motif enrichment on LA enriched peaks (n = 4,962). TFs ranked by normalized LA expression. The size of each point indicates the motif enrichment P value.

A similar analysis using Cre-inducible Sun1-sfGFP INTACT mice (24) and CM-specific Cre recombinase (Myh6-MCM) to isolate LA, right atrium, and ventricular CM nuclei uncovered, by unsupervised clustering, a strong correlation between chromatin accessibility and active histone modifications on promoters and enhancers (SI Appendix, Fig. S1A). A total 3,205 LA-specific peaks were identified, including several annotated to LA markers Nppa and Pitx2 (SI Appendix, Fig. S1 B and C).

AF-Associated Regions Located at 4q25 Possess Enhancer Characteristics.

The human 4q25 gene desert, containing multiple regions that are independently associated with AF, is conserved in mice and encompasses ∼1 Mb on mouse chromosome 3. Within this 1-Mb region, genome organization is also conserved since PITX2 and its neighboring genes are located in the same order and orientation in mice and humans. AF-associated regions are ∼150 kb upstream of PITX2 (Fig. 2A). To investigate whether this region contained enhancers, we aligned this region against a multitissue human DNase Hypersensitivity (DHS) dataset (25). We found DHS peaks enriched in fetal human heart samples, but not in noncardiac tissue (Fig. 2B). These fetal heart DHS peaks aligned with our adult LA CM-specific ATAC peaks.

Fig. 2.

Epigenomics guided interrogation of AF-associated 4q25 Locus. (A) Regional plot for PITX2 AF association. (B) Human LA-CM specific ATAC-seq and DHS (GSE18927), H3K4me1 ChIP-seq and chromatin HMM. Specific peaks indicated by orange bars on ChromHMM track. (C) Pitx2 human and mouse conservation. (D) Mouse genomic counterparts of AF-associated regions.

To investigate whether these peaks are potential enhancers, we aligned human LA CM ATAC-seq genomic tracks with human heart H3K4me1 chromatin immunoprecipitation sequencing (ChIP-seq), a marker of enhancer activity (26). ATAC peaks from LA aligned with the H3K4me1 peaks (Fig. 2B, boxed areas). ChromHMM, a machine-learning algorithm for characterizing chromatin states, predicted 2 cardiac enhancer blocks upstream of PITX2C (25). We narrowed down the 120-kb region into 2 blocks of 20 and 60 kb (Fig. 2B).

To identify the functional counterparts of these 2 putative enhancers in the mouse genome, we compared DNA sequence conservation (Fig. 2C). We identified 21 highly conserved DNA segments spread across these 2 blocks. Consistent with a previous report that cardiac enhancers are less conserved (27), we found only 1 of 3 human ATAC-seq peaks within the 60-kb block, and 2 of 4 in the 20-kb block contained highly conserved DNA sequences (Dataset S3). Nonetheless, we found the mouse counterpart of the 20-kb block contained 2 LA-enriched CM ATAC-seq peaks, while the 60-kb block was relatively inaccessible (Fig. 2D). These mouse LA-ATAC peaks were enriched for the active histone mark H3K27ac (27) (Fig. 2D). Motif enrichment using enhancers identified from both humans and mice revealed that consensus binding sites for Pitx2, Hnf4a, Nrf2, Ets, and Nkx factors were enriched (SI Appendix, Fig. S2), consistent with previous studies (28–30).

A Conserved 20-kb Enhancer Topologically Engages the Pitx2c Promoter.

The Pitx2 gene body is located on the boundary of 2 distinct topologically associated domains (TADs) (SI Appendix, Fig. S3). CTCF-mediated loop formation follows a simple rule, in which only a pair of distant CTCF-binding sites with convergent motif orientations form stable loops (31, 32). To determine TAD landscapes for the PITX2 locus, we determined CTCF motif orientation of all CTCF ChIP-seq peaks (Fig. 3A). CTCF motif polarity was highly conserved between mice and humans (Fig. 3B).

Fig. 3.

