Prenatal Exposure to a Maternal High-Fat Diet Affects Histone Modification of Cardiometabolic Genes in Newborn Rats

Infants born to women with diabetes or obesity are exposed to excess circulating fuels during fetal heart development and are at higher risk of cardiac diseases. We have previously shown that late-gestation diabetes, especially in conjunction with a maternal high-fat (HF) diet, impairs cardiac functions in rat-offspring. This study investigated changes in genome-wide histone modifications in newborn hearts from rat-pups exposed to maternal diabetes and HF-diet. Chromatin-immunoprecipitation-sequencing revealed a differential peak distribution on gene promoters in exposed pups with respect to acetylation of lysines 9 and 14 and to trimethylation of lysines 4 and 27 in histone H3 (all, false discovery rate, FDR < 0.1). In the HF-diet exposed offspring, 54% of the annotated genes showed the gene-activating mark trimethylated lysine 4. Many of these genes (1) are associated with the “metabolic process” in general and particularly with “positive regulation of cholesterol biosynthesis” (FDR = 0.03); (2) overlap with 455 quantitative trait loci for blood pressure, body weight, serum cholesterol (all, FDR < 0.1); and (3) are linked to cardiac disease susceptibility/progression, based on disease ontology analyses and scientific literature. These results indicate that maternal HF-diet changes the cardiac histone signature in offspring suggesting a fuel-mediated epigenetic reprogramming of cardiac tissue in utero.


Introduction
Diabetes complicates an estimated 17.8% of all pregnancies in the U.S. [1,2]. Its rising incidence parallels an increasing trend in obesity, as 34.4% of US women of childbearing age are obese [3]. Infants born to women with diabetes or obesity are at higher risk of cardiovascular disease (CVD) at birth and throughout life [4][5][6][7], purportedly through fuel-mediated influences on the developing heart. Both diabetes and hyperlipidemia influence cardiometabolic health in adults, playing a pivotal role in the pathophysiology of CVD over time [8][9][10]. Using a rat model, we have already demonstrated that in utero exposure to diabetes, especially in conjunction with a maternal high-fat (HF) diet, impairs cardiac function through metabolic disturbances in the developing fetal heart [11]. The objective of this ( Figure 1). Thereafter, blood glucose levels were kept between 200-400 mg/dL and ketones were kept to a minimum through monitoring and twice daily sliding scale insulin administration. Offspring delivered normally on gestational day 22. Newborn rat hearts were collected from four different groups: controls (control diet and citrate buffer exposed or CC), diabetes-exposed (control diet and STZ exposed or CS), diet-exposed (high fat and citrate buffer exposed or HC), and combination-exposed (high fat and STZ exposed or HS). Hearts were snap frozen and stored at −80 °C until analyses. Figure 1. Study timeline. Female rats were randomized into four groups. Two groups were fed high fat diet (HF) and two control diet (CD) for 28 days prior to their breeding and throughout the pregnancy. On gestational day 14, one each of HF and CD group received intraperitoneal injection of Streptozotocin (STZ, 65 mg/kg) and the remaining two received citrate buffer (CB) as a control. Pups were normally delivered on gestational day 22 and sacrificed for heart collection.

Chromatin Immunoprecipitation (ChIP) Assay
Chromatin immunoprecipitation followed by sequencing (ChIP-seq) was performed to generate extensive genome-wide data sets profiling four selected histone modifications across the four exposure groups. Frozen hearts (3-4 from litter mates) were pooled (~250 mg) and homogenized followed by chromatin isolation using P-2001 ChromaFlashTM Chromatin Extraction Kit (Epigentek, Farmingdale, NY, USA). Chromatin was sheared using Episonic2000 Sonication System (Epigentek, Farmingdale, NY, USA) in 300 μL of ChIP buffer, followed by a quality control of sheared chromatin using fluorescence quantification. A minimum of 4 μL of sheared chromatin was purified to obtain DNA, which was eluted with 20 μL of water. DNA fragment (100-300 bp) was quality-checked using a bioanalyzer. For each ChIP reaction, 10 μg of chromatin, 3 μg of histone 3 trimethylated at lysine 4 (H3K4me3) polyclonal antibody or 3 μg of histone 3 trimethylated at lysine 27 (H3K27me3) antibody or 3 μg of histone acetylated at lysine 9 and 14 (H3K9/14ac) polyclonal ChIP-grade antibody (all antibodies were from Epigentek, Farmingdale, NY, catalogue numbers are: A4021, A4039, A4033) and 6 μL of protein A/G beads were used. ChIPed deoxyribonucleic acid (DNA) was eluted in 20 μL of water. Non-ChIPed DNA was used as input controls for the test groups. Since acetylation is a gene-activating mark irrespective of the associated lysine residue and the same antibody targeted both residues, observed acetylation peaks for lysine 9 and 14 are presented together and are referred to as H3Ac, hereafter. Therefore, a total of 12 biological datasets plus four input control datasets were generated.

ChIP Sequencing
A ChIP-seq library was prepared using the EpiNext ChIP-Seq High Sensitivity Kit with amplification, DNA end polishing and adaptor ligation, following the manufacturer's instruction. 10 nM of sample libraries were provided for next generation sequencing on a HiSeq 2500 (sequencing was carried out by Epigentek Inc., Farmingdale, NY, USA). The basic analysis of ChIP-Seq was based on the published protocol [3] utilizing Bowtie, version 1.0.0 [4] and MACS, version 2.1.0 [5]. Raw reads were quality checked using FastQC, version v0.10.1 [6] and mapped onto the rat RN5 genome sequence using Bowtie, version 1.0.0 [4]. The option of "-m1" was activated so that only uniquely mapped reads were allowed to map. The mapping results in SAM were converted to BAM and sorted according to coordinates using samtools, version 0.1.19 [7]. Mapping results of each ChIP sample and the input sample were subjected to ChIP enriched peak calling. The option of "-broad" was activated Figure 1. Study timeline. Female rats were randomized into four groups. Two groups were fed high fat diet (HF) and two control diet (CD) for 28 days prior to their breeding and throughout the pregnancy. On gestational day 14, one each of HF and CD group received intraperitoneal injection of Streptozotocin (STZ, 65 mg/kg) and the remaining two received citrate buffer (CB) as a control. Pups were normally delivered on gestational day 22 and sacrificed for heart collection.

