ATAC-seq Reveals an Isl1 Enhancer that Regulates Sinoatrial Node Development and Function

Abstract

Rationale:

Cardiac pacemaker cells (PCs) in the sinoatrial node (SAN) have a distinct gene expression program that allows them to fire automatically and initiate the heartbeat. Although critical SAN transcription factors, including Isl1, Tbx3, and Shox2, have been identified, the cis-regulatory architecture that governs PC-specific gene expression is not understood, and discrete enhancers required for gene regulation in the SAN have not been identified.

Objective:

To define the epigenetic profile of PCs using comparative ATAC-seq and to identify novel enhancers involved in SAN gene regulation, development and function.

Methods and Results:

We used ATAC-seq on sorted neonatal mouse SAN to compare regions of accessible chromatin in PCs and right atrial cardiomyocytes. PC-enriched ATAC-seq peaks, representing candidate SAN regulatory elements, were located near established SAN genes and were enriched for distinct sets of TF binding sites. Among several novel SAN enhancers that were experimentally validated using transgenic mice, we identified a 2.9-kb regulatory element at the Isl1 locus that was active specifically in the cardiac inflow at E8.5 and throughout later SAN development and maturation. Deletion of this enhancer from the genome of mice resulted in SAN hypoplasia and sinus arrhythmias. The mouse SAN enhancer also directed reporter activity to the inflow tract in developing zebrafish hearts, demonstrating deep conservation of its upstream regulatory network. Finally, single nucleotide polymorphisms in the human genome that occur near the region syntenic to the mouse enhancer exhibit significant associations with resting heart rate in human populations.

Conclusions:

(1) PCs have distinct regions of accessible chromatin that correlate with their gene expression profile and contain novel SAN enhancers, (2) Cis-regulation of Isl1 specifically in the SAN depends upon a conserved SAN enhancer that regulates PC development and SAN function, and (3) a corresponding human ISL1 enhancer may regulate human SAN function.

Graphical Abstract

INTRODUCTION

Cardiac pacemaker cells (PCs) within the sinoatrial node (SAN) are the leading pacemakers of the heart during normal rhythm. PCs are elongated, densely innervated cardiomyocytes that exhibit robust electrical automaticity but weak contractility¹. Accordingly, their gene expression program diverges from that of working cardiomyocytes, with broad repression of genes required for rapid conduction and force generation, and activation of a distinctive gene set associated with innervation and electrical automaticity^{2, 3}.

In recent years, the detailed transcriptome of PCs has been defined, revealing a conserved gene regulatory network that controls mammalian PC development and function, with critical roles for transcription factors (TFs) Shox2⁴, Tbx3⁵, and Isl1^{2, 6}. However, the precise mechanisms that connect the expression of these regulators with their gene targets, and the mechanisms that control PC-specific expression of these TFs, remain incompletely understood, creating major barriers to understanding functional specialization in the cardiac conduction system in development and disease.

Epigenetic profiling has provided major insights into differentiation, functional specialization, and pathophysiology in a variety of tissues including the heart^7–9. Partly because of the difficulty of isolating large numbers of pacemaker cells, genome-wide profiling of histone marks or TF binding events with chromatin immunoprecipitation and sequencing (ChIP-seq) has been difficult to apply to PCs. Consequently, regulatory elements that direct gene expression specifically to PCs and their progenitors have not been identified, limiting our understanding of mammalian PC development and differentiation.

The Assay for Transposase-Accessible Chromatin (ATAC-seq) is a robust method to define regions of accessible chromatin in rare cell populations¹⁰. ATAC-seq is more sensitive for detection of active cis-regulatory elements in cardiomyocytes than H3K27Ac ChIP-seq, and requires much less starting material to yield comparable signal-to-noise ratio, making it a preferred method for epigenetic profiling of PCs¹¹. Here, we use ATAC-seq in purified PCs and right atrial cardiomyocytes (RACMs) to identify accessible chromatin regions that are specific to PCs. Analysis of these putative PC-specific regulatory elements identified enriched TF binding motifs, as well as several novel SAN enhancers that we experimentally validated. We then provide a detailed characterization of a 2.9-kb enhancer that activates Isl1 specifically in the venous inflow myocardium from embryonic day (E)8.5 onwards, with progressive restriction to the SAN as development proceeds. Deletion of this enhancer resulted in SAN hypoplasia and sinus arrhythmias. The mouse Isl1 SAN enhancer was also active in the cardiac inflow tract in transgenic zebrafish embryos, indicating deep conservation of the upstream regulatory architecture. Finally, SNPs located at the human region syntenic to the mouse Isl1 SAN enhancer were associated with variation in resting heart rate in a large human population, providing evidence for its functional role in human SAN.

METHODS

An extended Methods section is available in the online supplement. RNA-seq and ATAC-seq datasets have been deposited in the Gene Expression Omnibus (GSE148515). Additional data supporting this study are available from the corresponding author upon reasonable request.

RESULTS

Isolation of cardiac pacemaker cells using Fluorescence Activated Cell Sorting (FACS).

To isolate pure populations of pacemaker cells (PCs) and right atrial cardiomyocytes (RACMs), Hcn4-GFP or Hcn4-GFP/Myh6-mCherry neonatal hearts were imaged under live fluorescence and the GFP+ SAN head was identified at the junction of the superior vena cava and right atrium (Figure 1A,B). For each biological replicate, SAN-RA junctions from 5 neonatal hearts were pooled for digestion to a single cell suspension for flow sorting, yielding 2–5% GFP+ cells and 25–35% mCherry+ cells. Over 99% of GFP+ cells isolated from Hcn4-GFP/Myh6-mCherry mice were also mCherry+, as expected (Online Figure I, A). Differential expression analysis on bulk RNA-seq data from GFP+ and mCherry+ cells (n = 3 biological replicates) demonstrated both enrichment of established SAN genes and strong depletion of known atrial genes in the GFP+ cells, with comparable expression of core cardiomyocyte TFs Gata4, Tbx5, and Mef2C between the two populations (Figure 1C, Online Table I, Online Figure I, B-D). Based on the magnitude of the depletion of atrial makers Nppa, Nppb and Bmp10 in the GFP+ samples (Figure 1C), and on the low or undetectable expression signatures of non-myocyte cells (Online Figure I, E), we conclude that GFP+ cells are highly enriched for PCs.