Pitx2c topologically interacts with AF-associated enhancer in mouse heart. (A) CTCF-binding polarity. CTCF motif orientation (arrowheads). TAD organization highlighted (gray blocks). (B) CTCF-binding polarity in mouse heart. (C) 4C-seq across 2 Pitx2 viewpoints (highlighted). AF-associated region and Pitx2 gene body highlighted (cyan).

To obtain a higher-resolution view of enhancer–promoter interactome directed at Pitx2c, we performed chromosome conformation capture with high-throughput sequencing (4C-Seq) on postnatal day 8 mouse hearts using the Pitx2c promoter as the viewpoint. We found direct topological contacts between the 20-kb AF-associated enhancer and the Pitx2c promoter (Fig. 3C). Consistent with previous work, we also saw long-range Pitx2c promoter contacts with Playrr, a long non-coding RNA (33). The 4C-Seq revealed extensive contacts between Pitx2c and Enpep, as previously detected with 3C (Fig. 3C and SI Appendix, Fig. S4) (17). We also detected contacts between the Pitx2c promoter and a gene-dense region in the non–AF-associated TAD, containing genes like Egf and Lrit3. These data indicate that the topological architecture of the Pitx2 locus in cardiac tissue is complex, and the presence of long-range contacts between the 20-kb AF-associated enhancer region and the Pitx2c promoter was evident.

Deletion of a Noncoding 20-kb AF-Associated Region Predisposes Mice to AF.

To investigate how the AF-associated regulatory elements contribute to arrhythmogenesis, we used CRISPR/Cas9 editing to delete mouse counterparts of the aforementioned 20- and 60-kb enhancer blocks (Fig. 4 A and B). We obtained multiple homozygous mutant mouse lines for both the 20-kb deletion and the 60-kb deletion, which we refer to as Pitx2^ΔE20k and Pitx2^ΔE60k (Fig. 4 C and D). Both the Pitx2^ΔE20k and Pitx2^ΔE60kb mouse lines were viable and fertile.

Fig. 4.

Targeted genetic deletion of AF-associated regions. (A) Workflow for generation of enhancer deletions with CRISPR/Cas9 genomic editing. (B) Locations of deleted regions. (C) Genotyping strategy. (D) Representative genotyping from ΔE20K founders.

We subjected mice to programmed intracardiac stimulation and found that Pitx2^ΔE20k mice were prone to AF (Fig. 5 A–C). In contrast, Pitx2^ΔE60k mice did not have a phenotype in either incidence of induced AF or length of AF burden (Fig. 5 B and C). Male Pitx2^ΔE20k mice were more prone to AF than females (Fig. 5D). To address if the deletion of the 20-kb enhancer alters Pitx2 expression, we quantified LA Pitx2c expression via RNA-seq (Fig. 5E). In male homozygous Pitx2^ΔE20k mice, LA Pitx2c expression was reduced compared to controls, which we confirmed by quantitative RT-PCR (qRT-PCR) (SI Appendix, Fig. S4A). Pitx2c reduction was not observed in female mice (Fig. 5E). Enpep and Elovl6 were unchanged in male Pitx2^ΔE20k mice when compared to control males, but were elevated in female Pitx2^ΔE20k mice when compared to control females (SI Appendix, Fig. S4B). Differential expression analysis revealed that inflammation, cytoskeleton organization, and right atrial-identity genes, including Bmp10, were up-regulated in mutant LA tissue in both genders (SI Appendix, Fig. S4 C–F). Thus, the 20-kb AF-associated region encodes a Pitx2c enhancer that is required for the maintenance of Pitx2c expression in LA of adult male mice.

Fig. 5.

Pitx2 enhancer deletion #1 (ΔE20K) AF susceptibility. (A) Representative recordings of surface ECG (Lead II), atrial electrogram, and ventricular electrogram where arrow indicates end of intracardiac pacing. (B) Incidence of intracardiac pacing induced AF. (C) AF burden, calculated as average length of AF following pacing. (D) AF burden by sex. (E) Pitx2c expression level in LA.

A Distal 20-kb Enhancer Is Required to Maintain Pitx2 Gene Body Chromatin Status.

