A genomic link from heart failure to atrial fibrillation risk: FOG2 modulates a TBX5/GATA4-dependent atrial gene regulatory network

Abstract

Background:

The epidemiological link between heart failure (HF) and atrial fibrillation (AF) has been described as an epidemic, with up to half of HF patients progressing to AF. The pathophysiologic basis of AF in the context of HF is presumed to result from atrial remodeling. Upregulation of the transcription factor FOG2 (encoded by ZFPM2) is observed in human ventricles during heart failure and causes heart failure in mice.

Methods:

FOG2 expression was assessed in human atria. The effect of adult-specific FOG2 overexpression in the mouse heart was evaluated by whole animal electrophysiology, in-vivo organ electrophysiology, cellular electrophysiology, calcium flux, mouse genetic interactions, gene expression, and genomic function, including a novel approach for defining functional transcription factor interactions based on overlapping effects on enhancer non-coding transcription.

Results:

FOG2 is significantly upregulated in the human atria during heart failure. Adult, cardiomyocyte-specific FOG2 overexpression (OE) in mice caused primary spontaneous AF prior to the development of heart failure or atrial remodeling. FOG2-OE generated arrhythmia substrate and trigger in cardiomyocytes, including calcium cycling defects. We found that FOG2 repressed atrial gene expression promoted by TBX5. FOG2 bound a subset of GATA4 and TBX5 co-bound genomic locations, defining a shared atrial gene regulatory network. FOG2 repressed TBX5-dependent transcription from a subset of co-bound enhancers, including a conserved enhancer at the Atp2a2 locus. Atrial rhythm abnormalities in mice caused by Tbx5 haploinsufficiency were rescued by Fog2 haploinsufficiency.

Conclusion:

Transcriptional changes in the atria observed in human heart failure directly antagonize the atrial rhythm gene regulatory network, providing a genomic link between heart failure and atrial fibrillation risk independent of atrial remodeling.

Introduction

The epidemiological link between heart failure and atrial fibrillation (AF) is exceptionally strong, as up to half of all heart failure patients acquire AF¹. The co-occurrence of heart failure and AF causes an increased risk of stroke and all-cause mortality, making understanding the mechanisms underlying the epidemiologic link between HF and AF a high priority. Heart failure and AF share risk factors, and their frequent coexistence suggests the potential for shared pathophysiology^2–5. Genome-wide association studies (GWAS) for heart failure have identified some AF risk loci, suggesting some level of shared genetic risk². Existing paradigms for the increased risk of AF in the setting of heart failure rely on secondary effects of the physiological consequences of HF on atrial structure. There is a plausible relationship between several pathologic or physiologic consequences of heart failure, including structural atrial remodeling, mitral valve regurgitation, and neurohormonal responses, with increased AF risk^3–5. For example, increased atrial pressure promotes atrial remodeling, including fibrosis, which causes atrial rhythm instability and arrhythmia risk^6,7. While these secondary consequences of HF likely foment AF, the inciting mechanisms of AF in the context of HF remain poorly understood.

The recently discovered genetic basis underlying atrial fibrillation (AF) demonstrates a strong heritable component^8–15. GWAS have identified genetic variation associated with risk for AF in non-coding regions and enhancers at over 100 loci, including ion channels, gap junction proteins, and structural heart genes with known roles in cardiomyocyte electrophysiology^13,16. Furthermore, many transcription factors (TFs) with described roles in cardiac gene regulation, including TBX5, GATA4, and ZFPM2 (encoding FOG2), have also been implicated in human AF or PR interval prolongation (an AF risk factor) by human genetic studies^{11–13,17,18}. TBX5, a T-box transcription factor family member, is a significant driver of atrial rhythm homeostasis in the mature heart^19–22. Conditional deletion of Tbx5 from the adult mouse causes spontaneous atrial fibrillation, calcium handling deficits, and dysregulation of AF-associated genes that comprise a gene regulatory network for normal atrial cardiomyocyte electrophysiology ^22,23. TBX5 directly regulates enhancers for cardiac channel gene expression, including some that harbor genetic variation associated with AF risk at these loci, providing molecular evidence for a role in the genetic basis of AF ^{11–13,22,24,25}. GATA4 is a zinc finger transcription factor required for cardiac development and adult cardiac function^26–29. GATA4 and TBX5 interact synergistically during cardiac development^30–35 but antagonistically in atrial rhythm control³⁶. In the adult, the reduction of GATA4 dose rescues the effects of decreased TBX5 activity on downstream gene expression, cellular electrophysiology, and arrhythmia susceptibility³⁶. ZFPM2 encodes Friend of GATA-2 (FOG2), a transcriptional co-repressor required for normal cardiac and coronary artery development^37–40. FOG2 does not bind DNA directly but is recruited to genomic loci by GATA4 to repress gene expression by recruiting the nucleosome remodeling and histone deacetylase (NuRD) complex of chromatin remodelers^39–48. FOG2 expression is upregulated in heart failure in the ventricles of humans and mice^41,49. Additionally, transgenic FOG2 overexpression in mice during fetal development causes heart failure and decreased expression of the critical sarcoplasmic reticulum (SR) calcium ATPase, SERCA2, encoded by Atp2a2^41,49. These observations support a transcriptional model underlying AF risk, in which alterations in TF dose or in the enhancers they regulate affect gene expression of atrial rhythm effector genes such as ion channels, causing AF susceptibility.

We hypothesized that transcriptional responses to heart failure in the atria may be a primary cause of AF. Specifically, increased FOG2 overexpression in heart failure may contribute to AF susceptibility³⁶. We observed that FOG2 was significantly upregulated in the atria in human heart failure. We found that FOG2 overexpression in the adult mouse heart resulted in primary, spontaneous AF prior to any ventricular functional deficits. FOG2 overexpression altered atrial cardiomyocyte electrophysiology, including calcium handling.

Additionally, FOG2 overexpression caused misexpression of genes implicated in cardiac conduction and cardiomyocyte excitability that correlated closely with a Tbx5-dependent gene network for atrial rhythm control. FOG2 localized genome-wide at GATA4-occupied locations; however, FOG2 only repressed enhancers co-bound and activated by TBX5. Finally, we showed that Fog2 and Tbx5 interact genetically, as the atrial fibrillation propensity caused by Tbx5 haploinsufficiency was rescued by Fog2 haploinsufficiency. These observations identify a FOG2, TBX5, and GATA4 co-regulated atrial gene regulatory network for atrial rhythm control, providing insight into the transcriptional basis of AF susceptibility and a gene regulatory link between heart failure and AF risk.

Materials and Methods

Data availability

High-throughput data associated with this study, have been deposited in Gene Expression Omnibus (GSE198788), and Zenodo URL https://doi.org/10.5281/zenodo.8280395. Non-normalized values from luciferase reporter and ChIP-qPCR assays are available in supplemental tables 7, and 9, respectively.

Human heart CAGE data processing

Sequenced reads were processed as previously described⁵⁰. Briefly, samples from healthy and failing hearts (Supplementary Table 4) were sequenced in Illumina HiSeq2500. 25 M reads per sample were sequenced and checked for quality with FASTQC⁵¹, then trimmed for quality with fastx_trimmer, and adapters were removed with trimmomatic^52,53. Reads were mapped with BWA⁵⁴, and unmapped reads from BWA were aligned with HISAT2⁵⁵. The final set of usable reads was used to call CAGE peaks with CAGEr Bioconductor package (sequencingQualityThreshold = 10, 386 mappingQualityThreshold = 20, removeFirstG = TRUE, correctSystematicG = TRUE) to count 5’ prime end of the mapped CAGE reads at single base-pair resolution. Differential expression tests were performed with edgeR⁵⁶. Signal tracks were generated with deeptools and display TPM values, while boxplots (Figure 1B) were visualized with ggplot⁵⁷, using TPM values.

Figure 1. Atrial FOG2 overexpression is observed in human heart failure and causes atrial fibrillation in mice.

A. Genome browser shot (hg38) showing the location of the FOG2 promoter. Tracks show signal of normalized counts per million for combined replicates per tissue and per condition. The first track shows signal from healthy atrium, and the second track shows signal from failing atrium. The third track shows signal from healthy ventricles, while the fourth track shows signal from failing atrium. The fifth track shows RefSeq gene name.

B. Left panel shows boxplot comparing normalized tag counts from healthy atrium versus failing atrium. Statistics from differential expression test with edgeR. The right panel shows boxplot comparing normalized tag counts from healthy ventricle versus failing ventricle. Statistics from differential expression test with edgeR.

C. Schematic of the transgenes constructed to generate double transgenic (FOG2-OE) mice. The MHC-Tet transgenic line was derived from a transgenic expression cassette containing the αMHC promoter driving expression for the Tet-VP16 fusion protein as has been described¹¹⁸. A second transgene containing a modified αMHC promoter with the Tet operon driving expression of the FOG2 cDNA was used to generate the Tet-FOG2 transgenic line. In the presence of doxycycline, the Tet-VP16 fusion protein is unable to activate the expression of the Tet-FOG2 transgene. FOG2-OE mice were compared to litter-matched controls with neither the αMHC promoter nor the Tet-VP16 fusion protein (non-transgenic, NTG control).

D. qRT-PCR on RNA from NTG control or FOG2-OE atria using primers that recognize both endogenous and transgene-produced Zfpm2 (encoding FOG2) mRNA. Results were reported as the mean ± SEM for three independent RNA preparations. * indicates P = 0.001 from student’s t-test.

E. Western blot analysis using an anti-FOG2 or γ-tubulin antibody on atrial heart lysates from FOG2-OE or NTG control animals reared without doxycycline.

F. Densitometric quantitation was performed by normalizing FOG2 signal intensity to the signal of the γ-tubulin loading control. Results were reported as the mean ± SEM for three independent experiments. * indicates P = 0.001 from student’s t-test.

G. Representative ambulatory ECG recordings from FOG2-OE and NTG control mice at 4 weeks off doxycycline are shown. Scale bars = 100ms; n =3 animals/genotype.

H. Representative surface electrocardiograms (Lead II) and intracardiac electrograms (right atrium A EGM and right ventricle, V EGM) obtained from NTG control and FOG2-OE mice are shown. Tracings were obtained at 8 weeks off doxycycline; n=7 animals/genotype.

I. Percent of mice in normal sinus rhythm over time after discontinuation of doxycycline in NTG control (solid) and FOG2-OE (dashed) mice (n = 7-10 mice per genotype). Cardiac rhythm was examined under general anesthesia by surface electrocardiogram (ECG) and echocardiogram at 1, 4, 6, 8, 10, and 12 weeks post-doxycycline.

J. Time course of the development of LV dysfunction after discontinuation of doxycycline in FOG2-OE (dashed) vs. litter-matched NTG control (solid) mice, measured by percent fractional shortening. * indicates P = 0.05 from student’s t-test of NTG control vs. FOG2-OE at the specified time point, with P = 3.68x10⁻⁰⁸ from two-way ANOVA shown above.

K. Gross pathology of NTG control and FOG2-OE hearts from mice reared without doxycycline for 8 weeks post-weaning.

L. Histologic sections through hearts from NTG control and FOG2-OE animals reared without doxycycline, sectioned and stained with hematoxylin and eosin (H&E, top) or Masson’s Trichrome stain (bottom). Sections through FOG2-OE atria demonstrate thrombus formation (arrows) and significant fibrosis (blue staining in trichrome-stained sections). Magnification bars represent 12.5 mm.

M. Representative action potentials from isolated atrial myocytes of NTG control (control) and FOG2-OE (FOG2 overexpressing) mouse hearts using 1Hz current injection.

N. Action potentials demonstrate prolongation of the FOG2-OE atrial potential (red) in comparison to the NTG control action potential (black).

O-P. FOG2-OE atrial myocytes displayed increases in action potential duration at 50% (APD50) and 90% (APD90) recovery. * indicates P = 0.05 from student’s t-test.

Q. Representative tracings of calcium release in isolated atrial cardiomyocytes from isolated FOG2-OE and NTG control mouse hearts. Calcium was imaged with Fluo-4 dye, and myocytes were paced at 1 Hz. Representative abnormal afterdepolarizations were seen in FOG2-OE tracing, marked by arrows, which occurred in 14/35 FOG2-OE recordings vs. 0/16 in NTG control, P = 0.014 by two-tailed Fisher’s exact test.

