*3.4. Epigenome-Wide Association of Intronic DNA Methylation and Splicing in DCM vs. Control*

We first performed a genome-wide analysis of variance (ANOVA) and discovered that, on average, 2.71% (95%CI: 2.70–2.72, *n* = 2,684,933 tests) of the variance of PSI scores could be explained by the intronic DNA methylation on the same gene, which is sizable at the genomic scale. Thereafter, we carried out an epigenome-wide association study between methylation measurements and PSI scores of exons within the very same gene. Correlation coefficient, odds ratio, and *p* value of each association test were determined. Moreover, we compared the correlation coefficients between DCM and control samples, that is, we carried out statistical tests of significance for the difference between correlation coefficients in DCM patients and those in controls. As shown in the Manhattan plot for genome-wide statistical tests (Figure 6A), several loci with significantly different (FDR <0.05) correlation coefficients between DCM patients and controls were detected in the screening cohort, signifying a disease-dependent differential impact of DNA methylation on alternative splicing.

In the replication cohort, five exonic regions in *TTN-AS1* as well as one exonic region in *DCTN1* were validated with statistical significance (*p* < 0.05). However, only the five verified exonic regions in *TTN-AS1* could also be "directionally replicated" in the replication cohort (Figure 6B), with a positive correlation between PSI scores and methylation values in DCM patients, as well as a negative correlation between PSI scores and methylation values in healthy controls, as shown in Figure 7. Tables 2 and 3 list the odds ratios per 0.01 increments of DNA methylation (beta value) and *p* values derived from logistic regression models of both cohorts. The values in the parentheses indicate the 95% confidence interval. In all identified candidates, the odds ratios in DCM were less than one, indicating a negative association, that is, if there is an increase of DNA methylation in DCM patients, a decrease of PSI value is expected, and vice versa. On the other hand, in all identified candidates, the odds ratios in control samples were greater than one, suggesting a positive correlation. In terms of significance level, the positive association between DNA methylation and PSI value in control samples reached statistical significance both in screening and replication cohorts. However, the positive correlation between DNA methylation and PSI values in the DCM samples reached statistical significance only in the replication cohort, which could be due to less noise in the replication cohort owing to the stranded RNA-Seq, providing more of an advantage for antisense analysis.

**Figure 6.** (**A**) Manhattan plot summarizing genome-wide statistical tests of significance for the difference of correlation coefficients between DCM and control samples in the screening cohort. The red horizontal line represents the FDR of 0.05. (**B**) Genomic browser tracks showing the relative positions of the validated candidates from the epigenome-wide association study in *TTN-AS1*. The PSI scores of the validated exonic regions (green) in *TTN-AS1* were significantly associated with the methylation level of the highlighted locus (red). The first track is a reference transcript of *TTN*, the following three tracks are transcripts of *TTN-AS1*. The last four tracks were added to visualize the log-scaled RNA-Seq coverage in DCM and control, in both screening and replication cohorts. It should be noted that the RNA-Seq of the replication cohort was stranded, while the RNA-Seq of the screening cohort was unstranded. Hence, coverage in the screening cohort is noisier than in the replication cohort. Nevertheless, the candidates in *TTN-AS1* could be replicated in the replication cohort with statistical significance.

**Figure 7.** Visualization of DNA methylation measurements and PSI scores of validated genomic regions in *TTN-AS1*. For each validated candidate, all study subjects of the screening cohort were plotted by their methylation measurements (*X*-axis) and PSI scores (*Y*-axis). The conditions of the samples are color-coded (red: DCM, blue: Control). The depicted regression lines were computed using logistic regression and are also color coded (pink: DCM, light blue: Control). The same visualization for the replication cohort is presented below, showing the conserved principle. (**A**) Screening cohort; (**B**) Replication cohort.




