TBX5 and CHD4 coordinately activate atrial cardiomyocyte genes to maintain cardiac rhythm homeostasis

Abstract

Background:

Atrial fibrillation (AF), the most common sustained arrhythmia, affects 59 million individuals worldwide. The transcription factor TBX5 is essential for normal atrial rhythm. Its inactivation causes loss of atrial cardiomyocyte (aCM) enhancer accessibility, looping, transcriptional identity, and spontaneous AF. TBX5 interacts with CHD4, a chromatin remodeling ATPase canonically associated with the NuRD repressor complex.

Methods:

We investigated mechanisms by which TBX5 regulates chromatin organization by studying mice with aCM-selective inactivation of TBX5 or CHD4. We integrated multiple genomics approaches including concurrent single nucleus transcriptome and open chromatin profiling (snRNA-seq and snATAC-seq, respectively) and the genome-wide TBX5 and CHD4 chromatin occupancy assays.

Results:

We found that TBX5 recruits CHD4 to 33,170 genomic regions (TBX5-enhanced CHD4 sites). In addition to CHD4’s canonical repressive activity, we uncovered a CHD4 activator function predominantly at sites to which it was recruited by TBX5. TBX5-enhanced CHD4 recruitment increased local chromatin accessibility and promoted the expression of aCM identity genes. This mechanism of CHD4 recruitment by TBX5 was critical for sinus rhythm, as mice with CHD4 inactivation in aCMs had increased AF vulnerability. Assaying TBX5 binding in Chd4^AKO atria demonstrated that CHD4 also promotes TBX5 binding at over 10,000 genomic loci, including 3,051 TBX5-enhanced CHD4 sites. Consistent with its requirement to maintain normal atrial rhythm, CHD4 was implicated in the regulation of 42 genes linked to AF in humans. Nine had the hallmarks of TBX5-dependent, CHD4-mediated transcriptional activation.

Conclusions:

Our findings reveal that normal atrial rhythm requires CHD4, which activates and represses atrial genes in a context-dependent manner to maintain aCM gene expression, aCM identity, and atrial rhythm homeostasis.

Introduction

Atrial and ventricular cardiomyocytes (aCMs and vCMs) are highly specialized to achieve their distinct roles in the cardiac cycle. These specialized features are maintained and coordinated by the differential expression of thousands of genes^1,2. Chamber selective gene expression is, in turn, regulated by a/vCM-specific epigenetic regulatory programs and a/vCM specific chromatin organization^1,3. Abnormalities of aCM function can lead to atrial fibrillation (AF), the most common sustained human arrhythmia⁴, atrial impulse conduction becomes disorganized and forms fractionated, self-sustaining wavelets, causing uncoordinated atrial contraction and rapid and irregularly irregular activation of the ventricles⁵. AF elevates the risk of stroke and heart failure and increases health care costs in the United States by over $28 billion each year⁵.

The transcription factor (TF) TBX5 is essential for postnatal aCM identity^6–8. TBX5 inactivation caused loss of chromatin accessibility, the active enhancer mark H3K27ac, and promoter contacts of aCM enhancers⁹. TBX5 aCM inactivation caused the loss of atrial identity and spontaneous AF^9–11. The ability of TBX5 to maintain enhancer accessibility is likely facilitated by interactions with chromatin remodeling proteins. TBX5 interacts genetically and physically with two different chromatin remodelers. The Brg1/Brm-associated factor (BAF) complex facilitates the expression of direct TBX5 target genes Nppa and Bmp10, and BRG1 and TBX5 physically and genetically interact to promote normal heart development^12,13. TBX5 also physically interacts with chromodomain helicase DNA binding protein 4 (CHD4), the core nucleosome-remodeling enzyme of the repressive Nucleosome Remodeling and Deacetylase (NuRD) complex¹⁴. Chd4 deletion in developing and postnatal cardiomyocytes resulted in the upregulation of sarcomeric proteins from non-cardiomyocyte muscle lineages, including Acta1 from skeletal muscle and Myh11 from smooth muscle^15,16.

Here, we show that atrial-selective Chd4 knockout (Chd4^AKO) increased AF vulnerability and caused partially penetrant spontaneous AF. Mechanistically, TBX5 recruitment of CHD4 both activated and repressed genes in aCMs. Consistent with prior studies^15,16, TBX5 recruited CHD4 to repress non-CM lineage genes. In addition, we reveal an unappreciated gene-activating function of CHD4, in which TBX5-recruited CHD4 activated aCM genes. Our findings demonstrate that CHD4-mediated gene activation and repression are critical to aCM gene regulation and atrial rhythm homeostasis.

Methods

Please refer to the Supplemental Material for Detailed Methods and the Major Resources Table. Animal research was conducted under protocols approved by the Institutional Animal Care and Use Committees at Boston Children’s Hospital or the University of North Carolina, Chapel Hill.

Statistics

Statistical tests were performed in PRISM or R with P=0.05 as the significance threshold.

Data Availability

New sequencing data generated by the study has been deposited in GEO with accession number GSE284093 (Multiome), GSE284094 (Bulk RNA-seq), GSE284095 (ChIP-seq) and GSE296675 (CUT&RUN). All other data supporting the findings in this study are included in the main article and associated files. Resources used in the study are available upon request.

Results

BRG1 inactivation does not alter TBX5 target gene expression

TBX5 deletion caused loss of accessibility at enhancer elements promoting aCM-selective gene expression (Fig. 1A). To determine if TBX5 activity depends on BRG1, a TBX5 co-activator¹³, we injected P3 Brg1^flox/flox pups¹⁷ with adeno-associated virus serotype 9 (AAV9) with the aCM-selective Nppa promoter driving Cre (AAV9:Nppa-Cre)^9,18, or AAV9:Nppa-EGFP as a negative control. Brg1 inactivation in postnatal aCMs (Brg1^AKO) did not significantly affect expression of previously validated TBX5 targets Tbx5, Nppa, Nppb, Myl7, Scn5a, Atp2a2, and Ryr2 (Figure S1A)^9,11. Mice with aCM-selective inactivation of TBX5 (Tbx5^AKO) developed spontaneous AF by P21⁹. Electrocardiograms (ECGs) at P28 showed that Brg1^AKO remained in normal sinus rhythm (Figure S1B–C). We concluded that the TBX5-regulated postnatal aCM gene regulatory network is not dependent on Brg1.