Figure 1. Isolation of Neonatal Pacemaker Cells for ATAC-seq.

(A) Isolation of neonatal pacemaker cells (PCs) and right atrial cardiomyocytes (RACMs) from the sinoatrial node (SAN) using fluorescence activated cell sorting (FACS) on samples derived from 3 biological replicates, where each biological replicate contained pooled SAN tissue from 5 neonatal mice. (B) Ventral view of an Hcn4-GFP/Myh6-mCherry neonatal heart with SAN positive for both fluorescent markers (yellow) (C) Enrichment of known PC genes (green bars) and depletion of known RACM genes (red bars) from bulk RNAseq on sorted PCs and RACMs (each biological replicate is shown (PC1, PC2, and PC3), data are displayed as log(2) fold change of SAN/RACM transcripts per million (TPM)). Blue bars show genes that are not differentially expressed. (D,E) ATAC-seq signal for individual PC (n = 4) and RACM (n = 3) biological replicates at Hcn4 locus (D) and Nppa locus (E) Each biological replicate consisted of pooled SAN tissue from 1–2 litters of neonatal hearts. Abbreviations: san, sinoatrial node; svc, superior vena cava; la, left atrium; ra, right atrium; lv, left ventricle; rv, right ventricle.

Pacemaker cells demonstrate regions of differential chromatin accessibility.

Next, genomic DNA was isolated from flow sorted PCs and RACMs for ATAC-seq (~20,000 cells from 1 or 2 litters of neonates per biological replicate, Online Table II). Visualization of ATAC-seq signal at Hcn4 and Nppa loci revealed discrete peaks in non-coding regions near Hcn4 that were absent in RACMs and peaks near Nppa and Nppb that were absent in PCs (Figure 1D,E). Nonhierarchical clustering analysis on all peaks for each replicate showed that PC and RACM samples each clustered together, and that PC ATAC-seq peaks in our dataset were concordant with a previously published neonatal SAN ATAC-seq dataset¹² (Online Figure II A,B). To maximize capture of PC-specific regulatory elements, we then defined a set of consensus ATAC-seq peaks for each sample and examined the top 500 differentially accessible regions between PCs and RACMs (corresponding to a false discovery rate of 0.33, and uncorrected p < 1e-4, Figure 2A, Online Table III).

Figure 2. Proximity of Differential ATAC-seq Peaks to Differentially Expressed Genes and Enrichment of Cardiac TF Motifs and Binding Sites at ATAC-seq Peaks.

(A) Clustered heatmap with z-scores for the top 500 differentially accessible (DA) ATAC-seq peaks across all 7 samples. (B) Gene ontology (GO) terms for genes assigned to DA ATAC-seq peaks. (C) Motif enrichment in DA ATAC-seq peaks. Motifs and false discovery rates are for the underlined factors. (D) Overlap of published Mef2c, Tbx5, Gata4, Nkx2.5, and Shox2 ChIP-seq peaks with DA ATAC-seq peaks.

Differentially accessible regions in pacemaker cells are associated with differentially expressed genes.

In contrast to the global datasets of PC peaks and RACM peaks, which showed a distribution centered around basal promoters, the majority of the 500 differentially accessible regions that we identified were located between 50 and 500 kb from the nearest transcriptional start site (TSS), consistent with potential roles as tissue-specific regulatory elements (enhancers) (Online Figure II, C-E). To test this hypothesis, we used Genomic Regions Enrichment of Annotations Tool (GREAT) to assign differentially accessible regions to candidate genes¹³. Gene ontology (GO) analysis of this gene set (Figure 2B, Online Table IV) demonstrated a strong association with SAN development and function, and 108 out of 500 differentially accessible ATAC-seq peaks were associated with a gene that is differentially expressed in PCs vs RACMs (p < 1e-34), including Hcn4, Isl1, Shox2, and Tbx3 (Figure 2B).

These data provide strong in-silico evidence that the set of differentially accessible ATAC-seq peaks we identified here includes regulatory elements that control gene expression in PCs.

Motif enrichment analysis of differentially accessible regions.

We then used HOMER¹⁴ to identify enriched TF binding motifs in differentially accessible ATAC-seq peaks. In addition to over-representation of motifs for established cardiac TFs including Gata, Mef2, and T-box factors, several homeodomain TF binding sites were enriched including sites for Isl1, a critical PC TF, and Meis1, a TALE-class homeodomain protein involved in cardiac development¹⁵ (Figure 2C, Online Table V). ATAC-seq peaks in PCs were also enriched at experimentally validated ChIP-seq peaks for Mef2C, Tbx5, Gata4, and Tead from embryonic myocardium¹¹, and at Nkx2.5 and Shox2 binding sites from embryonic cardiac inflow myocardium¹⁶ (Online Figure III, A-B). Extending these analyses to the PC-enriched subset of differentially accessible ATAC-seq peaks showed robust enrichment of ATAC-seq signal at ChIP-seq peaks for Mef2C, Tbx5, and Gata4, but not as strongly for Nkx2.5 and Shox2 (Figure 2D). These data support existing models in which cardiac TFs including Tbx5, Mef2C, and Gata4 factors, but not Nkx2.5, positively regulate expression of genes that are restricted to PCs, and Shox2 represses atrial cardiomyocyte genes through competition with Nkx2.5¹⁶.

Selected ATAC-seq peaks are sufficient to direct reporter gene expression to SAN.

