DNA methylation profiling allows for characterization of atrial and ventricular cardiac tissues and hiPSC-CMs
Cardiac disease modelling using human-induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CM) requires thorough insight into cardiac cell type differentiation processes. However, current methods to discriminate different cardiac cell types are mostly time-consuming, are costly and often provide imprecise phenotypic evaluation. DNA methylation plays a critical role during early heart development and cardiac cellular specification. We therefore investigated the DNA methylation pattern in different cardiac tissues to identify CpG loci for further cardiac cell type characterization.
An array-based genome-wide DNA methylation analysis using Illumina Infinium HumanMethylation450 BeadChips led to the identification of 168 differentially methylated CpG loci in atrial and ventricular human heart tissue samples (n = 49) from different patients with congenital heart defects (CHD). Systematic evaluation of atrial-ventricular DNA methylation pattern in cardiac tissues in an independent sample cohort of non-failing donor hearts and cardiac patients using bisulfite pyrosequencing helped us to define a subset of 16 differentially methylated CpG loci enabling precise characterization of human atrial and ventricular cardiac tissue samples. This defined set of reproducible cardiac tissue-specific DNA methylation sites allowed us to consistently detect the cellular identity of hiPSC-CM subtypes.
Testing DNA methylation of only a small set of defined CpG sites thus makes it possible to distinguish atrial and ventricular cardiac tissues and cardiac atrial and ventricular subtypes of hiPSC-CMs. This method represents a rapid and reliable system for phenotypic characterization of in vitro-generated cardiomyocytes and opens new opportunities for cardiovascular research and patient-specific therapy.
KeywordsDNA methylation Cardiac tissue-specific DNA methylation Human induced pluripotent stem cell derived cardiomyocytes (hiPSC-CM) Engineered heart tissue (EHT) Bisulfite pyrosequencing 450K array
Atrial-ventricular DNA methylation
Atrioventricular septal defect
Congenital heart defect
Engineered heart tissues
Human-induced pluripotent stem cell-derived cardiomyocyte
Hypoplastic left heart syndrome
Hypoplastic right heart syndrome
Left ventricular aneurysm
Tricuspid valve atresia
DNA methylation plays a critical role during early mammalian development [1, 2] by regulating transcriptional processes [3, 4]. Studies have highlighted the impact of distinct DNA methylation patterns during cell type specification and organ maturation [5, 6, 7] of highly specialized organs, such as the human heart . The fine-tuned spatial and temporal DNA methylation process is not only relevant to early heart development [9, 10, 11], but it also influences cardiac disease and its progression   by impacting, among others, cardiomyocyte maturation . However, the epigenetic process shaping different cardiac cell types (e.g. pacemaker, atrial and ventricular cells) remains poorly understood.
Thorough insight into cardiac maturation and the molecular mechanisms driving cell type developmental and differentiation processes is essential to understand the different aspects of heart disease. Currently, the majority of the available cell culture disease models only unreliably mirror the in vivo condition. The introduction of human-induced pluripotent stem cell (hiPSC) technology  , including hiPSC cardiomyocytes (hiPSC-CMs) , opens new opportunities for cardiovascular research, drug-screening and patient-specific therapy . hiPSC-CMs and their technical improvements, such as 3D force-generating engineered heart tissues (EHTs)   , have dramatically increased our ability to reproducibly model different types of cardiac disease [22, 23]. Nevertheless, the precise molecular characterization of these cell cultures still remains challenging. Up to now, the distinction of different cardiac cell types was accomplished by analysing transcript levels of cardiac surface marker genes, histological stainings or electrophysiological techniques [23, 24, 25, 26, 27, 28]. However, many of these methods are costly and laborious proceedings.
We provide evidence for a distinct DNA methylation profile related to different cardiac tissue types. A key finding of our study is the discovery of a defined set of CpG dinucleotides (CpG loci) to reproducibly distinguish atrial and ventricular subtypes of hiPSC-CMs by their significant differential DNA methylation that further confirmed their previously measured phenotypic expression profiles. Using these techniques, we established a rapid and cost-effective method allowing the characterization of atrial and ventricular hiPSC-CMs for further use as cardiac disease-modelling systems.
Differential DNA methylation pattern between atrial and ventricular human cardiac tissues
Transcription factor binding sites (TFBS) subject to differential DNA methylation among the 168 CpGs were analysed by comparing the CpG loci with TF ChIP-seq performed by ENCODE (Factorbook Motifs ). Fifty-four percent (90/168) of differentially methylated CpGs were associated to different TFBSs in various cell lines (detailed information in Additional file 15: Table S2). The ten most common TFs showing TFBSs among the 168 CpGs were POLR2A, EZH2, EP300, GATA2, FOS, RUNX3, CEBPB, TCF7L2, MAX and TEAD4 (Additional file 6: Figure S6). There were no associations to any cardiac cell lines (AoAF, HCFaa, HCM, HPAF), nor to cardiac TFs (MEF2A, MEF2C, SRF). Over-representation analysis by Reactome Pathways Tool  linking genes with differentially methylated CpGs and associations to TFBSs resulted in the following significant (q < 0.01) pathway with ≥ 3 genes: ‘Developmental Biology’ (R-HSA-1266738, q = 6.3 × 10−3) (Additional file 15: Table S2). With two genes per pathway, ‘Signaling by FGFR in disease’ (R-HSA-1226099 (STAT3, FGFR2), q = 4.6 × 10−5) and ‘Activation of HOX genes during differentiation’ (R-HSA-5619507 (HOXA3, HOXB3), q = 7.4 × 10−4) were represented.