Chromatin Immunoprecipitation (ChIP) Assay
Chromatin immunoprecipitation followed by sequencing (ChIP-seq) was performed to generate extensive genome-wide data sets profiling four selected histone modifications across the four exposure groups. Frozen hearts (3-4 from litter mates) were pooled (~250 mg) and homogenized followed by chromatin isolation using P-2001 ChromaFlashTM Chromatin Extraction Kit (Epigentek, Farmingdale, NY, USA). Chromatin was sheared using Episonic2000 Sonication System (Epigentek, Farmingdale, NY, USA) in 300 µL of ChIP buffer, followed by a quality control of sheared chromatin using fluorescence quantification. A minimum of 4 µL of sheared chromatin was purified to obtain DNA, which was eluted with 20 µL of water. DNA fragment (100-300 bp) was quality-checked using a bioanalyzer. For each ChIP reaction, 10 µg of chromatin, 3 µg of histone 3 trimethylated at lysine 4 (H3K4me3) polyclonal antibody or 3 µg of histone 3 trimethylated at lysine 27 (H3K27me3) antibody or 3 µg of histone acetylated at lysine 9 and 14 (H3K9/14ac) polyclonal ChIP-grade antibody (all antibodies were from Epigentek, Farmingdale, NY, catalogue numbers are: A4021, A4039, A4033) and 6 µL of protein A/G beads were used. ChIPed deoxyribonucleic acid (DNA) was eluted in 20 µL of water. Non-ChIPed DNA was used as input controls for the test groups. Since acetylation is a gene-activating mark irrespective of the associated lysine residue and the same antibody targeted both residues, observed acetylation peaks for lysine 9 and 14 are presented together and are referred to as H3Ac, hereafter. Therefore, a total of 12 biological datasets plus four input control datasets were generated.

ChIP Sequencing
A ChIP-seq library was prepared using the EpiNext ChIP-Seq High Sensitivity Kit with amplification, DNA end polishing and adaptor ligation, following the manufacturer's instruction. 10 nM of sample libraries were provided for next generation sequencing on a HiSeq 2500 (sequencing was carried out by Epigentek Inc., Farmingdale, NY, USA). The basic analysis of ChIP-Seq was based on the published protocol [25] utilizing Bowtie, version 1.0.0 [26] and MACS, version 2.1.0 [27]. Raw reads were quality checked using FastQC, version v0.10.1 [28] and mapped onto the rat RN5 genome sequence using Bowtie, version 1.0.0 [26]. The option of "-m1" was activated so that only uniquely mapped reads were allowed to map. The mapping results in SAM were converted to BAM and sorted according to coordinates using samtools, version 0.1.19 [29]. Mapping results of each ChIP sample and the input sample were subjected to ChIP enriched peak calling. The option of "-broad" was activated to optimize the calling algorithm for broad binding regions. The called peaks were annotated to the nearest TSS (Transcript Starting Site) using ChIPpeakAnno, version 3.2.0 [30]. ChIP quality control was performed using ChIPQC, version 1.2.2 [31]. Raw nucleotide sequences are available in the National Center for Biotechnology Information (NCBI) gene expression omnibus database under the accession number GSE84831.

Bioinformatics Analyses
Enriched peaks were called and then annotated to the nearest transcription start site of the genes. Peaks were visualized using Integrative Genomics Viewer [32]. Genomic distribution of peaks were summarized in a heat map using Seqminer version 1.3.3 [33]. First, the genomic reference peak set (TSS.rat.Rnor_5.0.bed) was imported. Then the mapping results (BAM) were loaded; read densities were extracted and clustered (linear Kmeans clustering) using the default parameters. Differential binding analyses was carried out to determine the differential enrichment of cardiac-specific genes due to four different diet-conditions in all regions and also in the promoter region specifically. Downstream validation was carried out for the promoter regions by identifying CpG islands using online database [34] followed by a gene ontology analysis [35] to determine which biological processes were triggered by the four different diet/diabetes conditions using online database GOEAST [36]. The rat genome database [37] was extensively used to annotate the gene, overlapping quantitative trait loci (QTL) and disease ontologies. Briefly, the disease association of each gene in the genome was derived using Disease Ontology and peer-reviewed evidence from GeneRIF [38]. A condensed version of the Disease Ontology, Disease Ontology Lite, was used for the statistical analysis [39].

RNA Isolation and RT-PCR
Rat hearts were homogenized using Precellys 24 lysis and homogenization system (Bertin Technologies, Rockville, MD, USA). RNA was extracted using the RNeasy Fibrous Tissue Mini kit (Qiagen, Germantown, MD, USA) according to protocol. Ribonucleic acid (RNA) integrity was assessed by electropherograms via 2100 BioAnalyzer (Agilent Technologies, Santa Clara, CA, USA) and demonstrated RNA Integrity Numbers (RINs) from 9.7 to 10. RNA concentrations and absorbance ratios were measured utilizing an Epoch spectrophotometer (BioTek, Winooski, VT, USA). Average 260/280 ratios were 2.10 with a range from 2.023 to 2.124. RNA concentration ranged from 645 to 2115 ng/µL. Thereafter, 1 µg of RNA was used to synthesize cDNA using the iScript cDNA synthesis kit (BioRad, Hercules, CA, USA). Quantitative PCR was performed using a standard TaqMan approach performed in ABI7500 qPCR system using the Absolute Blue qPCR Mix (ThermoFisher, Waltham, MA, USA). Beta-2-microglobulin was used as a reference gene for normalizing gene expression levels. Probe/primer sets were obtained from Integrated DNA Technologies (Coralville, IA, USA) and ThermoFisher (Waltham, MA, USA).