Fig. 1. TBX5 recruits CHD4 to genomic loci.

A. Model of TBX5 gene regulation in aCMs. TBX5 inactivation results in the loss of enhancer accessibility and downregulation of aCM-selective genes. B. CHD4 ChIP-seq experimental design. Five animals were used per biological replicate, and two replicates were performed per group. C. Diffbind analysis of CHD4 binding in control and Tbx5^AKO samples. Significant regions (|Log₂FC| > 0.5, FDR<0.05) are colored red. On the right, the intersection of statistically significant regions with TBX5 occupancy data obtained from TBX5 bioChIP-seq in postnatal aCMs (GSE215065). D. Motif enrichment analysis of TBX5-enhanced CHD4 regions. q value, hypergeometric test with BH correction. E-F. Location of TBX5-enhanced and TBX5-impaired CHD4 regions with respect to genome annotations.

TBX5 enhances CHD4 chromatin occupancy in aCMs

We turned our attention to CHD4, another chromatin remodeler that interacts with TBX5¹⁴. To determine whether TBX5 actively recruits CHD4 in aCMs, we measured CHD4 occupancy in Tbx5^AKO and control aCMs (Fig. 1B). Neonatal Tbx5^flox/flox mice were treated at P3 with either AAV9:Nppa-Cre or AAV9:Nppa-EGFP. Nuclei marked by PCM1 (Pericentriolar Material 1), selectively expressed on cardiomyocyte nuclei, were isolated from 3-week-old atria in biological duplicate. We used chromatin immunoprecipitation followed by next generation sequencing (ChIP-seq) to measure CHD4 chromatin occupancy. Diffbind¹⁹ differential occupancy analysis identified 33,170 genomic loci enriched in control samples compared with Tbx5^AKO (TBX5-enhanced regions, Fig. 1C, Table S1). Of these, 12,134 (36.6%) regions were occupied by TBX5 in aCMs, based on our previously reported data¹. Far fewer CHD4 regions had greater CHD4 binding in Tbx5^AKO (363 TBX5-impaired CHD4 regions). 60,810 (64.5%) CHD4 regions were not significantly enriched in either sample type, consistent with CHD4’s recruitment by other TFs²⁰.

To identify regulators of the TBX5-enhanced CHD4 regions, we performed motif enrichment analysis. In addition to the TBX5 motif, we detected enrichment for the MEF2 and GATA4 motifs (Fig. 1D), cardiac TFs that interact with^21–23 and frequently co-occupy regions with TBX5^1,24,25. TBX5-enhanced CHD4 regions were also enriched for the motif of CTCF, a TF that interacted with CHD4 in cardiomyocytes by CHD4-IP/Mass Spectrometry²⁶.

TBX5-enhanced CHD4 binding regions were disproportionately located within gene promoters (± 1kb of the TSS, 30.7%; Fig. 1E). Other common locations included introns (38.5%) and distal intergenic regions (23.14%). TBX5-impaired regions were also enriched at promoter regions (12.4%), but to a lesser degree than TBX5-enhanced regions (Fig. 1F).

Together, these data indicate that TBX5 enhances CHD4 chromatin occupancy at regions co-occupied by TBX5.

CHD4 inactivation causes atrial remodeling and immune cell infiltration

To identify effects of CHD4 inactivation in postnatal aCMs, we injected AAV9:Nppa-Cre or AAV9:Nppa-EGFP (control) into P3 Chd4^flox/flox pups (Fig. 2A). By echocardiography, left ventricular size and function were comparable between AAV9:Nppa-Cre (Chd4^AKO) and control mice at 1 to 3 months (Fig. 2B). Heart sections from 2 to 3 month-old mice confirmed efficient and selective loss of CHD4 in Chd4^AKO aCMs (Fig. 2C) but not vCMs (Fig. 2D).

Fig. 2. Inactivation of CHD4 in aCMs results in atrial remodeling.

A. Schematic of the AAV9:Nppa-EGFP and AAV9:Nppa-Cre AAV constructs. Chd4^flox/flox mice that received AAV9:Nppa-EGFP and AAV9:Nppa-Cre were designated control and Chd4^AKO, respectively. B. Echocardiographic assessment of left ventricular size and function. Studies were performed at 4 weeks and 2–3 months after AAV injection. 4 weeks, n=13 per group. 2–3 mo, n=3 control and 4 Chd4^AKO. Two tailed t-test, ns, P>0.05. C-D. Confocal images of immunostained atrial and ventricular sections. White arrows in the Chd4^AKO images denote cells that retained CHD4 expression. Yellow boxed areas are enlarged in the second row of panels. Scale bars, 20 μm. E. Picrosirius red staining of control and Chd4^AKO heart sections (scale bars, 1 mm). Boxed areas are enlarged to the right (scale bars, 250 μm). Images are representative of three independent experiments. F. Quantification of fibrotic area. n=3 mice per group. Unpaired t-test. G. Confocal images of immunostained atrial sections from 3-month-old mice. Scale bar, 40 μm. Right, quantification of the percentage of CD45+ cells per section compared to the total number of cells. n=3 mice per group. Unpaired t-test.

Fibrotic remodeling is a hallmark of atrial dysfunction. We assessed fibrosis with picrosirius red/fast green staining on 2 to 3-month-old control and Chd4^AKO heart sections (Fig. 2E–F). Chd4^AKO mice had greater atrial fibrosis, whereas there was no difference in ventricular fibrosis. Staining for the pan-immune cell marker CD45 demonstrated greater numbers of immune cells in Chd4^AKO atria compared with controls (Fig. 2G), consistent with observations in human and mouse models of AF²⁷.

CHD4 inactivation increases atrial fibrillation vulnerability

SinceTbx5^AKO mice rapidly develop spontaneous AF⁹ with 100% penetrance, we investigated the rhythm of Chd4^AKO mice. We recorded surface ECGs and scored rhythm regularity using the standard deviation of the RR interval (SDRR; Fig. 3A). Below 3 months of age, SDRR was comparable between groups, suggesting that most mice remained in a regular rhythm. At 4 to 5-months-old, there was a tendency toward elevated SDRR (p=0.051). Analysis of the beat-to-beat variation in RR interval indicated irregularly irregular rhythm in Chd4^AKO mice with high SDRR, consistent with spontaneous AF (Fig. 3B). Out of 15 mice studied, one 5-week-old and three 5-month-old mice had markedly elevated SDRR, irregularly irregular rhythm, and no P waves, hallmarks of AF, compared to 0 of 12 in the control group (Fig. 3C).