**Table 3.** Odds ratios of the replicated candidates in the replication cohort.

#### **4. Discussion**

The present study utilized an epigenome-wide association approach to examine the interaction between DNA methylome and splicing of the transcriptome in the heart, as both biological processes were only recently shown to play an essential regulatory role in DCM. A significant positive correlation between intronic DNA methylation and usage of adjacent exons was detected. Moreover, we pinpointed and stringently validated several regions in *TTN-AS1* with a disease-dependent differential regulation of DNA methylation on alternative splicing. This is the first study to investigate in the full epigenome the complex yet highly ordered orchestration of methylome and transcriptome in the healthy human heart as well as in DCM.

In the past few years, GWAS have helped to identify several novel genomic regions associated with cardiac phenotypes. As a result, there has been a rapid progress in functional genetics to assist in the exploration of biological meaning of disease-associated genomic regions. Although genome studies massively advanced our knowledge of DCM, plenty of biological mechanisms of disease still need to be deciphered, and investigations on epigenetic–genetic and epigenetic–transcriptomic levels have been proposed to provide yet another crucial piece in disease etiology [28]. As an example, Wang et al. implemented the GWAS approach on an epigenomic dataset to identify signatures related to clinical parameters, such as from electrocardiograms (ECG). Eventually, they were able to experimentally validate the findings in iPSC cardiomyocytes [29]. The present study relied on human cardiac tissue as disease-relevant processes are often tissue- and species-specific [30]. While other studies attempted to investigate DNA methylation and alternative splicing in cell cultures, our approach is the first to inspect the relationship between DNA methylation and alternative splicing in heart muscle disorders using human cardiac tissues [31].

We discovered a positive correlation between DNA methylation of the flanking introns and the inclusion of the bordering exon across the whole genome. This relationship exists both in DCM and control subjects. There are three potential mechanisms underlying these findings. First, as early as in 1988, it was identified that splicing occurs during transcription [32]. Hence, it is possible that the detected association is mediated by some specific DNA-binding proteins, such as CTCF and MeCP2. These proteins can change their binding affinity to DNA by apprehending methylation signatures of DNA. When they are bound to DNA, they can impact the elongation rate of RNA polymerase II, influence the time for the splicing machinery to recognize weak splice-sites, and subsequently affect the inclusion of alternative exons [33–35]. Aside from the known proteins, other DNA-binding proteins with similar functions could still exist and be undiscovered so far. Second, it is also possible that DNA methylation-dependent recruitment of splicing factors takes place. For example, it has been reported that the adaptor protein HP1 can recruit splicing factors if bound to methylated DNA [36]. Interestingly, it has been suggested that histone modifications could facilitate the splicing factors to bind

to pre-mRNA [37], while there is literature reporting the strong link between histone modifications and DNA methylation [38–40], which is in line with our theory. Third, based on the evidence demonstrating DNA methylation's correlation with nucleosome occupancy as well as the regulatory role of DNA methylation on the modification of histones [41–46], it is reasonable to speculate that DNA methylation influences splicing through regulating the orchestration of chromatin remodeling and nucleosome positioning, especially the nucleosome positioning relative to the splice sites of interest [47], while more in-depth understanding of the interaction between DNA methylation and nucleosome occupancy is still needed.

In the present study, we identified numerous regions on *TTN-AS1* with a DCM-dependent differential regulation of DNA methylation on alternative splicing, while *TTN-AS1* is practically an inverse counterpart to *TTN*. *TTN-AS1* encodes Titin antisense 1, which is a long noncoding RNA (lncRNA) that produces an estimate of 80 different transcripts. In the literature, lncRNAs were reported to play a role in cardiac development and regeneration, in the pathogenesis of cardiovascular diseases, as well as in the doxorubicin-induced cardiac toxicity, which predisposes people to DCM [48–50]. Furthermore, disrupted splicing of lncRNAs were found to cause dysregulation of important cardiac proteins in mice, such as potassium voltage-gated channel proteins encoded by *Kcnq1* and *Kcna2* [51–53]. In addition, a cluster of antisense lncRNAs in the MYH7 locus was noticed to be essential in early development of cardiomyopathy under pressure-overload [54]. Interestingly, in recent studies, Titin antisense 1 was shown to act as a competing endogenous RNA (ceRNA) to sequester diverse microRNAs (miRNAs) and transcription factors, thus, consequently facilitating tumor aggressiveness in several cancers, including lung cancer [55], cervical cancer [56], esophageal cancer [57], gastric cancer [58], papillary thyroid cancer [59], colorectal cancer [60], and prostate cancer [61]. However, although *TTN-AS1* is highly expressed in the heart, there is minimal understanding of its cardiac role in the literature, and more investigations are required. The validated regions were found to locate at the counterpart locus of the genomic region encoding the A-band of titin. This finding is intriguing, because genetic variations in titin A-band are the leading cause of DCM [62,63]. Hence, based on our finding, it is not unreasonable to speculate that dysregulated splicing of Titin antisense 1 might be able to induce deleterious exon-skipping in the A-band of titin, which may mimic the pathologies seen in TTNtv, and that modification of this region might even be a therapeutic principle [64,65]. Hence, the current study provides new understanding of the regulation of this important gene locus.