Statistical Analyses
We compared how histone landscape differs among pups born to a healthy dam with those born to a diabetic or HF-diet exposed pregnancy. We also evaluated a combined effect of GDM and HF exposure in utero. Statistical analyses were performed with Sigma Plot (Systat Software, Inc., San Jose, CA, USA). To compare four exposed-groups, a two-way ANOVA followed by a post-hoc Tukey's test was performed. When interaction was significant, one-way ANOVA with Tukey's HSD test was performed. Principal coordinate analysis was used to show the difference among three histone modifications. The genome mapping statistics were extracted using Picard, version 1.90 [40]. Differential binding analyses were carried out using a DiffBind Bioconductor R package [41]. The differential binding comparisons use two matrices at a time where one matrix is fixed and the other one varied between the two groups being compared. For double exposed (HS) pups, we considered diabetes exposed pups (CS) as the comparator to account for the effects of HF-diet in the presence of diabetes. Significance of the association of the mapped histone peaks in gene ontology and disease ontology were determined using Fisher's exact test. Data were expressed as mean ± SEM, unless otherwise mentioned. A value of p < 0.05 or FDR (false discovery rate) <0.1 was considered significant.

Maternal Late Gestation Diabetes Increases Heart Weight in Offspring
We compared the average weight of the neonate hearts that were subjected to ChIP-seq experiments. Similar observations were previously reported with echocardiographic data in larger cohorts of rat pups using the same animal model ( Figure 1) [11,24]. Although we did not observe any significant interaction effect on neonatal heart weight due to combined diet and diabetes exposure, diabetes-exposed (CS) pups had significantly higher average heart weight compared to controls (CC) (55.64 ± 1.95 mg vs. 48.94 ± 2.03 mg, p < 0.05). A similar effect was observed in combination-exposed (HS) pups when compared with the diet-matched controls (HC) (52.92 ± 1.62 mg vs. 44.91 ± 2.30 mg, p < 0.05) (Figure 2a). Diabetes-exposed pups also had a significantly higher heart to body weight ratio when compared to controls (0.0089 ± 0.0004 vs. 0.0071 ± 0.0002, p < 0.05) ( Figure 2b). These data corresponded to elevated maternal late gestation glucose levels induced by STZ when compared to their respective CC and HC diet-matched control groups (305.30 ± 92.01 mg/dL vs. 82.23 ± 3.92 mg/dL and 295.42 ± 39.27 mg/dL vs. 88.30 ± 2.61 mg/dL; both p < 0.05) (Figure 2c). We did not observe any significant interaction effect on maternal late gestation glucose levels due to combined exposure. determined using Fisher's exact test. Data were expressed as mean ± SEM, unless otherwise mentioned. A value of p < 0.05 or FDR (false discovery rate) < 0.1 was considered significant.

Maternal Late Gestation Diabetes Increases Heart Weight in Offspring
We compared the average weight of the neonate hearts that were subjected to ChIP-seq experiments. Similar observations were previously reported with echocardiographic data in larger cohorts of rat pups using the same animal model ( Figure 1) [1,2]. Although we did not observe any significant interaction effect on neonatal heart weight due to combined diet and diabetes exposure, diabetes-exposed (CS) pups had significantly higher average heart weight compared to controls (CC) (55.64 ± 1.95 mg vs. 48.94 ± 2.03 mg, p < 0.05). A similar effect was observed in combination-exposed (HS) pups when compared with the diet-matched controls (HC) (52.92 ± 1.62 mg vs. 44.91 ± 2.30 mg, p < 0.05) (Figure 2a). Diabetes-exposed pups also had a significantly higher heart to body weight ratio when compared to controls (0.0089 ± 0.0004 vs. 0.0071 ± 0.0002, p < 0.05) ( Figure 2b). These data corresponded to elevated maternal late gestation glucose levels induced by STZ when compared to their respective CC and HC diet-matched control groups (305. 30 (Figure 2c). We did not observe any significant interaction effect on maternal late gestation glucose levels due to combined exposure. Physiological characteristics of pups and corresponding mothers. Higher neonatal heart weight (a) and neonatal heart to body weight ratio (b) that correspond to elevated maternal late gestation glucose levels (c) in diabetes exposed group; n = 10-12 (a,b), n = 2 (c). * p < 0.05, CC, controls; CS, diabetes exposed; HC, HF-diet exposed; HS, combination exposed.