Fig. 3. Chd4 inactivation in aCMs increases AF vulnerability.

A. Irregularity of cardiac rhythm was assessed from surface EKGs using the standard deviation of the R-R interval (SDRR). While SDRR was not significantly elevated at any timepoint, four mice (one at 4–5WO and 3 at 4–5MO) had elevated SDRR. Surface ECGs of these mice showed a loss of P waves, consistent with AF. Unpaired t-test. The dotted line represents the upper limit of normal for control mice (mean + 2SD). 1MO, n=4 control and 5 Chd4^AKO. 3MO, n=5 control and 6 Chd4^AKO. 4–5MO, n=3 control and 4 Chd4^AKO. B. Representative Poincaré plot of a control mouse and a Chd4^AKO mouse with high SDRR, showing the beat-to-beat variation in RR interval. C. Proportion of mice with spontaneous atrial arrhythmias. D. Representative data from programmed atrial stimulation experiments. The atrial rhythm was recorded by an intracardiac electrode, and the surface ECG was captured by limb leads. Following the pacing protocol, the atrial recording was evaluated for AF. A, atrial signal. V, ventricular signal. E. AF incidence following programmed atrial simulation of 2–3 mo mice. 1/11 control and 8/11 Chd4^AKO had AF for greater than 3 seconds following pacing. Fisher’s exact test. F. AF burden of the 2–3MO cohort of paced mice. AF burden is the total duration of AF following atrial pacing. Wilcoxon test.

We tested the AF vulnerability of Chd4^AKO mice at 23 months old by right atrial pacing (Fig. 3D). A greater proportion of Chd4^AKO were induced to develop sustained AF (lasting greater than 3 seconds, Fisher’s exact test P=0.0075) compared to control (Fig. 3E). The duration of induced AF was also increased in Chd4^AKO mice, including three mice that did not recover normal rhythm by 5 minutes, the study endpoint (Fig. 3F).

These data indicate that loss of Chd4 in aCMs increases AF vulnerability.

CHD4 and TBX5 coordinately regulate atrial gene expression

TBX5 recruitment of CHD4 to genomic loci suggested that these proteins cooperatively regulate gene expression in postnatal aCMs^15,20. To test this hypothesis, we performed bulk RNA-seq of P21 Chd4^AKO and control atria (n=4/group; Fig. 4A). Principal component analysis demonstrated a clear separation of the Chd4^AKO and control samples along PC1 (Fig. 4B). We identified 2,748 differentially expressed genes (DEGs; | Log₂FC | > 0.5 and P_adj <0.05), 1,077 and 1,671 with greater expression in controls or Chd4^AKO, respectively (Table S2, Fig. 4C).

Fig. 4. Bulk RNA-sequencing of P21 control and Chd4^AKO atria.

A. Experimental design. P21 Right and left atria from four control mice and four Chd4^AKO mice were used in the experiment. B. Principle component analysis of WT and Chd4^AKO RNA-seq samples. C. Differentially expressed genes between WT and Chd4^AKO atria. Significantly differentially expressed genes (DEGs) had |Log₂FC|>0.5 and P_Adj<0.05. P values, Wald test with BH correction. D. Heatmap of aCM-selective TBX5 target genes and genes upregulated in TBX5 KO aCMs in snRNA-seq data from WT and Tbx5^AKO aCMs compared to WT and Chd4^AKO atria. Gene symbols in purple were significantly differentially expressed in control vs KO in both Tbx5 and Chd4 knockout experiments. P values for TBX5 single cell experiment, Wilcoxon test with BH correction. E-F. GO enrichment terms enriched for control or Chd4^AKO DEGs. The X axis shows the ratio of the number of genes in the intersection of DEGS and the gene set to the total genes in the gene set. Datapoints are colored by the adjusted P value, and datapoint size corresponds to the total number of genes for each term. GO terms highlighted in red are of particular biological interest. Adjusted P values, hypergeometric test with BH adjustment.

To determine whether CHD4 regulated TBX5 target genes, we examined a curated list of aCM-selective genes that are downregulated (Sbk2, Sbk3, Myl7, Nppa, Myl4 and Fgf12) or upregulated (Lrrc4, Fbn2, Kcnq5, Casq1 and Nrg1) in Tbx5^AKO aCMs⁹ (Fig. 4D). For these selected targets, gene expression changes between WT and Chd4^AKO atria were largely comparable to differences between WT to Tbx5^AKO aCMs. GO term analysis revealed that genes more highly expressed in control samples were related to muscle function (Fig. 4E). Genes upregulated in Chd4^AKO samples were related to lymphocyte activation and atrial remodeling (Fig. 4F), consistent with the atrial remodeling, fibrosis, and immune cell infiltration that we observed in Chd4^AKO atria (Fig. 2E–G).

We compared the quantitative change in gene expression in the Chd4^AKO bulk RNA-seq and our previously reported TBX5 aCM knockout (Tbx5^AKO) single-nucleus RNA-seq (snRNA-seq) assays⁹ (Figure S2A). Most genes differentially expressed in both experiments were altered in the same direction by TBX5 or CHD4 inactivation (purple points in quadrants I and III). In each dataset, genes with aCM-selective expression were more highly expressed in WT compared with the AKO group (Figure S2B), consistent with CHD4 and TBX5 promoting their expression. In total, CHD4 and TBX5 coordinately promoted and repressed the expression of 513 and 633 genes, respectively (Figure S2C,D). This overlap between CHD4 and TBX5 DEGs was highly significant (hypergeomtric test: P<2.5E-34). Genes activated by both factors facilitate metabolism and nucleotide biosynthesis, whereas genes repressed by both CHD4 and TBX5 were related to extracellular matrix remodeling and TGFβ signaling (Figure S2E). Interestingly, GO analysis of genes upregulated in Chd4^AKO and downregulated in Tbx5^AKO (Figure S2E, QIV) revealed terms associated with ion transport and cardiac muscle contraction, perhaps explaining the lower incidence of spontaneous AF observed in Chd4^AKO mice compared to Tbx5^AKO mice.