Disruption of enhancer–promoter interactions can alter chromatin landscapes of topologically engaged regions (34). To determine whether the 20-kb enhancer acts as a functional cis-regulatory element for Pitx2, we crossed the Pitx2^ΔE20k, FVB background, to the Mus musculus castaneus (Cast) subspecies to evaluate haplotype-specific chromatin landscapes (35). We performed H3K4me1 and CTCF ChIP-Seq on hearts from hybrid mice, extracted enriched peaks, and calculated haplotype-specific reads (Fig. 6 A and B). In the 20-kb enhancer region, H3K4me1 ChIP reads at the Pitx2 gene body primarily originated from the control Cast allele (Fig. 6C). Interestingly, although we previously found direct contacts between the 20-kb enhancer region and several other distal gene promoters with 4C-seq, we did not observe such allelic H3K4me1 occupancy bias on other genes, such as Enpep or Sec24. Conversely, CTCF occupancy at both the Pitx2 gene body and distal enhancer regions showed no significant differences between the 20-kb enhancer-deficient FVB allele and control Cast allele (Fig. 6C). We performed CUT&RUN on control (FVB) and homozygous Pitx2^ΔE20k to more quantitatively assess chromatin compositional shifts (36). The levels of active histone marks, including H3K27ac and H2K4me1, were significantly altered across the Pitx2 gene body in Pitx2^ΔE20k mutants (Fig. 6 D and E and SI Appendix, Fig. S5 A and B). Conversely, CTCF and the inactive mark H3K27me3 were unchanged (SI Appendix, Fig. S5 C–F). These findings suggested that the 20-kb enhancer maintains the active chromatin state of Pitx2 via long-range cis-interactions.

Fig. 6.

AF-associated enhancer deletion alters local Pitx2 chromatin landscape. (A) Allele-specific H3K4me1 and CTCF ChIP-Seq in FVB ΔE20K/CAST hybrids. Total mixed reads shown in green track. The expected odds for individual allelic SNP was 50% (orange bar for FVB background, purple bar for CAST). (B) A CTCF-binding peak, also enriched for H3K4me1 occupancy, located in the Pitx2 intronic region. (C) Allele-specific H3K4me1 and CTCF-binding occupancy showing FVB (ΔE20K) strain-specific SNPs, rs221741451 (a), rs241380526 (b), and rs37111181 (c), rs39016764 (d). (D) CUT&RUN peak intensity of H3K27ac. (E) CUT&RUN of H3K27ac at Pitx2. **P < 0.01.

Pitx2 Enhancer–Promoter Interactions Are Mediated by CTCF.

TAD boundaries are commonly CTCF-rich (37, 38). Given that Pitx2 resides directly on a TAD boundary engaged by 2 distinct TADs, we wanted to dissect CTCF-dependent aspects of Pitx2 gene regulation. A CTCF-binding site separates the 2 Pitx2 promoters (Pitx2ab and Pitx2c). We hypothesized that this local CTCF site may play a bifunctional role in regulating both topological architecture and Pitx2c expression. Fortuitously, we had previously described a Pitx2^PZNeo allele (ref. 39, designated δab^hypoc), in which the Pitx2ab isoform-specific exon 3 and part of the third intron, including the CTCF-binding site, were replaced with an internal ribosomal entry site-LacZ sequence (Fig. 7A).

Fig. 7.

CTCF-binding site deletion caused Pitx2 expression reduction and AF predisposition. (A) Illustration of Pitx2^PZNeo allele. (B) Quantitative PCR results from LA tissue of Pitx2^PZNeo/PZNeo mice. (C) ECG tracing. (D) Incidence of intracardiac-pacing–induced AF.

qRT-PCR revealed that Pitx2c expression was reduced by about 50% in Pitx2^PZNeo/PZNeo LA (Fig. 7B). Programmed stimulation on Pitx2^PZNeo/PZNeo mice revealed that these mice were susceptible to atrial arrhythmias (Fig. 7 C and D). To determine the molecular consequences associated with a loss of local CTCF binding, we crossed Pitx2^PZNeo/PZNeo with Cast mice and performed allele-specific CTCF ChIP-seq, as described above (SI Appendix, Fig. S6). CTCF occupancy on 4 of 30 CTCF sites increased in Pitx2^PZNeo compared to the Cast allele (SI Appendix, Fig. S6A). All 4 sites were located distal to Pitx2. Taken together, the genetic deletion of a CTCF-binding site located within the third intron of Pitx2 caused a decrease in Pitx2c expression and an arrhythmogenic phenotype in mice.

