Cardiac cell type–specific gene regulatory programs and disease risk association

Single nucleus analysis of chromatin accessibility and transcriptome in adult human hearts

To assess the accessible chromatin landscape of distinct cardiovascular cell types, we performed snATAC-seq (27), also known as sci-ATAC-seq (21), on all cardiac chambers from four adult human hearts without known cardiovascular disease (table S1). We obtained accessible chromatin profiles for 79,515 nuclei, with a median of 2682 fragments mapped per nucleus (Fig. 1, A and B; fig. S1; and table S2). We also performed snRNA-seq for a subset of the above heart samples to complement the accessible chromatin data and obtained 35,936 nuclear transcriptomes, with a median of 2184 unique molecular identifiers and 1286 genes detected per nucleus (Fig. 1, A and C; fig. S2, A to F; and table S3). Using SnapATAC (28) and Seurat (29), we identified nine clusters from snATAC-seq (Fig. 1B) and 12 major clusters from snRNA-seq (Fig. 1C and fig. S2, G and H), which were annotated on the basis of chromatin accessibility at promoter regions or expression of known lineage-specific marker genes, respectively (Fig. 1, D and E, and table S4) (10, 11). For example, chromatin accessibility and gene expression of atrial and ventricular cardiomyocyte markers such as NPPA and MYH7 (30) were used to classify these two cardiomyocyte subtypes (Fig. 1, D and E). Although gene expression patterns of lineage markers strongly correlated with accessibility at promoter regions across annotated cell types (Fig. 1F) and single-cell integration analysis (29) revealed 93% concordance in annotation between snATAC-seq and snRNA-seq datasets (fig. S3 and table S3), some cellular subtypes identified from snRNA-seq including endocardial cells and myofibroblasts were not detected by snATAC-seq (Fig. 1F). Notably, cluster correlation and integration analysis showed that these cell types are present within the snATAC-seq data as part of the endothelial and smooth muscle clusters, respectively (Fig. 1F, fig. S3, and table S3). The discrepancy in clustering may be attributable to the conservative snATAC-seq clustering parameters or the sparse nature of snATAC-seq data (31). In addition, atrial and ventricular cardiomyocyte nuclei from the left and right regions of the heart could be further clustered by transcriptome but not chromatin accessibility (fig. S2, I and J). We noted that cell type composition varied substantially between biospecimens and donors, highlighting the importance of single-cell approaches to limit biases due to cell proportion differences in bulk assays (fig. S4 and tables S2 and S3). In summary, we identified and annotated cardiac cell types using both chromatin accessibility and nuclear transcriptome profiles.

Identification of cCREs in distinct cell types of the human heart

To discover the cCREs in each cell type of the human heart, we aggregated snATAC-seq data from nuclei comprising each cell cluster individually and determined accessible chromatin regions with MACS2 (32). We then merged the peaks from all nine cell clusters into a union of 287,415 cCREs, which covered 4.7% of the human genome (Fig. 2A and table S5). Sixty-seven percent of the cCREs identified in the current study overlapped previously annotated cCREs from a broad spectrum of human tissues and cell lines (fig. S5A) (5), and the union of heart cCREs captured 98.6 and 95.4% of candidate human heart enhancers reported in two previous bulk studies (fig. S5, B and C) (13, 15). Furthermore, 75% of cCREs in the union were at least 2 kilo–base pairs (kbp) away from annotated promoter regions, and 19,447 displayed high levels of cell type specificity [false discovery rate (FDR) < 0.01; Fig. 2B and table S6]. Gene ontology analysis (33) revealed that these cell type–specific cCREs were proximal to genes involved in relevant biological processes, including collagen fibril organization for cardiac fibroblast–specific cCREs (K1) and myofibril organization for ventricular cardiomyocyte–specific cCREs (K2; Fig. 2C and table S7). Using chromVAR (table S8) (34) and HOMER (table S9) (35), we detected cell type–dependent enrichment for 231 transcription factor–binding signatures, such as MEF2A/B, NKX2.5, and THR-β sequence motifs in cardiomyocyte-specific cCREs and TCF21 motifs in cardiac fibroblast–specific cCREs (Fig. 2D). To discover the transcription factors that may bind to these sites, we combined corresponding snRNA-seq data with sequence motif enrichments to correlate the expression of these transcription factors with motif enrichment patterns across cell types (Fig. 2E). As an example, we found strong enrichment of the binding motif for the macrophage transcription factor SPI1/PU.1 (36) in macrophage-specific cCREs, and SPI1 was exclusively expressed in macrophages (Fig. 2E and table S4). In addition, we observed that transcription factor family members were expressed in cell type–specific combinations. For instance, while GATA Family of transcription factors (GATA-binding factor) displayed similar motif enrichment patterns across sets of cell type–specific cCREs, we found that endothelial cells and cardiac fibroblasts expressed GATA2 and GATA6, respectively, whereas cardiomyocytes expressed both GATA4 and GATA6 and endocardial cells expressed GATA2, GATA4, and GATA6 (Fig. 2E and table S4). In summary, these results establish a resource of cCREs for interrogation of cardiac cell type–specific gene regulatory programs.