R. Time to 50% decay of calcium transient spikes was measured across n > 5 animals per group (P = 0.000048 , by two-tailed student’s t-est).

S. Representative line scans showing calcium sparks in unpaced NTG control and FOG2-OE atrial myocytes. A spark propagating into an abnormal afterdepolarization was observed (bottom).

T. Quantitation of sparks frequency as sparks/100um/s across n > 5 animals per group. * indicates P = 0.001 by student’s t-test.

Generation of tetracycline-responsive FOG-2 transgenic mice

All animal studies were carried out in adherence to protocols approved by the University of Chicago Institutional Animal Care and Use Committee and in compliance with the Guide for the Care and Use of Laboratory Animals published by the U.S. National Institutes of Health. Transgenic mice harboring a modified gene encoded the Tet-repressor protein fused to the VP16 transcriptional activator under the control of a weakened α-MHC promoter (MHC-Tet mice). A second transgene was generated containing the cDNA sequence of murine FOG2 (Accession #NM_011766) downstream of the Tet-responsive modified mouse α-MHC promoter. Mice harboring this transgene (Tet-FOG2 mice) were generated by the University of Chicago Transgenic/ES Cell Technology Mouse Core Facility. Transgenic founder lines were expanded based on the transmissibility of the transgene to offspring as well as demonstrated overexpression of FOG2 when mice harboring both the FOG2 transgene and the MHC-Tet transgene were reared without doxycycline in their drinking water.

Histological analysis of tissues

Hearts were harvested from mice and placed into phosphate-buffered saline (PBS) to facilitate the removal of blood prior to fixation in formalin and subsequent embedding in paraffin. Five-micron thick sections were mounted on glass slides and stained with either hematoxylin and eosin or Masson’s trichrome to detect fibrosis. Images were captured on a Zeiss Axioskop compound microscope using a QImaging Retiga Exi digital camera.

Kaplan-Meier survival curve

Mice were reared without doxycycline for up to one year to allow maximal expression of FOG-2 in cardiac myocytes in animals harboring Tet-FOG2 and MHC-Tet transgenes (FOG2-OE). Survival curves were also plotted for animals reared without doxycycline that harbored no transgenes (NTG control), only the Tet-FOG2 transgene or the MHC-Tet transgene. The colony was monitored for animal deaths. Statistical analysis to determine the significance of variance between Kaplan-Meier survival curves was performed using R package ggsurvival.

Protein expression analysis

Western blot analysis was performed as previously described^58,59. Briefly, mouse hearts were excised from FOG2-OE and NTG control animals seven days after doxycycline cessation. Atria was separated from ventricles, crushed using a cold mortar and pestle, and lysed using NTEN buffer [150 mM NaCl, 20 mM Tris Base, pH 8.0, 1.0 mM EDTA, 0.5% IGEPAL CA-630 (Sigma, St. Louis, MO, USA)] containing complete mini-EDTA–free protease inhibitor cocktail tablets (Roche Diagnostics, Indianapolis, IN, USA). Left and right atria were analyzed together, as were left and right ventricles. Atrial and ventricular lysates were resolved by 8% SDS-PAGE, followed by transfer to a nitrocellulose membrane. The membrane was blocked in blocking buffer (10 mM Tris, pH 7.5, 140 mM NaCl, 0.05% Tween-20, 5% powdered milk) for 1 hour at room temperature, followed by incubation with a 1:1000 dilution of anti-FOG-2 rabbit polyclonal antibody (M-247, Santa Cruz) or a 1:500 dilution of mouse monoclonal anti-γ-tubulin antibody (T6557, Sigma) in blocking buffer for 12 hours at 4 °C. The membrane was washed with TBST (10 mM Tris, pH 7.5, 140 mM NaCl, 0.05% Tween-20), incubated for 1 hour at room temperature with a 1:2500 dilution of goat anti-rabbit or rabbit anti-mouse antibody conjugated to horseradish peroxidase. The blot was then washed extensively and developed using a commercially available kit (ECL-plus, GE Healthcare, Piscataway, NJ). Quantitation was performed by densitometry using a Molecular Dynamics STORM 860 Phosphoimager. The intensity of the FOG-2 signal was normalized to that of γ-tubulin to control for variations in protein loading.

Electrocardiography

Surface electrocardiography was performed on NTG control and FOG2-OE mice at 1, 4, 6, 8, 10, and 12 weeks following removal of doxycycline. After anesthesia with 2% inhaled isoflurane / 100% O₂, subcutaneous sterile stainless steel needle electrodes were placed on the limbs to generate ECG leads I, II, and III. Precordial electrodes were placed at V1 and V6 positions. ECG recordings were obtained using ADInstruments BioAmp and Powerlab systems for at least thirty seconds per lead. Recordings were then subjected to a custom automated signal-averaging program, and average P wave duration, PR interval, QRS duration, and QT interval were obtained for leads I, II, and III. QT interval was sampled from lead II and was corrected for heart rate using the standard Bezet’s formula QTc (ms) = QT / √(R-R) (sec). The T-wave end was determined by the return to the isoelectric baseline.

Transthoracic echocardiography

Echocardiographic analysis of heart structure and function was performed at the same time points as surface electrocardiography using techniques as previously described⁵⁹. Mice were anesthetized using inhaled isoflurane (~2%) delivered via nose cone. The left precordial fur was removed with a topical depilatory agent. Body temperature was maintained using a heated imaging platform. The heart was imaged using a VisualSonics Vevo 770 device (VisualSonics, Toronto, Ontario, Canada) using a 30 MHz high-frequency transducer. Two-dimensional images were recorded in approximately the parasternal long- and short-axis projections with guided M-mode recordings at the mid-ventricular level in both views. Left ventricular internal dimensions at diastole and systole (LVIDd and LVIDs, respectively) were measured in at least three beats from each projection and averaged, and fractional shortening was calculated. When mice were in atrial fibrillation, at least five beats were analyzed and averaged. Echocardiographic measurements were obtained from M-mode images at the mid-papillary muscle level in the parasternal short-axis view and B-mode images acquired in the parasternal long- and short-axis views. Conventional measurements of the left ventricle included: end-diastolic diameter (LVEDD), end-systolic diameter (LVESD), posterior wall thicknesses, fractional shortening, calculated LV mass using the ASE correction⁶⁰ as well as LA diameter using the maximum anterior to posterior dimension in the parasternal long axis view. Mitral valve pulse wave Doppler recordings were used to generate the E/A wave ratio to estimate diastolic function.

Ambulatory cardiac rhythm recording

Mice were implanted with subcutaneous telemetry transmitters (ETA-F10; DSI, MN, USA) under general anesthesia and sterile conditions as previously described¹⁹. Arrhythmia analysis was done using Ponemah Physiology Platform software (DSI) from 24-hour recordings. Recordings were made from three NTG control and three FOG2-OE mice four weeks off doxycycline while ambulatory for 7-10 days.

Mouse atrial myocyte isolation

Mouse atrial myocytes were isolated from the combined left and right atrial appendages of NTG control (wild type) and FOG2-OE (FOG-2 overexpressing mice) as follows: briefly, mice were treated intraperitoneally with 200 IU of heparin. The hearts were removed en bloc, washed in calcium-free Tyrode solution, and attached via the aorta to a peristaltic Langendorff apparatus. Retrograde perfusion was carried out at 37° C using Ca²⁺-free Tyrode solution containing Butanedione monoxime, glucose, and penicillin/streptomycin for 1-2 minutes, followed by a switch to digestion buffer, having a similar Tyrode solution plus 40μM CaCl₂, and Collagenase type II. Perfusion of digestion buffer was carried out at 8 mL/min for 10 minutes. Perfusate was collected in a sterile tube for the final 2 minutes of perfusion. The atria were removed from the heart, placed in a digestion buffer along with previously collected perfusate at 37 °C, and pipetted gently every 5 minutes with a Pasteur pipette until completely dissociated. This solution was then centrifuged to a pellet, digestion buffer was removed, and stop solution containing bovine serum albumin and 50μM CaCl₂ was added. Cells were then plated on laminin-coated glass coverslips and maintained in Tyrode solution with 50μM CaCl₂.

Mouse atrial myocyte action potential measurement

Myocytes were isolated from pooled left and right atrial appendages of NTG control and FOG2-OE mice at seven days off-doxycycline. Briefly, membrane potential was recorded using a ruptured patch current clamp at 37°C using an Axon Axopatch 200B amplifier. Pipettes (1 to 2 MΩ) were filled with the following (in mmol/L): 20 KCl, 100 K-glutamate, 10 HEPES, 5.03 MgCl₂, 10 NaCl, 5 TrisATP, and 0.3 LiGTP, pH 7.4. During recording, cells were perfused in normal Tyrode solution containing the following (in mmol/L): 150 NaCl, 5.4 KCl, 10 glucose, 5 HEPES, 1 MgCl₂, 1 CaCl₂, pH 7.4. Action potentials were triggered using 8 pA x 1 ms current clamp pulses applied at 0.5 Hz. Recordings were carried out and analyzed with ClampFit software (Molecular Devices, LLC, Sunnyvale, CA, USA).

Coding RNA-seq library preparation and data analysis

Libraries were prepared from RNA starting with 1 μg per sample and using the mRNA-seq Sample Prep Kit (Illumina) as per recommended instructions. After RiboZero purification and removing only ribosomal RNA, barcoded libraries were prepared according to Illumina’s instructions (2013) accompanying the TruSeq RNA Sample prep kit v2 (Part# RS-122-2001). Libraries were quantified using the Agilent Bio-analyzer (model 2100) and pooled in equimolar amounts. The pooled libraries were sequenced with stranded 50-bp single-end reads on the HiSeq2500 in Rapid Run Mode following the manufacturer’s protocols (2013).

RNA library preparation was performed as previously discussed⁶¹. Briefly, 22 M to 30 M reads were mapped to the mouse genome with STAR (v 2.5.3a)⁶². Reads mapped to the mitochondrial genome and with a phred score <30 were excluded. Counts were retrieved with HTseq (v.0.6.0) in union mode⁶³. Lastly, counts were analyzed for differential expression with R (3.4) package DEseq2⁶⁴. Significance was considered reached when FDR was less than 0.1, and absolute log₂ fold change was greater than 0.5

Non-coding RNA-seq library preparation

Total RNA was extracted by TRIzol Reagent (Invitrogen), followed by ribosomal and polyA depletion. After RiboZero purification and oligo-dT depletion, RNA barcoded libraries were prepared according to Illumina’s instructions (2013) accompanying the TruSeQ RNA Sample prep kit v2 (Part# RS-122-2001). Libraries were quantified using the Agilent Bio-analyzer (model 2100) and pooled in equimolar amounts. The pooled libraries were sequenced with 50-bp stranded single-end reads on the HiSEQ4000 in Rapid Run Mode following the manufacturer’s protocols (2013).

Non-coding RNA-seq data analysis

170-186 million high-quality reads (quality score >30) for each sample were obtained. Fastq files were aligned to UCSC genome build mm10 using STAR, and between 168 million and 174 million reads were successfully mapped. Transcript assembly was performed by Stringtie (version 1.3.3) and merged with the Stringtie ‘merge’ function⁶⁵. Counts were retrieved from the merged transcriptome for differential expression testing, performed as above. False discovery rate (FDR) was calculated after removing the coding-gene transcripts, ncRNA that overlapped coding genes in the same strand, and RNAs shorter than 100 base pairs. Significance was considered reached when FDR was less than 0.1, and absolute log₂ fold change was greater than 0.5.

GO enrichment & pathway analysis

We used Metascape⁶⁶ to look for enriched GO terms from biological processes and Ingenuity Pathway analysis to look for overrepresented signaling pathways and disease-related pathways (Tox Function)⁶⁷. We visualized results from these two programs with R library ggplot2⁵⁷.

