DNA methylation reprograms cardiac metabolic gene expression in end-stage human heart failure

Abstract

Heart failure (HF) is a leading cause of morbidity and mortality in the United States and worldwide. As a multifactorial syndrome with unpredictable clinical outcomes, identifying the common molecular underpinnings that drive HF pathogenesis remains a major focus of investigation. Disruption of cardiac gene expression has been shown to mediate a common final cascade of pathological hallmarks wherein the heart reactivates numerous developmental pathways. Although the central regulatory mechanisms that drive this cardiac transcriptional reprogramming remain unknown, epigenetic contributions are likely. In the current study, we examined whether the epigenome, specifically DNA methylation, is reprogrammed in HF to potentiate a pathological shift in cardiac gene expression. To accomplish this, we used paired-end whole genome bisulfite sequencing and next-generation RNA sequencing of left ventricle tissue obtained from seven patients with end-stage HF and three nonfailing donor hearts. We found that differential methylation was localized to promoter-associated cytosine-phosphate-guanine islands, which are established regulatory regions of downstream genes. Hypermethylated promoters were associated with genes involved in oxidative metabolism, whereas promoter hypomethylation enriched glycolytic pathways. Overexpression of plasmid-derived DNA methyltransferase 3A in vitro was sufficient to lower the expression of numerous oxidative metabolic genes in H9c2 rat cardiomyoblasts, further supporting the importance of epigenetic factors in the regulation of cardiac metabolism. Last, we identified binding-site competition via hypermethylation of the nuclear respiratory factor 1 (NRF1) motif, an established upstream regulator of mitochondrial biogenesis. These preliminary observations are the first to uncover an etiology-independent shift in cardiac DNA methylation that corresponds with altered metabolic gene expression in HF.

NEW & NOTEWORTHY The failing heart undergoes profound metabolic changes because of alterations in cardiac gene expression, reactivating glycolytic genes and suppressing oxidative metabolic genes. In the current study, we discover that alterations to cardiac DNA methylation encode this fetal-like metabolic gene reprogramming. We also identify novel epigenetic interference of nuclear respiratory factor 1 via hypermethylation of its downstream promoter targets, further supporting a novel contribution of DNA methylation in the metabolic remodeling of heart failure.

INTRODUCTION

Even with the use of current guideline-directed medical therapy (63), heart failure (HF) is a progressive syndrome with a bleak prognosis. The clinical course of HF often entails frequent hospitalization, lifelong medical management, and in some cases the need for cardiac assist devices or heart transplantation (16, 52). Although controversial, HF is generally regarded an irreversible manifestation of longstanding cardiac pathology, where current medical approaches are limited to slowing its progression and managing symptoms of volume overload (63). Development of targeted therapies for HF is confounded by both etiological and clinical heterogeneity (32, 36). Efforts have therefore turned to defining the common molecular mechanisms that underlie the pathogenesis of end-stage HF regardless of etiology.

Among the hallmark features of HF are global changes to myocardial gene expression that involve reactivation of developmental pathways (14, 59). Familial sequencing analyses and genomewide association studies have, respectively, identified many single-gene mutations and genetic variants sufficient to confer susceptibility to HF. However, the clinical utility of these genomic discoveries is limited by both the infrequency of occurrence and a pleiotropy that exists even in single-gene etiologies (35, 51). Environmental and lifestyle influences have been since identified as modifiers of genetic risk (44), although the molecular mechanisms by which environmental stimuli alter cardiac gene expression remain elusive.

The field of epigenetics is uniquely poised to explain how changes to the cardiac microenvironment trigger a persistent, yet modifiable, transcriptional response. Although initially coined for its role in embryological development (60a), the epigenome remains malleable throughout adulthood, responding to both physiological (12) and pathological (22) stimuli. Many epigenetic mechanisms exist, which have been attributed a functional role in HF pathogenesis, from histone modifications (11) and long-noncoding RNAs (38) to direct covalent modifications of DNA. These modifications influence gene expression by either controlling the readability of genetic features (via histone and DNA modifications) or altering the stability of transcribed RNA (via noncoding RNAs).

The role of DNA methylation in HF has become the primary focus of our group because of its ability to regulate numerous biological processes essential for cardiac function. In cancer, aberrant DNA methylation is required for metabolic adaptation of malignant cells to the hypoxic tumor microenvironment (40, 48). We have recently reported comparable metabolic features of ischemic cardiomyopathy wherein the heart exhibits hypermethylation and gene silencing of oxidative pathways relative to nonischemic HF etiologies (46). However, the role of DNA methylation in regulating the transcriptional programs common to all etiologies of HF is unknown, as are the changes relative to nonfailing human hearts.