In order to investigate possible similarities in DNA methylation patterns potentially involved in developmental and differentiation processes, MBD-Seq data of DNA methylation in heart tissues of mice  was compared to the 450K data of the present study. All in all, 4 CpG loci of the 168 differentially methylated CpG loci showed an overlap to MBD-Seq reads of Sim et al. (2014) . The ± 60 bp flanking array sequences of three CpG loci—cg18177275, cg12924936 and cg13706058—overlapped with different query sequences of the MBD-Seq analysis, but the specific CpG loci were not targeted (Additional file 15: Table S2). Only one CpG locus, cg04115185, showed overlap to one MBD-Seq read of mouse P1 data (92% homology).
The 168 differentially methylated loci could be annotated to 78 RefSeq genes. Overlapping CpG loci of these 78 genes with predicted human heart enhancers from a study of Dickel et al.  resulted in 24 CpG loci (24/168 loci, OR = 0.96, p = 0.92) (for details, see Additional file 15: Table S2). Multiple CpG loci overlapping with heart enhancers could be identified in three genes: IRX4 (3 loci), NAV1 (2 loci) and TBX5 (2 loci), with confidence scores of each predicted enhancer  of 0.402, 0.539 and 0.372, respectively. Gene ontology analysis using PANTHER14.0  showed significant enrichment for genes involved in different developmental GO biological processes. Among the top ten with highest fold enrichment (Additional file 15: Table S2) ‘cardiovascular system development’ (GO:0072358, q = 1.96 × 10−2), ‘animal organ morphogenesis’ (GO:0009887, q = 2.02 × 10−4), ‘embryonic morphogenesis’ (GO:0048598, q = 2.95 × 10−2) and ‘positive regulation of cell differentiation’ (GO:0045597, q = 2.94 × 10−2) were observed.
Furthermore, expression of DNA methyltransferases was tested to analyse if atrial and ventricular cardiac tissues showed any differences in expression levels. Given the low residual tissue amounts, atrial and ventricular cardiac tissue from only one patient (patient 0126) was used for expression analysis of DNMT1 and DNMT3A. While the expression of DNMT1 showed no significant differences between atrial and ventricular tissues, DNMT3A revealed significantly higher expression in atrial tissue compared to ventricular tissue (Additional file 7: Figure S7).
Identification of candidate CpG loci for distinction of atrial and ventricular cardiac tissues
To assess the biological significance of the detected AVM pattern and to rule out cardiac disease-related effects, 13 non-failing heart tissue samples (four atrial and nine ventricular samples) were analysed as an independent validation set. Furthermore, four heart tissue samples (LA and RA tissue) from adult patients with heart failure were included (detailed sample information, see Additional file 14: Table S1). By DNA methylation analysis of these 16 CpG loci, we could consistently discriminate between atrial and ventricular tissues (mean Δ%-methylation = 25%) in this independent validation set (Fig. 5). Although variations in absolute methylation values occurred between the different sample cohorts, all 16 CpG loci showed (highly) significant DNA methylation differences. We could demonstrate a 100% predictability of heart tissue type classification in this independent sample cohort. The DNA methylation values of samples analysed by bisulfite pyrosequencing are listed in Additional file 18: Table S5 (p values Additional file 19: Table S6).
Specification of hiPSC cardiomyocyte subtypes
Subtype-specific DNA methylation of hiPSC cardiomyocytes
For appropriate cellular cardiac disease modelling, a decent phenotypic characterization of hiPSC-CMs is essential. Current methods to discriminate different cardiac cell types are mostly time-consuming and often provide imprecise phenotypic evaluation [21, 44]. Here, we identified highly significant differential DNA methylation patterns in atrial compared to ventricular human cardiac tissue samples by array-based genome-wide DNA methylation analysis. Systematic analysis of a small set of defined CpG loci enabled us to reproducibly detect atrial and ventricular tissues of independent sampled cohorts and to distinguish atrial and ventricular subtypes of hiPSC-CMs.