A potential limitation of the here conducted epigenome-wide approach is the sparse preexisting data on false-positive and false-negative discovery rates and adequate power calculations of epigenome-wide association analysis and in particular the here conducted multi-omics type of analysis [29]. Our approach, although explorative in the screening stage, used an independent validation step. Since cardiac tissue is highly limited for research studies, the sizes of both cohorts were still relatively small compared to traditional GWAS. However, the combination of information from *a priori* connected biological processes and coordinated molecular layers is able to reduce false-positive discoveries and add statistical power [8].

In conclusion, this study emphasizes the intricate interplay between the DNA methylation landscape and the mRNA splicing machinery. With a state-of-the-art epigenome-wide approach and utilization of human cardiac tissue as study material, a new understanding of the genome–epigenome relationship in DCM was presented. We showed that dysregulated methylation of the gene encoding Titin antisense 1 is associated with its splicing, which could induce pathological exon-skipping during the transcription of *TTN*.

**Supplementary Materials:** The supplementary materials are available online at http://www.mdpi.com/2077-0383/ 9/5/1499/s1. Figure S1: RNA-Seq in the replication cohort. Figure S2: PCA of RNA-Seq in both cohorts. Figure S3: PCA of DNA methylation measurement in both cohorts. Figure S4: Dispersion plot and MA plot of differential exon usage analysis in the screening cohort. Figure S5: Histograms showing the sequencing depth of both cohorts. File S1: Results of differential gene expression analysis. File Table S2: List of up- and down-regulated genes. File Table S3: Gene ontology analysis of up- and down-regulated genes. File S4: Gene ontology analysis of regions with cocurrent DEU and DMR.

*J. Clin. Med.* **2020**, *9*, 1499

**Author Contributions:** Conceptualization, B.M., J.H. and W.-T.G.; methodology, B.M. and W.-T.G.; software, W.-T.G., R.T., D.H.L., J.H.; validation, W.-T.G. and A.K.; formal analysis, W.-T.G.; investigation, B.M. and W.-T.G.; resources, H.A.K. and B.M.; data curation, J.H., F.S.-H., E.K. and O.S.S.; writing—original draft preparation, W.-T.G.; writing—review and editing, W.-T.G., B.M. and M.W.; visualization, W.-T.G.; supervision, B.M.; project administration, J.H., F.S.-H., E.K.; funding acquisition, B.M. and H.A.K. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by grants from the Deutsches Zentrum für Herz-Kreislauf-Forschung (German Center for Cardiovascular Research, DZHK), the German Ministry of Education and Research (CaRNAtion, FKZ 031L0075B), Informatics for Life (Klaus Tschira Foundation), the Faculty of Medicine of the Leipzig University (to M.M-W.), the European Union (Detectin-HF), and the Deutsche Forschungsgemeinschaft (DFG). B.M. was supported by an excellence fellowship of the Else Kröner Fresenius Foundation.

**Acknowledgments:** We thank Sabine Herch, Anne Marie Müller, Antje Weber, and Rouven Nietsch for excellent technical support.

**Conflicts of Interest:** All authors declare no conflicts of interest.