Therefore, in the current study, we used an integrative genomewide analysis to examine the relationship between cardiac DNA methylation and gene expression in human HF. Using whole genome bisulfite sequencing with paired RNA-sequencing analysis of cardiac tissue obtained from patients with end-stage HF, our analysis uncovers a marked shift in the transcriptional and epigenomic landscape consistent with cardiac metabolic remodeling. We also identify nuclear respiratory factor 1 (NRF1) as a candidate DNA methylation-sensitive transcriptional regulator of cardiac metabolism. We therefore propose that the metabolic consequences of HF are modulated, in part, by epigenetic programming.

MATERIALS AND METHODS

Ethics statement.

Human studies were approved by the University of Utah Institutional Review Board. Informed consent was obtained for the procurement of left ventricular assist device (LVAD) core biopsies, and a waiver of consent was granted for tissue samples received from nonfailing hearts of organ donors. Patient health information with demographics was acquired at the time of tissue acquisition. All human RNA-sequencing and whole genome DNA methylation data have been uploaded to the NCBI Gene Expression Omnibus database (GSE123976): https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE123976; https://zenodo.org/record/3333040.

Human cardiac left ventricular samples.

Apical biopsies of cardiac left ventricle were obtained during LVAD implantation at the University of Utah Hospital. All patients met the medical policy guideline of New York Heart Association class III/IV HF. Patient health information collected from all patients revealed a broad distribution of ages (34–62 yr), body mass indexes (BMI, 23–43 kg/m²), and HF durations (3 mo–20 yr) (Fig. 1A). Despite the notable heterogeneity in patient health metrics and history, cardiac functional parameters reflected equivalent degrees of severe systolic dysfunction with left ventricular ejection fractions 10–25% and elevated cardiac filling pressures (Fig. 1B) among HF subjects. Because age and BMI are known drivers of DNA methylation (18), we inspected the available health data of nonfailing (NF) donor hearts for notable differences [Supplemental Table S1 (Supplemental data for this article can be found at 10.5281/zenodo.3364044)]. We found no differences in age or BMI between HF patients and NF donor hearts (Fig. 1, C and D).

Fig. 1.

Pre-left ventricular assist device (LVAD) patient characteristics and control nonfailing hearts. A: table outlining patient characteristics collected from electronic medical records. B: cardiac functional parameters based on right heart catheterization and echocardiography immediately before LVAD procedure. Comparison between heart failure (HF) and nonfailing control hearts of patient age (in years, C) and body mass index (BMI, D) at the time of tissue procurement. BNP, brain natriuretic peptide; RAAS, renin-angiotensin-aldosterone system; HR, heart rate; RA, right atrium; PCWP, pulmonary capillary wedge pressure; CO, cardiac output; M, male; F, female; LVEF, left ventricular ejection fraction; LVEDD, left ventricular end-diastolic dimension; CON, control; Y, yes; N, no; ICM, ischemic cardiomyopathy; NICM, nonischemic cardiomyopathy; ns, not significant. Statistical significance was assessed using Mann-Whitney nonparametric U-test, and data are reported as means ± SE.

Whole genome bisulfite sequencing analysis.

All sequencing was done at the University of Utah core laboratories, with subsequent bioinformatics performed at the University of Alabama at Birmingham. DNA was first isolated from cardiac whole tissue using the DNeasy Mini Kit (Qiagen, Valencia, CA). DNA quality was assessed using Qubit dsDNA HS Assay to ensure uniform concentrations of dsDNA were used. Extracted DNA was then bisulfite reduced using the Agilent SureSelect XT Human Methyl-Seq Enrichment System. Sequencing of bisulfite-reduced samples was then performed using HiSeq 125 Cycle Paired-End Sequencing analysis. Briefly, sequencing libraries (25 pM) were chemically denatured and applied to an Illumina HiSeq v4 paired-end flow cell using an Illumina cBot. Hybridized molecules were clonally amplified and annealed to sequencing primers with reagents from an Illumina HiSeq PE Cluster Kit v4-cBot (PE-401-4001). Following transfer of the flowcell to an Illumina HiSeq 2500 instrument (HCS v2.2.38 and RTA v1.18.61), a 125-cycle paired-end sequence run was performed using HiSeq SBS Kit v4 sequencing reagents (FC-401-4003).