By analysing genome-wide DNA methylation profiles in different anatomical regions of the human heart using a discovery set of 49 cardiac tissue samples, we identified distinct differential DNA methylation patterns in atrial and ventricular primary cardiac tissues using the 450K array platform. Up to now, only a few studies exist, analysing the DNA methylation in human heart tissue [12, 13, 14, 45]. Especially, the limited availability of primary heart tissues and the small amounts of material excised routinely during surgery are challenging aspects in such studies. The presented data of this study are—to our knowledge—the first revealing distinct DNA methylation patterns in different human cardiac tissues. We were able to detect exceedingly high delta β-values (mean Δβ = 0.4) among the 168 significantly differentially methylated CpG loci (q ≤ 1 × 10−6, σ/σmax > 0.4) in atrial compared to ventricular cardiac tissues. The highly differential atrial-ventricular methylation (AVM) pattern of the 168 CpG loci might reflect an important influencing factor for cardiac tissue differentiation. In order to investigate if the atrial and ventricular cardiac tissues showed any differences in the expression of DNA methyltransferases, qPCR analyses of DNMT1 and DNMT3A were performed. While the expression of DNMT1 showed no significant differences, DNMT3A revealed significantly higher expression in atrial tissue compared to ventricular tissue (Additional file 7: Figure S7). This might be due to individual effects of the analysed sample, as previous GTEx  experiments showed no significant differences in expression of DNMT3A between atrial and ventricular tissue, and studies in mice revealed that deficiency of de novo CpG methylation capacity mediated by DNMT3A and DNMT3B was dispensable for pathological mechanisms in heart failure . Unfortunately, no additional tissue material was available from cardiac samples analysed in this study to perform further analyses of changes in the activity of the DNA methyltransferases.
Although the 168 differentially methylated CpGs were not associated to TFBSs of any specific cardiac TFs, developmental (cell specific) TFs like GATA2, FOS and TCF7L2 were among the ten most common TFs (Additional file 6: Figure S6). Also, over-representation analysis by Reactome Pathways Tool  resulted in significant enrichment in the pathway ‘Developmental Biology’ (R-HSA-1266738, q = 6.3 × 10−3) of genes with differentially methylated CpGs and associations to TFBSs. This indicates that the differential DNA methylation of the 168 CpGs is associated with—and might influence—developmental processes involved in differentiation towards atrial and ventricular cells. DNA methylation is not necessarily associated with the inhibition of TF binding , and to further elucidate the impact of methyl-TFBSs on atrial and ventricular differentiation processes, prospective investigations have to be done in a cell specific manner in vivo.
The 168 CpG loci did not overlap with known age-related CpGs  or DMRs (differentially methylated regions) identified in epigenome-wide association scans of age or with age-related phenotypes according to Bell et al. (2012) , so that, with reasonable certainty, we could exclude the possibility of age-related methylation effects. However, until now, only limited data is available regarding age-related changes in human heart tissues, and in the study of Horvath , the heart tissues showed a relatively low correlation of predicted DNA methylation age and chronological age. Therefore, we performed an in-depth literature research which thus far did not reveal any of the 168 CpG loci to be associated with gender or age effects.
Considering regulatory features, a significant enrichment (OR = 3.6, p = 5.9 × 10−16) of CpGs overlapping with predicted enhancer elements determined by ENCODE Project Consortium  (50% of CpGs) could be shown (Additional file 5: Figure S5). As enhancers are key regulatory elements that control the process of establishing a tissue-specific transcriptional programme and can be regulated by DNA methylation , this might indicate a relation of differential DNA methylation in differentiation processes towards atrial or ventricular tissues. Although only a few cardiac enhancers have thus far been identified, being less evolutionarily defined than other tissue-specific enhancers , we could identify 24 CpG loci overlapping with predicted human heart enhancers from a study of Dickel et al.  which underpins the hypothesis of differential DNA methylation-driven atrial and ventricular tissue differentiation. However, enrichment of regulatory features and certain gene pathways in 450K array data has to be regarded with caution. Selection of loci of 450K array was defined by a set of content categories identified by a consortium of epigenetics researchers , a design that is biassed due to preselection of probes that interrogate only certain CpG sites, therefore, the design is not hypothesis neutral. For future comprehensive studies of all heart tissue-specific differentially methylated loci, whole genome bisulfite sequencing would be the ideal method as it best represents regions of lower CpG density (e.g. intergenic ‘gene deserts’ or distal regulatory elements) that potentially control tissue-specific gene expression .
Methylation data of dynamic changes in the cardiac methylome during postnatal development are available and in order to compare changes in DNA methylation potentially involved in developmental and differentiation processes, the MBD-Seq data of Sim et al. on cardiac left ventricle of mice  was compared to the 450K data of the present study. Previous analyses of Zhou et al. (2017) of cross-species DNA methylation (rat, mouse, human) revealed that a significant proportion of tissue-specific DNA methylation is conserved  and considering that protein-coding genes and gene-regulatory regions (both genomic regions with CpG loci primarily targeted by 450K array) show high sequence similarities between mouse and human   , informative results were expected. However, even with less stringent approaches, only one CpG locus, cg04115185, laying in a highly conserved non-coding region, showed overlap to one MBD-Seq read (left ventricle of mouse P1). This CpG locus was hypermethylated in ventricular tissue compared to atrial tissue in the present study. If one regards the technique of MBD-Seq, capturing hypermethylated regions, cg04115185 exhibited the same DNA methylation in mouse and human left ventricle in a conserved non-coding region with IRX4 as its nearest gene (390 kb distance). A high fraction of sequences that are conserved across multiple species resides in non-coding regions  and might be associated with the control of early development  and tissue-specific gene expression . Therefore, differential DNA methylation of cg04115185 might potentially be involved in regulation of the ventricular-specific IRX4 gene  which plays an important role in regulating chamber-specific gene expression in the developing heart [60, 61]. All in all, the little overlapping of the two data sets could either be due to differences in DNA methylation analysis platforms, as MBD-Seq is a Methyl-CpG-binding domain-based capture method biassed towards hypermethylated regions , or due to cross-species differences between mouse and human concerning DNA methylation patterns. However, no study exists yet, to our knowledge, that investigates DNA methylation in primary human heart tissue from neonates and infants as it was conducted in the present study.