Multiomic analysis of control and Chd4^AKO mice atria

Given CHD4’s role in chromatin remodeling, we performed concurrent snRNA-seq and snATAC-seq on Chd4^AKO and control atria. In total, four Chd4^AKO samples and two control samples were processed, with each sample consisting of a right and left atrium from two mice, one male and one female (Table S3). For the Chd4^AKO samples, 7 of the 8 mice had inducible (4; 3.5 months-old) or spontaneous (3; 5 months old) AF. Control samples were from 5-month-old wild type mice. Following doublet removal and rigorous quality control (see Methods), we obtained 27,263 high-quality nuclei (Figure S3A–B). The snRNA and snATAC data were integrated using a weighted nearest neighbors (WNN) approach, grouped by Louvain clustering, and visualized using Uniform Manifold and Approximation Projection (UMAP; Figure S3C). Data were highly concordant between genotypes and we observed minimal batch effect between samples (Figure S3D,E).

Cell states were annotated using atrial cell marker genes (Figure S4A). We removed a small cluster of likely residual doublets with high expression of epicardial and immune cell markers. A small cluster of vCMs was identified in control samples (Figure S4A,B). aCM_1, the largest aCM cluster, contained predominantly control nuclei, as did smaller aCM clusters aCM_3 and aCM_4 (Figure S4C). In contrast, the second largest aCM cluster, aCM_2, derived nearly exclusively from Chd4^AKO samples (Magenta dotted line, Figure S3D; Figure S4B). There was also genotype-specific enrichment for endocardial (Endocardium_3, control; Endocardium_2, Chd4^AKO) and macrophage (Macrophage_3, control, Macrophage_2, Chd4^AKO) clusters.

CHD4 controls the aCM identity gene program

To determine if aCM_2 represented Chd4^AKO aCMs, we projected Cre expression onto the WNN UMAP (Fig. 5A,B). Cre expression was largely confined to aCM_2, identifying them as Chd4^AKO aCMs. We identified DEGs between aCM_2 and aCM_1, the largest control aCM cluster (Fig. 5A,B). 2,256 DEGs ( |Log₂FC| > 0.5 and P_adj<0.05) were identified, with 431 and 1825 higher aCM_1 and aCM_2, respectively (Table S4). DEGs from this snRNAseq comparison agreed with those from the bulk RNA-seq of control and Chd4^AKO atria (Pearson r=0.835; P=1.9E-302. Figure S5A), as did GO terms enriched for DEGs more highly expressed in control or Chd4^AKO aCMs (Fig. 5D). Chd4^AKO DEGs were related to chamber remodeling, while control DEGs were related to muscle function, cell junction assembly, and cardiac rhythm.

Fig. 5. Combined snRNA-seq and snATAC-seq identifies distinct WT and CHD4^AKO aCM cell states.

A. Concurrent snRNA-seq and snATAC-seq comparison of Chd4^AKO (n=4) and control (n=2) atria. Weighted nearest neighbor (WNN) UMAP integration of both data types is split by sample type. aCM_1 and aCM_2 predominantly originated from Chd4^AKO and control atria, respectively. B. The expression of Cre was projected onto the WNN UMAP. Cre is predominately detected in aCM_2. C. Pseudobulk RNA-seq analysis of genes differentially expressed (|Log2FC|>0.5 and Adjusted P-value <0.05) between aCM_1 (control) and aCM_2 (Chd4^AKO). P values, Wilcoxon test with BH correction. D. Gene ontology terms enriched for DEGs more highly expressed in Chd4^AKO or control. P values, hypergeometric test with BH correction. E. Altered expression of aCM and vCM selective genes in Chd4^AKO or control aCMs. Red and blue dots indicate aCM-selective and vCM-selective DEGs, respectively, as defined in Cao et al. 2023 (GSE215065). Contingency tables and the results of the Fisher test are shown to the right. F. Expression of smooth muscle and skeletal muscle genes in Chd4^AKO and control aCMs. The left-most column indicates normal tissue-selective expression, based on GTEX (Figure S5). The next column indicates fold-change in gene expression from aCM_1 vs. aCM_2. The main heatmap indicates row-scaled expression in bulk RNA-seq from Chd4^AKO versus control atria. Genes colored purple were significantly differentially expressed in the snRNA-seq or bulk RNA-seq experiments, respectively. G. Aggregate scores were determined using the expression of skeletal muscle genes Neb, Atp2a1, Kcnq5, Tnnt3, and Tnnt1, or smooth muscle genes Csrp1, Flna, Slit2, Myl9, and Cald1, and projected onto the UMAP. A violin plot was created showing the aggregate score in cells from aCM_1 and aCM_2.

To determine whether CHD4 regulates aCM identity, aCM- and vCM-selective genes were overlaid with DEGs from the aCM_2 vs aCM_1 comparison (Fig. 5E). aCM-selective genes were overrepresented among genes activated vs repressed downstream of Chd4 (26% (112/413) vs. 12% (213/1825); Fisher’s exact test P=9.1E-13). In contrast, we did not observe significant enrichment of vCM-selective genes among genes activated or repressed downstream of Chd4.

In fetal and postnatal cardiomyocytes, CHD4 binds cardiac TFs, including TBX5, to repress the expression of skeletal and smooth muscle genes^15,16. To determine whether this function is conserved in aCMs, we examined the top ~100 DEGs ranked by significance in aCM_1 and aCM_2 for their expression levels in human left ventricle, atrial appendage, skeletal muscle, and coronary arteries, and their hallmark muscle cells from the Genotype-Tissue Expression Project (GTEX), a bulk and snRNA-seq atlas of human gene expression (Figure S5B,C). Top-ranked DEGs activated by CHD4 were highly expressed in atrial appendage and ventricle, with few showing enrichment in coronary artery or skeletal muscle, or their related myocyte subtypes. In contrast, top-ranked DEGs repressed by CHD4 were enriched in skeletal muscle tissue and myocytes (e.g., Myl3, Tnnt3, Tnnt1, Tnni1, Atp2a1, Acta1, Casq1, Ypl2, Scn1b) or in coronary artery tissue and smooth muscle myocytes (e.g., Prdm16, Eps8l2, Pygl, Atp2a3). Heatmaps generated from the Chd4^AKO snRNA-seq and bulk RNA-seq datasets (Fig. 5F) confirmed that Chd4 inactivation downregulated aCM-selective genes and upregulated skeletal and coronary artery tissue genes. Aggregate scores of genes specifically expressed in skeletal muscle myocytes (Neb, Atp2a1, Kcnq5, Tnnt3, Tnnt1) or smooth muscle myocytes (Csrp1, Flna, Slit2, Myl9, Cald1) compared to aCMs, identified from the GTEX snRNA-seq atlas (Figure S5B,C), confirmed their upregulation in aCM_2 Chd4^AKO aCMs compared to control aCM_1 (Fig. 5G). These data revealed that CHD4 is critical for preventing the misexpression of genes from other muscle lineages in postnatal aCMs.