To test whether differentially accessible ATAC-seq peaks are bona fide SAN enhancers, we used enSERT, a site-specific genomic integration method that allows high-throughput experimental validation of putative enhancers in mouse embryos¹⁷. Because ATAC-seq was performed at neonatal stages but enSERT-generated transgenic embryos were harvested at E11.5, we intersected differentially accessible ATAC-seq peaks with embryonic mouse heart H3K27Ac ChIP-seq datasets from ENCODE ¹⁸ (Online Figure III, C) and selected 16 regions with co-localization of PC-specific ATAC-seq peaks and H3K27Ac ChIP-seq peaks (Online Table VI). 5 out of 16 of these ATAC-seq-predicted, candidate enhancers exhibited cardiac activity in consistent patterns, of which 3 had activity in the SAN primordium (at loci for Hcn4, Rgs6, and Ptgfr, all PC-enriched genes), and 2 were specific for the venous inflow and PC primordium (Rgs6 and Ptgfr, Figure 3A-C and Online Figure IV). These data demonstrate that a subset of PC-specific ATAC-seq peaks can function as enhancers that direct gene expression directly to the SAN, and that the entire dataset of differentially accessible ATAC-seq peaks is likely to contain many additional SAN enhancers.

Figure 3. Selected PC-Enriched ATAC-seq Peaks Are Sufficient to Direct Reporter Activity to the Cardiac Venous Inflow and Sinoatrial Node.

(A) ATAC-seq signal in pacemaker cells (PC, green track) and right atrial cardiomyocytes (RACM, red track) at the Rgs6 locus (top) and the Ptgfr locus (bottom) with differentially accessible (DA) ATAC-seq peaks highlighted in yellow. (B) 3 founders from transient transgenic enhancer reporter mice generated with enSERT for the DA ATAC-seq peak at Rgs6 locus (VISTA ID mm2011) and (C) Ptgfr locus (VISTA ID mm2010), harvested at E11.5 and stained with X-gal. Lower panels show a magnified view of the venous inflow (arrow) and right atrium (ra). (D) DA ATAC-seq peak at the Isl1 locus is highlighted in yellow. (E-J) Whole mount images of ISE-LacZ embryos harvested and stained with salmon-gal at E8.5 (E) or bluo-gal at the developmental time points shown (F-J). (K) Double immunohistochemistry for Hcn4 and β-gal in the sinoatrial node (SAN) of E14.5 ISE-LacZ embryo. (L) Whole-mount image of an adult ISE-LacZ SAN stained with bluo-gal. (Abbreviations: ra, right atrium, rv, right ventricle, lv, left ventricle, la, left atrium, sv, sinus venosus, oft, outflow tract, san, sinoatrial node)

Spatiotemporal activation pattern of an Isl1 locus SAN enhancer.

To identify additional enhancers at loci of critical PC TFs, we searched for differentially accessible ATAC-seq peaks within previously annotated TADs that contain Shox2, Tbx3, and Isl1 (Online Figure V). While Tbx3 and Shox2 TADs each had numerous individual ATAC-seq peaks in PCs that were not present in RACMs, close examination of ATAC-seq signal over the entire Isl1 TAD revealed a single previously uncharacterized peak located ~120 kb downstream of Isl1 that was strongly differentially accessible between PCs and RACMs (Online Figure V, Figure 3D). To test whether the region defined by this peak might be an SAN-specific Isl1 enhancer, we generated Isl1 Locus SAN Enhancer (ISE)-LacZ and ISE-mCherry, stable transgenic mouse lines in which a 2.9-kb region encompassing the Isl1 PC-specific ATAC-seq peak was cloned upstream of hsp68LacZ or hsp68mCherry. Of 7 transgenic mouse lines examined (2 ISE-LacZ and 5 ISE-mCherry), all exhibited restricted reporter activity in the SAN (Online Figure VI, A-C).

Onset of ISE-LacZ enhancer activity occurred at E8.5 in the cardiac inflow region (Figure 3E). At E9.5 enhancer activity was stronger and was restricted to the right inflow myocardium (Figure 3F and Online Figure VI, D). By E10.5 and E11.5, enhancer activity tracked the SAN primordium, with extension into the right venous valve (Figure 3G and Online Figure VI, E). By E12.5 and until neonatal stages, enhancer activity was largely restricted to the definitive SAN at the SVC-right atrial junction with extensions into the SAN tail and into the crista terminalis (Figure 3H-J, Online Figure VI, F). Notably, from the time of the formation of the SAN primordium and onward, ISE was active in the SAN region but not in other Isl1+ second heart field-derived tissues. Immunohistochemistry for Hcn4 and β-gal at E14.5 in ISE-LacZ hearts demonstrated restriction of β-gal expression to the SAN (Figure 3K). Adult hearts had persistent SAN activity (Figure 3L), with some additional low-level activity in other areas of the myocardium (Online Figure VI. G-I). Overall, our developmental time course demonstrates that ISE exhibits a remarkable degree of specificity for the SAN throughout development and maturation.

Isl1 Locus SAN enhancer-mCherry+ cells are functional Hcn4+ pacemaker cells.

To test whether ISE activity co-localized with Hcn4 expression in PCs, we crossed ISE-mCherry with Hcn4-GFP. Whole mount fluorescent imaging demonstrated co-localization of mCherry signal with GFP at E8.5 in the cardiac inflow tract (Figure 4A,A’), the sinus venosus at E10.5 (Figure 4B,B’), and the SAN at E14.5 and P2 (Figure 4C,C’,D, and Online Figure VI, J-L). Whereas mCherry expression was restricted to the junction of the sinus venosus and right atrium during development, GFP was expressed more broadly, as has been previously described for Hcn4^{19, 20}, suggesting that ISE is activated by a more specific regulatory module than Hcn4. We did note a dim mCherry+/GFP- nodal extension that was visible in E14.5 embryos and in post-natal hearts (Figure 4C,C’).

Figure 4. SAN Enhancer Activity Co-localizes with Hcn4 Expression and Functional Cardiac Pacemaker Cells.