To evaluate sequencing quality, FastQC (0.11.7) was used both before and after adapter trimming via TrimGalore (0.4.4). The bisulfite-reduced and sequenced reads were then aligned to the CT- and GA-converted human hg38 (GRCh38.p12) genome assembly via Bismark (v0.20.0) to quantify relative alignment of methylated and unmethylated cytosine-phosphate-guanine (CpG), respectively. Bias because of PCR artifacts was minimized by removing duplicated reads. We then quantified differential DNA methylation on a CpG site-specific (DMC) basis to exploit the base-pair resolution afforded by whole genome bisulfite sequencing (WGBS) using the R package methylKit (1.8.0). Alignments were first filtered for those with <99.9% methylated CpGs and sequencing depth >10× to remove PCR-biased and low-coverage CpG sites, respectively. To determine statistical significance of differential methylation, an overcorrection adjusted χ²-test was used, as recommended by Wreczycka et al. (62) To further adjust P values for multiple testing, the false discovery (FDR) adjusted P value was used to assign statistical significance (54). From this analysis, we identified 1,910 DMCs (FDR <0.05) associated with 1,346 known genes (Supplemental Table S2).

RNA-sequencing analysis.

Next-generation RNA sequencing analysis was used to quantify differential gene expression in the same cardiac samples submitted for WGBS analysis. RNA isolation was accomplished using the RNeasy Fibrous Tissue Mini Kit (Qiagen) and analyzed to ensure RNA integrity numbers >7. High-throughput next-generation RNA sequencing was performed using the Illumina HiSeq 2500 sequencer at the University of Utah core laboratories. Before alignment, adapters and low-quality (PHRED <20) sequences were trimmed from reads files using TrimGalore (0.5.0). Sequencing data were then aligned to the human hg38 genome (GRCh38.p12) using STAR (2.5.3a). DESeq2 (1.22.1) was then used within R (3.5.1) to perform differential gene expression. To account for sample size limitations, dispersion estimates were first determined via maximum likelihood (30), which aggregates dispersion estimates based on expression strength. Genewise dispersion estimates were then adjusted via the empirical Bayes method to provide normalized count data for genes proportional to both dispersion and sample size. Differential gene expression was then measured between HF (n = 7) and NF (n = 3) from quantile-normalized read counts via log₂ (fold change) using the Wald test, with statistical significance assessed by Benjamini-Hochberg P value correction (Supplemental Table S3). Comparative analysis was then performed to identify differentially expressed genes (DEGs) with overlapping and inverse changes in promoter methylation (DMCs) (Supplemental Table S4).

Bioinformatics and data visualization.

A detailed bioinformatics protocol is included as an online supplement, along with the R coding scripts and session information (see Supplemental Methods). Analysis was performed as a single workflow to ensure reproducibility, with functional and network GSEA completed within Qiagen’s Ingenuity Pathway Analysis software (Qiagen, Redwood City, CA). Heatmap generation and hierarchical clustering were performed using Ward’s minimum squared variance algorithm, and dendrograms were constructed by Euclidean distance via R package pheatmap (1.0.10).

Cell culture and transient transfection.

The H9c2 cells (rat cardiomyoblasts) were cultured in 10% fetal bovine serum (GIBCO, Grand Island, NY) containing DMEM (GIBCO), 5 mM glucose, and 1× penicillin and streptomycin (Sigma-Aldrich, St. Louis, MO) in a humidified atmosphere with 5% CO₂ at 37°C. Plasmids expressing constitutively active human myc-tagged DNA methyltransferase (DNMT) 3A (pcDNA3.1-DNMT3A, Addgene plasmid no. 35521) and empty vector (pcDNA3.1-EMPTY) were used as previously described (5). For transient expression of pcDNA3.1-DNMT3A or pcDNA3.1-EMPTY, H9c2 cells were transfected with Lipofectamine 3000 (Invitrogen, Carlsbad, CA) according to the manufacturer’s instructions.

Quantitative real-time PCR.

Total RNA from cell cultures of H9c2 rat cardiomyoblasts was isolated using TRIzol reagent (Invitrogen), as described previously (24). cDNA was synthesized by reverse transcription PCR using High-Capacity cDNA Reverse Transcriptase (Applied Biosystems), DNase I (Invitrogen) treatment, and RNase inhibitor (Invitrogen). cDNA was amplified using gene-specific RT-qPCR primers (Supplemental Table S5) to determine mRNA expression. The mRNA expressions were standardized and normalized to the 36B4 reference gene.

Statistics.

Statistical significance was determined via unpaired two-tailed Bonferroni-adjusted P value <0.05. For all targets, an unpaired Student’s t-test was performed using Tukey’s correction for multiple comparisons. Statistical significance was concluded based on a false discovery-adjusted P value <0.05.

RESULTS

Whole genome analysis of DNA methylation in human HF.