ChIP-seq data processing

The Zfpm2^flbio allele was created by knocking the FLAG and bio epitope tags onto the C-terminus of Zfpm2 in ES cells. After the generation of mice, the Frt-neo-Frt resistance cassette was removed using a germline FLP deleter strain (JAX #016226). The mice were bred to homozygosity with Rosa26^BirA, which expresses the BirA enzyme that biotinylates the bio epitope tag. Homozygous mice survived normally and are available at MMRRC (#037509). FOG2 bioChIP-seq was performed as described⁶⁸. For each biological replicate, about 20 E12.5 Zfpm2^flbio/flbio;Rosa26^BirA/BirA (037509-JAX) heart ventricles were harvested and crosslinked in 1% formaldehyde in PBS for 10 minutes at room temperature. Chromatin was fragmented to 300-800 bp using a microtip sonicator (Qsonica, S-3000). Biotinylated FOG2 and bound chromatin were pulled down by incubation with streptavidin beads (ThermoFisher Scientific, Cat#11206D). Libraries were constructed using the KAPA HyperPrep ChIP-seq library preparation kit (Roche, Cat#07962347001). Sequencing (50 nt single end) was performed on an Illumina HiSeq 2000.

FOG2 bioChIP-seq in HL1 cells was performed as described⁶⁹. About 2 x 10⁷ HL1 cells were infected with rtTA-IRES-BirA and bio-tagged Fog2 adenoviruses at an MOI of 100 and cultured in a medium supplemented with 1 μg/ml doxycycline to induce the expression and biotinylation of Fog2^flbio. Cells were crosslinked with 1% formaldehyde for 10 minutes at room temperature and then harvested for bioChIP-seq 48 hours after doxycycline induction. BioChIP was performed as described above. BioChIP-seq libraries were constructed using the Illumina ChIP-seq library preparation kit. 20M 25 nt single-end reads were obtained for each sample.

GATA4, TBX5, and H3K27Ac Fastq files from previously generated ChIP-seq datasets were downloaded from Gene Expression Omnibus and processed identically as previously described⁷⁰. Reads from FOG2 ChIP-seq were processed the same: briefly, were aligned to UCSC mouse genome mm10 with Bowtie2 version 2.3.4 in end-to-end mode⁷¹. Mismatched reads, PCR duplicates, ENCODE blacklisted regions, and reads with quality <30 were removed with Samtools⁷² version 1.5. Peaks and fold enrichment tracks were generated with Macs2⁷³ version 2.11. TF ChIP-seq assays were combined with the HOMER⁷⁴ tool mergePeaks (4.3) to make a master list and visualized with ChIPpeakAnno⁷⁵. Known and de novo motif scanning was performed using HOMER⁷⁴ (4.3). Differential motif analysis was performed with gimmeMotif⁷⁶. We downloaded the processed data sets from the ENCODE^76,77 portal (https://www.encodeproject.org/), and identifiers have been added to Supplementary table 6. TF-binding peaks were intersected with ncRNAs using the Bioconductor package GenomicRanges,⁷⁸ allowing for a 1000 bp gap upstream the 5′ end of the RNA and 50 bp into the RNA. All regions were overlapped with the 5′ regions of the ncRNA background; the fold change distribution was plotted with ggplot2⁵⁷ per binding category.

LiftOver and conservation analysis

We used the liftOver application to map candidate enhancer elements from mouse to human genome hg38 as described⁷⁹. We used all the summits from whole heart ATAC-seq⁸⁰, as well as the subset that intersected with differentially transcribed enhancers. We then intersected mapped elements to the human left atrium ATAC-seq from ENCODE. From three available assays, we created a consensus list by combining all peaks with the HOMER tool mergePeaks v4.1⁷⁴ and keeping only those regions that contained at least 2 of 3 called peaks. We used R programs GenomicRanges⁷⁸ and VennDiagram⁸¹ to create and visualize the intersection, respectively. Conservation of regions was analyzed with precalculated phastCons 60-way vertebrate scores from UCSC⁸². These scores were then assigned to each mouse region tested using deeptools multiBigWigSummary⁸³. We used R programs ggbeeswarm and rstatix to create the visualization and perform a Wilcoxon Rank Sum test.^84,85

Cloning of candidate regulatory elements

Candidate regulatory elements were amplified from HEK293T or C57/B6 mouse genomic DNA. The sequence was verified and cloned into the pGL4.23 enhancer luciferase response vectors (Promega, Madison, WI, USA) with the minimal promoter.

In vitro luciferase response assays

HEK293T cells were co-transfected with the enhancer-luciferase construct (600ng) and appropriate combinations of Tbx5, Gata4, and Fog2 overexpression vectors (400ng of each) in the pcDNA3.1 backbone using FuGENE (Promega) as previously described¹⁹. The total DNA transfected was normalized with a blank vector when necessary. HL-1 cells were transfected with the enhancer-luciferase vector alone (600ng) or in combination with the Fog2 overexpression vector (800ng). Cells were cultured for 48 hours after transfection, then lysed and assayed using the Dual-Luciferase Reporter Assay system (Promega).

Chromatin immunoprecipitation

Chromatin immunoprecipitation was performed as previously, with minor modifications⁸⁶. Briefly, to prepare chromatin extract, HL1 cells, grown in a standard cell culture T-25 flask, were transfected with 4.5ug of pcDNA-Fog2 expression construct, cultured for 48 hours, and then harvested. Cells transfected with empty pcDNA3.1 vector (800ng) were used as a control. 2 x 10⁷ cells were crosslinked in 1% formaldehyde for 10 minutes at room temperature, with rocking. The reaction was then quenched with 125 mM glycine. The cross-linked cells were incubated in cell lysis buffer (50 mM Tris-HCl, pH 7.5; 140 mM NaCl; 1 mM EDTA, pH 8.0; 1% NP-40) with protease and phosphatase inhibitors for 10 minutes at 4 °C, with rocking. The cells were then sonicated in nuclear lysis buffer (50 mM Tris-HCl, pH 7.5; 2 mM EDTA, pH8.0; 0.5% sodium deoxycholate; 1% SDS) with protease and phosphatase inhibitors for 15 minutes at 4 °C. Next, chromatin extract was cleared by centrifugation at 14,000g, 4 °C for 10 min. For immunoprecipitation, the chromatin extract was incubated with 10μg of the anti-TBX5 antibody (Santa Cruz Biotechnology sc-17866; Lot #G1516) at 4 °C for >12 hours. Normal rabbit IgG was used as a negative IP control. The immunocomplexes were captured by Protein G-conjugated magnetic beads (Life Technologies, 1003D), washed in sequence by low salt buffer (20 mM Tris-HCl, pH 8.0; 500 mM NaCl; 2 mM EDTA, pH 8.0; 0.1% SDS; 1% Triton X-100), high salt buffer (20 mM Tris-HCl, pH 8.0; 150 mM NaCl; 2 mM EDTA, pH 8.0; 0.1% SDS; 1% Triton X-100), LiCL buffer (20 mM Tris-HCl, pH 8.0; 2 mM EDTA, pH 8.0; 0.5% sodium deoxycholate; 0.1% NP-40; 250 mMLiCL), and TE (10 mM Tris-HCl, pH 8.0; 1 mM EDTA, 0.01% Tween-20). The captured chromatin, and input samples, were eluted in ChIP Elution Buffer (100 mM NaHCO_3, 1% SDS) at 65 °C. After RNase and proteinase K treatment and reverse cross-linking, DNA was purified using a PCR cleanup kit (Qiagen, Valencia, CA, USA). To determine fold enrichment, qPCR was performed using input controls compared with DNA bound to immunoprecipitated proteins, using 1) primers specific to the sites of interest in the Ryr2 or Atp2a2 cis-regulatory elements, 2) as well as primers to a negative control (site not bound by TBX5 therefore, not expected to be enriched). The sequences of the primers used in the ChIP-PCR analysis have been shown in Supplemental Table 8. DNA concentration (N₀) after ChIP was calculated using R software, and statistical analysis was carried out as described in “Statistical analyses’ section. Data are presented as [N₀ChIP: N₀Input]_{pcDNA-Fog2_transfected cells (cells overexpressing Fog2)} divided by [N₀ChIP: N₀Input]_{empty_pcDNA_transfected cells (control cells).}

Statistical analyses

All values were summarized as mean ±SD unless specified. For all tests, significance was considered at P-value < 0.05 unless specified. Survival curve P-values were calculated with R package ggsurvival. We used the Mantel-Cox log-rank test to compare curves. For genomic experiments, we used the R package rstatix for all comparisons except for differential expression testing, which used the DESeq2 algorithm. DESeq2 uses Benjamini-Holchberg (FDR) for multiple testing correction. Correlation analyses were performed with the R base function cor.test using the Pearson correlation. Gene ontology performed by Metascape used hypergeometric distribution to calculate p-values, and Bonferroni correction for p-values⁶⁶. Ingenuity Pathway Analysis, and Tox function analysis uses Fisher’s Exact Test to calculate P-values, and Benjamini-Holchberg for multiple testing correction⁶⁷. When analyzing specific subsets within a dataset containing multiple groups, we conducted pairwise comparisons using the student-t-test test unless stated otherwise. In cases where it was suitable, entire groups were compared using either one-way or two-way ANOVA. We applied the Benjamini-Hochberg method to correct for multiple tests involving multiple groups. Group variance comparison was performed with Brown-Forsythe test. Random permutation tests were performed with R package regioneR where the P-value is the proportion of samples that have a test statistic larger than the observed data. Significance in figure panels indicated as **** , ***, **, * correspond to P < 0.0001, P < 0.001, P < 0.01, P<0.01, and P<0.05, respectively. The statistical test for each panel is also described in the figure legends. ChIP-qPCR was analyzed as previously shown⁸⁷. Briefly, cycle threshold values (C_t) were calculated by subtracting IP C_t value from the input C_t value (delta C_t) for anti-TBX5 antibody and IgG antibody, then fold enrichment is expressed as 2^{normalized delta Ct} where normalized C_t is the difference between delta C_t of antibody of interest and delta C_t mock IgG antibody.

Results

FOG2 is upregulated in the atria of failing human hearts

We interrogated FOG2 expression in a CAGE dataset from human atria and ventricular samples from functionally normal and failing human hearts.⁵⁰ FOG2 expression was compared between healthy (n=36) and failing (n=15) atria and healthy (n=34) and failing (n=18) ventricles (Figure 1A). FOG2 expression was significantly higher in the atria of failing versus normal hearts (log fold change = 0.59, FDR = 0.007; Figure 1B, left). FOG2 also trended higher in the ventricles of failing versus normal hearts (log fold change = 0.43, FDR = 0.07) (Figure 1B, right). These results are consistent with prior observations showing increased FOG2 expression in ischemic and non-ischemic hearts compared to non-failing hearts (Figure S1D and S1E)^41,49. These data indicated that heart failure causes significantly increased expression of FOG2 in the atria.

Cardiac overexpression of FOG2 causes atrial fibrillation

We examined the effect of cardiomyocyte-specific FOG2 overexpression on heart function in the adult mouse. Conditional overexpression of FOG2 was achieved with a bigenic Tet-off system, in which the removal of doxycycline activated a Tet-VP-16 fusion protein expressed from the cardiac-specific αMHC promoter, driving cardiomyocyte-specific overexpression of FOG2 (Figure 1C). Transgenic mice with both the Fog2 and Tet-VP-16 alleles (Double Transgenic, DTG; FOG2 overexpression, FOG-OE) were compared to littermate controls with neither transgenic allele (Non-Transgenic, NTG control; controls). At 3 weeks of age, mice were weaned and removed from doxycycline treatment. After 1 week off-doxycycline, atria of FOG2-OE animals had 6-fold elevated levels of Fog2 mRNA (Figure 1D) and 3-fold elevated levels of FOG2 protein (Figure 1E–F) compared to NTG controls (both P < 0.001). Significant elevations of Fog2 transcript and FOG2 protein were also observed in the ventricles of FOG2-OE mice (5-fold and 2-fold, respectively, both P < 0.01 Figure S1). This degree of FOG2-OE in the atria is similar to that observed in humans with heart failure^41,49. FOG2-OE animals were at a significant survival disadvantage (P < 0.0006, log-rank test) relative to NTG control and single transgenic mice (animals with only the MHC-FOG or the MHC-Tet-VP-16 transgenes), with approximately 50% of mice dying after 8 months (Figure S2).