Discussion

The connection between distal noncoding genetic variants and PITX2 expression and AF predisposition has not been previously functionally verified. Using cell-type–specific epigenomics and comparative genomics, we identified 2 conserved putative enhancer elements that are associated with AF. Deletion of one of these regulatory elements in mice recapitulated the AF predisposition phenotype. This Pitx2 regulatory element makes contact with Pitx2c, and the deletion of this element leads to dysregulation of the chromatin environment on the Pitx2 gene body.

Pitx2 is expressed in multiple cell types at several developmental time points. During heart development Pitx2c is expressed in the LA, atrioventricular canal, cardiac outflow tract, right ventricle, left superior caval vein, and pulmonary vein myocardium (39, 40). Although it is surprising that homozygous ΔE20K and ΔE60K mice were phenotypically normal in the absence of physiologic stress, recent studies found that multiple enhancers with comparable activity proximal to the same gene are pervasive and function to confer phenotypic robustness (41). Previous work, looking at the 85-kb AF-associated region at 4q25, investigated human regulatory sequences in mouse cell lines and transgenic mice (SI Appendix, Fig. S7) (17). The authors concluded that these regulatory elements do not behave as cell-type–specific enhancers (17). It is likely that the activity of noncoding regulatory elements in mice and humans may have diverged. Indeed, as much as 30% of mammalian enhancers show evidence for functional gains and losses of TF-binding sites (42). Future work to dissect the regulatory elements driving tissue-specific Pitx2 expression will make use of combinatorial pair-wise enhancer deletions.

We performed CM-enriched ATAC-seq on human LA tissue from “healthy” patients to prioritize the AF-associated risk variants found at 4q25. We did not observe a large overlap of AF risk variants with LA CM ATAC peaks most likely because the AF risk variants are markers of genomic locations, defined by linkage disequilibrium, rather than causative variants. If these nonaccessible sites do contain causative SNPs, then several possibilities exist to explain their function, including developmental activity, cell type and/or spatial specificity, and/or function during a response to some external stimulus or injury. Consistent with the presence of a stimulus response element, mouse models of neonatal cardiac injury have shown that Pitx2 expression is induced by injury and is required for mitochondrial functional maintenance, the prevention of deleterious fat accumulation, and reactive oxygen species handling (30, 43). Alternatively, these unoccupied AF risk variants may initiate ectopic TF binding, which promotes deleterious gene expression patterns. Looking forward, the completion of large-scale cell-type–specific and tissue region resolved datasets in combination with high-throughput functional genomics experiments aimed at molecularly defining the roles of the 4q25 AF risk variants will help to resolve this important matter (44).

Materials and Methods

Detailed materials and methods provided in SI Appendix.

Mouse Alleles and CRISPR/Cas9-Mediated Deletion Lines.

The Pitx2^PZNeo allele and INTACT mouse were described (24, 39).

Sequencing and Analysis.

ChIP-seq was performed as described (30). ATAC-seq libraries were generated as described (20). The 4C protocol was followed as described (45). The 4C-seq data analysis pipeline was previously described (45).

Cardiomyocyte Nuclear Isolations.

Nuclear isolation was done as described (24).

Programmed Intracardiac Pacing.

Programmed intracardiac pacing was performed (11). Recordings of surface electrocardiogram (ECG) were done, and intracardiac electrograms were performed in anesthetized mice via right jugular vein catheterization.

Published Datasets Used in This Study.

Human DHS-seq data was obtained from Gene Expression Omnibus (GEO) with entry GSE18927 (25). Mouse heart H3K27Ac ChIP-seq was extracted and reprocessed from GSE52386 (27).