The transcriptional effects of DNA methylation are highly dependent on the position of differentially methylated single CpG sites (DMCs) relative to known genic features, with promoter methylation contributing an inverse regulatory impact on the adjacent gene’s expression (3, 21). Therefore, we examined the methylation dynamics in HF based on DMC (FDR <0.01) proximity to annotated gene regions (Fig. 2A). A single peak emerged, centering at the transcription start site with a modest plateau within the proximal (<1.5 kb) promoter. Hierarchical clustering and heatmap visualization revealed robust separation between HF and NF, with no apparent outliers or confounding variables among patients (Fig. 2B). Additionally, patient age appeared to cluster within the HF group, suggesting that biological age contributes to differences in DNA methylation, as previously reported (18). However, the effects of age on differential DNA methylation were insufficient to describe the global epigenomic changes in HF relative to NF.

Fig. 2.

Whole genome DNA methylation analysis of human heart failure. A: distribution of differentially methylated 50-bp regions (DMRs) with respect to the genomic region 5 kB from transcription start sites (TSS), with dotted line reflecting the “proximal” promoter (1.5 kB from TSS). B: heatmap and hierarchical clustering via Euclidean distance of DMCs [determined using χ²-test with correction for false discovery (FDR) <0.01] in heart failure (HF) relative to nonfailing control hearts, as determined by whole genome bisulfite sequencing analysis. C: Pathway enrichment of genes with hypermethylated (C) and hypomethylated (D) promoter DMCs (determined using χ²-test with correction for FDR <0.01]. CON, control; DMCs, differentially methylated cytosine.

Gene set enrichment analysis (GSEA) was performed using genes with differentially methylated promoters. Before GSEA, however, we first separated the hypermethylated from hypomethylated DMCs to interpret the impact of methylation on the enriched pathway(s). This approach revealed a diametric metabolic shift, with hypermethylated gene promoters associated with mitochondrial compartmentalization and oxidative pathways (Fig. 2C). Among the hypermethylated genes associated with oxidative metabolic processes were acetyl-CoA acetyltransferase 1, isopentenyl-diphosphate Δ-isomerase, farnesyl diphosphate synthase, and 3-hydroxy-3-methylglutaryl-CoA synthase 1. In contrast, promoter hypomethylation enriched genes involved in glycolysis and anaerobic metabolic processes (Fig. 2D), where genes responsible for this enrichment included phosphofructokinase (PFKL and PFKP), enolase (ENO1, ENO2, and ENO3), and glyceraldehyde-3-phosphate dehydrogenase (GAPDH). These observations together suggest that differential DNA methylation accompanies the known shift in cardiac gene expression to reactivate “fetal-like” glycolytic preference in HF (47).

NRF1 motif interference by DNA hypermethylation.

Promoter DNA methylation has been found to impact gene expression via competition with CpG-containing transcription factor-binding sites (Fig. 3A) (39). To identify promoter response elements disproportionately affected by differential DNA methylation in HF, we analyzed the genomic positions of differential promoter methylation using the Hypergeometric Optimization of Motif Enrichment (15). Because CpG islands are conserved regions of sequence bias, we selected background regions with equivalent CpG content for use in the hypergeometric motif enrichment. In contrast, analysis of hypomethylated DMCs enriched the upstream transcription factor II-B (TFIIB) recognition element (BREu), a cis-regulatory element that primes transcription via TFIIB (27). In contrast, analysis of hypermethylated DMCs enriched the binding site for NRF1 (Fig. 3B), a nodal transcriptional regulator of mitochondrial biogenesis (45).

Fig. 3.

Identification of DNA methylation-sensitive motifs. A: model showing the functional impact of response element methylation. B: de novo motif enrichment of hypermethylated DMCs identifying nuclear respiratory factor 1 (NRF1) motif based on the JASPAR database of position-weighted matrixes (PWMs). C: motif enrichment analysis identifies upstream transcription factor (TF) II-B recognition element PWM as the most enriched among hypomethylated DMCs. D: pathway analysis of NRF1-binding sites, as determined by pathway enrichment for differential promoter methylation of NRF1 gene targets, as determined using NRF1 chromatin immunoprecipitation-sequencing data set performed by ENCODE Project Consortium (11a). DMCs, differentially methylated cytosine.