FOG2-OE resulted in progressive atrial arrhythmias. The effect of cardiac FOG2-OE on cardiac rhythm and function was assessed in FOG2-OE vs. NTG control mice over a 16-week time course. Irregularly irregular QRS complex patterns and absence of P-waves were observed beginning three weeks after FOG2-OE (Figure 1G–H), when there was no change in LV systolic function as indicated by fractional shortening (FS) (Figure 1J). The frequency and severity of atrial arrhythmias progressively increased over time. After four weeks of FOG2-OE, 37% of FOG2-OE mice demonstrated atrial arrhythmias, whereas, after 6 weeks, as many as 62% displayed atrial arrhythmias (Figure 1I). The mice remaining in sinus rhythm at 6 weeks of FOG2-OE showed significant P-wave and PR-interval prolongation (Supplemental Table 1). Intracardiac electrogram recordings at this time point showed a lengthening of the AH Interval, indicative of slowed atrial conduction (Supplemental Table 2). In addition, FOG2-OE mice were susceptible to AF induction by pacing at 6 weeks (P = 0.05) (Supplemental Table 2). No other alterations in cardiac electrophysiology were identified between NTG control and FOG2-OE mice by evaluation of ECG and intracardiac EP parameters (Supplemental Tables 1–2). By 10 weeks, all FOG2-OE mice had converted from normal sinus rhythm to sustained AF (Figure 1I, dashed), demonstrated by an irregularly irregular pattern of atrial activity by surface ECG, ambulatory telemetry (Figure 1G–H) and intracardiac electrograms (Figure 1H). All NTG control mice remained in normal sinus rhythm for the duration of the study (Figure 1H, solid). We conclude that FOG2-OE in cardiomyocytes of the adult heart causes spontaneous, sustained AF.

Atrial fibrillation occurs prior to atrial remodeling and ventricular dysfunction in FOG2 overexpression mice

Atrial fibrillation preceded other functional abnormalities in FOG2-OE mice. LV fractional shortening, LV diastolic internal diameter, and LA diameter were indistinguishable between NTG control and FOG2-OE mice with FOG2-OE at 6 weeks when 60% of FOG2-OE mice displayed AF (Figure 1I, Supplemental Table 3). Longitudinal analysis revealed that AF onset always preceded LV dysfunction and atrial enlargement in individual mice (Figure S3A–B, representative M-mode echocardiograms Figure S3C). No significant differences in LV mass between NTG control and FOG2-OE animals were observed over the course of the study (Supplemental Table 3, Figure S3D), and no significant left atrial enlargement or dilation was observed until after 8 weeks of FOG2-OE (Figure 1K, S3E, Table 3). Histopathology of hearts at 8 weeks of FOG2-OE revealed extensive fibrosis and intra-atrial thrombi in the enlarged atria (Figure 1K–L). A complete evaluation of functional parameters by echocardiography showed no other changes between NTG control and FOG2-OE mice for the duration of the study (Supplemental Table 3). We conclude that atrial arrhythmias preceded the onset of functional deficits or pathologic remodeling in FOG2-OE mice, indicating that atrial arrhythmias are a primary effect of FOG2-OE.

FOG2 overexpression causes disrupted calcium handling

We hypothesized that FOG2-OE disrupted atrial cardiomyocyte electrophysiology leading to atrial arrhythmias. We compared cellular electrophysiology in atrial cardiomyocytes after 7 days of FOG2-OE versus NTG control mice. At this early time point, all hearts were in normal sinus rhythm. We investigated arrhythmia substrates, including conduction speed and action potential duration, and arrhythmia triggers, including early after-depolarizations and delayed after-depolarizations (EADs, DADs). FOG2-OE mice showed a significantly prolonged action potential duration (APD) compared to NTG control atrial cardiomyocytes at both 50% and 90% repolarization (P < 1x10⁻⁴ for APD50 and P < 9x10⁻⁴ for APD90, respectively, vs NTG control) (Figure 1M–P). A significant decrease in maximum systolic potential but no change in resting diastolic potential was observed after 7 days of FOG2-OE (Figure S4). Furthermore, EADs and DADs were frequently observed in FOG2-OE atrial cardiomyocytes but never in NTG control littermate controls (14/35 in FOG2-OE vs. 0/16 in NTG control, P = 0.014Figure 1Q, P = 0.000048 1R).

Given the association between inappropriate depolarizations and abnormal cardiomyocyte Ca²⁺ handling in AF¹⁰, we hypothesized that FOG2-OE altered atrial calcium fluxes. FOG2-OE mice showed faster calcium decay (P = 3.3x10⁻⁴; Figure 1S) and increased spark frequency (P = 3.4x10⁻⁴; Figure 1T), indicative of altered SR calcium handling. These abnormal calcium sparks can underlie abnormal depolarizations, suggesting that altered calcium handling contributed to the mechanism of AF pathogenesis in FOG2-OE mice.

FOG2 overexpression disrupts the expression of a network of atrial rhythm genes

Given that FOG2 is a transcriptional repressor, we investigated the atrial gene expression alterations caused by FOG2-OE. We performed transcriptional profiling on the left atrium from FOG2-OE versus control mice (FOG2-OE n=4, NTG control n=3) by RNA-seq at 7 days of FOG2-OE, prior to the onset of cardiac arrhythmias, dysfunction, or remodeling. A FOG2-OE-specific effect on gene expression was observed by principal component analysis, in which PC1 corresponded to genotype and explained 51% of the variance between samples (Fig S5A–B). FOG2-OE resulted in 1020 significantly dysregulated genes in the adult atrium, including 575 down-regulated and 445 up-regulated genes (|log₂FC| > 0.5; FDR < 0.1) (Figure 2A and Figure S5B). IPA analysis on dysregulated genes identified cardiac arrhythmia terms. Specifically, “Cardiac Arrhythmia” was the most enriched Tox term containing a union of 48 genes over 83 terms (red, Figure 2A; Benjamini–Hochberg corrected p-value: 4.31x10⁻¹⁴ to 2.47x10⁻¹¹)⁶⁷. The top 4 terms by p-value were “Arrhythmia” (P = 3.97x10⁻¹⁴), “Supraventricular arrhythmia” (P = 1.36x10⁻¹¹), “Atrial fibrillation” (P = 2.48x10⁻¹⁰), and “Cardiac fibrillation” (P = 1.12x10⁻⁹) (Figure 2B). Gene Ontology (GO) analysis identified pathways related to cardiac electrophysiology, including “cardiac conduction,” “cardiac contraction,” and “cation transport” (Figure S5C). We validated disrupted expression of genes critical to atrial calcium handling identified by qPCR in independent samples (Figure S5D) and compared the expression of various ion channels that have been disrupted after FOG2-OE (Figure S5E). We conclude that FOG2-OE caused altered left atrial gene expression indicative of cardiac arrhythmias several weeks before the onset of an overt arrhythmic phenotype (Figure 2B).

Figure 2. FOG2 overexpression disrupts a network of Tbx5-dependent atrial rhythm control genes.

A. Volcano plot of FOG2 mRNA-seq genes. Downregulated genes (negative log₁₀ false discovery rate greater than 1 and absolute log₂ fold change less than than −0.5) in maroon, and upregulated genes (negative log₁₀ false discovery rate greater than 1 and absolute log₂ fold change greater than 0.5) in blue. Dysregulated ion channel genes are highlighted in green. Significance cut-offs have been marked with grey dashed lines.

B. Bar graph of FOG2 dysregulated genes associated with IPA Cardiac Arrhythmia Tox functions. The horizontal axis shows negative log₁₀ of Benajamini-Holchberg adjusted p-value from Fisher’s exact test.

C. Venn diagram of Fog2, Gata4, and Tbx5 dysregulated genes. Shared genes in the Venn diagram include all those determined to be significant by adjusted p-value and fold change increase and decrease.

D. Scatter plot with fitted lines showing the correlation of log₂ fold change between Fog2 vs. Tbx5 and Fog2 vs. Gata4. Fog2 genes intersected with Tbx5 genes, shown in purple, while Fog2 genes intersected with Gata4, shown in brown. Correlation values from Pearson’s test with comparative p-value from Fisher’s z-transformation.

E. Dot plot showing the expression values in scaled CPMs, per cell type of ZFPM2, TBX5, and GATA4.

F. Enriched ingenuity canonical signaling pathways for Fog2 vs. Gata4 and Fog2 vs. Tbx5 dysregulated mRNA-seq genes. Heatmap values are determined from negative log₁₀ of Benajamini-Holchberg adjusted p-value in red. Column names on the bottom show the source experiment, while row names show the signaling pathway name.

G. Heatmap of −log₁₀ P-values from enriched biological process terms for genes dysregulated in Fog2, Gata4, and Tbx5 mRNA-seq experiments. Column names on the bottom show the source of the experiment, while rows show biological process terms from GO, REACTOME, and KEGG.

FOG2-OE-dependent gene expression overlapped with Tbx5-dependent gene expression

We compared the transcriptional consequences of FOG2-OE with that of Gata4²⁹, and Tbx5²² loss of function (LOF) in the murine left atrium, given the known physical interaction between FOG2 and GATA4 and between the NuRD complex, of which FOG2 is a member, and TBX5. Comparison of FOG2-OE-, Gata4- and Tbx5-dependent transcripts identified significant overlap between all three datasets (Figure 2C), with stronger overlap of dysregulated transcripts between Tbx5-LOF and FOG2-OE (OR=2.5, P = 4x10⁻⁵⁷) than Gata4-LOF and FOG2-OE (OR=1.6, P = 2x10⁻⁰⁹). Although GATA4 is a major FOG2 binding partner in the heart, we observed a modest correlation between Fog2-OE and Gata4-LOF-dependent LA gene expression (R=0.15; Figure 2D, brown). In contrast, we observed a significantly higher correlation between FOG2-OE- and Tbx5 LOF-dependent gene expression (R=0.41; Figure 2D, purple; Fishers’ r to z P < 0.00001). We performed ingenuity pathway analysis (IPA) on genes down-regulated by the loss of Gata4, Tbx5, and OE of FOG2. We found an enrichment for “Calcium Signaling” across the three experiments but more so on the group of genes down-regulated by Tbx5 and FOG2-OE relative to those genes down-regulated by Gata4 LOF (Figure 2G). GO terms on genes co-regulated by all three TFs included “heart development” and “heart contraction” (Figure 2H). Interestingly, only FOG2-OE-dependent and Tbx5-dependent genes shared GO terms for cardiac conduction and rhythm terms, including regulation of the action potential (adjusted P = 6.16x10⁻⁰⁶; Figure 2H) and calcium signaling pathways (adjusted P = 5.10x10⁻⁰⁵; Figure 2H). This suggested that FOG2, GATA4, and TBX5 coordinately regulated cardiomyocyte gene expression. We therefore assessed the cell-type specificity of Fog2, Tbx5, and Gata4 expression in cell types from the human heart using a large-scale single-cell human heart transcriptional profiling dataset⁸⁸. We observed that ZFPM2 is expressed predominantly in muscle cells, including cardiomyocytes, vascular smooth muscle, and fibroblasts. Interestingly, ZFPM2 was only co-expressed with GATA4 and TBX5 simultaneously at high levels in atrial cardiomyocytes (Figure 2E). Therefore, the intersection of FOG2, GATA4, and TBX5 appears to be atrial cardiomyocyte-specific in the heart. Overall, this analysis suggested that FOG2 and TBX5 oppositely modulated atrial gene expression for atrial rhythm homeostasis.