Further bisulfite pyrosequencing analyses enabled us to identify 16 CpG loci allowing for reproducible verification of the AVM pattern. Given that heart tissue samples are rather difficult to obtain and only small amounts of material are excised routinely during surgery, we could only use the remaining sample material of seven atrial and four ventricular samples (verification set) from the discovery set to verify the 450K array results using bisulfite pyrosequencing. The 16 candidate CpG loci showed heart tissue-predictive capability as we could consistently detect the AVM pattern in independently sampled cardiac tissues. These cardiac tissues consisted of pathologic tissues from cardiac patients (CHDs, HF) and this may affect the results, which is why we also included cardiac tissues from non-transplantable heart-healthy donor hearts (validation set). Although we could partially detect variations in absolute methylation values between sample cohorts (discovery, validation and verification set) that might be due to differences in clinical phenotypes or technical issues, all 16 candidate CpG loci exhibited significant DNA methylation differences between atrial and ventricular tissues in each sample cohort. This tissue-predictive capability of the 16 CpG loci was even valid in the non-pathologic samples from non-transplantable heart-healthy donor hearts and therefore occurring irrespectively from cardiac phenotype. Due to the small amounts of material excised routinely during surgery no further cell sorting was possible, therefore our analyses are based on bulk cardiac tissue samples. These comprised mostly the muscular parts, the myocardium. The human heart contains many different cell types, while the volume fraction of the heart occupied by CMs accounts for 70–80% . We therefore assumed that our methylation patterns represent cardiac tissue type-specific signatures reflecting the morphological and functional differences of atrial and ventricular myocardium and predominantly their CMs. In particular, with respect to electrophysiological and contractile properties, atrial and ventricular CMs differ significantly . Considering the many different epigenetic processes that have been implicated in influencing cardiac gene expression in development and disease [11, 14, 65, 66], the observed differential AVM pattern might represent an epigenetic signature for differentiation of atrial and ventricular subtypes of CMs.
We hypothesized that the AVM pattern in cardiac tissues could also be identified in atrial and ventricular-like subtypes of in vitro-derived hiPSC-CM populations, allowing us to distinguish hiPSC-CMs by their DNA methylation. Differentiated in the absence of retinoic acid, hiPSC-CMs from 2D monolayers and 3D EHTs showed a phenotype resembling (immature) ventricular CMs, since we detected the ventricular-specific marker MLC2v . Further evidence for their ventricular phenotype is provided by electrophysiological lack of increase in Im by the acetylcholine analogue carbachol  as acetylcholine-activated potassium currents typically exist in atrial and not in ventricular tissue . Besides ventricular-like hiPSC-CMs, we were able to differentiate atrial-like hiPSC-CMs by activating retinoid signals  during cardiac specification based on published protocols [24, 25, 39]. The impact of retinoic acid on cardiomyocyte subtype specification in hiPSC-CM differentiation is demonstrated by differential expression of a set of established marker genes (Fig. 6). The higher expression of atrial natriuretic peptide, SLN, MLC2A, COUP-TFI, COUP-TFII, KCNA5, KCNJ3, SK2 and SK3 genes indicates an atrial phenotype [24, 39, 40, 41, 43]. In addition to expression analyses, we investigated the potential of using 16 candidate CpG loci to detect the cellular identity of atrial and ventricular-like cells of in vitro-derived hiPSC-CMs by testing their methylation profiles. We observed significant differential AVM patterns at 11 CpG loci in atrial and ventricular-like hiPSC-CM subtypes, reflecting similar methylation signatures of human atrial and ventricular tissue. In order to verify specificity of AVM patterns of the 16 candidate CpG loci, we tested the DNA methylation using bisulfite pyrosequencing in further cell lines which included endothelial cells from the original hiPSC line (hiPSC-EC), human cardiac VICs and a non-cardiac cell line MCF7 (breast cancer cell line). In hiPSC-ECs, seven CpG loci revealed similar DNA methylation values ± 7% comparable to atrial hiPSC-CMs (Additional file 13: Figure S13) which could be explained by their same hiPSC origin. All in all, there was no concordance between DNA methylation values of hiPSC-ECs and atrial/ventricular heart tissue. The remaining CpG loci though showed considerably lower or higher DNA methylation values or intermediate DNA methylation status in hiPSC-ECs as compared to aCMs/vCMs and atrial/ventricular tissues, respectively. Hence, no DNA methylation pattern could be detected in hiPSC-ECs that would recapitulate the AVM pattern. The same applied to MCF7 cell line which showed very high DNA methylation values (median 77.6% over the 16 CpG loci) or human cardiac VICs with low DNA methylation values (median 9.4%)—both revealing no correlation to atrial or ventricular heart tissue DNA methylation. All in all, the observed AVM patterns appear to be specific for atrial and ventricular heart tissues. It is noteworthy that additionally, three CpG loci (located on TRAPPC9, MYLK and LINC00982), although not significant, showed also comparable AVM patterns in hiPSC-CMs. The slight but not significant DNA methylation differences between atrial and ventricular-like hiPSC-CMs at these three loci and the inversely