In Utero Exposure to Diabetes or HF-Diet Associates with Differential Histone Modifications in Offspring Cardiac Tissue
ChIPed DNA sequence revealed differential peak distribution among the diet groups. The highest concentration of enriched peaks (around 40% enrichment compared to input DNA) were overlapping or upstream (−5 kb or gene promoters) of the transcription start site (TSS) for all three modifications in the treatment groups ( Figures A1-A3). About 29, 11, and 36% of approximately 41,000; 73,000; and 54,000 enriched peaks for H3Ac, H3K27me3, and H3K4me3 respectively overlapped to known exonic regions when mapped to the rat genome (data not shown). The genome occupancy of the gene-activating histone marks (H3Ac, H3K4me3) were distinct from the suppressive mark H3K27me3 by 24% and 20% on axes 1 and 2, respectively ( Figure 3). The 5 kb upstream regions in the combination-exposed newborn heart genome had relatively higher concentrations of gene-activating marks (H3Ac and H3K4me3) and lower concentration of repressive marks (H3K27me3) as measured in terms of mean peak densities (tag/50 bp) compared to diabetes or diet-exposed hearts as shown by the Seqminer density plots (Figure 4). In the diet-exposed group, the two gene-activating marks, H3Ac and H3K4me3, showed small but higher association between each other (Pearson's r = 0.29) than between the gene-activating and gene suppressive methylation marks (Pearson's r = 0.09) ( Figure A2). Together, the effect of the diabetes and HF diet affected the Figure 2. Physiological characteristics of pups and corresponding mothers. Higher neonatal heart weight (a) and neonatal heart to body weight ratio (b) that correspond to elevated maternal late gestation glucose levels (c) in diabetes exposed group; n = 10-12 (a,b), n = 2 (c). * p < 0.05, CC, controls; CS, diabetes exposed; HC, HF-diet exposed; HS, combination exposed.

In Utero Exposure to Diabetes or HF-Diet Associates with Differential Histone Modifications in Offspring Cardiac Tissue
ChIPed DNA sequence revealed differential peak distribution among the diet groups. The highest concentration of enriched peaks (around 40% enrichment compared to input DNA) were overlapping or upstream (−5 kb or gene promoters) of the transcription start site (TSS) for all three modifications in the treatment groups ( Figures A1-A3). About 29, 11, and 36% of approximately 41,000; 73,000; and 54,000 enriched peaks for H3Ac, H3K27me3, and H3K4me3 respectively overlapped to known exonic regions when mapped to the rat genome (data not shown). The genome occupancy of the gene-activating histone marks (H3Ac, H3K4me3) were distinct from the suppressive mark H3K27me3 by 24% and 20% on axes 1 and 2, respectively ( Figure 3). The 5 kb upstream regions in the combination-exposed newborn heart genome had relatively higher concentrations of gene-activating marks (H3Ac and H3K4me3) and lower concentration of repressive marks (H3K27me3) as measured in terms of mean peak densities (tag/50 bp) compared to diabetes or diet-exposed hearts as shown by the Seqminer density plots ( Figure 4). In the diet-exposed group, the two gene-activating marks, H3Ac and H3K4me3, showed small but higher association between each other (Pearson's r = 0.29) than between the gene-activating and gene suppressive methylation marks (Pearson's r = 0.09) ( Figure A2). Together, the effect of the diabetes and HF diet affected the gene-upregulatory histone marks in a way distinct from its influence on the gene-suppressive H3K27me3 mark.
Nutrients 2017, 9,407 6 of 20 gene-upregulatory histone marks in a way distinct from its influence on the gene-suppressive H3K27me3 mark. Axes showing % of variations. CC, controls; CS, diabetes exposed; HC, HF-diet exposed; HS, combination exposed. . High fat and diabetes-induced genome wide differential histone marks. Heat maps showing read density along with corresponding mean density (tag/50 bp) plots in ±5 kb of any known transcriptional start site for H3Ac, H3K4me3, and H3K27me3 modifications clustered and visualized as four panels in the order of CC (control), CS (diabetes exposed), HC (diet exposed), and HS (combination exposed) from left to right. From the vertical view, clusters with the central density profile are located at the top, followed by clusters with peaks in the −5 kb region (marked within the black lines), those with peaks in the +5 kb region, those with diffuse read density, and those without significant read density. Axes showing % of variations. CC, controls; CS, diabetes exposed; HC, HF-diet exposed; HS, combination exposed.
Nutrients 2017, 9, 407 6 of 20 gene-upregulatory histone marks in a way distinct from its influence on the gene-suppressive H3K27me3 mark. Axes showing % of variations. CC, controls; CS, diabetes exposed; HC, HF-diet exposed; HS, combination exposed. . High fat and diabetes-induced genome wide differential histone marks. Heat maps showing read density along with corresponding mean density (tag/50 bp) plots in ±5 kb of any known transcriptional start site for H3Ac, H3K4me3, and H3K27me3 modifications clustered and visualized as four panels in the order of CC (control), CS (diabetes exposed), HC (diet exposed), and HS (combination exposed) from left to right. From the vertical view, clusters with the central density profile are located at the top, followed by clusters with peaks in the −5 kb region (marked within the black lines), those with peaks in the +5 kb region, those with diffuse read density, and those without significant read density. . High fat and diabetes-induced genome wide differential histone marks. Heat maps showing read density along with corresponding mean density (tag/50 bp) plots in ±5 kb of any known transcriptional start site for H3Ac, H3K4me3, and H3K27me3 modifications clustered and visualized as four panels in the order of CC (control), CS (diabetes exposed), HC (diet exposed), and HS (combination exposed) from left to right. From the vertical view, clusters with the central density profile are located at the top, followed by clusters with peaks in the −5 kb region (marked within the black lines), those with peaks in the +5 kb region, those with diffuse read density, and those without significant read density.

Differential Binding Analyses of Diabetes and Diet Induced Histone Modifications
Mapping the differential histone modifications to the annotated rat genome (version 5.0) helped identify associated genes (referred to as candidate genes) and to further characterize the peak occupancy in the −5 kb region of TSS as well as all regions using differential binding analyses (Table 1). In absolute terms, however, for gene-activating H3Ac, the total number of peaks decreased by 52% in the diabetes-exposed group and 53% in the diet-exposed group respectively compared to controls. For H3K4me3, which is also an activating mark, the diabetes-exposed offspring acquired an additional 33% and diet-exposed offspring acquired an additional 73% more peaks. For repressive H3K27me3, diabetes-exposed and diet-exposed offspring acquired an additional 56% and 70% peaks, respectively ( Figure 5). Compared to controls, the combination-exposed hearts had an additional 89% K4me3 peaks ( Figure 5). CC, controls; CS, diabetes exposed; HC, HF-diet exposed; HS, combination exposed.