FOG2 localizes to cardiac enhancers at loci with FOG2-OE-dependent gene expression

We interrogated FOG2 genomic occupancy to identify locations of potential FOG2 activity as a transcription factor. Although FOG2 does not directly interact with DNA, it binds chromatin through GATA-family TFs, including GATA4^39,89. We performed ChIP-seq to identify genomic locations of FOG2 association. ChIP-seq was performed in fetal mouse heart tissue in which the endogenous Zfpm2 locus has been modified so that it expresses FOG2 with a C-terminal AviTag, which is selectively biotinylated by E. coli biotin ligase, expressed from the Rosa26 locus. Biotin-mediated ChIP-seq yielded 6166 regions with significant FOG2 occupancy genome-wide reproduced in biological duplicates. We observed FOG2 binding at known FOG2-OE-dependent loci. For example, FOG2 has been previously reported to regulate expression of SERCA2 (Atp2a2), and we observed reproducible FOG2 binding at a known Atp2a2 enhancer⁴¹ (Figure 3A). FOG2 binding was enriched at gene promoters (one-way ANOVA P = 1.79x10⁻³³) that were down-regulated by FOG2-OE to a greater degree than at unaffected gene promoters (P = 8.69x10⁻¹³) or up-regulated gene promoters (P = 6.77X10⁻⁰⁹), consistent with FOG2’s known role as a transcriptional repressor (Figure 3B). We noted that FOG2 binding occurs predominantly at regions distal to gene TSSs at approximately 12,000bp (Figure S6A), suggesting a gene regulatory role at enhancers. We therefore performed IPA analysis on the FOG2-OE-dependent genes nearest to a FOG2 binding event (within 200 kb of TSS) and observed enrichment for Cardiac Arrhythmia functions from IPA Tox lists. (Figure 3C). De novo motif analysis at locations of FOG2 binding (±100bp from the called summit) identified strong enrichment for the GATA motif, a reassuring observation given that FOG2 is known to be recruited to chromatin by its well-characterized physical interaction with GATA4/5/6 (P = 1x10⁻⁴⁷⁵, Figure 2D first row, and reverse complement P = 1x10⁻²⁵⁸ third row). De novo motifs matching cardiac TFs TEAD and T-BOX were also significantly enriched, as well as their known consensus sequences. (Figure 3D, Supplementary table 5)^69,90–94. This observation suggested that FOG2 directly regulates the expression of genes involved in atrial rhythm.

Figure 3. FOG2 binds to a subset of GATA4 and TBX5 co-bound genomic locations.

A. Genome browser shot of FOG2 ChIP-seq (2 replicates) at the Atp2a2 locus. Replicates 1 and 2 at the first two tracks and input control and combined replicate fold enrichment are shown in the bottom two tracks.

B. Average signal in reads-per-million over input from FOG2 ChIP-seq at the transcription start site of FOG2-OE-dependent genes (left panel). Top significance bar indicates significance from one-way ANOVA test, and individual comparisons using Tukey Honest Significant Differences test. The red line shows the signal at FOG2-repressed genes, the blue line at FOG2-upregulated genes, and the grey are unchanged genes. The right panel shows the same as boxplots with results from the Tukey Honest Significant Differences test comparing FOG2-repressed genes (red) vs. unchanged genes (grey) and FOG2-upregulated genes (blue).

C. Bar graph of genes nearest to FOG2 ChIP-seq associated with IPA Cardiac Arrhythmia Tox functions. The horizontal axis shows negative log₁₀ of Benajamini-Holchberg adjusted p-value from Fisher’s exact test.

D. De novo motif discovery from all FOG2-bound regions. The first column shows an enriched novel position weight matrix, the second column shows p-values from hypergeometric test, and the third column shows the best match to known motifs.

E. Heatmap of fold enrichment signal for fetal FOG2 (F), GATA4 (G), TBX5 (T), and H3K27AC ChIP-seq experiments at TF-bound regions grouped as FGT (dark blue), GT, (teal), FG (orange), FT (pink), G (yellow), (light blue), and F (red).

F. Heatmap of −log₁₀ P-values from hypergeometric test of enriched GO terms and pathways. Genes used were those in the union set that lost expression in the Gata4, Tbx5, and Fog2 mRNA-seq analyses and were closest to a TF binding event. Column names show the TF binding combination that was nearby the dysregulated gene which contributed to the GO terms.

G. Violin plot of fold enrichment signal from H3K27AC ChIP-seq experiments at TF-bound regions for fetal FOG2 (F), GATA4 (G), TBX5 (T), and their groups. Significance from Wilcoxon-Mann-Whitney test.

FOG2 localized to genomic locations of GATA4 co-occupancy to regulate TBX5-dependent gene expression

To assess the possibility that FOG2, GATA4, and TBX5 co-regulate enhancers for cardiac gene expression, we intersected FOG2 binding locations from ChIP-seq with published GATA4 and TBX5 ChIP-seq datasets from E12.5 mouse heart⁷⁰ (Figure 3E). We identified a union set of 61,148 genomic locations for FOG2 (F), GATA4 (G), and/or TBX5 (T) binding as defined by greater than 2-fold enrichment over the background. We note the caveat of using assays sourced from fetal mouse tissues to explore a phenotype observed in the adult mice, so we also performed ChIP-seq of FOG2 in HL1 cells. While technically limited, we found that the FOG2-bound sites in HL1s intersected with those bound in-vivo (Figure S6B). Furthermore, we note centrally enriched FOG2 binding signal in fetal tissue at sites defined by binding in HL1 cells (Figure S6C). We compared the activity of FOG2-bound regions from fetal heart and HL-1 atrial cardiomyocyte cell line with the heart enhancer landscape defined by H3K27Ac binding at E11.5, E14.5, and P56. FOG2-bound regions in fetal hearts and HL-1 cells show consistent H3K27Ac signal across time (Figure S6C right). To enrich for elements from the mature heart, we subset the combinatorial consensus FOG2-, GATA4-, and TBX5-co-bound regions by accessible chromatin locations from the adult mouse whole heart defined by ATAC-seq⁸⁰ (Figure 3E). We found that the vast majority of FOG2 binding sites overlapped with GATA4 binding sites (21,849/26,159), consistent with the described capacity of GATA4 to recruit FOG2 to GATA4-bound genomic locations^31,33,35,95. TBX5 and GATA4 physical interactions are also well described^31,33,35,95, and over 70% of TBX5 binding sites coincided with GATA4 binding sites (30,372/42,179) (Figure 3E). Interestingly, TBX5 / GATA4 co-bound sites were split almost equally into FOG2-bound (14,873/30,372) and FOG2-unbound (15,499/30,372) locations (Figure 3D).

We sought to understand whether the distinct bins of FOG2, GATA4, and TBX5 binding revealed distinct subnetworks of gene regulation. Differential motif analysis of the sites bound or co-bound between each bin revealed enrichment of sites enriched for the GATA, TBX, and TEAD motif. Notably, however, we found the motif most enriched in the regions bound by all three TFs (FGT) was T-BOX, suggesting regulation of TBX5 targets by FOG2 (Figure S7B). We asked if specific biological processes were direct FOG2, GATA4, and TBX5 targets, linking Fog2-OE-dependent, Gata4-LOF-dependent, and Tbx5-LOF-dependent genes to cis-regulatory elements bound by FOG2, GATA4, and TBX5 within 200kb. We found that dysregulated genes nearest to TBX5 binding sites, with or without GATA4 or FOG2 binding (T-, GT- and FGT-bound elements), were significantly overrepresented for general cardiac terms, including muscle system process heart contraction (Figure 3F). Dysregulated genes affiliated with TBX5 and GATA4 binding without FOG2 were overrepresented for metabolic terms (Figure 3F). In contrast, genes nearest to FOG2, TBX5, and GATA4 co-bound sites were enriched for cardiac conduction terms, including cardiac conduction, potassium channel activity, and His-to-Purkinje communication (Figure 3F). These data indicated that FOG2 was recruited by GATA4 to genomic locations that included a specific subset of TBX5 / GATA4 co-bound enhancers that mediate the expression of genes specifically crucial to cardiac rhythm.

We examined the relationship of FOG2, TBX5, and/or GATA4 localization with the activity of cardiac regulatory elements by comparing their binding locations with H3K27ac levels in adult wild-type whole hearts (Figure 3E)⁷⁰. TBX5 binding, with or without GATA4 co-binding, globally identified locations with strong H3K27ac levels (Figure 3E, G). Regions of GATA4 binding alone, FOG2 binding alone, and GATA4/FOG2 co-binding all showed modest enrichment for H3K27ac (Figure 3E). Notably, regions that bound FOG2, TBX5, and GATA4 (FGT; 16,892 sites) showed stronger signal from H3K27ac than regions that bound FOG2 and GATA4 without TBX5. (P = 6.8x10⁻²⁸ Figure 3E, G). We evaluated the impact of FOG2 binding at TBX5-GATA4 co-bound enhancers. We found that these locations had significantly lower H3K27ac levels than TBX5-GATA4 co-bound or TBX5 singly bound enhancers (P = 3.3x10⁻¹³, = 3.3x10⁻³³, respectively Figure 3G). These observations suggested that FOG2 localized with GATA4 and TBX5 at more active cardiomyocyte enhancers than those where FOG2 and GATA4 bound without TBX5, and further suggests that FOG2 modulates transcriptional control at active regulatory elements.

Given that FOG2 is a transcriptional repressor and its binding pattern at sites marked for transcription, we explored whether it bound chromatin marked by other modifications in various contexts. From ENCODE, we retrieved multiple sets of regions that mark active or repressive transcription as defined by ChIP-seq from various histone modifications. Specifically, we selected: H3K27ac, H3K27me3, H3K36me3, H3K4me1, H3K4me2, H3K4me3, and H3K9me3 from multiple tissues (Supplemental table 6). All these genomic intervals from multiple assays and tissues were grouped as a single index with GIGGLE⁹⁶. We then performed a large-scale comparison of FOG2-bound regions against our index that revealed a specific enrichment for regions marked for active transcription across various tissues (Figure S7A). Predictably, the heart was the most enriched tissue, along with the lung, consistent with previous studies implicating FOG2 in lung development⁹⁷.

We assessed FOG2 genomic localization with respect to other known cardiac TFs. ChIP-seq suggested that FOG2 interacts with GATA and other members of the mature cardiomyocyte TF kernel, including T-box TFs. Given this enrichment, we asked if the regions bound by FOG2 were co-bound by other cardiac TFs, defined by ChIP-seq. We generated a consensus set of binding sites from TF ChIP-seq experiments, including TEAD, MEF2C, MEF2A, NKX2-5, and SRF⁷⁰. FOG2-bound regions significantly intersected with this consensus set of regions (Figure S6D, P = 0.0009). Lastly, we asked if each TF would demonstrate fold enrichment signal across sites of FOG2 binding. We found that all TFs had centrally defined FOG2 signals, with GATA4, TBX5, and TEAD being the strongest and SRF being the weakest (Figure S6D). These observations suggested that FOG2 functions in the context of the broader cardiomyocyte transcriptional kernel to regulate cardiomyocyte gene expression.

FOG2-OE-dependent non-coding transcriptional profiling identifies FOG2-OE-dependent enhancers

Our findings indicated that although FOG2 has been described as a transcriptional repressor^39,40, we demonstrate its affiliation with highly active enhancers based on chromatin marks. We hypothesized that FOG2 buffers highly active enhancers of cardiac rhythm genes to prevent the untoward consequences of their overexpression. We, therefore, sought to directly interrogate the impact of FOG2-OE on its dependent enhancer set and define the left atrial FOG2-OE-dependent gene regulatory network. Enhancer transcription is strongly correlated with enhancer activity, and differential enhancer transcription represents a quantitative metric of relative changes in enhancer activity^23,98,99. We have previously applied TF-dependent non-coding RNA (ncRNA) abundance to quantitatively assess TF-dependent enhancer function^23,61. We applied FOG2-OE-dependent ncRNA profiling to define FOG2-OE-dependent cis-regulatory elements. We performed FOG2-OE-dependent ncRNA transcriptional profiling on the left atrium at 7 days status post-induction (FOG2-OE), when coding transcriptional profiling was performed and prior to arrhythmia onset, compared to control (NTG control) mice (n=3 each; Figure 4A, Figure S8A). De novo transcript assembly identified 25,529 ncRNAs, of which 2,203 were significantly dysregulated by FOG2 overexpression (log₂FC| > 0.5, FDR < 0.1; Figure 4A, B, S8A). Principal component analysis of ncRNA abundance in NTG control versus FOG2-OE atria demonstrated a strong FOG2-OE-dependent effect, in which the first principal component corresponded to genotype and explained 50% of the variance between samples (Figure S8B). To enrich for ncRNAs associated with FOG2-OE-dependent enhancers, we focused on differentially expressed ncRNAs transcribed from accessible chromatin regions. The putative promoters (1000 bp upstream of the TSS) of 369 out of 1,223 FOG2-dysregulated ncRNAs intersected with open chromatin defined by ATAC-seq from adult mouse heart⁸⁰ and were located within 200 kb of a FOG2-OE-dependent coding transcript. We further found that these differentially transcribed enhancers were enriched for ATAC-seq and H3K27ac signal relative to the background set of ncRNAs (P = 2.2x10⁻¹⁶ and P = 1 x10⁻⁹, respectively; both student-t-test Figure 4E, 4F). Of this subset of FOG2-OE-dependent ncRNA transcripts, 86 were upregulated (FOG2-OE-activated), and 283 were downregulated (FOG2-OE-repressed). The top GO terms for FOG2-OE-dependent genes nearest to FOG2-OE-dependent ncRNAs included terms specific for cardiac rhythm, including “cation membrane transport” and “Arrhythmogenic right ventricular cardiomyopathy” (Figure 4C). De novo motif analysis on enhancers defined by FOG2-OE-dependent ncRNAs demonstrated overrepresentation for Tgif2/TBX5, MEF2, and GATA motifs (P = 1 x10⁻²⁵, 1 x10⁻²² and 1 x10⁻¹³ respectively Figure 4D). These analyses indicated that FOG2 altered enhancer function at cardiac rhythm loci important for cardiac conduction and rhythm control.