Whereas BREu hypomethylation reflected a nonspecific upstream priming of genes involved in anaerobic glycolysis, NRF1 binding is known to occur in a methylation-dependent manner to regulate downstream genes involved in mitochondrial energetics (10). Therefore, we used a NRF1 chromatin immunoprecipitation-sequencing data set generated by Myers et al. as part of the Encyclopaedia of DNA Elements (11a; GSE96424) to determine the putative functional consequences of DMC-associated NRF1 target silencing. Enrichment of genic and intergenic DMPs failed to enrich either NRF1 or BREu (see Supplemental Results), supporting the specificity of these transcriptional elements within the promoter region. We overlapped the empirically derived NRF1 binding sites with the genomic positions of promoter-associated DMCs that we had identified, followed by gene set enrichment. We found “fatty acid biosynthesis” and “fatty acid β-oxidation” pathways as enriched among genes containing hypermethylated NRF1-binding sites in their proximal promoters (Fig. 3D). These observations together suggest that, in human HF, DNA hypermethylation may control cardiac metabolism in part by disrupting NRF1-dependent oxidative metabolism.

Gene expression analysis validates transcriptional hallmarks of HF.

To identify changes to the cardiac epigenome that correspond with changes in gene expression, we performed RNA-sequencing analysis in the sample cardiac biopsies used for WGBS. As visualized with a volcano plot, a significant reduction of many mitochondrially encoded genes was observed (Fig. 4A), consistent with mitochondrial dysfunction described in HF (49). Hierarchical clustering and heatmap visualization revealed a clear separation between HF and NF, even at the P < 0.01 level and no apparent outliers (Fig. 4B). Quantification of β-myosin heavy chain [MHC (MYH7)], α-MHC (MYH6), atrial natriuretic peptide (NPPA), B-type natriuretic peptide (NPPB), and sarcoplasmic reticulum Ca²⁺-ATPase expression validated our analysis with robust induction of NPPA (28.6-fold, FDR <0.01) and NPPB (54.3-fold, FDR <0.01), with suppression of α-MHC (MYH6, 8.7-fold, FDR <0.01) (Fig. 4C). Additionally, to determine whether the changes in gene expression might represent alterations in tissue composition, we performed an in silico method of cellular deconvolution by comparing the expression profile of our samples with those of isolated cell types using Xcell (1). No differences were detected between failing and nonfailing hearts (Supplemental Table S6).

Fig. 4.

Transcriptional reprogramming of the failing human heart. A: volcano plot showing −log₁₀(P value) and log₂(fold change) of genes changing in heart failure (HF, n = 7) relative to nonfailing control hearts (NF, n = 3). B: heatmap and hierarchical clustering via Euclidean distance of differentially expressed genes [DEGs; determined based on a Wald test (P < 0.05) with Benjamini-Hochberg post hoc adjustment] in HF relative to NF, as determined by next-generation RNA-sequencing analysis. C: differential expression of cardiac biomarkers of heart failure [MYH7, MYH6, atrial natriuretic peptide (NPP) A, NPPB, sarcoplasmic reticulum Ca²⁺-ATPase (ATP2A2)] relative to NF. D: gene set enrichment analysis of differentially expressed genes, illustrating the DEGs responsible for enriching the “fatty acid β-oxidation” GO term. E: genes responsible for enriching the oxidative phosphorylation pathway, where blue color reflects decreased gene expression in HF relative to NF. CON, control; MYH, myosin heavy chain.

Gene set enrichment and pathway analysis of DEGs again demonstrated a metabolic shift consistent with robust suppression (z score = −5.0) of oxidative phosphorylation (Fig. 4D). When we examined the DEGs responsible for driving enrichment of oxidative phosphorylation, we found nuclear- and mitochondrially encoded genes that were all suppressed in HF (Fig. 4E): complex I (NDUFB4, NDUFB1, NDUFB6, NDUFB11, NDUFA3, NDUFA5, NDUFA12, ND1, ND2, ND3, NDUFV3, NDUFS8, and ND4L), complex II (SDHB), complex III (CYTB, UQCRFS1, and UQCRB), and complex IV (COX1, COX2, COX3, and COX4I2), and complex V (ATP5MC1 and ATP5MF). Together, these observations strongly demonstrate the presence of established transcriptional changes in HF, most notably the suppression of mitochondrial energetics (64).

Epigenetic cross talk in human HF.

Because de novo DNA methylation is an enzyme-mediated process, we sought to determine whether differential DNA methylation in HF might itself be transcriptionally regulated. We found increased expression of DNMT3A (1.3-fold) and DNMT3B (2.1-fold) (Fig. 5A), the DNMTs responsible for de novo DNA methylation, whereas DNMT1 was unchanged (refer to Supplemental Table S3). Expression of the enzymes responsible for oxidizing and thereby actively removing cytosine methylation from DNA (ten-eleven translocases) was also unchanged (Fig. 5B); however, reduced expression of an epigenetic coactivator required for DNA demethylation, growth arrest and DNA damage-inducible-β, was observed (34, 55).

Fig. 5.