methylated CpG loci at GALNT2 and WWP1 (opposite AVM patterns of hiPSC-CMs in comparison with cardiac tissue samples) might be due to the characteristics of in vitro-generated hiPSC-CMs, which are not to be equated with native myocardium and do not show a fully mature phenotype . Cardiac maturation has long been investigated in numerous studies [70, 71, 72]. Nevertheless, the identification of techniques to differentiate hiPSC-CMs to mature cardiomyocytes is only in its initial stages [69, 73]. Therefore, deviations from the observed AVM patterns could be explained by the differences in the degree of differentiation between hiPSC-CMs and native myocardial samples. However, on account of the largely highly significant DNA methylation differences measured in atrial and ventricular-like hiPSC-CM subtypes showing similar AVM patterns as in human cardiac tissues, these 11 CpG loci could represent potential loci for subtype characterization of in vitro-generated cardiomyocytes. Taking into consideration that previous techniques for phenotyping of hiPSC-CMs are mostly time-consuming approaches, testing DNA methylation of only a small set of CpG loci might provide a rapid and cost-effective option to identify the differentiation state of in vitro-generated cardiomyocytes. Compared to this, immunofluorescence stainings (fixation, antibody incubations) [26, 74] and electrophysiological techniques (e.g. action potential (AP) and ionic currents) [23, 27, 28] of hiPSC-CMs are laborious and costly proceedings. Differentiation protocols and cell culture conditions may influence AP phenotypes of hiPSC-CMs allowing merely imprecise phenotypic evaluation   using this electrophysiological technique. Analyses of the expression of key structural and functional genes in hiPSC-CMs using qPCR is a standard method of hiPSC-CM molecular profiling . One could argue that the method of bisulfite pyrosequencing we put forward here is just as time-consuming as qPCR techniques since both require pretreatment of starting material. But unlike DNA which is used for bisulfite sequencing, RNA is severely delicate once extracted from its cellular environment and the linearity of the reverse-transcription step to create cDNA may be unsteady, as secondary structures and primer-independent cDNA synthesis can influence the outcome  . In general, targeting the genome and its modifications like DNA methylation results in robust data  while transcriptome data is context-dependent with varying mRNA complement and level depending on physiological state and changes in cell culture conditions . Therefore, determining the DNA methylation in few candidate CpG loci might provide a valuable alternative to molecularly profile in vitro-generated cardiomyocytes.
In conclusion, we have investigated the genome-wide DNA methylation pattern of human cardiac tissues from different anatomical regions of the heart. We subsequently assessed the DNA methylation level at candidate CpG loci in further independently sampled cardiac tissues and in vitro-generated hiPSC-CM subtypes. We identified distinct differential DNA methylation patterns in atrial compared to ventricular human cardiac tissues. A key finding of our study is the potential of using a small number of candidate CpG loci allowing for lineage commitment verification of atrial and ventricular-like hiPSC-CM subtypes. We showed that the current hiPSC lines do not fully recapitulate the epigenetic DNA modification of human atrial and ventricular heart tissue. However, our method is applicable to guide this process, enabling to distinguish cardiac tissue subtypes by analysing only few CpG loci. Thus, this method might serve as a rapid approach for characterization of in vitro-generated cardiomyocytes, potentially improving prospective research of hiPSC-CMs and patient-specific therapy.
Human heart tissue samples
Heart tissue samples were obtained from paediatric patients with congenital heart disease (CHD) and from adult patients with arrhythmic heart defects as well as from non-failing (NF) heart samples of ejected donor hearts which could not be transplanted for technical reasons. Interatrial septum (IAS) samples were obtained from 25 patients with hypoplastic left heart syndrome (HLHS) and from nine patients with hypoplastic right heart syndrome (HRHS), including tricuspid valve atresia (TA). Furthermore, samples from the right atrium (RA) of one patient with HLHS and of three patients with atrioventricular septal defect (AVSD) were obtained during open-heart surgery. The tissue samples were routinely excised in the first weeks of life during surgery and immediately snap-frozen in liquid nitrogen, ensuring an ex vivo time of less than 5 min. Care was taken to use mainly the muscular part rather than the endocardial part of the cardiac samples in this study. Heart explants from one patient with HLHS and transposition of the great arteries (TGA) and from one patient with left ventricular aneurysm (LVA) were available to excise tissue samples (biological triplicates and duplicates) from the myocardium of left and right atrium (LA, RA) and left and right ventricle (LV, RV). Furthermore, IAS, LA and LV tissue samples could be obtained postmortem (12 hpm) from one patient with familial dilated cardiomyopathy (DCM). Myocardial samples from LA and RA from adult patients with heart failure (HF) were obtained to additionally analyse heart tissue from adult donors. Moreover, LA, RA, LV and RV samples from seven non-transplantable donor hearts were investigated to include non-failing heart samples. A table listing the type of heart tissue, diagnosis, gender and age of patients and controls is given in the supplementary data (Additional file 14: Table S1).