Figure 4. FOG2 modulates enhancer transcription at GATA4 and TBX5 co-bound genomic locations.

A. Volcano plot (left) and heatmap (3 FOG2-OE and FOG-WT replicates, right) of Fog2 ncRNA-seq. Dysregulated genes (negative log₁₀ false discovery rate greater than 1 and absolute log₂ fold change greater than 0.5).

B. Heatmap of dysregulated ncRNA expression showing samples by condition and scaled by row. Hierarchical clustering was performed column-wise

C. Bar graph showing enriched biological process terms for nearest Fog2-dysregulated genes to a Fog2 dysregulated ncRNA. The horizontal axis shows the negative log₁₀ of the Benajamini-Holchberg adjusted p-value from hypergeometric test.

D. De novo motif enrichment analysis from HOMER of open-chromatin regions that intersected with dysregulated Fog2 ncRNA. The first column shows the motif position weight matrix, the second column shows the P-value from hypergeometric test, and the third column shows the best match to the motif.

E. Boxplot from metagene analysis of open-chromatin signal from ATAC-seq reads per million at Fog2-dysregulated ncRNAs. The grey boxplot shows regions from background regions, and the brown boxplot shows signals at regions from dysregulated ncRNAs. Significance from student’s t-test.

F. Boxplot from metagene analysis of active enhancer signal from H3K27ac ChIP-seq reads per million at Fog2-dysregulated ncRNAs. The grey boxplot shows regions from background regions, and the brown boxplot shows signals at regions from dysregulated ncRNAs. Significance student’s from t-test.

G. Heatmap showing the cluster analysis of shared Fog2-dysregulated ncRNA vs. TBX5-dysregulated ncRNAs.

H. Scatter plot of log₂ fold change values from Fog2 ncRNA-seq vs. Tbx5 ncRNA-seq, with Pearson product-moment correlation. The vertical axis shows fold change from ncRNAs dysregulated from Tbx5 knockout, while the horizontal axis shows fold change from ncRNAs dysregulated from Fog2 overexpression. ncRNAs that intersected with open chromatin are highlighted in blue.

I. Bar graph showing enriched biological process terms for Fog2-dysregulated genes nearby a Fog2-dysregulated ncRNA intersected with a Tbx5-dysregulated ncRNA. The horizontal axis shows the negative log₁₀ of the Benjamini-Holchberg adjusted P-value from hypergeometric test.

J. Violin plot with an inner boxplot of ncRNA log2 fold change values for each TF-binding combination. Blue violin plots show the distribution of Tbx5 dysregulation values, while mean lines are highlighted in red with zero lines marked by a dashed line.

K. Violin plot with an inner boxplot of ncRNA log2 fold change values for each TF-binding combination. Red violin plots show fold change distribution for ncRNA from FOG2 overexpression. Mean lines per category are highlighted in red, with zero lines marked by a dashed line.

FOG2 and TBX5 co-regulate a shared atrial rhythm gene regulatory network

Based on their highly correlated effect on atrial gene expression, we hypothesized that FOG2 and TBX5 coordinately controlled a common set of enhancers at critical atrial rhythm loci, defining a co-TF-regulated gene regulatory network (Figure 3D, 3F). We defined candidate TBX5 / FOG2 co-modulated enhancers by the overlap of FOG2-OE- and Tbx5-dependent ncRNAs in the left atrium at accessible genomic locations ²³. We found a significant overlap between FOG-OE- and Tbx5-dependent ncRNAs (412, OR = 5.16, P = 2.2 x10⁻¹⁶⁾; we further identified 154 candidate co-regulated enhancers that demonstrated Tbx5- and FOG2-OE-dependent ncRNA dysregulation and were associated with a local TBX5- and FOG2-OE-dependent gene expression (P = 2.2 x10⁻¹⁶, OR = 2.820257, Figure S8D). We evaluated the coordinate effects of TBX5 and FOG2 on enhancer function. Hierarchical cluster showed that FOG2-OE samples clustered with Tbx5 knockout samples and FOG2 NTG control samples clustered with Tbx5 control samples, indicating that gain of FOG2 and loss of Tbx5 caused dysregulation of co-regulated left atrial enhancers in the same direction (Figure 4G). Specifically, 145 of 154 Tbx5- and FOG2-OE-dependent ncRNAs were coordinately regulated. Furthermore, the quantitative degree of TF-dependent ncRNA dysregulation was highly concordant (Cor = 0.69, P = 1.7 x10⁻²³, 4H). We hypothesized that the shared FOG2-OE and TBX5-dependent ncRNAs defining regulatory elements associated with FOG2-OE and TBX5-dependent coding genes were functionally relevant to AF risk. Shared genes were overrepresented in the GO term Cardiac Muscle Cell Action Potential (adjusted P = 9.44 x10⁻⁰⁸; Figure 4I). These findings identified a common set of cis-regulatory elements activated by TBX5 and repressed by FOG2 identified by differential enhancer transcription critical to cardiac rhythm control.

We integrated TBX5 and FOG2 genomic binding and differential ncRNA transcription to define a set of co-bound FOG2- and Tbx5-dependent enhancers comprising a co-regulated GRN. TF-dependent ncRNA abundance in FOG2-OE vs. WT mice and Tbx5 control vs. KO were binned by FOG2, TBX5, and GATA4 occupancy. TBX5-dependent enhancers were identified among all TF binding patterns that included TBX5, and none that omitted TBX5, indicating a strong effect of TBX5 on enhancer transcription (Figure 4J). This indicated that TBX5 occupancy was a major driver of TBX5-dependent enhancer activity. In contrast, FOG2 caused transcriptional changes only for enhancers co-bound by FOG2, GATA4, and TBX5 (FGT, P < 8.7 x10^−07; Figure 4K). Notably, this enhancer set included the two conserved TBX5 and FOG2 co-bound and modulated enhancer elements at calcium handling loci Ryr2 and Atp2a2 (Figure 5D, E). Together, these data demonstrate that FOG2 and TBX5 directly regulate a shared set of cis-regulatory elements to define a specific atrial GRN critical to cardiac rhythm control.

Figure 5. FOG2-OE- and Tbx5-dependent ncRNAs identify shared enhancers conserved in humans.

A. Venn diagram showing the intersection of mouse elements that were mapped to the human genome vs. open chromatin regions from human left atria. The red circle encompasses the sites from human left atria from a consensus set made up of ENCODE data. The light blue circle shows mouse regions from a whole heart open chromatin that were mapped to the human genome. The inner dark blue circle shows a subset of regions from the light blue circle. These regions intersected in mice with a differentially transcribed enhancer that was nearby a FOG2-dependent gene.

B. Violin plot showing the pre-calculated from 60-way vertebrate Phastcons score at FOG2-dependent enhancer regions in dark blue and mouse open chromatin regions in light blue. Statistics were from Wilcoxon Rank Sum test between background and FOG2-dependent regions.

C. Boxplots comparing normalized tag counts from healthy versus failing hearts at conserved elements mapped from mouse genome to human genome. The left column shows the comparison of elements near RYR2 (top) and ATP2A2(bottom) in the left atria. The right column shows the same for the left ventricle. Statistics from t-test.

D. Left panel shows a genome browser shot of Atp2a2 and the Atp2a2-enhancer locus. First 4 tracks: wild-type and mutant FOG2 and TBX5 polyA-depleted CPM normalized bigWigs. The next 4 tracks in red show FOG2 (HL1 and Fetal), TBX5, and GATA4 fold enrichment adjusted for library size from ChIP-seq. The last two tracks in ATAC-seq fold enrichment were adjusted for library size (purple), followed by mm10 60-way conservation score (black). The right panel shows the same at Ryr2 and Ryr2-enhancer locus.

E. Left panel shows genome browser shot ATP2A2 and the ATP2A2-enhancer locus in the human genome with mapped mouse elements in the first two top tracks followed by peak calls and the signal from ATAC-seq from the human left atrium, followed by a signal from H3K27ac from the left ventricle and right atrium. The right panel shows the same at RYR2 and RYR2-enhancer locus.

F. Left panel shows a boxplot showing relative luciferase activity of Ryr2 enhancer in HL1 cells. Student t-test between wild-type and FOG2 and two other experiments: wild-type, wild-type and FOG2. Student t-test between GATA4 mutant and FOG2, and wild-type, as well as GATA4 mutant vs wild-type and FOG2. Middle panel shows box plot showing relative luciferase activity of Ryr2 enhancer in HEK293T cells. Top significance bar indicates significance from two-way ANOVA test. Student’s t-test between TBX5/GATA4/FOG2 and TBX5, TBX5/GATA4/FOG2, as well as a student t-test between TBX5/GATA4Mutant/FOG2 and TBX5. Right panel shows results from ChIP-qPCR of TBX5 in the presence and absence of FOG2-OE in HL1 cells. Student’s t-test between delta fold enrichment of control TBX5-antibody, FOG2-transfected TBX5-antibody.

G. Left panel shows a boxplot showing the relative luciferase activity of Atp2a2 enhancer in HL1 cells. Student’s t-test between wild-type and FOG2 and two other experiments: wild-type, wild-type, and FOG2. Student t-test between GATA4 mutant and FOG2, and wild-type, as well as GATA4 mutant vs wild-type and FOG2. The middle panel shows a boxplot from the relative luciferase activity of Atp2a2 enhancer in HEK293T cells. Top significance bar indicates significance from two-way ANOVA test Student’s t-test between TBX5/GATA4/FOG2 and TBX5, TBX5/GATA4/FOG2, as well as a student’s t-test between TBX5/GATA4Mutant/FOG2 and TBX5. The right panel shows results from ChIP-qPCR of TBX5 in the presence and absence of FOG2-OE in HL1 cells. Student-t-test between delta fold enrichment of control TBX5-antibody, FOG2-transfected TBX5-antibody.

FOG2-OE- and Tbx5-dependent ncRNAs identify shared enhancers conserved in humans

Since FOG2 and TBX5 have both been associated with human electrocardiographical parameters and cardiac rhythm abnormalities, we asked if FOG2- and TBX5-dependent enhancers could be identified in humans by mapping the transcribed mouse enhancer locations into the human genome. We previously observed that enhancers defined in mice by Tbx5-dependent ncRNAs were well-conserved in humans²³. We found that FOG2-OE-dependent ncRNA-defined enhancer elements were highly conserved relative to mouse open chromatin by analysis of pre-calculated Phastcons conservation scores (P = 2.2x10⁻¹⁶; Figure 5B). Next, we mapped the differentially transcribed ATAC-seq summit of each FOG2-OE-dependent ncRNA from mm10 into hg38⁷⁹ and extended the hg38 summit by 250 bp in each direction to define a putative TF-dependent transcribed region in human genome coordinates. We then asked if these elements were enriched in human-left atrial open-chromatin data. We observed a significant overlap between lifted-over FOG2-OE-dependent enhancers and open chromatin regions from human left atria from ENCODE (98/142, P = 0.009 by random permutation Figure 5A). Notably, the FOG2 and TBX5 co-modulated enhancers located at Atp2a2 and Ryr2 in mice were conserved in humans (Figure 5E). Furthermore, analysis of human CAGE data revealed a loss of transcription in left atria tissue from failing hearts at the conserved elements near RYR2 and ATP2A2 when compared to healthy individuals (Figure 5C, left top and left bottom ; P = 0.01 both). No change was observed in the expression of this element in left ventricle samples. Together, these data suggested that differentially transcribed enhancer elements from FOG2-OE were well-conserved cardiac regulatory regions affecting human gene expression relevant to atrial rhythm control.