Differential expression of DNA methyltransferases (DNMTs) in human heart failure. A: gene expression via RNA-sequencing normalized counts of de novo DNMT3A and DNMT3B. B: DNA demethylases ten-eleven transferase (TET) 1, TET2, TET3, and TET-binding protein growth arrest and DNA damage inducible-β (GADD45B). Differentially expressed class II histone-modifying deacetylases (HDACs, C), class III HDACs (“sirtuins,” D), and histone acetyltransferases (HATs, E) found to be differentially expressed in heart failure (HF, n = 7) relative to nonfailing control hearts (NF) (n = 3). CON, control; SIRT, sirtuin. Differential gene expression was based on Mann-Whitney nonparametric U-test with Benjamini-Hochberg adjustment, *P < 0.05.

Because of the known possibility of epigenetic cross talk in other biological settings (26), we inspected differential expression of histone-modifying acetyltransferases (HATs) and deacetylases (HDACs). We observed that several class II (HDAC4 and HDAC7) and III [sirtuin (SIRT) 1, SIRT2, SIRT7] HDACs were induced (Fig. 5, C and D), whereas several HATs were suppressed (Fig. 5E). These observations are consistent with prior reports demonstrating a central role for both nuclear (class II) and mitochondrial (class III) HDACs in the pathogenesis of HF (61). Taken together, these data support the presence of epigenetic cross talk between differential DNA methylation and histone acetylation dynamics.

Differential promoter methylation reflects metabolic changes of HF.

Whereas motif competition is one mechanism by which promoter-associated DNA methylation silences gene expression, regional methylation changes have been shown to reflect robust changes in genomic architecture via heterochromatin formation (42). To identify genes likely regulated by methylation-associated variations to gene accessibility, we examined the density of DMCs using a circular genome plot, which identified dense regions of differential methylation throughout the genome largely centered around CpG islands (Fig. 6A). Among the top differentially regulated genes, with the largest quantity of promoter DMCs, we found two members of the class II HDAC family (HDAC4 and HDAC7) associated with widespread promoter hypomethylation along with previously noted gene induction (Fig. 6B).

Fig. 6.

Combined analysis identifies epigenetically regulated metabolic intermediates. A: circular genome plot showing the regional methylation distribution and inversely regulated increased (yellow) and decreased (blue) differentially expressed genes (DEGs) and promoter differentially methylated 50-bp regions (DMRs). Gene symbols were highlighted, which met high-stringency selection of differential gene expression (|fold change| >2, FDR <0.05) and inversely changing promoter DMCs (|methylation change| >10%, FDR <0.05). B: bar graph showing inverse regulation of epigenetic regulators histone-modifying deacetylase (HDAC) 4, HDAC7, and sirtuin (SIRT) 1. Black bars illustrate differential promoter methylation, with gray axes reflecting fold changes in gene expression for heart failure relative to nonfailing control hearts. C: gene expression fold change (gray) and differential promoter methylation (black) of genes that encode metabolic intermediates. All bar graphs depict average fold changes or %methylation ± SD. DMCs, differentially methylated cytosine.

Because metabolic changes appeared both transcriptionally mediated and epigenetically encoded, we sought to determine whether differential promoter methylation of metabolic intermediates was associated with inverse gene expression in the same genes. Consistent with our hypothesis, inspection of metabolic genes identified a signature of hypomethylation and induction of glycolytic intermediates (PFKL, PFKFB3, and ENO2), with hypermethylation and transcriptional silencing of genes involved in tricarboxylic acid cycle (SDHB, IDH1), oxidative phosphorylation (COX17, COX6A2, UQCRB, ATP5MC1, NDUFS8, NDUFB6, NDUFV3, NDUFA5, and NDUFB5), and fatty acid oxidation (HADHB, HADHA, ACSL1, ACSL5, and ACAA2) pathways (Fig. 6C). A few exceptions to this global shift toward glycolytic metabolism existed, with tricarboxylic acid intermediate oxoglutarate dehydrogenase as the most profoundly induced (6.8-fold, FDR < 0.001) and hypomethylated (14%, P < 0.01). Nevertheless, these observations demonstrate that differential DNA methylation largely corresponds with the recurrence of a metabolic transcriptional program known to occur in HF (47).

Experimental cross-validation and in vitro analysis.

Although cardiac DNA methylation in HF has remained understudied, numerous gene expression analyses have been performed. We thus incorporated a recent transcriptome-wide analysis of human HF by Sweet et al. (56) to determine whether the DEGs associated with inverse promoter DMCs reflect conserved transcriptional changes of human HF. We identified 2,170 promoter DMCs associated with 740 conserved DEGs (Supplemental Table S7). Gene set enrichment analysis again exposed a disproportionate abundance of metabolic genes containing NRF1 promoter binding sites (Fig. 6, B and C). These observations support that our analysis offers generalizable insights into the transcriptional, and therefore likely the epigenetic, underpinnings of human end-stage heart failure.