Cardiac cell type–specific gene regulatory programs implicated in chamber-specific structure and function

Each cardiac chamber performs a unique role that is crucial to system-level heart function (37). To investigate the gene regulatory programs underlying chamber-specific gene expression and cellular functions in distinct cardiac cell types, we tested cCREs for differential accessibility across five of the most abundant cell types of the heart: cardiomyocytes, cardiac fibroblasts, endothelial cells, smooth muscle cells, and macrophages. We found 16,451 differentially accessible (DA) cCREs between pooled atria and ventricles, most of which were detected in cardiomyocytes (Fig. 3, A to C, and table S10). Specifically, 11,159 cCREs displayed differential accessibility between the right atrium and the right ventricle, and 12,962 cCREs exhibited differential accessibility between the left atrium and the left ventricle (fig. S6, A to C, and table S10). Comparing the left and right sides of the heart, we identified 101 DA cCREs between the right and left ventricles (fig. S6D) and 2687 DA cCREs between left and right atria, which, in contrast to comparisons between atria and ventricles, were found primarily in cardiac fibroblasts (fig. S6E and table S10).

Using coaccessibility analysis (38) to link distal DA cCREs (~88% of all DA cCREs) to their putative target genes (table S11; median distance, 88.7 kbp), we observed that distal DA cCREs in cardiomyocytes between atria and ventricles were associated with chamber-specific gene expression of their putative target genes (Fig. 3D; fig. S6, B to E; and table S12), and genes near these DA cCREs were enriched for chamber-specific biological processes (Fig. 3E; fig. S6, B to E; and table S13). Specifically, distal DA cCREs with higher accessibility in atrial cardiomyocytes were associated with genes such as PITX2, a transcriptional regulator of cardiac atrial development (39), as well as the ion channel subunit SCN5A, which regulates cardiomyocyte action potential (Fig. 3E and table S13) (40). Furthermore, we found distal DA cCREs with higher accessibility in atrial cardiomyocytes at the HAMP gene locus, which encodes a key regulator of ion homeostasis and was recently described as a potential previously unknown cardiac gene in the right atrium by single-nucleus transcriptomic analysis (10, 11). Conversely, genes near distal DA cCREs with higher accessibility in ventricular cardiomyocytes were enriched for biological processes such as trabecula formation and ventricular cardiac muscle cell differentiation. For example, several distal DA cCREs with increased accessibility in ventricular cardiomyocytes compared to atrial cardiomyocytes were linked to the promoter region of MYL2, which encodes the ventricular isoform of myosin light chain 2 (Fig. 3F and table S4) (41), a regulator of ventricular cardiomyocyte sarcomere function.

In addition, analysis of distal DA cCREs in cardiac fibroblasts revealed that putative target genes were involved in distinct biological processes between right and left atria. In particular, we found that DA cCREs with higher accessibility in right atrial cardiac fibroblasts were proximal to genes involved in heart development, heart growth, and tube development, whereas DA cCREs with higher accessibility in left atrial cardiac fibroblasts were adjacent to genes involved in biological processes such as wound healing and vasculature development (fig. S6E and table S13). We further found a cardiac fibroblast–specific DA cCRE with higher accessibility in left atria at the fibrinogen FN1 gene locus, potentially indicating a more activated fibroblast state (10, 42). Supporting these findings, we identified several other DA cCREs with higher accessibility in left atrial cardiac fibroblasts adjacent to genes involved in generation of extracellular matrix (ECM) such as MMP2 and FBLN2 (table S13). These observations are consistent with previous findings that a higher fraction of ECM is produced in fibroblasts of the left atrium (10).