Cultivation of hiPSC-derived cardiomyocytes and engineered heart tissues
Expansion of undifferentiated human-induced pluripotent stem cells (hiPSCs), cardiomyocyte (CM) differentiation and generation of 3D-engineered heart tissues (EHTs) were performed as recently described . In brief, expansion of undifferentiated hiPSCs was performed in FTDA medium. Embryoid body (EB) formation was induced in stirred suspension cultures (spinner flasks). Mesodermal induction was achieved using BMP-4 (10 ng/ml), activing A (3 ng/ml) and bFGF (5 ng/ml) in the absence of insulin in RPMI medium . Specification of cardiac differentiation of mesodermal progenitors was performed by WNT signal inhibition (XAV939, 1 μM). This resulted in a population of a primarily ventricular cardiomyocyte (ventricular-like) phenotype. Based on previous reports [25, 39] differentiation of atrial cardiomyocytes was achieved by addition of retinoic acid (1 μM) for the first 3 days of Wnt signalling inhibition. By fluorescence-activated cell sorting (FACS), hiPSC-CM differentiation efficiency of cardiac troponin T positive cells was analysed to ensure that similar differentiation efficiency could be obtained in the absence and presence of retinoic acid. At the end of cardiac differentiation, EBs were enzymatically dispersed with collagenase . The dissociated cells were mixed with fibrinogen (Sigma F4753) and thrombin (100 U/ml, Sigma Aldrich T7513) to generate EHTs (1 × 106 cells/EHT) , a synchronously beating syncytium of hiPSC-CMs in two elastic silicone posts [26, 80].
Phenotypic characterization of atrial and ventricular hiPSC cardiomyocytes
In order to characterize cultivated atrial and ventricular-like subtypes of hiPSC-CMs (2D cultures and EHTs), expression analyses of atrial and ventricular-specific genes were performed using quantitative real-time PCR (qPCR)  technique. Subtypes of hiPSC-CMs were maintained under standard cell culture conditions for 2 weeks before performing qPCR. After proteinase K (Thermo Scientific) digestion, extraction of total RNA was performed with RNeasy Mini Kit (Qiagen) according to the manufacturer’s instructions. qPCR experiments were performed on atrial and ventricular hiPSC-CMs from three independent hiPSC-CM generations, each as duplicates. For assessing gene expression by qPCR, cDNA was synthesized from approximately 200 ng of total RNA. RNA was reverse-transcribed into cDNA using high-capacity cDNA reverse transcription kit (Applied Biosystems). qPCR was performed using Maxima SYBR Green/ROX (Thermo Scientific) on an ABI Prism instrument (Applied Biosystems). Each reaction was performed in triplicates and non-template reaction (replacing cDNA with water) was used as negative control. The cycling parameters were 50 °C for 2 min followed by 95 °C for 10 min, 15 s at 95 °C and 1 min at 60 °C for 40 cycles. mRNA-specific CT values were normalized with CT values for human GUSB (beta glucuronidase). Relative differences between atrial and ventricular samples were calculated with ∆∆Ct method for relative quantifications. Primer sequences are enclosed in the Additional file 17: Table S4. Candidate markers for atrial and ventricular phenotype were chosen based on previous publications . Statistical analyses were performed with GraphPad Prism software 5.0. Data are expressed as mean ± SEM in bar graphs. Differences between groups were analysed by unpaired t test. Results were considered statistically significant if the p value was less than 0.05.
Isolation of DNA
Genomic DNA from frozen heart tissue samples and cardiomyocyte cell lines (2D monolayer, 3D EHT) was extracted using Gentra Puregene DNA isolation reagents (Qiagen) according to the manufacturer’s protocol (5–10 mg tissue). Following fluorometric quantification by Qubit dsDNA BR Assay (Life Technologies), DNA integrity was visually inspected by agarose gel electrophoresis.
Illumina Infinium HumanMethylation450 Assay
The DNA methylation analysis of human heart tissue samples was conducted using the Infinium HumanMethylation450 BeadChip (Illumina), which interrogates the methylation level of 485,577 loci. Bisulfite conversion of isolated DNA (1 μg) was performed using the Zymo EZ DNA Methylation Kit (Zymo Research) according to the manufacturer’s instructions. Bisulfite-converted DNA was eluted in 15 μl ddH2O. Isothermal amplification, enzymatic fragmentation, hybridization (for 20 h) onto the HumanMethylation450 Bead Chips (Illumina) and subsequent scanning of immunohistochemistry staining using an iScan Microarray Scanner (Illumina) were performed following the manufacturer’s protocol as described earlier . Samples were randomly distributed across arrays to limit batch effects.