(A,A’) Whole-mount fluorescence imaging of ISE-mCherry/Hcn4-GFP embryos at E8.5 (arrows denote inflow tract), (B,B’) E10.5, and (C,C’) E14.5 imaged for mCherry alone (top row) and mCherry/GFP merge (bottom row). ‘*’ denotes an mCherry+/GFP- SAN extension into the right atrium (D) Postnatal ISE-mCherry activity at P2 and (E) in adult isolated pacemaker cells (PCs) with typical PC morphology. (F) Spontaneous action potentials (APs) recorded from an adult mCherry+ isolated pacemaker cell. (G) Firing rate, maximum diastolic potential (MDP), action potential (AP) amplitude, early diastolic depolarization rate (EDDR), upstroke velocity (dV/dt max) and AP duration at 50 percent and 90 percent repolarization in 12 isolated adult mCherry+ pacemaker cells. Error bars indicate standard deviation. (H) Whole cell voltage clamp on an mCherry+ adult cell with funny current (I(f)) elicited by a series of hyperpolarizing voltage pulses from a holding potential of −40 mV to −130 mV in 10 mV increments. (I) Mean I(f) current density at the indicated voltages in 11 mCherry+ PCs. Error bars indicate SEM. Abbreviations: sv, sinus venosus; ra, right atrium; la, left atrium; san, sinoatrial node.

To test whether adult mCherry+ cells are bona fide cardiac pacemaker cells, we isolated single SAN cells from the intercaval regions of 5 adult ISE-mCherry mice. mCherry+ cells exhibited typical pacemaker cell morphologies (Figure 4E), although we also observed enhancer activity in some adult peri-nodal RACMs. However, action potentials recorded from spontaneously beating mCherry+ SAN myocytes were typical for PCs, with a depolarized maximum diastolic potential and spontaneous diastolic depolarization (Figure 4F,G). In addition, mCherry+ SAN cells exhibited robust funny current (the current carried by Hcn channels) as would be expected for PCs (Figure 4H, I).

Isl1 Locus SAN enhancer regulates Isl1 expression and SAN development.

To test directly whether ISE regulates Isl1 expression in PCs and SAN development, we deleted a 2.7-kb region encompassing ISE in mice using CRISPR/Cas9 (Figure 5A). Homozygous ISE null mice (hereafter, Isl1^ΔISE/ΔISE) assessed at F2 and later generations did not demonstrate statistically significant embryonic or perinatal lethality (Figure 5B). To test for reduction of Isl1 expression in Isl1^ΔISE/ΔISE SAN, we used immunohistochemistry to detect Isl1 protein specifically in mutant PCs as compared to wild-type (WT) PCs by co-staining embryonic SAN for Isl1 and Hcn4 at E14.5. Isl1^ΔISE/ΔISE Hcn4+ PCs exhibited grossly reduced Isl1 expression as compared to WT PCs (Figure 5C). To quantify this difference, we measured the ratio of nuclear Isl1 signal to background fluorescence in Isl1^ΔISE/ΔISE and in WT littermates (35 PCs and 35 non-PCs from 7 sections per heart, 2 biological replicates per genotype), and found a highly significant reduction in Isl1 protein expression in PCs from Isl1^ΔISE/ΔISE as compared to WT hearts, although Isl1^ΔISE/ΔISE hearts did have detectable Isl1 signal above background, unlike non-PCs (Figure 5D). Importantly, these data establish that ISE regulates Isl1 expression in the SAN in-vivo.

Figure 5. Deletion of the Isl1 SAN Enhancer Leads to Reduced Isl1 Expression in SAN and Abnormal SAN Development.

(A) (Top) Deletion of a 2.7-Kb genomic segment containing the Isl1 locus sinoatrial node enhancer (ISE) (Middle) Genotyping strategy to identify the mutant allele with an agarose gel (Bottom) on an F4 intercross confirming germline transmission of the deletion allele. (B) Genotypes from 88 weaned pups of Isl1^ΔISE/+ intercrosses with the indicated frequencies. (C) Immunohistochemistry of WT (top) and Isl1^ΔISE/ΔISE (bottom) SAN at E14.5 stained for Hcn4 (red), Isl1 (green), and DAPI. Area of detail is shown in the right panels. Pacemaker cell nuclei in the WT are Isl1+ (turquoise, denoted by white arrows) whereas the Isl1^ΔISE/ΔISE has less Isl1 signal in PC nuclei. (D) Quantification of green fluorescent signal (Isl1) in 35 Hcn4+ pacemaker cell nuclei and 35 Hcn4- non-PC nuclei, normalized to tissue background green fluorescent signal, from 7 sections each from two different sets of WT and Isl1^ΔISE/ΔISE littermates (denoted ‘replicate 1’ and ‘replicate 2’). ‘****’ denotes p < 0.0001 and ‘***’ denotes p < 0.001. (E) Junction of the superior vena cava (svc) and right atrium (ra) showing the location of the sinoatrial node (san) and the sectioning plane used to perform pacemaker cell counts and cross sectional area measurements. Z=0 denotes the center of the sinoatrial node head. (F) Quantification of PC nuclei in 12 sections each at the indicated z-distance from the center of the SAN head. Counts were performed for 2 pairs of E14.5 littermates (denoted ‘Replicate 1’ and ‘Replicate 2’). (G). Estimation of area of the SAN head for the sections analyzed in (F). ‘*’ denotes p < 0.05.