Using motif enrichment analysis, we inferred candidate transcriptional regulators involved in chamber-specific cellular specialization, including TBX5 (T-Box Transcription Factor 5), GATA4, and TGIF1 (TGFB Induced Factor Homeobox 1) for atrial cardiomyocytes and nuclear factor of activated T cells (NFAT), ERRG (Estrogen-Related Receptor Gamma), HAND1 (Heart- and Neural Crest Derivatives-Expressed Protein 1), and HAND2 for ventricular cardiomyocytes (Fig. 3G and table S14). While the TBX5 DNA binding motif was strongly enriched in both right and left atrial cardiomyocyte DA cCREs, the NFAT5 motif ranked highest in left ventricular cardiomyocyte DA cCREs and the TBX20 motif was strongly enriched in right ventricular cardiomyocyte DA cCREs (fig. S6, B and C, and table S14). Furthermore, cardiac fibroblast DA cCREs with higher accessibility in the right atrium were enriched for the binding motif of forkhead transcription factors (fig. S6E), whereas cardiac fibroblast DA cCREs with higher accessibility in the left atrium were enriched for the homeobox transcription factor CUX1 motif (fig. S6E and table S14). Together, we identified cCREs and candidate transcription factors associated with specific cardiac chambers, particularly within cardiomyocytes and cardiac fibroblasts.

Cell type specificity of candidate enhancers associated with heart failure

Recent large-scale studies profiling the H3K27ac histone modification in human hearts have uncovered candidate enhancers associated with heart failure (15, 18). However, because these studies either examined heterogeneous bulk heart tissue (15, 18) or focused solely on enriched cardiomyocytes (16), it remains unclear what role, if any, additional cardiac cell types and cCREs may contribute to heart failure pathogenesis. Using our cell atlas of cardiac cCREs, we revealed the cell type specificity of candidate enhancers showing differential H3K27ac signal strength between human hearts from healthy donors and donors with dilated cardiomyopathy (heart failure) (Fig. 4 and fig. S7, A to E) (15). We observed that a large fraction of candidate enhancers that displayed increased activity (45%) during heart failure were accessible primarily in cardiac fibroblasts (Fig. 4A, K2-4_up, and table S15), whereas most of those exhibiting decreased activity (67%) were accessible primarily in cardiomyocytes (Fig. 4B, K1-3_down, and table S15). Candidate enhancers with increased activity in cardiac fibroblasts were proximal to genes involved in ECM organization and connective tissue development (Fig. 4A, K2-4_up, and table S16), whereas those exhibiting decreased activity in cardiomyocytes were proximal to genes involved in the regulation of heart contraction and cation transport (Fig. 4B, K1-3_down, and table S16). For example, several of these cardiac fibroblast candidate enhancers were present at loci encoding the ECM proteins lumican (LUM) and decorin (DCN) and coaccessible with the promoters of these genes (Fig. 4C). Consistent with these findings, both genes were primarily expressed in cardiac fibroblasts (fig. S7F and table S4), and LUM has been reported to exhibit increased expression in failing hearts compared to control hearts (15). On the other hand, several cardiomyocyte candidate enhancers displaying decreased activity in heart failure were coaccessible with the promoter region of IRX4 (Fig. 4D), which encodes a ventricle-specific transcription factor (43) and is primarily expressed in cardiomyocytes of the left ventricle (fig. S7G and table S4).

To identify potential transcription factors regulating these pathologic responses during heart failure, we performed motif enrichment analysis in cell type–specific subsets of enhancers showing differential H3K27ac signal strength between healthy and failing hearts (table S17). For candidate enhancers exhibiting increased activity in heart failure, we identified enrichment of not only bHLH motifs such as AP4 in cardiac fibroblast candidate enhancers, which matched previous bulk analysis (Fig. 4E, K2-4_up) (15), but also TEAD3 and MYF6 motifs in cardiomyocyte candidate enhancers (Fig. 4E, K1_up). Conversely, for candidate enhancers displaying decreased activity in heart failure, we observed enrichment of nuclear receptor motifs such as glucocorticoid response element in cardiomyocyte candidate enhancers, which is consistent with previous findings (Fig. 4F, K1-3_down) (15), as well as other motifs that were not detected in bulk analyses, such as the bZIP transcription factor CEBPA for cardiac fibroblast candidate enhancers (Fig. 4F, K4_down). Thus, these results show that this cardiac cell atlas of cCREs may be used to assign disease-associated candidate enhancers from bulk assays to their affected cell types and infer transcriptional regulators involved in lineage-specific disease pathogenesis.