Processing and quality control of Illumina Infinium HumanMethylation450 data
Signal intensities and raw methylation values were extracted from the GenomeStudio™ Software (version 2011.1, Methylation Analysis Module version 1.9.0, Illumina) for each CpG without any data processing. Methylation levels (β-values) are given as ratios of fluorescent signal intensities between methylated and unmethylated alleles, ranging from β = 0 (unmethylated) to β = 1 (completely methylated). Hybridization quality was analysed using detection p values calculated by the GenomeStudio™ Software for each CpG. To estimate the percentage of loci showing a median detection p value < 0.01 for each sample, a loci call rate (LCR) was calculated as LCR = ((number of loci with detection p value < 0.01)100) (485,577 total number of loci)−1. Samples with LCR > 98% were included in subsequent data processing. Next, 450K array data was normalized using the normalization function of RnBeads  R package, a Subset-quantile Within Array Normalization (SWAN) method. Following this, the data was filtered by (1) probes mapping to sex chromosomes, to avoid gender specific bias; (2) probes harbouring SNPs with an allele frequency (AF) > 0.05 as reported by 1000G rel. 20110521; (3) probes comprising annotated SNPs (1000G rel. 20110521) within 3 bp of the interrogated CpG having an AF > 0.05; and (4) cross-reactive probes according to Chen at al. . Following these preprocessing steps, β-values of biological replicates from one patient with HLHS-TGA (3× LV, 3× RV, 3× LA, 3× RA) and one patient with LVA (3× LV, 2× RV, 3× LA, 3× RA) were combined to average values each. Thus, the final 450K dataset consisted of 44 atrial heart datasets (35× IAS, 3× LA, 6× RA) and five ventricular datasets (3× LV, 2× RV).
Unsupervised principal component analyses (PCA) of preprocessed and quality filtered 450K methylation data was performed using R statistical software  to obtain an overview of preliminary data. To identify differences in DNA methylation of distinct cardiac tissue types (IAS, RA, LA, RV, LV), analysis of variance (ANOVA) test with Benjamini-Hochberg  FDR multiple testing correction was applied using OMICS explorer (version 2.3, Qlucore) software. A variance filter of σ/σmax > 0.4 was applied. To keep the number of false positives as small as possible, a stringent FDR of < 1 × 10−6 was used. The ANOVA was plotted as PCA using OMICS explorer (version 2.3, Qlucore) for 3D plots to visualize the segregation pattern of different cardiac tissue types and to analyse their epigenetic distance and relatedness. Based on these preliminary results, due to the observed distribution pattern, a Student’s t test was applied to compare the two segregating groups—atrial versus ventricular heart tissue samples using OMICS explorer (version 2.3, Qlucore) software. Showing a multiple test adjusted FDR of q < 1 × 10−6 and σ/σmax > 0.4, CpG loci were considered being significantly differentially methylated between atrial and ventricular samples. Following this pairwise group comparison analysis, a hierarchical cluster analysis (heat map depiction) was applied to visualize the statistical results using OMICS explorer (version 2.3, Qlucore) software. To minimize effects due to intra-group heterogeneity and small group sizes, Welch two-sample t test was applied to analyse the significance of differential methylation of the 16 candidate CpG loci in the different analysis cohorts (Additional file 19: Table S6). Group data was compared using unpaired Welch two-sample t tests and were presented as standard box-and-whiskers plots (whiskers, 5th–95th percentile). A p value of p < 0.05 was considered to be statistically significant. Graph Pad Prism 5 (GraphPad Software, San Diego, CA, USA) was used for data analysis of qPCR experiments.
MBD-Seq data of DNA methylation in heart tissues during postnatal development of mice  was compared to the 450K data of the present study, in order to investigate possible similarities between changes in DNA methylation potentially involved in developmental and differentiation processes. Over 60 million reads of MBD-Seq data from P1 (GSM1462877) and P14 (GSM1462880) mouse cardiac left ventricle were compared to the 450K array target sequences (± 60 bp flanking the CpG) of 168 differentially methylated CpGs in atrial and ventricular human heart tissues. In a second step, a less stringent approach was conducted by comparing ± 10 bp around CpG sites of MBD-Seq query sequences to the 450K target sequences.
Furthermore, the correlation between methylation values measured by 450K array (discovery set) and those measured by bisulfite pyrosequencing (verification set) was analysed using linear regression statistics (method Pearson). For quality control of 450K array by means of reproducibility of measured β-values, the correlation between biological replicates was assessed and depicted in scatter plots using R statistical software . Gene ontology analyses were performed using Reactome Pathways Tool  and PANTHER over-representation test  (PANTHER 14.0). Enrichment analyses were performed as follows: Fisher’s exact (p < 0.05) and multiple test adjustment by Benjamini & Hochberg method .