FOG2 represses TBX5-dependent activity of GATA4 co-bound enhancers

We directly assessed interactions between FOG2, TBX5, and GATA4 on co-bound and transcribed conserved Tbx5- and Fog2-dependent enhancers at Atp2a2 and Ryr2 (Atp2a2: mm9 Chr5: 122970476-122971591; Ryr2 : mm9 Chr13: 12223550-12226563) (Figure 5D left, 5D, right) ^10,22,36. These cis-regulatory elements have been shown to be TBX5-dependent in HL-1 cardiomyocytes, which endogenously express the atrial cardiomyocyte transcriptional kernel, including TBX5 and GATA4^22,36,100. Consistently, both wild-type cis-regulatory elements demonstrated robust activity by luciferase reporter assays in HL-1 cells (Ryr2: mm9 Chr13: 12223550-12226563; Atp2a2: mm9 Chr5: 122970476-122971591) (9.79±1.64-fold and 12.39±1.4-fold respectively, P = 8.02x10⁻⁰⁸ and 1.12x10⁻¹² respectively, compared to scramble vector control; Supplemental Table 7). FOG2 overexpression strongly suppressed enhancer activity in each case (P = 9.3 x10⁻⁰⁸ and p< 8.2 x10⁻¹¹, respectively, vs. no overexpression control; Figure 5F, G left). These findings validate the conclusions from the genome-wide in vivo enhancer transcription assay and identify FOG2-repressed TBX5-dependent enhancers at Ryr2 and Atp2a2.

We assessed the independent and combinatorial impact of TBX5, GATA4, and FOG2 on the Atp2a2 and Ryr2 enhancers in-vitro (Figure 5F, G). Both wild-type enhancers demonstrated robust transcriptional activation in response to Tbx5 expression in dual luciferase reporter assay in HEK (human embryonic kidney 293T) cells (Figure 5F, G middle, in the presence of exogenous TBX5: 5.65±1.14-fold and 6.16±2.80-fold, P = 1,21x10⁻⁰⁸ and 4.12x10⁻⁰⁶ respectively, compared to scramble control vector). The addition of GATA4 expression did not affect TBX5-dependent activation (TBX5 and GATA4: 4.59±1.93-fold and 6.07±2.87-fold, P = 2.74x10⁻⁰⁵ and 4.78x10⁻⁰⁶ respectively, compared to scramble control, P = N.S. and P = N.S. compared to TBX5 alone). Overexpression of FOG2 strongly suppressed enhancer activation was observed in the presence of exogenous TBX5 and GATA4 (P = 1.8x10⁻⁰⁶ and 4.4x10⁻⁰⁶, respectively, versus no FOG2 overexpression, Figure 5F, G middle). However, mutation of the GATA4 binding sites (AGATAA to AAAAAA) abrogated FOG2-dependent repression of both the Atp2a2 and Ryr2 enhancers in HL-1 cells (P = 1.4x10⁻⁰⁶ and P = 3.6x10⁻¹⁰, respectively, for GATA4 mutant versus wild-type Atp2a2 and Ryr2 enhancers with FOG2 OE, Figure 5F, G left). Mutation of the GATA4 sites did not impact the enhancer activation in the absence of FOG2 OE in HL-1 cells (P = N.S., P = 0.0046, respectively, for GATA4 mutant versus wild-type Atp2a2 and Ryr2 enhancers without FOG2 OE, Figure 5F, G left). Furthermore, mutation of the GATA4 binding sites abrogated FOG2-dependent repression of TBX5-dependent activation in HEK cells (P = 3.6x10⁻¹⁰, and = 3.0x10⁻⁰⁶, respectively, for GATA4 mutant versus wild-type Atp2a2 and Ryr2 enhancers with TBX5, GATA4 and FOG2 OE, Figure 5F, G middle), but did not impact TBX5-dependent activation itself (P = N.S., and P = N.S., respectively, for GATA4 mutant versus wild-type Atp2a2 and Ryr2 enhancers with TBX5, GATA4 without FOG2 OE, Figure 5F, G middle) in HEK293T cells. These results indicated that FOG2 represses TBX5-dependent activation of TBX5 and GATA4 co-bound Atp2a2 and Ryr2 enhancers and that the functional TBX5/FOG2 interaction required GATA4 co-binding, supporting a model in which GATA4 recruits FOG2 to modulate TBX5-dependent expression of genes critical to cardiac rhythm.

To determine whether FOG2 represses TBX5-dependent transcriptional activation by competing TBX5 off chromatin or an alternate genomic mechanism, we evaluated the TBX5 binding to Ryr2 and Atp2a2 cis-regulatory elements (CREs) in the presence and absence of FOG2 overexpression. We performed ChIP experiments on HL1 cardiomyocytes overexpressing FOG2 using anti-TBX5 antibody and observed significant enrichment of Ryr2 and Atp2a2 CREs in the TBX5-immunoprecipitation fraction compared to the IgG-immunoprecipitation fraction by qPCR (P = 1.4x10⁻⁰³ and P = 1.1x10⁻⁰⁴, respectively by student’s t-test. Table 10). The negative control DNA was not significantly enriched in the TBX5 immunoprecipitation fraction either from control HL1 cardiomyocytes or FOG2 overexpressing HL1 cardiomyocytes (Figure E, F right), confirming the specificity of our IPs. We observed no significant differences in the enrichment of TBX5 binding to the Ryr2 or Atp2a2 enhancers in HL1 cells with FOG2 overexpression compared to control HL1 (no FOG2 overexpression, Figure 5E, F). This result suggested that the presence or absence of FOG2 does not displace or prevent TBX5 binding.

Tbx5 drives Zfpm2 expression in an incoherent feed-forward loop

Specific transcriptional network architectures impart specific network gene expression characteristics. An incoherent feed-forward loop transcriptional architecture imparts transcriptional stability on the shared targets of a transcriptional activator and a transcriptional repressor if the activator also drives expression of the repressor¹⁰¹. We asked if TBX5 drove FOG2 expression and found that Zfpm2 (encoding FOG2) was significantly downregulated in the left atria of adult-specific Tbx5 knockout mice (log2 fold change = 0.55, adjusted P = 0.01; Figure 6A). Furthermore, the Zfpm2 locus had seven co-TBX5 and GATA4¹⁰² bound regions with high H3K27ac signal in the adult mouse atrium, suggesting the possibility that TBX5 and GATA4 directly drive Zfpm2 expression (Figure 6B).

Figure 6. Tbx5 and Zfpm2 interact genetically for AF inducibility in mice.

A. Box plots with data points showing the difference between Zfpm2 expression from normalized RNA-seq in control (blue) and Tbx5 knockout (red). Significance represents P-value < 0.05 from the student’s t-test.

B. Genome browser shot showing the gene body of Zfpm2 in mm10 genome build. Tracks show read out in fold enrichment and peak calls from TBX5, GATA4, and H3K27ac ChIP-seq.

C. Boxplot showing the P-wave duration in milliseconds for wild-type (R26^CreERt2black) adult specific Tbx5 haploinsufficient mice (Tbx5^fl/+; R26^CreERt2 second from left), Zfpm2 haploinsufficient mice (Zfpm2^+/tm1Jml; R26^CreERt2 third from left), and compound heterozygote mice (Tbx5^fl/+; Zfpm2^+/−; R26^CreERt2 rightmost bar). P-value from one way ANOVA test on top with by student’s t-test P-values between wild-type, Tbx5 haploinsufficient, and Zfpm2 haploinsufficient mice.

D. Boxplot showing PR interval prolongation in milliseconds for wild-type (R26^CreERt2black) adult specific Tbx5 haploinsufficient mice (Tbx5^fl/+; R26^CreERt2 second from left), Zfpm2 haploinsufficient mice (Zfpm2^+/tm1Jml; R26^CreERt2 third from left), and compound heterozygote mice (Tbx5^fl/+; Zfpm2^+/−; R26^CreERt2 rightmost bar). P-value from one-way ANOVA on top with student’s t-test P-values between wild-type, Tbx5 haploinsufficient, Zfpm2 haploinsufficient mice, and compound heterozygote mice.

E. Atrial electrogram and surface ECG for wild-type mice (R26^CreERt2).

F. Atrial electrogram and surface ECG for Zfpm2 haploinsufficient mice (Zfpm2^+/tm1Jml; R26^CreERt2) shows no induction of atrial fibrillation.

G. Atrial electrogram and surface ECG for Tbx5 haploinsufficient mice (Tbx5^fl/+; R26^CreERt2) shows induction of atrial arrhythmia.

H. Atrial electrogram and surface ECG for Tbx5, Zfpm2 compound heterozygote mice (Tbx5^fl/+; Zfpm2^+/−; R26^CreERt2) shows no induction of atrial fibrillation.

I. Table of mice induced into atrial fibrillation by genotype with P-values from Fisher’s Exact test.

The incoherent feed-forward loop transcriptional architecture is predicted to diminish the dispersion of network gene expression to impart stability of downstream target gene expression¹⁰¹. We hypothesized that TBX5, GATA4, and FOG2 co-bound enhancers would display diminished activity dispersion compared to enhancers with an alternate pattern of TBX5, GATA4, and/or FOG2 binding. We found that H3K27ac signal at TBX5, GATA4, and FOG2 co-bound regions (Figure 3G) demonstrated diminished dispersion compared to other patterns of TF binding by the Brown-Forsythe test¹⁰³. Specifically, TBX5, GATA4, and FOG2 co-bound regions demonstrated diminished dispersion compared to GATA4 and TBX5 co-bound regions (P < 3.41x10⁻²¹), FOG2 and GATA4 co-bound regions (P < 3.54x10⁻¹²⁸), and TBX5 singly bound regions (P < 7.97x10⁻¹²). There was no significant difference in the dispersion between GATA4 and TBX5 co-bound regions compared to TBX5 bound regions (P = N.S.). The diminished activity dispersion of TBX5/GATA4/FOG2 co-bound enhancers may buffer the expression of their target cardiac rhythm gene set.

Tbx5 and Zfpm interact genetically for AF susceptibility in mice

A TBX5/FOG2 incoherent feed-forward loop suggested the possibility that Tbx5 and Zfpm2 may genetically interact. We hypothesized that the cardiac rhythm pathophysiology caused by reduced Tbx5 dose may be mitigated by reduced native Zfpm2 dose. We, therefore, examined Tbx5 and Zfpm2 loss-of-function alleles for a genetic interaction for atrial rhythm in vivo. We interrogated littermate adult-specific Tbx5 haploinsufficient mice (Tbx5^fl/+; R26^CreERt2)¹¹, Zfpm2 haploinsufficient mice (Zfpm2^+/−; R26^CreERT2)²¹, compound heterozygote (Tbx5^fl/+; Zfpm2^+/−; R26^CreERT2), and control (R26^CreERT2) mice at 8-weeks of age following tamoxifen treatment at 6-weeks of age. Tbx5 haploinsufficiency caused cardiac conduction abnormalities, including P-wave and PR-interval prolongation (P = 0.0303 and 0.0467 respectively compared to R26^CreERT2 control, Figure 6C–D), and atrial arrhythmias induced by intracardiac pacing (8/12 Tbx5^fl/+; R26^CreERt2 mice compared with 0/9 R26^CreERt2, P = 4.6 x10⁻³; Figure 5C–D), consistent with prior reports^22,36. Fog2 haploinsufficiency also caused P-wave and PR-interval prolongation (P = 0.0077 and 0.011 vs. R26^CreERt2 controls, Figure 6C–D). However, these mice showed no AF susceptibility through intracardiac pacing (Figure 5E–G. 0/8, Zfpm2^+/−; R26^CreERt2 P = NS). However, reduced Zfpm2 dose rescued AF inducibility in Tbx5 haploinsufficiency (2/12 inducible, vs 8/12 inducible, Tbx5^F/+;R26^CreEET2; P = 0.019) (Figure 6I). Furthermore, reduced Zfpm2 dose normalized the prolonged P-wave duration (P = NS vs. R26^CreEET2 and P = 0.6873 vs. Tbx5^F/+;R26^CreEET2) but not prolonged PR interval (P = 1x10⁻⁴ vs. R26^CreEET2 and P = 0.1244 vs Tbx5^F/+;R26^CreEET2) observed in Tbx5 haploinsufficient mice (Figure 6C–D). Thus, reduced FOG2 dose mitigated the physiologic effects of reduced TBX5 dose on atrial rhythm stability in vivo, supporting the essential requirement of the functional FOG2 / TBX5 genomic interactions for normal atrial rhythm control.