Interpreting noncoding risk variants of cardiac diseases and traits

Noncoding genetic variants contributing to risk of complex diseases are enriched within cCREs in a tissue and cell type–dependent manner (24, 44–47). To examine the enrichment of cardiovascular disease variants within cCREs active in cardiac cell types, we performed cell type–stratified LD (linkage disequilibrium) score regression analysis (48) using genome-wide association study (GWAS) summary statistics for cardiovascular diseases (Fig. 5A) (49–53) and control traits (fig. S8A and table S18) by measuring the enrichment of disease-associated variants within all cCREs identified for each cell type. This analysis revealed significant enrichment of AF-associated variants in both atrial (Z = 5.61, FDR = 1.9 × 10⁻⁶) and ventricular cardiomyocyte cCREs (Z = 6.80, FDR = 2.8 × 10⁻⁹), varicose vein-associated variants in endothelial cell cCREs (Z = 4.36, FDR = 3.9 × 10⁻⁴), and coronary artery disease–associated variants in cardiac fibroblast cCREs (Z = 3.29, FDR = 1.7 × 10⁻²; Fig. 5A). Notably, except for AF, these associations were not significant in a pseudo-bulk heart dataset created by combining chromatin accessibility profiles from all cardiac cell types (Fig. 5A). Furthermore, cardiovascular disease variants were not significantly enriched in accessible chromatin from noncardiac tissues (3, 5, 17) or human lung cell types (54), with the exception of a significant enrichment of varicose vein–associated variants in endothelial cells (fig. S8, B and C).

Next, to identify likely causal AF risk variants in cardiomyocyte cCREs, we first determined the probability that variants were causal for AF [posterior probability of association (PPA)] at 111 known loci using Bayesian fine mapping (55). We then intersected fine-mapped AF variants with cCREs detected in atrial and/or ventricular cardiomyocytes and identified 38 variants with PPA > 10% in cardiomyocyte cCREs, including previously reported variants at the HCN4 (13) and SCN10A/SCN5A (56) loci (table S19). We further prioritized AF variants for molecular characterization on the basis of their overlap with cCREs that were primarily accessible in cardiomyocytes, evolutionarily conserved, and coaccessible with promoters of genes expressed in cardiomyocytes. To experimentally validate the molecular functions of cCREs containing AF variants, we used a human pluripotent stem cell (hPSC)–derived cardiomyocyte differentiation model system (Fig. 5B) (57). From the variant prioritization analysis, we found a cCRE in the second intron of the potassium channel gene KCNH2 (HERG), which was coaccessible with the KCNH2 promoter (Fig. 5C) and harbored two variants, rs7789146 and rs7789585, with a combined PPA of 28% (Fig. 5C and fig. S9A). KCNH2 was primarily expressed in ventricular and atrial cardiomyocytes in the adult human heart (fig. S9B). The cCRE appeared to be activated during hPSC-cardiomyocyte differentiation, as evidenced by an increase in H3K27ac signal that correlated with KCNH2 expression (Fig. 5C). Supporting its in vivo role in regulating gene expression in mammalian hearts, a genomic region (hs2192) (58) containing this cCRE was previously shown to drive LacZ reporter expression in mouse embryonic hearts (Fig. 5D) (58).

A cardiomyocyte enhancer of KCNH2 is affected by noncoding risk variants associated with AF

To investigate whether these AF variants may affect enhancer activity and thereby regulate KCNH2 expression and cardiomyocyte electrophysiologic function, we initially carried out reporter assays using a hPSC cardiomyocyte model system. Results from these studies confirmed that in D15 hPSC cardiomyocytes, the KCNH2 enhancer carrying the rs7789146-G/rs7789585-G AF risk allele displayed significantly weaker enhancer activity than when containing the nonrisk variants (Fig. 5E and fig. S9C), thus supporting the functional significance of these AF variants. We next used CRISPR-Cas9 genome editing strategies to remove the enhancer and performed quantitative polymerase chain reaction (qPCR) and electrophysiologic assays to examine its role in KCNH2 expression and function. Supporting the aforementioned findings, CRISPR-Cas9 genome deletion of this cCRE in hPSC cardiomyocytes resulted in decreased KCNH2 expression in an enhancer dosage–dependent manner (Fig. 5F and fig. S9D). Similar to human cardiomyocytes with loss of KCNH2 function due to mutations in the KCNH2 coding sequence (59) or gene knockdown (60), cellular electrophysiologic studies demonstrated that these cCRE-deleted hPSC cardiomyocytes displayed a significantly prolonged action potential duration (Fig. 5, G and H), thus suggesting that cardiac repolarization abnormalities in atrial cardiomyocytes may lead to AF in an analogous manner to ventricular arrhythmias due to long QT syndrome (59). Together, these results highlight the utility of this single-cell atlas for assigning noncoding cardiovascular disease risk variants to distinct cell types and affected cCREs and functionally interrogating how these variants may contribute to cardiovascular disease risk.