Filtering for candidate CpG loci
In order to analyse further heart tissue samples regarding the aspect of differential DNA methylation between heart tissue from atrial and ventricular origin, candidate CpG loci were filtered to be subsequently analysed by bisulfite pyrosequencing technique. Methylation differences between atrial and ventricular tissues are presented as delta β-values (or Δ% methylation in case of bisulfite pyrosequencing data), as absolute values of atrial subtracted by ventricular methylation values.
To select candidate CpG loci, the 450K array data was filtered by (1) upper 10% quantile of CpG loci with greatest delta β-values (Δβ) among the 168 differentially methylated CpG loci between atrial (LA, RA, IAS) and ventricular (LV, RV) samples, identified by Student’s t test; (2) association of CpG to UCSC RefSeq gene region between TSS200 (0–200 bases upstream of the transcriptional start site) up to the 3′UTR; (3) if more than one CpG site among the upper 10% quantile is associated to the same UCSC RefSeq gene, then the one displaying the higher Δβ-value was chosen and the next significant CpG locus with high Δβ-value was included in the candidate CpG list; and (4) if no primer design using PyroMark Assay Design 2.0 software (Qiagen) was possible (e.g. due to high CpG density), the next significant CpG locus was selected according to the filtering method of (1) and (2), respectively.
Bisulfite pyrosequencing of candidate CpG loci
To verify the 450K array data and to validate the methylation pattern in further heart tissue samples, bisulfite pyrosequencing was performed using a Pyromark Q96 ID sequencer (Qiagen). PyroMark Assay Design software (version 2.0, Qiagen) was applied for primer design (primer list Additional file 16: Table S3). To ensure methylation-independent amplification, primers were designed to hybridize with CpG-free sequences. Human high methylated genomic DNA (80-8061-HGHM5, EpigenDx) served as methylated control, and whole-genome-amplified DNA (WGA-DNA) served as unmethylated control. The WGA-DNA was prepared using a pool of ten female and male healthy DNA control samples, which was amplified using Illustra GenomiPhi™ V2 DNA Amplification Kit (GE Healthcare) and cleaned up using Wizard® DNA Clean-Up System (Promega). DNA samples were bisulfite converted using the Zymo EZ DNA Methylation Kit (Zymo Research) as aforementioned. Bisulfite-converted DNA (1 μl) was applied for PCR using PyroMark PCR Kit reagents (Qiagen). Biotinylated PCR products underwent washing and creation of single-strand structure by usage of the Vacuum Prep Tool (Biotage) and PyroMark Gold 96 Reagents Kit (Qiagen) according to the manufacturer’s instructions. Bisulfite pyrosequencing reactions and quantification of methylation (ranging from 0% to 100%) were performed on a Pyromark Q96 ID sequencer (Qiagen). Quality control included analysis of accordance of histogram and measured DNA methylation peak signals as well as inspection of DNA methylation values of methylated (> 70%) and unmethylated (< 10%) controls at analysed CpG loci.
The authors would like to thank the patients and their families for their support and participation and the technical staff of the molecular genetic laboratories of the Institute of Human Genetics for expert assistance.
This study was supported by funding from the German Center for Cardiovascular Research, the 675351 AFib-TrainNet, the DFG HA 3423/5-1 and the Competence Network for Congenital Heart Defects funded by the Federal Ministry of Education and Research (BMBF). OA recived funding from the the German Center for Lung Research (DZL; 82DZL001A5).
Availability of data and materials
The datasets of 450K array analyses generated during and/or analysed during the current study are available from the corresponding author on reasonable request.
KH, KR, DS and ML contributed to the sample preparation. KH, KR and DS helped in the methylation experiments. KH, OA and RS contributed to the method development. KH, EA and M-PH helped in the statistical analyses. ML and AH contributed to the hiPSC cultivation and qPCR experiments. JS, TA, TP, KH, A-KK, HM and AH helped in the sample recruitment. M-PH, A-KK, H-HK, AC, HM and AH contributed to the phenotyping. KH and M-PH helped in the study design. KH, ML, M-PH and AH contributed to the writing. M-PH, OA and AH to the supervision. All authors read and approved the final manuscript.
Ethics approval and consent to participate
The study protocol was approved by the Competence Network & Registry for Congenital Heart Defects, Germany.
Consent for publication
The authors declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
- 29.CDC. National Center on Birth Defects and Developmental Disabilities. Centers for Disease Control and Prevention, https://www.cdc.gov/ncbddd/index.html (2018, Accessed 16 Oct 2018).
- 47.Nührenberg TG, Hammann N, Schnick T, et al. Cardiac myocyte de novo DNA methyltransferases 3a/3b are dispensable for cardiac function and remodeling after chronic pressure overload in mice. PLoS ONE. 10. Epub ahead of print 22 June 2015. https://doi.org/10.1371/journal.pone.0131019.PubMedPubMedCentralCrossRefGoogle Scholar
- 56.Initial sequencing and comparative analysis of the mouse genome. Nature. 2002;420:520.Google Scholar
- 85.Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Ser B Methodol. 1995;57:289–300.Google Scholar
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.