Previous reports in mice have demonstrated a correlation between DNMT3A expression and metabolic gene expression during postnatal development (13, 42); however, no studies have yet provided experimental evidence to support a causal role of DNMT3A in the regulation of this cardiac gene program. We thus acquired and used Lipofectamine-based transfection of the pcDNA3.1-myc-DNMT3A overexpression plasmid (Addgene plasmid no. 35521) in H9c2 rat cardiomyoblasts (Fig. 7D), as previously described (5). We discovered that overexpression of DNMT3A is sufficient to suppress numerous oxidative metabolic genes in a manner consistent with hearts of HF subjects relative to NF. Specifically, overexpression of DNMT3A was sufficient to suppress ACSL1, ACSL2, HADHA, and NDUFA5 mRNA levels; by contrast, expression of NRF1 and COX6A2 was unaffected. These data therefore provide novel support that DNMT3A negatively regulates oxidative metabolic gene expression in cardiomyocytes.

Fig. 7.

Validation of DNA methyltransferase (DNMT) 3A-dependent differential gene expression in human heart failure (HF). A: Venn diagram illustrating the overlap of differentially expressed genes via RNA-sequencing analysis using Sweet et al. (56) and the current analysis (“Pepin et al.”). B: table summarizing conserved differential regulation of metabolic genes with inversely methylated promoters in human HF relative to nonfailing control hearts. C: gene set enrichment analysis of differentially coexpressed genes using the KEGG pathways database, reporting only statistically significant pathway enrichment (P < 0.05). D: empirical enrichment analysis comparing overlapping differentially expressed genes (DEGs) with the Encyclopaedia of DNA Elements consortium chromatin immunoprecipitation-sequencing data sets. E: quantification of mRNA via RT-PCR following 48 h transfection of empty pcDNA (pcDNA-EMPTY) or myc-tagged human DNMT3A (pcDNA-DNMT3A) in H9c2 rat cardiomyoblasts. ns, Not significant. *P < 0.05, statistical significance was assessed using Mann-Whitney nonparametric U-test, and data are reported as means ± SE.

DISCUSSION

Despite efforts to develop individualized therapies for HF, progress is hindered by the phenotypic and clinical heterogeneity noted, even among single-gene etiologies. Interim focus has resultantly shifted toward defining the convergent mechanisms of pathogenesis. Among its hallmark pathological features is a global reprogramming of cardiac metabolism, wherein the failing heart reactivates glycolytic genes while silencing those involved in oxidative metabolism (58). Although the molecular machinery responsible for governing this metabolic switch remains unknown, in the current study, we demonstrate that the cardiac epigenome contributes to this metabolic reprogramming.

The healthy heart accommodates an extraordinary range of hemodynamic requirements by rapidly adjusting both myocardial contractility and its metabolic resources. As a “metabolic omnivore,” the healthy adult heart adapts to both the nutrient microenvironment and its energetic demands, yet it primarily consumes fatty acids under baseline conditions (20). Specifically, the normoxic healthy heart derives over 95% of its ATP from oxidative phosphorylation, of which 70% originate from fatty acid substrates (19). HF has long been known to disrupt cardiac energy homeostasis by inducing mitochondrial toxicity (29), impairing the heart’s capacity for oxidative metabolism (53). Despite its increased energetic demands, the failing heart displays preference for glycolytic and ketone metabolism in a manner reminiscent of the developing fetal heart (2, 9, 43). This fetal-like metabolic preference for carbohydrates is mediated by the reactivation of glycolytic genes and concurrent suppression of oxidative enzymes (25, 57). In addition to this global shift in cardiac metabolism, the failing heart loses its ability to alter its preference of metabolic substrates, relying on glycolytic fuel to meet its high energetic requirements (23). Lowes et al. determined that, following a period of LVAD-induced cardiac unloading, the heart retains its preference for glycolytic metabolism (31). Although currently untested, this loss of metabolic “plasticity” may occur via epigenetic mechanisms, thereby preventing cardiac recovery following cardiac unloading.