Experimental design

We performed snATAC-seq to define a comprehensive catalog of cCREs for the cell types in four regions of nonfailing human hearts and generated in parallel snRNA-seq datasets for a subset to delineate gene expression patterns. We used the cCRE catalog to computationally assign dynamic enhancers in failing hearts to cell types and to assign cardiovascular disease risk variants to cCREs in individual cardiac cell types. Last, we applied reporter assays, genome editing, and electrophysiological measurements in in vitro differentiated human cardiomyocytes to validate the molecular mechanisms of cardiovascular disease risk variants.

Human tissues

Adult human heart tissues were procured at the time of organ donation using an Institutional Review Board protocol (no. 101021) approved by the University of California, San Diego. Donated hearts were perfused with cold cardioplegia before cardiectomy and then explanted immediately into an ice-cold physiologic solution as we previously described (67). Full-thickness samples from each chamber were obtained, and epicardial fat was rapidly removed before immediately flash-freezing samples in liquid nitrogen. Samples were received from the United Network for Organ Sharing. Limited clinical data were obtained for each heart per approved Institutional Review Board protocol (table S1). All samples were stored at −80°C until processing.

Single-nucleus ATAC-seq

Combinatorial barcoding snATAC-seq was performed as described previously (21, 27, 28) with slight modifications. Nuclei were isolated in gentleMACS M tubes (Miltenyi) on a gentleMACS Octo dissociator (Miltenyi) using the “Protein_01_01” protocol in magnetic-activated cell sorting (MACS) buffer [5 mM CaCl₂, 2 mM EDTA, 1× protease inhibitor (Roche, 05-892-970-001), 300 mM MgAc, 10 mM tris-HCl (pH 8), and 0.6 mM dithiothreitol (DTT)]. Nuclei were pelleted with a swinging-bucket centrifuge (500g, 5 min, 4°C; 5920R, Eppendorf) and resuspended in 1 ml of nuclear permeabilization buffer [1× phosphate-buffered saline (PBS), 5% bovine serum albumin (BSA), 0.2% IGEPAL CA-630 (Sigma-Aldrich), 1 mM DTT, and 1× protease inhibitor]. Nuclei were rotated at 4°C for 5 min before being pelleted again with a swinging-bucket centrifuge (500g, 5 min, 4°C; 5920R, Eppendorf). After centrifugation, permeabilized nuclei were resuspended in 500 μl of high-salt tagmentation buffer [36.3 mM tris acetate (pH 7.8), 72.6 mM potassium acetate, 11 mM Mg acetate, and 17.6% N,N′-dimethylformamide] and counted using a hemocytometer. Concentration was adjusted to 2000 nuclei/9 μl, and 2000 nuclei were dispensed into each well of a 96-well plate per sample (96 tagmentation wells per sample; samples were processed in batches of two to four samples). For tagmentation, 1 μl of barcoded Tn5 transposomes (table S20) was added using a BenchSmart 96 (Mettler Toledo), mixed five times, and incubated for 60 min at 37°C with shaking (500 rpm). To inhibit the Tn5 reaction, 10 μl of 40 mM EDTA (final 20 mM) was added to each well with a BenchSmart 96 (Mettler Toledo), and the plate was incubated at 37°C for 15 min with shaking (500 rpm). Next, 20 μl of 2× sort buffer (2% BSA and 2 mM EDTA in PBS) were added using a BenchSmart 96 (Mettler Toledo). All wells were combined into a separate fluorescence-activated cell sorter tube for each sample and stained with DRAQ7 at 1:150 dilution (Cell Signaling Technology). Using an SH800 (Sony), 20 nuclei per sample were sorted per well into eight 96-well plates (total of 768 wells) containing 10.5 μl of Elution Buffer (EB; Qiagen) [25 pmol of primer i7, 25 pmol of primer i5, and 200 ng of BSA (Sigma-Aldrich)]. During the sort, nuclei with two to eight copies of DNA (2-8n) were included because cardiomyocyte nuclei in human hearts are often polyploid (16). Preparation of sort plates and all downstream pipetting steps were performed on a Biomek i7 automated workstation (Beckman Coulter). After addition of 1 μl of 0.2% SDS, samples were incubated at 55°C for 7 min with shaking (500 rpm). One microliter of 12.5% Triton X was added to each well to quench the SDS. Next, 12.5 μl of NEBNext High-Fidelity 2× PCR Master Mix (New England Biolabs) were added, and samples were PCR-amplified [72°C 5 min, 98°C 30 s, (98°C 10 s, 63°C 30 s, 72°C 60 s) × 12 cycles, held at 12°C]. After PCR, all wells were combined. Libraries were purified according to the MinElute PCR Purification Kit manual (Qiagen) using a vacuum manifold (QIAvac 24 Plus, Qiagen), and size selection was performed with SPRISelect reagent (Beckman Coulter; 0.55× and 1.5×). Libraries were purified one more time with SPRISelect reagent (Beckman Coulter; 1.5×). Libraries were quantified using a Qubit fluorimeter (Life Technologies), and a nucleosomal pattern of fragment size distribution was verified using a TapeStation (High Sensitivity D1000, Agilent). Libraries were sequenced on a NextSeq 500 sequencer (Illumina) using custom sequencing primers with the following read lengths: 50 + 10 + 12 + 50 (read 1 + index 1 + index 2 + read 2). Primer and index sequences are listed in table S20.