Differential Binding Analyses of Diabetes and Diet Induced Histone Modifications
Mapping the differential histone modifications to the annotated rat genome (version 5.0) helped identify associated genes (referred to as candidate genes) and to further characterize the peak occupancy in the −5 kb region of TSS as well as all regions using differential binding analyses (Table  1). In absolute terms, however, for gene-activating H3Ac, the total number of peaks decreased by 52% in the diabetes-exposed group and 53% in the diet-exposed group respectively compared to controls. For H3K4me3, which is also an activating mark, the diabetes-exposed offspring acquired an additional 33% and diet-exposed offspring acquired an additional 73% more peaks. For repressive H3K27me3, diabetes-exposed and diet-exposed offspring acquired an additional 56% and 70% peaks, respectively ( Figure 5). Compared to controls, the combination-exposed hearts had an additional 89% K4me3 peaks ( Figure 5). Table 1. Genome-wide, differential binding of enriched histone modification peaks. CC, controls; CS, diabetes exposed; HC, HF-diet exposed; HS, combination exposed.

Figure 5.
Enriched histone modification peaks overlapping 5 kb upstream region of known transcriptional start site. All H3Ac (Ac), H3K4me3 (K4), and H3K27me3 (K27) enriched peaks detected on the rat genome are shown for each exposure group. CC, controls; CS, diabetes exposed; HC, HF-diet exposed; HS, combination exposed.
Differential binding analyses of 5 kb upstream of TSS revealed two and three H3Ac peak losses respectively in the diabetes-exposed and diet-exposed groups compared to controls (both, FDR < 0.1, Table 1, Supplementary Table S1). For H3K4me3 modification, 2 and 28 peaks were gained in diabetes-exposed and diet-exposed groups, respectively (all, FDR < 0.1, Table 1, Supplementary Table  S1). Of note, the 28 differentially enriched H3K4me3 peaks in the diet-exposed group were annotated to 28 candidate genes, 15 of which were related to 'metabolic process' in the gene ontology analysis (Tables 2 and 3, Figure 6). This indicates that in utero exposure to a maternal HF-diet induced epigenetic changes that contributes to fetal programming of cardiac metabolism. H3K27me3 showed a mixed response: six peaks were significantly gained for the diabetes-exposed and diet-exposed groups, while they also lost three and one peaks, respectively (Table 1). Differentially bound peaks in regions other than −5 kb of TSS are shown in Table 1, but were not further analyzed in gene ontology or disease ontology.

Figure 5.
Enriched histone modification peaks overlapping 5 kb upstream region of known transcriptional start site. All H3Ac (Ac), H3K4me3 (K4), and H3K27me3 (K27) enriched peaks detected on the rat genome are shown for each exposure group. CC, controls; CS, diabetes exposed; HC, HF-diet exposed; HS, combination exposed. Differential binding analyses of 5 kb upstream of TSS revealed two and three H3Ac peak losses respectively in the diabetes-exposed and diet-exposed groups compared to controls (both, FDR < 0.1, Table 1, Supplementary Table S1). For H3K4me3 modification, 2 and 28 peaks were gained in diabetes-exposed and diet-exposed groups, respectively (all, FDR < 0.1, Table 1, Supplementary  Table S1). Of note, the 28 differentially enriched H3K4me3 peaks in the diet-exposed group were annotated to 28 candidate genes, 15 of which were related to 'metabolic process' in the gene ontology analysis (Tables 2 and 3, Figure 6). This indicates that in utero exposure to a maternal HF-diet induced epigenetic changes that contributes to fetal programming of cardiac metabolism. H3K27me3 showed a mixed response: six peaks were significantly gained for the diabetes-exposed and diet-exposed groups, while they also lost three and one peaks, respectively (Table 1). Differentially bound peaks in regions other than −5 kb of TSS are shown in Table 1, but were not further analyzed in gene ontology or disease ontology.  Bolded genes either gained activating H3K4me3 or lost repressive H3K27me3 mark. CC, controls; CS, diabetes exposed; HC, HF-diet exposed; HS, combination exposed. Table 3. Gene ontology of identified genes in the HF-diet exposed group.  Figure 6. Gene ontology classification of genes with H3K4me3 enrichment histone modification due to high fat exposure. A representative gene ontology analysis of candidate genes corresponding to H3K4me3 peak enrichment in the diet exposed group (HC) is shown. Gene ontology of 'metabolic process' is chosen for the downstream analyses.