In our analysis, we found the binding site for NRF1 as disproportionately hypermethylated within promoter-associated CpG sites. This finding is supported by numerous studies that establish NRF1 as a DNA methylation-sensitive regulator of mitochondrial biogenesis in other tissues (10, 17, 60). Although NRF1 expression was itself unchanged at the transcriptional level, NRF1 activity is primarily regulated by posttranslational addition of β-linked N-acetylglucosamine (O-GlcNAc) via both functional inhibition and protein destabilization by O-GlcNAc transferase (4, 50). This observation supports the coupled regulation of glucose and fatty acid substrate utilization, termed the Randle Cycle, since protein O-GlcNAcylation occurs in proportion to intracellular glucose levels as a nutrient sensor (28, 41). In HF, global increases in cardiac O-GlcNAcylation have been observed, since glucose is shunted through “bio-orthogonal” pathways of glycolysis (33). Altogether, these observations support the role of NRF1 as a positive regulator of cardiac oxidative metabolism that is epigenetically interrupted by DNA methylation in HF.

The current study identifies notable changes in cardiac DNA methylation in HF, but we recognize that the epigenome comprises numerous additional modifications that contribute to an overarching epigenetic landscape. DNA methylation has been shown to accompany macrostructural changes to chromatin architecture, thereby contributing to global changes to gene accessibility alongside direct modifications to histones (8, 26). In heart failure, epigenetic cross talk has been demonstrated between histone marks and noncoding RNAs (38). Consistent with this notion, we have previously supported the colocalization of the polycomb repressor complex 2 to differentially methylated gene promoters in ischemic cardiomyopathy (46). Therefore, future investigations should further characterize these epigenetic interactions in HF as a regulatory syncytium.

Although we provide novel evidence supporting an epigenetic contribution to the metabolic reprogramming in HF, key limitations exist that should be carefully considered in its interpretation. First, the cost of whole genome bisulfite sequencing and paired RNA sequencing limited our analysis of patient samples; although we provide cross-validation of RNA-sequencing data, large-cohort follow-up studies using WGBS or array-based methylation analyses will permit a more generalizable interpretation of our findings. The patients from whom cardiac samples were acquired reflect a diverse range of ages, comorbidities, and pharmacological regimens. Although we identify a conserved epigenomic signature in end-stage heart failure, understanding the contributions of age, medications, and comorbidities remains an ongoing focus of investigation that the current study was underpowered to explore. Given the demonstrated impact of heart failure on many cardiac cell types (37), isolated, cell type-specific experimentation should be performed to define individual contributions of these cell populations. Last, the current study provides correlative evidence linking promoter-associated DNA methylation to gene expression in human heart failure; therefore, future studies should examine the mechanistic relationship between DNA methylation and cardiac gene expression in HF, as well as potential effects of patient sex and race.

Conclusion.

Although future studies are needed to examine the causal nature of cardiac DNA methylation in HF pathogenesis, our findings support the overall hypothesis that cardiac metabolic genes are epigenetically reprogrammed in HF (Fig. 8). Specifically, we provide novel evidence that DNA methylation of the NRF1 response element likely contributes to oxidative metabolic gene suppression in end-stage human HF. Although these metabolic changes may initially protect the myocardium from energy collapse following hypoperfusion, the incorporation of epigenetic changes may explain its intractable clinical course.

Fig. 8.

Working model of metabolic reprogramming in heart failure. Heart failure is known to shift cardiac metabolism toward a more glycolytic state. Observations made in the current study suggest that convergent epigenetic mechanisms contribute to alterations in metabolic flux to both silence oxidative metabolic gene expression and reactivate a fetal-like metabolic state. HDAC, histone-modifying deacetylases; DNMT, DNA methyltransferase; NRF1, nuclear respiratory factor 1; FAO, fatty acid oxidation.

GRANTS

Financial support to O. Wever-Pinzon was provided by the University of Utah (Research Incentive Seed Grant Program) and the University of Utah Program in Personalized Health and National Center for Advancing Translational Sciences of the National Institutes of Health (NIH) under Award Number 1UL1-TR-002538. Financial support to S. G. Drakos was provided by American Heart Association Heart Failure Strategically Focused Research Network, 16SFRN29020000, NIH Grants R01-HL-135121 and R01-HL-132067, and the Nora Eccles Treadwell Foundation. Financial support for this work was provided to A. R. Wende by NIH Grant R01-HL-133011. Training support was provided to M. E. Pepin by NIH Grant F30-HL-137240.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

M.E.P., S.D., M.T.-F., J.C.F., A.R.W., and O.W.-P. conceived and designed research; M.E.P., C.-M.H., C.H.S., and O.W.-P. performed experiments; M.E.P. and C.-M.H. analyzed data; M.E.P., A.R.W., and O.W.-P. interpreted results of experiments; M.E.P. prepared figures; M.E.P. and A.R.W. drafted manuscript; M.E.P., S.D., M.T.-F., A.R.W., and O.W.-P. edited and revised manuscript; M.E.P., S.D., M.T.-F., C.H.S., J.C.F., A.R.W., and O.W.-P. approved final version of manuscript.