A dual function for CHD4 in enhancer accessibility maintenance and repression

Next, we examined differences in chromatin accessibility between aCM_1 and aCM_2 (Fig. 6A). There were 13,935 differentially accessible regions (DARs) between control and Chd4^AKO aCMs (aCM_1 vs. aCM_2), with 2,471 and 11,464 regions having greater accessibility in aCM_1 (control DARs) or aCM_2 (Chd4^AKO DARs) aCMs, respectively (Table S5). The greater number of regions with increased accessibility following Chd4 inactivation is consistent with its canonical function as a repressor in the NuRD complex²⁸. Most control DARs (83.6%; 2,061 regions) CHD4-bound, suggesting CHD4 directly participates in maintaining the accessibility of these regions. In contrast, CHD4 bound only 41.5% (4,760 regions) Chd4^AKO DARs, suggesting that CHD4 governs the accessibility of these regions by a mix of direct and indirect mechanisms. To determine how region accessibility affected gene expression, we compared the transcript levels of genes near control and Chd4^AKO DARs in aCM_1 and aCM_2 (Fig. 6B). Compared to genes neighboring regions that were not differentially accessible, genes near control DARs overall were expressed at greater levels in aCM_1, and genes near Chd4^AKO DARs were expressed at greater levels in aCM_2, consistent with control and Chd4^AKO DARs acting as enhancer or repressor cis-regulatory elements, respectively.

Fig. 6. Transcriptional activation by TBX5-recruited CHD4.

A. Differentially accessible regions (DARs) in aCM_1 vs. aCM_2. DARs were defined by |Log₂FC|>0.25 and P_adj<0.05. Adjusted P values, LRT with BH correction. B. The Log₂FC of gene expression in aCM_1/aCM_2 is plotted for genes near control DARs, Chd4^AKO DARs, or non-DAR regions. Wilcoxon rank-sum test. C-D. DARs colored by their overlap with TBX5-enhanced (C) and TBX5-impaired (D) CHD4 regions. E. The proportion of control or Chd4^AKO DARs overlapping with TBX5-enhanced, TBX5-impaired, and TBX5-independent regions. F-G. Control or Chd4^AKO DARs with TBX5-enhanced CHD4 occupancy were classified as neighboring aCM-selective, vCM-selective, or non-selective (not differentially expressed between aCMs and vCMs) genes. Regions were further classified by association with DEGs that were upregulated in aCM_1 (control), aCM_2 (Chd4^AKO), or neither. Numbers indicate number of regions.

To determine if control DARs act as aCM enhancers regulated by CHD4 and TBX5, we analyzed the active enhancer mark H3K27ac at control and Chd4^AKO DARs in Tbx5^AKO aCMs, Chd4^AKO aCMs, and respective control aCMs. Control DARs exhibited H3K27ac signal in WT aCMs, consistent with enhancer activity of these regions (Figure S6A). Either Tbx5 or Chd4 inactivation reduced both H3K27ac signal intensity and chromatin accessibility, indicating that both factors stimulate enhancer activity. Unlike control DARs, Chd4^AKO DARs had little detectable H3K27ac signal in WT aCMs (Figure S6B), consistent with Chd4^AKO DARs functioning as repressor rather than enhancer elements in control aCMs.

TBX5 enhanced CHD4 binding at 33,170 genomic loci and impaired CHD4 binding at 363 genomic loci (Fig. 1C). Intersecting these regions with control and Chd4^AKO DARs revealed that TBX5 enhanced CHD4 binding at a greater proportion of control (1,168/2,471, 47.3%) compared to Chd4^AKO (1,302/14,464, 11.4%) DARs (Fig. 6C–E), while impairing CHD4 binding at only 4 (0.2%) control and 19 (0.1%) Chd4^AKO DARs, respectively (Fig. 6D–E). Since TBX5 is required to activate aCM-selective genes in aCMs, we assessed the relationship of control and Chd4^AKO DARs to the regulation of aCM-selective genes. Control DARs neighbored 3-fold more aCM-selective genes than vCM-selective genes (15.7% vs. 4.2%; Figure S6H), greater enrichment than observed for Chd4^AKO DARs (11.1% vs. 6.2%; Figure S6H). These data suggest that CHD4 cooperates with TBX5 to activate aCM-selective genes in aCMs.

To further evaluate the effect of TBX5-enhanced control and Chd4^AKO DARs on the expression of aCM- and vCM-selective genes, we analyzed the association of these regions with aCM-, vCM-, or non-selective genes and the changes in gene expression caused by the inactivation of Chd4 (Fig. 6F–G). Of the 206 TBX5-enhanced control DARs neighboring aCM genes, 97 (47.1%) were associated with aCM-selective genes with greater expression in aCM_1 compared to aCM_2 (Fig. 6F). Other functionally important genes without chamber-selective expression that were activated by TBX5-enhanced control DARs included Ryr2 and Rbm20, both implicated in AF^29–31. Together, these data underscore the importance of TBX5-enhanced CHD4 binding to activate aCM-selective genes and genes important for atrial rhythm homeostasis.

The parallel analysis of the 1,302 TBX5-enhanced Chd4^AKO DARs revealed that 79 regions neighbored vCM-selective genes, 36 (45.5%) of which were repressed downstream of CHD4 (Fig. 6G). Of the 162 TBX5-enhanced Chd4^AKO DARs associated with aCM-selective genes, 59 (36.4%) were repressed by CHD4. The repressed aCM-selective genes included Atp2a1 and Myl1, which are expressed at greater levels in aCMs than vCMs but are more often associated with skeletal muscle, and Tgfb2, associated with chamber remodeling. These data indicate that TBX5-enhanced CHD4 binding and repression contributes to aCM gene regulation.