Quantitative Association of Diet and Diabetes Induced Histone Modifications with Candidate Genes and QTLs Related to Cardiometabolic Functions
We carried out a quantitative genetic analysis using histone modification occupancy levels on the identified genomic regions focusing on cardiac or metabolic function related genes. Gene ontology revealed that Hspa1a and Hspa1b are associated with "primary metabolic process" against HF diet-induced H3K4me3 modification (log odd ratio 1.38, FDR = 0.06) ( Table 3). Also, "positive regulation of cholesterol biosynthetic process" (FDR = 0.0295) was associated to Fdps and Por (Table 3  and Supplementary Table S2). Although, the candidate genes belonged to distant genomic regions, 15 of them overlapped with a total of 455 QTLs, primarily consisting of 88 QTLs for blood pressure (19.34%), 24 QTLs for body weight (5.27%), and 20 QTLs for serum cholesterol (4.40%) ( Table 4). Particularly, Fdps overlapped with a total of 33 QTLs, of which 22 (66.67%) were related to blood pressure. Atp5g3 overlapped with a total of 22 QTLs, of which 9 (40.91%) were linked to body weight. Similarly, Por and Ero1lb overlapped with a total of 12 and 19 QTLs, respectively, out of which 2 (16.67%) and 3 (15.79%) were related to serum cholesterol, respectively, ( Table 4 and Supplementary Table S3), suggesting the putative role of HF diet in hypercholesterolemia. Further, enrichment of genes in disease ontology revealed that the Hspa1a gene was related to multiple CVD indications such as coronary disease, atherosclerosis, and acute coronary syndrome, as well as disease progression of systolic heart failure (Supplementary Table S4). Hspa1b, a close relative of Hspa1a, was associated to obesity, type 2 diabetes mellitus, and hyperlipidemia. Fold enrichment (using FunDO [42] on 20 April 2016) of these two genes against histone peak occupancy in diet and diabetes exposed pups comparing with the control pups were 17.099 against type 2 diabetes mellitus (FDR = 0.018). Further, H6PD was related to experimental diabetes mellitus and inborn errors on metabolism (Supplementary Table S4), supporting the potential role of prenatal HF diet either alone or in combination with maternal diabetes on the fetal metabolic programming of adult-onset diabetes and CVD. However, the presented data from this exploratory study is newborn cardiac-tissue specific. The proof of concept generated here will be reexamined in the future in other physiologically relevant tissues in a time dependent manner through adulthood of the offspring generation. Table 4. Quantitative trait loci (QTL) related to metabolic processes that correspond to differentially bound histone peaks in the HF-diet exposed group.

Gene Expression Validation of Selected Candidate Genes
Gene expression validation was carried out for selected genes for which promoters gained epigenetic mark H3K4me3 (the highest differential peak enrichment was observed for this mark as shown in Tables 1 and 2) compared to the respective control group. Gene selection was based on two criteria: those with higher levels of changes and also with functional relevance to our previously reported phenotype [11]. Due to the gains of the H3K4me3 mark in the diet exposed group for all genes and in the diabetes exposed group for one of the five selected genes (Hsp1a1), expression levels of these selected genes were anticipated to be upregulated in a corresponding manner. qRT-PCR results showed a trend for increased expression for all five genes in diet-exposed group as expected, with statistical significance achieved for Atp5g3 only. Likewise, for Hsp1a1 an increased expression trend was observed in the diabetes exposed group (Figure 7). A small sample size for newborn hearts (n = 4) may have, at least partially, contributed to the lack of statistical significance for some of the candidate genes. Although, the combination exposed group did not always show a synergistic impact on gene expression, this is in line with the observations for epigenetic marks ( Figure 5). Additional investigation is needed in the future to understand the observations for the combination exposed group.

Discussion
Nutrient-mediated fetal metabolic programming due to a maternal HF diet and diabetes increases the risk of CVD at birth and throughout life. To our knowledge, this is the first report of genome-wide mapping of histone modifications in the cardiac tissue of offspring exposed to diabetic pregnancy and a prenatal HF diet, and the results suggest an epigenetic basis for fuel-mediated cardiac tissue reprogramming in utero. Furthermore, these results support our previous phenotypic observations that a maternal HF diet in conjunction with maternal diabetes causes mitochondrial dysfunction with impaired glycolysis, oxidative phosphorylation, and fatty acid oxidation in the exposed offspring heart at birth [11]. Cellular bioenergetics play a pivotal role in heart health, thus, understanding how these and additional candidate histone marks are causally linked to disease development could allow precise early-detection and prevention strategies for cardiometabolic dysfunction at birth and in later life.
Histone modifications are typically reversible and precede more permanent gene-suppressive DNA methylation [22] and, for these reasons, are ideal candidates for preventive interventions. Among various known PTMs, methylation, and acetylation of histone H3 lysine residues have been the most thoroughly investigated in terms of biological implications. In this pilot work, we presented four major histone PTMs that involve H3 methylation and acetylation and are well-characterized in both human and rat [43][44][45]. Histone acetylation at all lysine residues opens and activates chromatin by responding to acetyl-CoA levels, which relates to the metabolic state of the host. Therefore, characterizing histone acetylation following metabolic perturbations, as in our in-utero model, may provide a mechanism for cells to initiate metabolic adaptations [46]. In contrast, site-specific methylations can have both an activating and a suppressive role in gene transcription and are among the most stable, frequently inherited PTMs. We chose to study the histone trimethylation marks at lysines 4 and 27, because we wanted to study both an activating (H3K4me3) and a repressive (H3K27me3) methylation mark. In particular, changes in H3K4me3 levels can have stable, longlasting effects over several generations, even after the initial source of perturbation is no longer present [14,44,45,47].