Single-nucleus RNA-seq

Nuclei were isolated from heart tissue using a gentleMACS dissociator (Miltenyi). Frozen heart tissue (~40 mg) was suspended in 2 ml of MACS dissociation buffer [5 mM CaCl₂ (G-Biosciences, R040), 2 mM EDTA (Invitrogen, 15575-038), 1× protease inhibitor (Roche, 05-892-970-001), 3 mM MgAc (Grow Cells, MRGF-B40), 10 mM tris-HCl (pH 8) (Invitrogen, 15568-075), 0.6 mM DTT (Sigma-Aldrich, D9779), and ribonuclease (RNase) inhibitor (0.2 U/μl; Promega, N251B) in water (Corning, 46-000-CV)] and placed on wet ice. Next, samples were homogenized using a gentleMACS dissociator (Miltenyi) with gentleMACS M tubes (Miltenyi, 130-096-335) and the Protein_01_01 protocol. Suspension was filtered through a 30 μM CellTrics filter (Sysmex, 04-0042-2316). M tube and filter were washed with 3 ml of MACS dissociation buffer and combined with the suspension. Suspension was centrifuged in a swinging-bucket centrifuge (Eppendorf, 5920R) at 500g for 5 min (4°C, ramp speed of 3/3). Supernatant was carefully removed and pellet was resuspended in 500 μl of nuclei permeabilization buffer[(0.1% Triton X-100 (Sigma-Aldrich, T8787), 1× protease inhibitor (Roche, 05-892-970-001), 1 mM DTT (Sigma-Aldrich, D9779), RNase inhibitor (0.2 U/μl; Promega, N251B), and 2% BSA (Sigma-Aldrich, SRE0036) in PBS]. Sample was incubated on a rotator for 5 min at 4°C and then centrifuged at 500g for 5 min (Eppendorf, 5920R; 4°C, ramp speed of 3/3). Supernatant was removed and pellet was resuspended in 600 to 1000 μl of sort buffer [1 mM EDTA and RNase inhibitor (0.2 U/μl) in 2% BSA (Sigma-Aldrich, SRE0036) in PBS] and stained with DRAQ7 (1:100; Cell Signaling Technology, 7406). Seventy-five thousand nuclei were sorted using an SH800 sorter (Sony) into 50 μl of collection buffer [RNase inhibitor (1 U/μl) and 5% BSA (Sigma-Aldrich, SRE0036) in PBS]. Sorted nuclei were then centrifuged at 1000g for 15 min (Eppendorf, 5920R; 4°C, ramp speed of 3/3), and supernatant was removed. Nuclei were resuspended in 18 to 25 μl of reaction buffer [RNase inhibitor (0.2 U/μl) and 1% BSA (Sigma-Aldrich, SRE0036) in PBS] and counted using a hemocytometer. Twelve thousand nuclei were loaded onto a Chromium controller (10x Genomics). Libraries were generated using the Chromium Single Cell 3′ Library Construction Kit v3 (10x Genomics, 1000078) according to the manufacturer specifications. Complementary DNA was amplified for 12 PCR cycles. SPRISelect reagent (Beckman Coulter) was used for size selection and cleanup steps. Final library concentration was assessed by the Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific), and fragment size was checked using TapeStation High Sensitivity D1000 (Agilent) to ensure that fragment sizes were distributed normally around 500 bp. Libraries were sequenced using a NextSeq 500 or HiSeq 4000 (Illumina) using these read lengths: read 1, 28 cycles; read 2, 91 cycles; and index 1, 8 cycles.