Together, these data reveal dual repressive and activating functions of CHD4 in aCMs. TBX5 recruits CHD4 to a set of genomic loci to activate a subset of aCM-selective genes. At the same time, TBX5 recruits CHD4 to distinct loci to repress gene expression, consistent with prior studies¹⁵.

CHD4 enhances TBX5 binding at specific loci

Given the loss of accessibility at control DARs (Fig. 6A), we asked if CHD4 inactivation altered the binding of TFs to these regions. To test this hypothesis, we applied a recently reported deep learning framework, Seq2PRINT³², to pseudobulk snATAC data from aCM_1 and aCM_2 aCMs, as well as Tbx5^AKO and control aCMs, to predict TF binding. We initially used Seq2PRINT to study the promoter and two putative enhancers of Myl4 accessible regions bound by TBX5 (Figure S7A). The obtained footprints were less than 100 bp in size, as expected for TFs. CHD4 or TBX5 inactivation reduced the density of a subset of these footprints compared to control (arrows, Figure S7A, top panel). Predictive TF binding scores were high at the enhancer regions within the TBX5 ChIP summit. These scores were reduced in Tbx5^AKO (Figure S7A, middle and bottom panels), consistent with loss of TBX5 binding. The TF binding score was also lower in aCM_2, predicting that inactivation of Chd4 reduced TBX5 binding.

To determine if the predicted loss of TF binding was broadly applicable to regions that lose accessibility in Chd4^AKO aCMs, we calculated TF binding scores for TBX5, MEF2C, GATA4, and NKX2–5 at TBX5-enhanced control DARs (Figure S7B–C, quantified in Figure S7D). Binding scores of these four TFs were significantly lower in Tbx5^AKO aCMs (Figure S7C–D), consistent with their collaborative binding with TBX5.^1,24 Inactivation of Chd4 reduced binding scores of TBX5 and MEF2C but not GATA4 or NKX2–5 (Figure S7B,D). This result predicts that CHD4 promotes TBX5 chromatin occupancy.

To experimentally test this prediction, we performed Cleavage Under Target & Release Using Nuclease (CUT&RUN) for TBX5 in control and Chd4^AKO atria (Fig. 7A). We used two primary antibodies recognizing TBX5 and performed each reaction in biological duplicate, producing four control samples and four Chd4^AKO samples. TBX5-bound regions identified by either antibody were enriched for similar motifs, including the TBX5 motif (Fig. 7B). PCA analysis of binding intensities the union of regions bound across all samples showed good agreement between antibodies. Furthermore, samples clustered by genotype, indicating that Chd4 inactivation significantly altered TBX5 chromatin occupancy (Fig. 7C).

Fig 7. CHD4 promotes TBX5 genomic occupancy.

A. Experimental design of the TBX5 CUT&RUN. Atria were pooled from five ~P21 control or Chd4^AKO mice per replicate. 500,000 nuclei were used as an input for each reaction. Two antibodies were used in separate CUT&RUN pulldown experiments. Two biological replicates were performed for each antibody for each genotype, for a total of n=4 control samples and n=4 Chd4^AKO samples. B. Motif enrichment analysis was performed on peaks called from control samples for each antibody using HOMER. q values, hypergeometric test with BH adjustment. C. Principal component analysis was performed using binding intensity of each sample across a unified peak set (all peaks called for each sample). Samples clustered by genotype, rather than by primary antibody used in the experiment. D. CHD4 binding signal (from CHD4 aCM ChIP-seq) and TBX5 binding signal (from TBX5 CUT&RUN) at control and Chd4^AKO DARs. Wilcoxon Rank Sum test of signal intensity at the center 100 bp of the region. Solid and dotted lines indicate biological duplicate samples. E-F. Diffbind analysis comparing four control to four Chd4^AKO samples. E, Binding signal for each TBX5 replicate at regions with differential TBX5 binding affinity in control and Chd4^AKO samples, determined by DiffBind. F, TBX5 differential chromatin occupancy. Left, regions with greater TBX5 binding in control or Chd4^AKO ( |log₂FC| >1, FDR<0.05) are colored blue and black, respectively. Bottom five panels, volcano plots of differentially bound regions. Orange regions, regions co-bound by CHD4. red regions, regions with greater CHD4 binding in WT aCMs compared to Tbx5^AKO aCMs. Purple regions are control DARs, with greater accessibility in control (aCM_1) compared to Chd4^AKO (aCM_2) aCMs. Green regions are Chd4^AKO DARs, with greater genomic accessibility in Chd4^AKO aCMs (aCM_2) compared to control (aCM_1). FDR calculated with BH. G. The overlap between TBX5-enhanced CHD4 binding sites and CHD4-enhanced TBX5 binding sites. Hypergeometric test. H. Data from TBX5 CUT&RUN and H3K27ac CUT&RUN in control and Chd4^AKO, CHD4 ChIP-seq from control and Tbx5^AKO aCMs, and snATAC data from Chd4^AKO (aCM_2), Tbx5^AKO, and control aCMs (aCM_1 and control TBX5 aCMs) from each multiomics experiment, are visualized at intronic enhancer regions of aCM-selective genes Nav3 and Psd3. Asterisks represent Padj<0.05 in signal intensity for each experiment.

We compared TBX5 occupancy in control or Chd4^AKO aCMs (Fig. 7D). At control DARs, Chd4 inactivation reduced TBX5 occupancy (Fig. 7D, right panels), mirroring the effect of Tbx5 inactivation on CHD4 occupancy (Fig. 7D, left panels). In comparison to control DARs, binding differences at Chd4^AKO DARs were small. We used Diffbind¹⁹ to better quantify differences in TBX5 binding between control and Chd4^AKO samples. At the union of peak regions across all the samples (61,935 regions; Fig. 7E–F, Table S6), 10,154 regions had enhanced TBX5 binding in control aCMs compared to Chd4^AKO aCMs, and only 819 regions had increased TBX5 binding in Chd4^AKO aCMs. Of the 10,154 regions with CHD4-enhanced TBX5 binding, 4,642 were bound by CHD4, and TBX5 enhanced CHD4 binding at 3,051 of these regions (Fig. 7E), i.e., at these 3,051 regions, TBX5 and CHD4 promoted each other’s binding. The overlap between CHD4-enhanced TBX5 regions and TBX5-enhanced CHD4 regions was highly significant (Fig. 7G).