To test further whether ISE regulates Isl1 transcription in cis, we crossed Isl1^ΔISE/+ with Isl1^{3xFLAG/3xFLAG}, a targeted mouse line generated in our lab with 3xFLAG inserted before the Isl1 stop codon (Online Figure VII, A). RNA was isolated by manual dissection from the inflow regions of 8 E11.5 embryos from 2 litters, yielding Isl1^3xFLAG/ΔISE and Isl1^3xFLAG/+ RNA samples (n = 4 for each). To quantify the effect of DISE on Isl1 expression in cis, we designed PCR primers specific for the Isl1–3xFLAG mRNA (transcribed from the Isl1 allele in trans to DISE) and WT Isl1 mRNA (transcribed from the Isl1 allele in cis to either DISE or WT ISE). While levels of Isl1–3xFLAG mRNA were similar between the two groups (Online Figure VII,B), expression of WT Isl1 mRNA was reduced by 43% in E11.5 sinus venosus from Isl1^3xFLAG/^ΔISE as compared to Isl1^3xFLAG/+ (Online Figure VII, C), demonstrating that ISE regulates Isl1 transcription via a cis-regulatory mechanism.

Because Isl1 positively regulates PC proliferation⁶, we also assessed PC number and SAN size in Isl1^ΔISE/ΔISE hearts and WT littermates at E14.5. First, the number of Hcn4+ nuclei was counted in a series of sections taken from a 200 μm long region at the SAN head at 20 βm cranial to caudal intervals (Figure 5E). The results of this analysis for 2 biological replicates showed a reduction in Hcn4+ nuclei in the SAN head region in Isl1^ΔISE/ΔISE as compared to WT hearts at E14.5 (Figure 5F). In parallel, we observed a similar reduction in SAN volume as assessed by serial measurement of cross-sectional area of the Hcn4+ region in these sections (Figure 5G). These results demonstrate that ISE plays a role in establishing SAN cellularity and size, likely via regulation of Isl1 expression level during a critical developmental window.

Isl1 Locus SAN enhancer regulates SAN function.

Cardiac morphology and function as determined by echocardiography were similar between adult Isl1^ΔISE/ΔISE mice and their WT littermates (Online Table VII). To test for an effect of enhancer deletion specifically on heart rhythm, transmitters were implanted in 9 Isl1^ΔISE/ΔISE adult mice and 11 WT littermates. Continuous electrocardiograms were recorded for 48 hours in all mice and average heart rate was determined hourly and averaged across mice within each group and sex (Figure 6A). These data demonstrated a slower average heart rate in mutant as compared to WT hearts, particularly during low activity periods (8AM – 8PM, highlighted in grey in Figure 6A). To explore this difference further, heart rates during low activity periods for each mouse were averaged over one-minute intervals and collated into 10 beat per minute bins to generate a histogram depicting the amount of time spent at different heart rates (Figure 6B). Averaging histograms across mice within each genotype and sex showed a significant shift in mutant mice towards slower heart rates, with more time per mouse below 400 bpm in males and below 500 bpm in females (p < 0.05 for both). Assessment of intrinsic heart rate using atropine and propranolol, as well as maximum heart rate using isoproterenol, showed reductions for Isl1^ΔISE/ΔISE versus WT littermates, but only reached statistical significance for maximum heart rate in females (Online Figure VIII). Detailed analysis of beat-to-beat variability of RR intervals in the time domain demonstrated increased variability in female Isl1^ΔISE/ΔISE versus WT mice during high activity periods (Online Figure IX).

Figure 6. Abnormal SAN Function in Adult Isl1^ΔISE/ΔISE Mice.

(A) Average hourly heart rate in adult Isl1^ΔISE/ΔISE (red, n = 3 for female and n = 6 for male) and WT littermate mice (black, n = 6 for female and n = 5 for male). Error bars denote SEM. The low activity period is highlighted in gray (B) Heart rates were averaged over 1-minute intervals for each mouse studied in (A) (n = 11 WT and 9 Isl1^ΔISE/ΔISE, analyzed separately by sex), and then binned into 10 beat per minute bins for generation of heart rate histograms for WT and Isl1^ΔISE/ΔISE females (top) and males (bottom) Vertical scale bar = 25 counts. Comparison was made between WT and Isl1^ΔISE/ΔISE for total counts less than 400 bpm (male) or 500 bpm (female) for each mouse. ‘*’ denotes p < 0.05 for Mann-Whitney test. (C) A 2-second electrocardiogram tracing from a WT mouse (top) and with 2 examples of sinus arrhythmias observed in 2 different Isl1^ΔISE/ΔISE mice: a sinus pause (middle tracing), and an episode of bradycardia (bottom tracing) (D) Quantification of numbers of sinus arrhythmias recorded per hour over a 24-hour period in n = 11 WT and n = 9 KO mice. (E) Quantification of total percentage of time spent with heart rate less than 250 beats per minute (bradycardia) for 11 WT and 9 mutant mice. Comparison of WT and Isl1^ΔISE/ΔISE distribution were made with a Mann-Whitney test with the p value shown.

We also observed spontaneous arrhythmias in some mutant mice, including sinus pauses (defined as a two-fold or greater increase in the P-P interval from one beat to the next – Figure 6C, middle panel) and bradycardic episodes (defined as any period with heart rates less than 250 bpm—Figure 6C, lower panel). For each mouse, an investigator blinded to genotype counted the number of these events over a 24-hour period (Figure 6D). We found that while WT mice had between 0 and 2 episodes per hour, mutant mice had significantly more frequent episodes (p < 0.01). In addition, the cumulative duration of these episodes was also significantly greater in the mutant mice than in the WT mice (Figure 6E, p < 0.01), demonstrating that removal of ISE, in addition to affecting Isl1 expression and SAN development, also caused changes in SAN function.

To test whether episodes of SAN dysfunction in Isl1^ΔISE/ΔISE could arise from abnormal cellular electrophysiology in Isl1^ΔISE/ΔISE PCs, we used whole cell patch clamp to record action potentials from isolated PCs from 2 WT and 2 Isl1^ΔISE/ΔISE adult mice (Online Figure X). We found no significant differences in firing rate, minimum diastolic potential, and early diastolic depolarization, suggesting that ISE likely affects SAN function through its effects on SAN morphogenesis rather than by determining functional properties of differentiated PCs.

Evolutionary conservation and upstream regulation of Isl1 locus SAN enhancer.