Discussion

FOG2 upregulation establishes a transcriptional model for AF risk and molecularly links heart failure to AF

FOG2 is upregulated in the atria in human heart failure, an impetus for modeling the impact of FOG2 overexpression in the adult mouse heart. Conditional overexpression of FOG2 in cardiomyocytes of the adult mouse heart caused primary atrial fibrillation, including arrhythmia substrate, prolonged cardiomyocyte action potential, and arrhythmia trigger, including increased calcium cardiomyocyte sparks. FOG2 intersected with an atrial gene regulatory network that controls the expression of atrial rhythm and AF-associated genes. FOG2 and TBX5 functionally interact, genetically and genomically, and FOG2 represses TBX5-driven gene expression to define a FOG2/TBX5 co-modulated gene regulatory network for control of atrial gene expression and rhythm. These findings offer a genomic link between a specific atrial gene expression alteration caused by heart failure and the cause of atrial fibrillation.

Transcription factor interactions in AF pathogenesis

GWAS continue to discover transcription factor loci that underlie disease susceptibility ^13,16,18. Zfpm2 (encoding FOG2) has been associated with conduction defects, including PR interval prolongation, an AF risk factor^{11–13,17,18}. FOG2 joins multiple atrial TFs that have been implicated in AF risk, including TBX5, GATA4, PITX2, and others¹¹. Downstream targets of these TFs include cardiac channel and rhythm effector genes also implicated in AF risk^{11–13,22,24,25,104}. These observations are consistent with a transcriptional model for disease risk, in which alterations in TF expression cause disrupted downstream gene expression and electrophysiology, increasing AF susceptibility. Recent functional genomics studies support this model: Atrial TFs such as TBX5 and PITX2 control enhancers that are linked to genes essential for atrial rhythm control and that contain GWAS-associated SNPs that modulate their activity^22,105,106.

Functional TF interactions modulate atrial gene expression and impact TF dosage effects on atrial gene expression and AF risk^17,22,23,36. For example, TBX5, a transcriptional activator, and PITX2, a transcriptional repressor, function together in an incoherent feed-forward loop that tunes the expression of common downstream rhythm targets²². Here, we find that TBX5 and FOG2 also form an incoherent feed-forward loop in which TBX5 drives FOG2 expression. Overexpression of FOG2, a transcriptional repressor, modulates the activity of TBX5, an activator, at enhancers of cardiac rhythm genes, for example, those important for atrial cardiomyocyte calcium handling. Genetic interaction studies, in which the atrial arrhythmia propensity caused by Tbx5 haploinsufficiency was rescued by FOG2 haploinsufficiency, demonstrate that FOG2 and TBX5 normally co-modulate atrial gene expression to maintain normal rhythm (Figure 5). Together these studies uncover the intricate interplay between transcriptional activators and repressors in the tight control of gene expression for cardiomyocyte excitability.

Functional TBX5 / FOG2 interactions define a gene regulatory network for atrial rhythm loci

Combinatorial transcription factor (TF) interactions drive tissue and context-dependent gene expression¹⁰⁷. Efforts to define TF interactions have relied on physical interactions or co-localization at genomic locations; however, neither approach identifies functional TF interactions indicative of shared impact on enhancer activity, the sine-qua-non of a TF interaction¹⁰⁸. Identification of a shared gene regulatory network between TFs defined by shared functional activity on a group of cis-regulatory elements is a longstanding goal of transcriptional biology. Enhancers transcribe non-coding RNAs, and enhancer-bourne RNA production quantitatively reflects enhancer activity^109–111. We have previously applied TF-dependent enhancer-generated ncRNA profiling to define TF-dependent enhancer activity¹¹⁰. We applied FOG2-OE-dependent ncRNA profiling to identify FOG2-dependent enhancers and found that FOG2 only affects enhancer transcription at a subset of GATA4 and TBX5 co-bound enhancers. The intersection of TBX5 activated and FOG2 repressed enhancer transcription defined a co-modulated gene regulatory network essential for atrial rhythm control. Shared cis-regulatory elements were enriched for enhancers conserved in humans, including at calcium handling genes Ryr2 and Atp2a2 implicated in human AF (Figures 4, 5). Beyond the specific regulatory networks defined in this study, comparing TF-dependent differential enhancer transcription will enable functional TF interactions to be identified based on shared activity on cis-regulatory elements in other native contexts.

Our results support a model in which GATA4 acts as a TF scaffold, recruiting FOG2 to modulate TBX5-dependent expression of genes critical for cardiac rhythm. T-Box and FOG transcription factors were not previously known to interact, and there is no evidence that they physically associate. However, TBX5 and FOG2 have been independently described as components of a transcriptional kernel shared with GATA4 in the heart¹¹². Gene expression alterations caused by FOG2-OE overlapped more with Tbx5 than Gata4-dependent gene expression. However, FOG2 only altered enhancer activity at locations co-occupied by TBX5 and GATA4 (Figure 4), suggesting that the functional TBX5 / FOG2 interaction is mediated via shared interactions with GATA4 (Figure 7). Furthermore, FOG2-TBX5 co-regulated enhancers control a specific gene set for cardiac rhythm control. By defining a shared gene regulatory network between TFs previously linked to AF by genetic association, this work clarifies the molecular basis of atrial gene expression and how its perturbation can cause disease risk.

Figure 7. Model for link from heart failure to AF: FOG2 modulates TBX5 function at GATA4 co-bound enhancers to control atrial rhythm gene expression and atrial fibrillation risk.

A. In the wild-type condition, normal expression of FOG2 buffers TBX5-dependent activation to enable appropriate expression of cardiac rhythm/calcium handling genes and normal atrial rhythm.

B. In heart failure, overexpression of FOG2 causes decreased TBX5-dependent expression of cardiac rhythm/calcium handling genes and increased atrial fibrillation risk.

Mechanism of FOG2-dependent repression of cardiac rhythm gene expression

GATA4-dependent recruitment of FOG2 to buffer TBX5 activity is consistent with the observation that although TBX5 and GATA4 interact cooperatively for cardiac development, they interact antagonistically for AF risk in the adult heart^{31,33,35–48,95,113}. We note that FOG2 binding was not enriched for histone modifications associated with repression. In fact, although TBX5, GATA4, and FOG2 co-bound enhancers have less H3K27ac than TBX5 and GATA4 co-bound enhancers alone, they are still highly H3K27 acetylated and transcribed in-vivo (Figure 3E). This suggests that although TBX5, GATA4, and FOG2 co-bound enhancers are active, FOG2 binding buffers their activity to prevent overexpression of dosage-sensitive cardiac rhythm control genes. The incoherent feed-forward loop transcriptional architecture imparts transcriptional stability, as we have previously shown for TBX5 and PITX2²². We observe that this architecture between TBX5 and FOG2 results in diminished dispersion of target enhancer activity at TBX5, GATA4, and FOG2 co-bound locations compared to enhancers with other patterns of TBX5, GATA4, and/or FOG2 binding (Figure 3G). A balance between transcriptional activators and repressors that buffer transcriptional enhancer activity may be a general facet of the transcriptional kernel controlling cardiac rhythm effector gene expression.

The genomic mechanism by which FOG2 diminishes the activity of active TBX5-driven enhancers may involve shared interactions with the NuRD complex. FOG2 recruits the NuRD complex through physical interactions with GATA transcription factors in the heart, notably GATA4^39–48. NuRD components interact with both FOG2^46,112,114 and TBX5.¹¹² FOG2 interacts with MTA and RbAp subunits of NuRD complex, required for FOG2-mediated transcriptional repression in-vitro⁴⁶. An unbiased proteomics screen for TBX5-interacting proteins identified NuRD components CHD4 and MTA1, and the region of TBX5 necessary and sufficient for NuRD interaction is affected by TBX5 mutations that cause Holt – Oram syndrome in humans¹¹². Disruption of the FOG2-NuRD or TBX5-NuRD interactions causes similar heart defects in mice^46,112,115. These data support a model in which FOG2 recruits NuRD complex proteins to GATA4-TBX5 co-bound locations to mediate acetylation levels and Pol II binding to buffer TBX5-dependent transcriptional output of rhythm control genes¹¹⁶. In heart failure, overexpression of FOG2 may increase NuRD activity at TBX5-dependent enhancers, increasing repressive tone at cardiac rhythm control genes to predispose to AF.

A gene regulatory link from heart failure to atrial fibrillation

Heart failure and AF are strongly epidemiologically linked. Approximately one-third of AF patients have heart failure, and up to half of all heart failure patients have AF, although the molecular mechanisms underlying this association remain unclear^1,41,49,117. The pathophysiologic link from heart failure to atrial fibrillation has been presumed to be based on shared structural or physiologic endpoints of each disease, such as structural atrial remodeling, a known consequence of both heart failure and atrial fibrillation. Here, we provide a gene regulatory link in which a transcriptional effect of heart failure, upregulation of FOG2 in the atria of human heart failure (Figure 1), is sufficient to cause primary atrial fibrillation. AF is a primary result of FOG2-OE, which causes completely penetrant primary spontaneous AF preceding the onset of functional ventricular deficits (Figure 1) or structural atrial remodeling (Figure 1) in each case. FOG2 directly intersects with the atrial transcriptional kernel required for tight control of atrial rhythm gene expression, providing a direct genomic mechanism. The subset of TBX5 and GATA4-co-bound genomic locations that FOG2 selects for regulation includes genes essential for normal atrial cardiomyocyte electrophysiology and rhythm control. The identification of FOG2 as a necessary member of the atrial transcriptional kernel and the genomic intersection between FOG2, TBX5, and GATA4 functional provides a molecular and genomic link between the consequences of heart failure and the cause of atrial fibrillation. This model suggests that AF can be a primary consequence of the atrial transcriptional response to HF rather than as a secondary consequence of atrial remodeling, providing a genomic mechanism that may underpin the epidemiological association between HF and AF.

Clinical Perspective.

1). What is new?

• FOG2 overexpression, present in human heart failure, causes primary atrial fibrillation in mice prior to the onset of atrial remodeling or ventricular dysfunction.

• FOG2 binds and directly represses TBX5-driven GATA4-bound enhancers that comprise a gene regulatory network that controls cardiac rhythm gene expression and is conserved in humans.

• TBX5 and FOG2 interact in the normal heart to buffer cardiac rhythm gene expression for atrial rhythm control.

2). What are the clinical implications?

• Clinical approaches to managing AF are based on disease duration, rather than on underlying mechanisms, therefore new pathophysiologic insights are essential to guide improved therapeutic strategies to AF.

• The epidemiological link between heart failure and atrial fibrillation (AF) is exceptionally strong, as up to half of all heart failure patients acquire AF, currently without mechanistic explanation.

• This work proposes a genomic explanation for the onset of AF in the context of heart failure, based on a primary genomic response of atrial myocardium rather than secondary to atrial remodeling.

Nonstandard Abbreviations and Acronyms

ATP2A2

calcium-ATPase 2

AF

Atrial fibrillation

CAGE

Cap analysis of gene expression

ChIP

Chromatin immunoprecipitation

DTG

Dual trans-genic

eRNA

Enhancer RNA

GO

Gene ontology

GRN

Gene regulatory network

GWAS

Genome-wide association studies

H3K27ac

Acetylation of histone H3 at lysine 27

H3K4me

Methylation of histone H3 at lysine 4

HF

Heart failure

lncRNA

long-noncoding RNA

ncRNA

noncoding RNA

NTG

Non transgenic

OE

overexpression

RNA-seq

RNA sequencing

RYR2

Ryanodine receptor 2

TF

Transcription Factor