We visualized these chromatin changes at the regulatory regions of two aCM-selective genes that are downregulated in Chd4^AKO and Tbx5^AKO aCMs: Neuron navigator 3 (Nav3), which regulates heart development, contraction, and morphology in zebrafish,³³ and Pleckstrin and Sec7 Domain containing 3 (Psd3), which neighbors a variant linked to AF by a recent genome-wide association study³⁴ (Fig. 7H). Chd4 inactivation reduced TBX5 occupancy as well as H3K27ac at these regions. Similarly, Tbx5 inactivation reduced CHD4 occupancy of these regions. Either Chd4 or Tbx5 inactivation reduced accessibility of these regions in aCMs compared to control.

Together, these data indicate that CHD4 and TBX5 function together to bind regulatory regions in the aCM genome and promote the expression of aCM genes.

CHD4 co-occupancy with other cardiac transcription factors

Other CHD4-interacting TFs have important roles in atrial gene regulation, including GATA4 and NKX2–5²⁰. Using previously generated biotin-mediated ChIP-seq data¹, we examined TBX5, GATA4, MEF2A, MEF2C, NKX2–5, SRF and TEAD1 occupancy at Chd4^AKO and control DARs in WT aCMs (Figure S8A). TBX5 was the most prominently bound TF at both control and Chd4^AKO DARs – TBX5 bound 53% of control DARs. NKX2–5 and GATA4, known CHD4 interactors, co-occupied 26.8% and 11.1% of control DARs, respectively, and TEAD1 co-bound 24.0% of control DARs. These data reveal TBX5 as the primary factor that recruits CHD4 as an activator, with other TFs playing lesser roles. A smaller percentage of Chd4^AKO DARs were bound by TBX5 (19%) or other cardiac TFs, suggesting that CHD4 recruitment to repressed loci predominantly occurs through mechanisms independent of binding to these major cardiac TFs.

To identify other TFs that participate in CHD4 recruitment, we performed motif enrichment analysis on CHD4-occupied control and Chd4^AKO DARs (Figure S8B; full list in Table S7). Many motifs (91) were enriched in both control and Chd4^AKO DARs, including motifs for TBX5, MEF2A/C, and GATA2/4/6. Among the 30 motifs uniquely enriched in control DARs, the top ranked motif belonged to the glucocorticoid receptor, implicated in cardiac fibrosis by interactions with the mineralocorticoid receptor (NR3C2), another control-enriched motif³⁵. Among the 147 motifs enriched in Chd4^AKO DARs, top uniquely enriched motifs belonged to Fos and AP-1, transcriptional regulators of immediate early stress responses, and ETS-family transcriptional regulators, implicated in atrial remodeling and arrhythmia³⁶.

CHD4 regulates AF-associated genes

TBX5-enhanced CHD4 binding activates and represses hundreds of genes in aCMs, including genes selectively expressed in aCMs, and the AF phenotype of Chd4^AKO mice demonstrates that these genes support atrial rhythm maintenance. Thus, we compared genes that are upregulated or downregulated in aCMs by inactivation of CHD4 or TBX5 with 241 genes associated with human AF by GWAS or familial studies⁴ (Figure S9A). 42 human AF genes were regulated by both CHD4 and TBX5, including MYL4, which was enhanced by CHD4 and TBX5 (Figure S9B).

The Myl4 locus contains two predicted enhancers with TBX5 and CHD4 occupancy, chromatin accessibility, and flanking H3K27ac signal (blue highlights, Figure S9C). H3K27ac HiChIP from aCMs⁹ revealed that Myl4_enh2 contacts the Myl4 promoter and Myl4_enh1 adjoined a loop anchor also contacting the promoter. Correlation between the accessibility of these regions and Myl4 expression further supported the function of these regions as Myl4 enhancers (Signac linkage, Figure S9C). Tbx5 inactivation reduced CHD4 occupancy, chromatin accessibility, and flanking H3K27ac signal at Myl4_enh1, Myl4_enh2, and the Myl4 promoter (green highlight, Figure S9C). Inactivating CHD4 similarly reduced TBX5 occupancy, flanking H3K27ac, and genomic accessibility of the regions.

To further investigate CHD4 and TBX5 regulation of Myl4_enh1, we constructed a reporter AAV in which Myl4_enh1 and a minimal promoter drive atrial expression of GFP (Figure S9D). The vector additionally contains an RNA polymerase III promoter driving a noncoding Broccoli transcript, which serves as an internal control. The Myl4_enh1 reporter AAV was injected with (AKO) or without (control) AAV9:Nppa-Cre into Chd4^flox/flox or Tbx5^flox/flox mice, and enhancer activity was measured by the level of GFP transcript normalized to Broccoli. Inactivation of either Chd4 or Tbx5 significantly reduced Myl4_enh1 enhancer activity (Figure S9E). Collectively, these data indicate that TBX5 and CHD4 coordinately stimulate Myl4_enh1 and Myl4_enh2 to activate Myl4 transcription.

Together these results support TBX5 and CHD4 regulation of multiple genes implicated in human AF.

Discussion

This study established that CHD4 is essential to maintain aCM function and atrial rhythm homeostasis. Mechanistically, we showed that CHD4 has dual activities in aCM gene regulation. First, consistent with prior studies,^16,20 we confirmed that CHD4 transcriptionally represses genes not normally expressed in cardiomyocytes, such as those characteristic of skeletal and smooth muscle (Fig. 8, CHD4 repressor role). Second, we revealed that CHD4 transcriptionally activates genes, including several selectively expressed in aCMs (Fig. 8, CHD4 activator role).

Fig. 8. A model of TBX5-associated CHD4 activator and repressor functions.

Top, CHD4 activator function. TBX5 recruits CHD4 to enhancer elements and promoters to maintain their accessibility. Inactivating CHD4 results in the loss of CHD4 and TBX5 from these regions, and the loss of genomic accessibility, leading to enhancer inactivation and gene downregulation. Inactivating TBX5 similarly results in the loss of CHD4, because TBX5 is necessary to recruit CHD4 to these regions. Enhancers similarly close and the gene is downregulated. Bottom, CHD4 repressor function. TBX5 recruits CHD4 to limit enhancer accessibility. The loss of CHD4 results in increased accessibility and gene activation. The loss of TBX5 reduces CHD4 recruitment, similarly causing increased accessibility and gene activation. No significant loss of TBX5 to Chd4^AKO DARs was observed in the Chd4^AKO.