Using the ECR Browser²¹ to visualize the degree of evolutionary conservation between mouse ISE and other mammalian species, we found that ISE exhibited >70% conservation in humans and opossum, suggesting that the regulatory network upstream of the enhancer may also be deeply conserved (Figure 7A, top track). Furthermore, alignment of ISE ATAC-seq signal with ChIP-seq datasets from embryonic myocardium¹¹ demonstrated that ISE contains experimentally validated binding sites for Gata4, Tbx5, and Tead, factors with conserved functions across vertebrate heart development (Figure 7A). Heart rhythm in zebrafish is controlled by a small population of automatically firing cells at the junction of the sinus venosus and atrium that expresses Isl1, Shox2, and Hcn4, similar to the mammalian SAN^{22, 23}. To test whether the regulatory network upstream of ISE is present in the zebrafish SAN, we cloned mouse ISE into a zebrafish enhancer-reporter vector and injected this construct into fertilized zebrafish eggs derived from a Tg(Myl7:mCherry-NTR) zebrafish line, in which mCherry is expressed throughout the myocardium. 18 out of 160 injected embryos examined at 48 hours post fertilization (hpf) exhibited restriction of GFP expression to the junction of the sinus venosus and atrium, consistent with deep evolutionary conservation of the regulatory network controlling enhancer activity (Figure 7B, top panel).

Figure 7. Evolutionary Conservation and Upstream Regulation of Isl1 SAN Enhancer Function.

(A) View of ISE with alignment to human, mouse, and opossum genomes (above, areas in red indicate greater than 70% conservation, scale from 50% to 100%) and previously published embryonic heart ChIP-seq data for Gata4, Tbx5, and Tead. (B) Maximum intensity projections of confocal z-stacks of 48 hpf Tg(myl7:mCherry-NTR) zebrafish hearts after injection of ISE:GFP reporter construct (top panel), or a 1.0-kb deletion fragment (Fr1, lower panel). Abbreviations: san, sinoatrial node, sv, sinus venosus, a, atrium, avc, av canal, v, ventricle. (C) Enhancer-reporter analysis of founders from injections of ISE and a 1-kb fragment (Fr1) in 48 hpf zebrafish hearts with the indicated patterns of reporter activity. Numbers in parentheses indicate the embryos observed to have the indicated expression pattern/total embryos injected that survived to 48 hpf.

To test whether the regions within ISE containing the Gata4, Tbx5, and Tead TF binding sites are sufficient for enhancer function, we injected a 1-kb fragment containing the cardiac TF binding sites (Fr1). In 19 out of 120 embryos examined, Fr1 directed reporter expression to the entire heart and sinus venosus without restriction of enhancer activity to the SAN region (Figure 7B, bottom panel, and Figure 7C). We did not observe any consistent pattern of extra-cardiac GFP expression in zebrafish embryos injected with either construct. These results suggest a model in which the Gata4, Tbx5, and Tead may be sufficient for enhancer activation but are insufficient to confer enhancer specificity to the SAN. Finally, we tested Fr1-hsp68-LacZ in mouse transient transgenic assays and did not observe any consistent pattern of reporter activity, although 1 out of 3 transgenic founders had enhancer activity throughout the heart, suggesting that activation and restriction of enhancer activity in the murine SAN likely requires more than one region of the 2.9-kb full length enhancer (Online Figure XI).

SAN ATAC-seq peaks occur at sites that regulate variation in human SAN function.

It is widely believed that non-coding variants are major contributors to genetically determined variation in complex human traits such as sinus node function and disease susceptibility^{17, 24}. We therefore hypothesized that single nucleotide polymorphisms occurring near sites of accessible chromatin in PCs and RACMs might exhibit associations with parameters that reflect sinus node function such as resting heart rate. To test this possibility with our ATAC-seq dataset, we first defined the set of human genomic regions that are syntenic to the murine ATAC-seq peaks we identified in RACMs and PCs, and interrogated resting heart rate data from the UK Biobank²⁵, a population-based cohort that included 434,532 genotyped participants meeting inclusion criteria (Online Figure XII, A). We then compiled a list of genotyped SNPs that overlap or are within 0.5 kb at either end of each of these regions of human genomic sequence (335,353 in total for all ATAC-seq peaks). After adjusting for age, sex, and body mass index, we identified SNPs associated with resting heart rate located in or near our ATAC-seq peaks, including numerous associations near genes that have been previously identified in heart rate GWAS analysis, such as MYH7, TTN, and KIAA1755, among many others (Figure 8A, Online Table VIII).

Figure 8. Association of Pacemaker Cell ATAC-seq Peaks with Resting Heart Rate Associated Single Nucleotide Polymorphisms.

(A) Manhattan plot showing association of 335,353 SNPs within or ± 0.5 kb away from a region syntenic to any mouse ATAC-seq peak with human resting heart rate ascertained from the UK Biobank cohort. Blue line indicates significance threshold of Bonferroni adjusted p value of 0.01 (corresponding to uncorrected p value 1.2×10⁻⁸) (B) Manhattan plot similar to (A) including 239 SNPs within or ± 0.5 kb away from the top 500 differentially accessible (DA) ATAC-seq peaks. Blue line indicates Bonferroni adjusted p value of 0.01 (corresponding to uncorrected p value 4.2×10⁻⁵). (C) Manhattan plot showing association of SNPs at the ISL1 locus with resting heart rate. The human genomic region syntenic to the mouse Isl1 SAN enhancer is highlighted.