Discussion
Nutrient-mediated fetal metabolic programming due to a maternal HF diet and diabetes increases the risk of CVD at birth and throughout life. To our knowledge, this is the first report of genome-wide mapping of histone modifications in the cardiac tissue of offspring exposed to diabetic pregnancy and a prenatal HF diet, and the results suggest an epigenetic basis for fuel-mediated cardiac tissue reprogramming in utero. Furthermore, these results support our previous phenotypic observations that a maternal HF diet in conjunction with maternal diabetes causes mitochondrial dysfunction with impaired glycolysis, oxidative phosphorylation, and fatty acid oxidation in the exposed offspring heart at birth [11]. Cellular bioenergetics play a pivotal role in heart health, thus, understanding how these and additional candidate histone marks are causally linked to disease development could allow precise early-detection and prevention strategies for cardiometabolic dysfunction at birth and in later life.
Histone modifications are typically reversible and precede more permanent gene-suppressive DNA methylation [22] and, for these reasons, are ideal candidates for preventive interventions. Among various known PTMs, methylation, and acetylation of histone H3 lysine residues have been the most thoroughly investigated in terms of biological implications. In this pilot work, we presented four major histone PTMs that involve H3 methylation and acetylation and are well-characterized in both human and rat [43][44][45]. Histone acetylation at all lysine residues opens and activates chromatin by responding to acetyl-CoA levels, which relates to the metabolic state of the host. Therefore, characterizing histone acetylation following metabolic perturbations, as in our in-utero model, may provide a mechanism for cells to initiate metabolic adaptations [46]. In contrast, site-specific methylations can have both an activating and a suppressive role in gene transcription and are among the most stable, frequently inherited PTMs. We chose to study the histone trimethylation marks at lysines 4 and 27, because we wanted to study both an activating (H3K4me3) and a repressive (H3K27me3) methylation mark.
In particular, changes in H3K4me3 levels can have stable, long-lasting effects over several generations, even after the initial source of perturbation is no longer present [14,44,45,47].
Increased occupancy of H3K4me3 was observed in the promoter regions of various genes associated with metabolic stress and cardiac dysfunction in the HF-diet group, including those known to be involved in mitochondrial injury (Atp5g3, [48]), autophagy (Ei24, [49]), experimental diabetes mellitus (H6pd, [50], Erol1b, [51]), percentage body fat (SLC11A2, [52]), oxidation of inflammatory lipid substrates (CYP4F, [53]), and hypertrophic cardiomyopathy and chronic heart failure (Trim63, [54]). Of significant interest is increased H3K4me3 on promoter regions of the Atp5g3 gene. This gene encodes a portion of ATP synthase subunit c, part of the mitochondrial proton channel, and regulates both ATP synthesis and proton leak to influence the balance between cellular energy production from respiration and ROS production [55]. Using qPCR, we confirmed a significant cardiac-tissue specific increase in expression of ATP5G3 in HF diet and combination-exposed newborns. Interestingly, hearts of offspring in these two groups also had significantly more lipid peroxidation [11]. Findings suggest a role for H3K4me3 of ATP5g3 in proton leak and oxidative stress in the developing heart. Our predicted gene expression results are further supported by proteomics literature, such as ATP synthase protein [56,57] and LIM-binding domain (LBD) protein [57,58] in a similar diabetes-induced rat model, and heat shock proteins (HSPs) in human patients with heart failure [59].
Two functionally related genes, Hspa1a and Hspa1b, also showed increased H3K4me3 occupancy in both the HF and diabetic groups. Hspa1a upregulation is linked with HF diet associated oxidative stress in patients with metabolic syndrome [60], correlates with HbA1c levels in maternal diabetes [61], and is proposed as an independent prognostic marker in patients with heart failure and cardiac arrest [59,62]. Our disease ontology analysis comparing HF-diet-induced, H3K4me3-occupied gene promoters with those of the control group linked Hsp1a1 and Hspa1b activation with severe disease progression in systolic heart failure and CVD. Although qPCR did not confirm a tissue-specific increase in expression of Hspa1a in newborn offspring, H3K4me3 of Hsp1a1 may still serve as a valuable prognostic indicator for heart health later in life. Following expression over time may be important for further understanding the role of heat shock proteins in developmentally programmed CVD following in utero exposure to maternal HF diet.
In addition, enrichment of the Por and Fdps gene promoters with the H3K4me3 marks was correlated with "metabolic process" and "positive regulation of cholesterol biosynthesis process" in gene ontology analyses. Por is involved in cholesterol and bile acid synthesis [63], while Fdps activation is associated with hypercholesterolemia [64]. Enrichment was also found on the Cyp4f18 promoter, a member of the cytochrome p450 complex involved in regulation of lipotoxicity. Interestingly, all of these annotated genes against differential H3K4me3 peak enrichment in the HF-diet group overlapped with 'blood pressure' QTLs, while most of them overlapped with 'cholesterol synthesis' QTLs. Newborn rat offspring exposed to a maternal HF diet had significantly increased cardiac lipid deposition [11]. While qPCR did not confirm an increase in tissue-specific Fdps expression possibly due to a smaller sample size, there was a trend towards increased cardiac expression of Cyp4f18 in HF-diet exposed offspring. Future studies should investigate the role of epigenetics in cardiac lipotoxicity through these or other candidate genes.
Since the activating H3k4me3 and repressive H3K27me3 modifications coexist to form bivalent promoters in early life, promoters may also respond to developmental stimuli and change their histone modification state through H3K27me3 demethylation [47]. When we compared offspring exposed to late-gestation diabetes born to dams on control vs. a HF diet, a lysine demethylase enzyme-encoding gene, Kdm6b, was associated with significant H3K27me3 loss, suggesting potential upregulation of this enzyme. Other notable candidate genes with loss of H3K27me3 occupancy are ABCB9 (associated with type 2 diabetes [65]); Sts (associated with HF diet and ob/ob models of obesity and type 2 diabetes [66]); and Ucn2 (associated with heart failure patients [67]). Additionally, there was significant H3K27me3 loss in the promotor region of ATP5G2 which encodes another subunit in the ATP synthase proton pore. Loss of H3K27me3 could lead to tissue-specific upregulation and potentiate the effects of H3K4me3 induced ATP5G3 expression found in HF-diet exposed offspring. This may explain why combination-exposed offspring have the most severe mitochondrial dysfunction [11]. We also observed differential H3K27me3 marks in HF diet and STZ in combination when compared to STZ treatment alone (Table 2), suggesting HF diet has a major role in the combined treatment.
Enrichment of H3K27me3 indicates potential gene silencing, and was observed in the HF-diet exposed offspring for AK3, a mitochondrial matrix protein that is highly expressed in the heart [68]) and negative regulation of adipogenesis (STAT5B, [69]). Expression of these genes may further explain the lipotoxic phenotype found in the hearts of HF diet-exposed newborn offspring [11]. In this context, we did not observe higher body weight in the HF-diet exposed newborns, although the hearts were larger and had significant lipid deposition [11]. Of note here is a previous study that used a similar rat model and found that HF-diet exposure in utero led to higher body weights and greater adiposity much later in the life of the offspring, in spite of having similar birth weights as control rats [70]. Histone modification predominantly H3K4 and H3K9 methylation closely links an oxidative stress pathway with underlying biochemical mechanisms of cardiovascular disease [71]. The fetal exposure to elevated glucose and free fatty acids likely generate cardiac mitochondrial reactive oxygen species, resulting in oxidative damage to the cardiomyocyte mitochondria. Although the mitochondrial numbers may increase as a part of compensation, overall mitochondrial efficiency goes down in diabetic hearts, leading to mitochondrial dysfunction playing a potentially pathogenic role in elevating risk of adult onset diabetes [72]. This is consistent with our previous findings [11] where we reported that HF diet exposure was associated with a higher copy number, but diabetes exposure alone (CD-STZ) was associated with a lower mitochondrial copy number than control. Overall, in the previous study we showed a significant interaction effect causing the poorest mitochondrial respiration in HF diet and STZ treatment combined group in the same rat model used in this study [11].