hPSC culture

An engineered H9-hTnnTZ-pGZ-D2 hPSC transgenic reporter line was purchased from WiCell and maintained on Geltrex (Gibco) precoated tissue culture plates in E8 medium (68) containing Dulbecco’s modified Eagle’s medium/F12, l-ascorbic acid-2-phosphate magnesium (64 mg/liter), sodium selenium (14 μg/liter), fibroblast growth factor 2 (100 μg/liter), insulin (19.4 mg/liter), NaHCO₃ (543 mg/liter), transferrin (10.7 mg/liter), and transforming growth factor–β1(2 μg/liter). Cells were passaged every 3 to 5 days upon reaching ~80% confluency. For single-cell passaging experiments, cells were incubated with prewarmed TrypLE Select Enzyme, no phenol red (1 ml per well of a six-well plate) for 2 to 3 min in a 37°C, 5% CO₂ incubator. Following incubation, cells were triturated to create a single-cell suspension and cultured in E8 medium supplied with Rock inhibitor (69) for 18 to 24 hours after split, followed by daily feeding with E8 medium.

In vitro cardiomyocyte differentiation

The H9-hTnnTZ-pGZ-D2 cell line was differentiated into beating cardiomyocytes using a previously reported Wnt-based monolayer differentiation protocol (70). Briefly, the H9-hTnnTZ-pGZ-D2 cell line was cultured in E8 medium for 3 to 10 passages. Before differentiation, hPSCs were seeded at a density of 350,000 to 400,000 cells per well of a 12-well plate and cultured for 2 days. For direct differentiation, cells were treated with 10 μM CHIR99021 (Fisher Scientific, no. 442350) in RPMI/B-27 without insulin. Fresh RPMI/B-27 without insulin media was replaced at post 24 hours, and cells were then cultured for 2 days. At day 3, cells were treated with 5 μM IWP2 (TOCRIS, no. 353310) in conditional medium and RPMI/B-27 without insulin 1:1 mix medium for another 2 days. At day 5, cells were exposed to fresh RPMI/B-27 without insulin media again for 2 days. Then, fresh RPMI/B-27 with insulin media was used and replenished every 2 days. Contracting cardiomyocytes were usually observed at days 7 and 8. D25 in vitro cardiomyocytes were purified using a PSC-derived cardiomyocyte isolation kit, human (Miltenyi Biotec, 130-110-188) and used for real-time qPCR (RT-qPCR).

Luciferase reporter assay

A genomic region harboring the KCNH2 intronic enhancer (containing the risk allele rs7789146-G/rs7789585-G) was amplified by nested PCR (KCNH2-E-cF, CTGGCTGAAGACACCTTACTTT; KCNH2-E-cR, ACGGAGCAGTCAAGGAAAC and KCNH2-In-cF, CGGGGTACCCCTCCGTAAATGAGGTGCTATC; KCNH2-In-cR, CCCTCGAGACGGAGCAGTCAAGGAAAC) using the genomic DNA of H9-hTnnTZ-pGZ-D2 transgenic cells as a template and cloned into pGL4.23 [luc2/minP] (Promega, catalog no. E8411) luciferase reporter vector. Synthetic DNA containing the KCNH2 intronic 5′-half enhancer (rs7789045-rs7789690) with the nonrisk/nonrisk allele (rs7789146-A/rs7789585-A), nonrisk/risk allele (rs7789146-A/rs7789585-G), and risk/nonrisk allele (rs7789146-G/rs7789585-A) were purchased from Integrated DNA Technologies. KCNH2 intronic 3′-half enhancer (rs7789654-rs7790480) was amplified by PCR (KCNH2-R-In-cF, GCTGTGCAGTGTCAGGTTAT; KCNH2-In-cR, CCCTCGAGACGGAGCAGTCAAGGAAAC). Then, the whole KCNH2 intronic enhancer with the nonrisk/nonrisk allele (rs7789146-A/rs7789585-A), nonrisk/risk allele (rs7789146-A/rs7789585-G), and risk/nonrisk allele (rs7789146-G/rs7789585-A) were generated by second PCR (KCNH2-In-cF, CGGGGTACCCCTCCGTAAATGAGGTGCTATC; KCNH2-In-cR, CCCTCGAGACGGAGCAGTCAAGGAAAC). One day before transfection, 3 × 10⁵ of D15 in vitro differentiated cardiomyocytes were plated in a Geltrex-coated 24-well plate. Cardiomyocytes were transfected with 500 ng of pGL4.23 plasmid (either empty, KCNH2 enhancer with G/G allele, A/A allele, or mix) and 10 ng of TK:Renilla-luc as an internal control using Lipofectamine Stem Transfection Reagent (Invitrogen, no. STEM00003). Media were replaced with fresh media at 24 hours after transfection. At 72 hours after transfection, media were removed and the cells were washed with PBS. Luminescence was measured using a Dual-Luciferase Reporter Assay System (Promega, no. E2920) according to the manufacturer’s protocol.