Next, to test whether PC-enriched enhancers might harbor associations as well, we identified 239 SNPs that were within or ±0.5 kb outside of human genomic regions syntenic to PC-enriched ATAC-seq peaks. In this analysis, we identified SNPs at the HCN4 and RGS6 loci, genes with established roles in heart rate regulation (Figure 8B). However, by limiting our analysis to 239 SNPs and adjusting our significance cutoff accordingly, our analysis also uncovered a SNP at the ISL1 locus in a region of DNA syntenic to ISE that has a significant association with resting heart rate. To test whether other SNPs in the region surrounding the ISE might also exhibit associations with resting heart rate, we interrogated a broader region at the Isl1 locus, and we identified another SNP in close proximity to the ISE with a stronger association with resting heart rate (Figure 8C). Although the effect sizes are small (0.1 bpm/allele for rs1423611 and 0.2 bpm/allele for rs57848528, Online Figure XII B,C), the finding of multiple SNPs in close proximity to ISE that associate with variation in resting heart rate in a human population provides evidence supporting the functional conservation of this enhancer. Finally, we note that in a separate ATAC-seq dataset derived from human induced pluripotent stem cells differentiated into pacemaker-like cells, there was a discrete ATAC-seq peak at the region syntenic to the ISE, further supporting a role for ISE in regulating human heart rhythm (Vincent Christoffels, personal communication).

DISCUSSION

In this work, we have defined the epigenetic profile of cardiac pacemaker cells and experimentally validated a novel set of SAN enhancers, including a deeply conserved PC-specific enhancer for Isl1 (ISE) that is required for normal SAN development and function. Our findings provide new insights into transcriptional regulation in PCs and lay a foundation for future studies on PC differentiation, SAN development and SAN disease.

Comparison of epigenetic profiles of PCs and RACMs.

Our comparative ATAC-seq demonstrated that while the vast majority of ATAC-seq peaks are shared between RACMs and PCs, PCs had many more differentially accessible regions than RACMs. Notably, except for relatively few genes such as Nppa and Bmp10 that are not expressed in PCs, the majority of genes associated with the working cardiomyocyte program are downregulated in PCs rather than turned off completely², which may not require closure of open chromatin. In contrast, the PC gene program includes many transcripts that are not expressed at all in RACMs, requiring opening of genomic regulatory regions that are otherwise closed in working cardiomyocytes. Defining the epigenetic profile of the progenitor populations for PCs and RACMs might provide further insight into the events that underlie acquisition of the distinct PC epigenetic profile.

Transcriptional regulation in PCs.

While PCs and RACMs both express cardiac TFs Tbx5, Mef2C, and Gata4 at comparable levels, PCs express lower levels of Nkx2.5 and also express Tbx3, Tbx18, Shox2, and Isl1 ²⁶. Whereas both Tbx3 and Shox2 repress RACM genes in PCs^5,16, loss of function experiments support a positive role for Isl1 in transcriptional regulation in PCs⁶. Our motif analysis, in which we identified enrichment of Isl1 binding sites specifically in PC-enriched ATAC-seq peaks, is also supportive of such a role. Furthermore, we identified other LIM homeodomain binding motifs (Lhx1 and Lhx3) and motifs for TALE-class homeodomain proteins (including Meis1) in PC-specific ATAC-peaks, raising the possibility of combinatorial regulation of PC gene expression by a variety of different homeodomain proteins. In this context, a role for Shox2 in activating PC-specific expression cannot be ruled out, although we observed limited co-localization of PC-enriched ATAC-seq signal at Shox2 ChIP-seq peaks. On the other hand, involvement of cardiac TFs such as Gata4, Tbx5, and Tead in activation of ISE, along with the strong enrichment of these and other cardiac factors such as Mef2 in our motif analysis, raises the possibility that in some instances PC-restricted expression may be achieved largely through repression in working cardiomyocytes as opposed to action of distinct sets of transcriptional activators in PCs. The epigenetic profile presented here provides a foundation for future studies directed at testing these possibilities.

Identification of novel SAN enhancers.

In addition to ISE, 3 out of 16 putative enhancers we tested were sufficient to direct SAN-specific reporter activation at E11.5. All 3 (Hcn4, Rgs6, and Ptgfr) were located in proximity to genes enriched in PCs. Deletion of a region containing the enhancer at the Hcn4 locus reduces Hcn4 expression and affects SAN function in mice, supporting annotation of this element as a bona fide Hcn4 SAN enhancer²⁷. The other putative enhancers that we tested were not sufficient to direct reporter expression to the SAN, suggesting that additional regions may be required for full regulatory function or that these regions are active at different developmental stages. Nevertheless, the SAN enhancers we identified are of tractable size (1–3 kb), raising the possibility of deeper exploration of upstream transcriptional regulation in PCs in future work. In addition, using a candidate gene approach, we identified regulatory elements controlling expression of the key PC-specific TFs, including Isl1, Shox2, and Tbx3. Beyond our work here with the ISE, several of the elements at Shox2 and Tbx3 loci that we identified are also present as ATAC-seq peaks within the syntenic regions in human iPS-derived cardiac pacemaker-like cells and are required for expression of Shox2 and Tbx3 in the SAN in mouse embryos (Vincent Christoffels, personal communication), providing definitive experimental validation of the regulatory elements we identified with ATAC-seq.

Modular control of Isl1 during cardiogenesis.

While Isl1 is located within a megabase-sized gene desert that encompasses hundreds of highly conserved non-coding regions, ISE was the only highly differentially accessible region within the entire TAD in neonatal SAN. A broad Isl1 second heart field enhancer was previously identified in the region immediately downstream of the Isl1 3’-untranslated region²⁸. This enhancer was active as early as E7 in second heart field cardiac progenitors and depended on conserved forkhead TF binding sites. In contrast, activity of ISE was detectable at E8.5 in a restricted region near the cardiac inflow tract. Enhancer activity became progressively more restricted to the SAN as development progressed, with persistent SAN restriction in the perinatal period. Thus, the Isl1 SAN enhancer demonstrates modular control of Isl1 expression in different expression domains within the heart during development. Nevertheless, deletion of ISE did not result in complete loss of Isl1 expression, which suggests either that other regulatory elements that we did not identify are involved in persistent Isl1 expression in the SAN, or that there are shadow enhancers or other mechanisms that can activate Isl1 expression in the SAN in the absence of ISE.