Conclusions
This study is the first to show that prenatal exposure to HF diet affects acetylated and methylated chromatin states in a site-specific manner along the rat genome in the offspring cardiac tissue and that these changes are present at birth. Together with others reporting liver-tissue-specific fetal metabolic programming [73][74][75], the results discussed here clearly demonstrate that epigenetic cues for metabolic disease risk are established in utero. These results further corroborate our previous reports on hyperinsulinemia and hyperlipidemia observed in rat offspring born to diabetic and/or HF-fed dams, particularly during pregnancy [11]. However, here we did not observe any synergy between prenatal exposure of a HF-diet and diabetes. The changes resulting from HF and diabetes exposure were distinct and the most hits against "metabolic process" in gene ontology analyses of differentially bound peaks came from H3K4me3 gains in the HF-exposed (HC) group when compared to the control (CC) group. Nevertheless, the results explain variable phenotypic findings in offspring from various groups. Moreover, data presented emphasize the critical role of the prenatal environment in epigenetically laying the foundation for the lifelong cardiometabolic health of offspring born following pregnancy affected by HF diet, diabetes, or both. Knowledge gained from this study will aid in a deeper mechanistic investigation of developmental programming of cardiometabolic disease and, if translatable, will set the foundation for the early identification and prevention of heart disease in at-risk infants. Understanding the role of diet-induced early epigenetic cues in the pathogenesis of disease is essential to halting the progression before symptoms ever develop and could be a key to decreasing the global burden of metabolic diseases.
Supplementary Materials: The following are available online at www.mdpi.com/2072-6643/9/4/407/s1. These are large Excel Spreadsheets that do not convert to PDF format correctly. Table S1: Differential binding analyses for upstream and overlap-start peaks in different diet groups. Table S2: Gene ontologies of candidate genes with H3K4me3 modifications due to maternal high fat diet. Table S3: Quantitative trait loci (QTL) overlapped with candidate genes with H3K4me3 modifications due to maternal high fat diet. Table S4: Disease ontologies of candidate genes with H3K4me3 modifications due to maternal high fat diet. Appendix A Figure A1. Peaks detection in ChIP sequencing: Representative genomic region of a negative strand (chromosome 7) near the transcriptional start site of Atp5g2 gene, illustrating 5 kb upstream peaks in cyan (acetylation), red (H3K427me3), blue (H3K4me3), and grey (Input control) colors. CC, controls; CS, diabetes exposed; HC, HF-diet exposed; HS, HF-diet and diabetes exposed or combination exposed. Figure A1. Peaks detection in ChIP sequencing: Representative genomic region of a negative strand (chromosome 7) near the transcriptional start site of Atp5g2 gene, illustrating 5 kb upstream peaks in cyan (acetylation), red (H3K427me3), blue (H3K4me3), and grey (Input control) colors. CC, controls; CS, diabetes exposed; HC, HF-diet exposed; HS, HF-diet and diabetes exposed or combination exposed.  Figure A2. Genome wide differential histone marks in H3Ac, H3K4me3, and H3K27me3 modifications. Read density in ±5 kb of any known transcriptional start site for each marker are clustered and visualized in a heat map that consists of three panels in the order of Ac, K27, and K4 (from left to right). A selected cluster with differential density profile centralized at the transcription start sites are marked under the blue lines, which shows a slight correlation (Pearson coefficient, r = 0.29) between K4 and Ac modifications while no correlation (Pearson coefficient, r = 0.09) is observed between K4 and K27 modifications due to maternal high fat( HF) diet. HC, HF-diet exposed treatment. Figure A2. Genome wide differential histone marks in H3Ac, H3K4me3, and H3K27me3 modifications. Read density in ±5 kb of any known transcriptional start site for each marker are clustered and visualized in a heat map that consists of three panels in the order of Ac, K27, and K4 (from left to right). A selected cluster with differential density profile centralized at the transcription start sites are marked under the blue lines, which shows a slight correlation (Pearson coefficient, r = 0.29) between K4 and Ac modifications while no correlation (Pearson coefficient, r = 0.09) is observed between K4 and K27 modifications due to maternal high fat( HF) diet. HC, HF-diet exposed treatment. Figure A3. Genic distributions of differential histone marks. (a) H3Ac, (b) H3K4me3, and (c) H3K27me3 in four diet groups around the TSS in four diet groups are plotted. Distances to the nearest TSS for peaks identified and plotted in pie charts. CC, controls; CS, diabetes exposed; HC, HF-dietexposed; HS, diet and diabetes exposed or combination exposed; TSS: transcription start site; US: upstream; DS: downstream; IF: include feature; I: inside; OE: overlap end; OS: overlap start.