CRISPR-mediated genome editing experiments

To interrogate the functional significance of the AF-associated risk variant–containing cCRE at the KCNH2 locus, the cCRE sequence was genetically deleted in H9-hTnnTZ-pGZ-D2 transgenic hPSCs using an efficient CRISPR-Cas9–mediated knockout system (57, 71). Two adjacent guide RNAs (gRNAs) (KCNH2-enh gRNA-1, CTCATTTACGGAGGAGCGCA; KCNH2-enh gRNA-2, TACAGTGGCCTTCTAGACGA) targeting the cCRE were designed using a web-based software tool CRISPOR (72) based on targeting region of interest and minimizing potential off-target effects. The identified gRNAs were then synthesized in vitro using the GeneArt Precision gRNA Synthesis kit (Invitrogen) according to the manufacturer’s protocol. One day before transfection, 1.5 × 10⁵ H9-hTnnTZ-pGZ-D2 hPSCs were seeded in 12-well plates. A pair of ribonucleoprotein complexes containing 1.2 μg of Cas9 protein (New England Biolabs) and 400 ng of in vitro transcribed gRNA was then transfected (73, 74) using Lipofectamine stem transfection reagent (Invitrogen). Seventy-two hours after the transfection, cells were diluted and clonally expanded another 7 days. Colonies were picked and lysates were prepared after the first passage for genotyping (75) (KCNH2-enh extended forward primer, ACACCTTACTTTGGGTGAGAAG; KCNH2-enh extended reverse primer, AGACAGAGCACAGACCTAGAA; KCNH2-enh internal forward primer, GCTGTGCAGTGTCAGGTTAT; KCNH2-enh internal reverse primer, TCTCCCTCCTTCTCTCTCATTC). After confirmation of genome-edited clones by Sanger sequencing, two transfected wild-type (WT) clones, two heterozygote clones, and two homozygote clones were selected for further functional analysis.

Real-time quantitative polymerase chain reaction

Total RNA was isolated from the cells using TRIzol reagent (Invitrogen). One microgram of total RNA was reverse-transcribed using the iScript Reverse Transcription Supermix kit (Bio-Rad) for RT-qPCR. RT-qPCR was performed using PowerUP SYBR Green Master Mix (Applied Biosystems) in the CFX Connect Real-Time System (Bio-Rad). The results were normalized to the TBP gene. The primers used for RT-qPCR are listed in table S21.

Electrophysiology of cardiomyocytes

Both WT and KCNH2 enhancer knockout D15 in vitro cardiomyocytes were purified using the PSC-derived cardiomyocyte isolation kit, human (Miltenyi Biotec, 130-110-188) and cultured for another 10 to 20 days in a low density before electrophysiological measurements. The single-pipette, whole-cell patch current-clamp technique was used for recordings. Action potentials were recorded with a patch-clamp amplifier (Axopatch 200B, Axon), and experiments were performed at a temperature of 35° ± 0.5°C. Current-clamp command pulses were generated by a digital-to-analog converter (DigiData 1440, Axon), which was controlled by the pCLAMP software (10.3; Axon). Pipettes (resistance, 3 to 5 megohm) were pulled using a micropipette puller (Model P-87, Sutter Instrument Co.). Several minutes after seal formation, the membrane was ruptured by gentle suction to establish the whole-cell configuration for voltage clamping. Subsequently, the amplifier was switched to the current-clamp mode. Cells were paced with 1 Hz and injected current stimuli from 3 to 15 nA for 5-ms duration. Cells were superfused with extracellular solution containing the following: 140 mM NaCl, 5.4 mM KCl, 1.8 mM CaCl₂, 1.0 mM MgCl₂, 5.5 mM glucose, and 5.0 mM Hepes (pH 7.4 adjusted with NaOH). Pipette solution contained the following: 120 mM K-gluconate, 10 mM KCl, 5 mM NaCl, 10 mM Hepes, 5 mM phosphocreatine, 5 mM ATP-Mg₂, and amphotericin (0.44 μM; pH 7.2 adjusted with KOH).

Data analysis