CHD4 was required to repress sarcomeric genes of non-cardiac muscle lineages, consistent with prior reports^16,20. CHD4 canonically represses genes as a component of the NuRD complex, although some CHD4 functions are reported to be independent of the NuRD complex³⁷. TBX5-dependent CHD4 recruitment accounted for a small fraction of genes repressed by CHD4 in aCMs; only 11.4% of regions that gained accessibility in Chd4^AKO aCMs had TBX5-dependent CHD4 occupancy. Other cardiac TFs likewise co-occupied a minority of Chd4^AKO DARs. These data demonstrate that CHD4’s previously described function of repressing non-cardiomyocyte sarcomere genes^15,16 in fetal and ventricular CMs is active in adult aCMs.

We found that CHD4 has a novel transcriptional activator function in aCMs tied to its requirement to maintain accessibility at 2,471 loci. This CHD4 activating function was closely aligned with TBX5, as TBX5 enhanced CHD4 occupancy at 47% of these control DARs. These data support a model in which TBX5 recruits CHD4 through their previously described physical interaction¹⁴, increasing chromatin accessibility and activating gene transcription. CHD4-mediated recruitment of TBX5 and other TFs is a second mechanism by which CHD4 may activate gene transcription. CHD4 promoted TBX5 chromatin occupancy at 10,154 regions. Indeed, CHD4 and TBX5 reinforced each other’s binding at 3,051 regions.

Our results establish CHD4 as the first chromatin remodeling enzyme associated with AF and a key epigenetic regulator required for atrial homeostasis. Ablation of Tbx5 in postnatal aCMs caused rapid development of spontaneous AF due to loss of aCM identity⁹ and misexpression of Ca²⁺ handling genes^10,11. Here we show that CHD4 cooperates with TBX5 to maintain aCM enhancer accessibility and activate the expression of aCM-selective genes. Of 241 genes associated with increased AF risk by human genetics^38–40, 95 are regulated by TBX5 (26), CHD4 (27), or both (42). Although these studies have not implicated CHD4 and other NuRD complex members in human AF, they have implicated cardiac TFs that bind CHD4 (TBX5, GATA4, and NKX2–5).^34,41 Of the AF risk genes co-regulated by CHD4 and TBX5, nine (Tbx5, Mtss1, Myl4, Kcnq1, Nppa, Dpf3, Gja1, Myo18b, and Casz1) had the hallmarks of being stimulated by the TBX5-dependent CHD4 gene activation mechanism: co-occupied by TBX5-enhanced CHD4 binding, loss of accessibility with Chd4 inactivation, and downregulation with inactivation of either Tbx5 or Chd4. That TBX5-CHD4 activates Tbx5 itself suggests an intriguing feed-forward loop that reinforces aCM identity.

Patients initially develop paroxysmal AF and progress to sustained AF⁴². The mechanism by which episodes of AF remodel the atria’s electrical properties to facilitate and sustain AF are unclear, but likely involves epigenetic alterations to the atrial gene program. CHD4 is well positioned to be a central epigenetic factor that participates in the transcriptional reprogramming that accompanies AF progression. Additional studies are needed to test this intriguing hypothesis.

Limitations

While our study shows that TBX5 and CHD4 coordinately promote the accessibility of control DARs and activate aCM gene expression, additional studies are required to elucidate the mechanisms by which CHD4 and TBX5 directly and indirectly cooperate to promote aCM gene transcription. Although the data are consistent with CHD4 and TBX5 functioning as components of a molecular complex that promotes gene transcription, we cannot exclude alternative models in which TBX5 and CHD4 act sequentially. Another important question is whether CHD4 activates transcription as a component of the NuRD complex, which canonically represses transcription⁴³, or as a component of a novel regulatory complex. While TBX5 was the predominant cardiac TF bound at control DARs occupied by CHD4, CHD4 binds other cardiac TFs including GATA4 and NKX2–5²⁰, and it will be important to determine if CHD4 can likewise partner with these other TFs to activate transcription.

Clinical Perspective.

What Is New?

• TBX5, a transcription factor implicated in human AF by genome-wide association studies, regulates atrial cardiomyocyte gene transcription by recruiting a chromatin remodeling enzyme, CHD4, to control the accessibility of cis-regulatory elements.

• Preventing this mechanism through CHD4 deletion in atrial cardiomyocytes resulted in partially penetrant spontaneous AF and increased AF vulnerability in mice.

• In addition to its established role as a transcriptional repressor, we identified a novel activity of CHD4 as a transcriptional activator in aCMs, which promotes the binding of TBX5.

What are the Clinical Implications?

• CHD4 is a new component of the atrial gene regulatory network, which becomes dysregulated in AF.

• As an epigenetic remodeling enzyme, CHD4 may contribute to epigenetic memory, which is likely central to the inexorable progression of paroxysmal to permanent AF.

Nonstandard Abbreviations:

aCM

atrial cardiomyocyte

AAV9

adeno-associated virus serotype 9

AF

atrial fibrillation

AKO

atrial knockout

Brg1 ^AKO

Brg1 inactivation in postnatal aCMs

Chd4 ^AKO

atrial-selective Chd4 knockout

CUT&RUN

Cleavage Under Target & Release Using Nuclease

DEG

differentially expressed gene

DAR

differentially accessible region

ECG

electrocardiogram

GTEX

Genotype-Tissue Expression Project

H3K27ac

Histone H3 acetylated on Lysine 27

NuRD

Nucleosome Remodeling and Deacetylase complex

SDRR

standard deviation of the RR

snATACseq

single-nucleus Assay for Transposase-Accessible Chromatin using sequencing

snRNAseq

single-nucleus RNA sequencing

Tbx5 ^AKO

atrial-selective Tbx5 knockout

TF

transcription factor

TSS

transcription start site

UMAP

Uniform Manifold and Approximation Projection

vCM

ventricular cardiomyocyte

WNN

weighted nearest neighbors