ISE was active in adult hearts at low levels in areas outside the SAN including ventricular myocardium, ventricular conduction system, and perinodal right atrial myocardium, where Isl1 is not expressed. While this activity most likely reflects removal of ISE from its native genomic context in the hsp68 vector, enhancer utilization changes with cardiac maturation, raising the possibility that ISE may function differently in the adult heart¹¹.

Role of Isl1 SAN Enhancer in SAN development and function.

Complete loss of Isl1 in the developing SAN results in reversion of PC gene expression to an atrial-like pattern², with early-onset severe bradycardia and SAN hypoplasia⁶. Although we observed only partial loss of Isl1 expression in Isl1^ΔISE/ΔISE SAN, there was still significant SAN hypoplasia, indicating that SAN growth is sensitive to Isl1 dosage. This finding fits with studies that have defined E11 to E14 (when ISE activity is strongest) as the period when the SAN grows most rapidly²⁹. In addition, we have previously found that Isl1 expression in PCs as assessed by RNA-seq peaks at E14.5². Based on these findings, we propose a model in which ISE generates a burst of Isl1 expression in the developing SAN to support PC proliferation during a critical window. Accordingly, loss of ISE results in SAN hypoplasia but does not disrupt early events leading to SAN formation.

Our findings of increased sinus arrhythmias and slower heart rate in adult Isl1^ΔISE/ΔISE mice despite normal PC cellular electrophysiology demonstrate that developmental hypoplasia can lead to dysfunction of the adult SAN. Notably, we observed a larger heart rate decrease and greater change in beat-to-beat variation in female mice as compared to male mice. This could have resulted from sex differences in body mass leading to increased thermal stress in female mice at room temperature. Alternatively, our findings could have resulted from intrinsic sex differences in SAN structure and function that render females more sensitive to loss of ISE. Further investigation of sex differences, as well as more detailed analysis of heart rate variability, would be required to test these possible explanations more directly. Although the functional phenotype in Isl1^ΔISE/ΔISE mice was mild despite loss of an important regulatory element, modest changes in SAN function under resting conditions may affect survival and fitness in mice subjected to stress or increased cardiac demand, explaining the deep evolutionary conservation of ISE.

Relationship of SAN enhancers identified in mice to human heart rhythm.

Because mouse hearts function at much higher rates than human hearts, sequence level conservation of a non-coding genomic element such as ISE does not imply conservation of function. Although a naturally occurring deletion or disruption of ISE co-segregating with abnormal SAN function in a human family would provide definitive evidence of a conserved role for ISE, recent data suggest that common sequence variation in enhancers accounts for some variation and disease susceptibility in human populations^17,30. For example, the human region syntenic to the Hcn4 enhancer we identified here harbors SNPs that affect susceptibility to atrial fibrillation²⁷. In addition, increasingly large cohorts of patients with well-annotated clinical data have enabled GWAS-based discovery of numerous variants that have an impact on heart rhythm ^{31, 32}. Here we tested whether limiting the SNP analysis to a smaller search space could uncover more modest associations that might not reach statistical significance in genome-wide analysis. Our finding of 2 SNPs in the region of ISE that are associated with small changes in resting heart rate suggests that this approach may be fruitful for regulatory elements more broadly, and provides some evidence that ISE plays a role in human heart rhythm regulation.

NOVELTY AND SIGNIFICANCE.

What Is Known?

• Cardiac pacemaker cells (PCs) in the sinoatrial node (SAN) have a unique gene expression program that allows them to function as leading pacemakers of the heart

• Identification of tissue-specific gene regulatory elements through genome-wide profiling of chromatin accessibility has provided critical insights into how regulatory networks function in other cardiac cell types but has not yet been carried out on PCs.

What New Information Does This Article Contribute?

• PCs and right atrial cardiomyocytes (RACMs) were isolated with fluorescence activated cell sorting from neonatal mouse hearts and genome-wide maps of chromatin accessibility were generated using the assay for transposase accessible chromatin (ATAC-seq)

• Comparison of ATAC-seq peaks in PCs and RACMs revealed differentially accessible chromatin regions in PCs that functioned as tissue specific enhancers in-vivo, including a deeply evolutionarily conserved enhancer that regulates Isl1 transcription in PCs and is required for normal SAN development function.

• Human single nucleotide polymorphisms (SNPs) located near regions syntenic to accessible chromatin loci in PCs are associated with resting heart rate in a large human population, including small associations for SNPs near the Isl1 SAN enhancer.

The sinoatrial node (SAN) contains specialized cardiac pacemaker cells (PCs) that control heart rhythm though coordinated electrical automaticity, ensuring that heart rate and cardiac output are matched to metabolic demand. SAN failure is common and cannot be prevented or reversed, necessitating electronic pacemaker implantation for affected patients. Despite the importance of SAN function to cardiovascular disease, the gene regulatory networks that drive functional specialization of PCs in the SAN remain incompletely understood, limiting the development of biologically-based treatments for SAN failure. To address this knowledge gap, we compared the landscape of accessible chromatin in purified PCs with that of right atrial cardiomyocytes, revealing regions of differentially accessible chromatin that functioned as SAN enhancers in-vivo, including a deeply conserved regulatory element for the transcription factor Isl1 that was required for normal SAN development and function. Regions of the human genome that are syntenic to the differentially accessible chromatin regions from mouse PCs contained numerous SNPs that were associated with human heart rate, providing evidence that regulation of gene expression in PCs has a deeply conserved architecture that is relevant to human heart rhythm. Taken together, these data provide a new foundation for investigations of transcriptional mechanisms in SAN development and disease.

Nonstandard Abbreviations and Acronyms:

SAN

Sinoatrial node

RACM

Right atrial cardiomyocyte

PC

Pacemaker cell

TF

Transcription factor

ISE

Isl1 locus SAN enhancer

ATAC-seq

Assay for transposase-accessible chromatin