Genomic Reorganization of Lamin-Associated Domains in Cardiac Myocytes is Associated with Differential Gene Expression and DNA Methylation in Human Dilated Cardiomyopathy

Abstract

Rationale:

Lamin A/C (LMNA), a nuclear membrane protein, interacts with genome through lamin-associated domains (LADs) and regulates gene expression. Mutations in the LMNA gene cause a diverse array of diseases, including dilated cardiomyopathy (DCM). DCM is the leading cause of death in laminopathies.

Objective:

To identify LADs and characterize their associations with CpG methylation and gene expression in human cardiac myocytes in DCM.

Methods and Results:

LMNA chromatin immunoprecipitation-sequencing, reduced representative bisulfite sequencing, and RNA-sequencing were performed in 5 control and 5 LMNA-associated DCM hearts. LADs were identified using Enriched Domain Detector (EDD) program.

Genome wide 331 ± 77 LADs with an average size of 2.1 ± 1.5 Mbp were identified in control human cardiac myocytes. LADs encompassed ~20% of the genome and were predominantly located in the heterochromatin and less so in the promoter and actively transcribed regions. LADs were redistributed in DCM as evidenced by a gain of 520 and loss of 149 genomic regions. Approximately, 4,500 coding genes and 800 long non-coding RNAs (lncRNAs), whose levels correlated with the transcript levels of coding genes in cis, were differentially expressed in DCM. TP53 was the most prominent amongst the dysregulated pathways. CpG sites were predominantly hypomethylated genome-wide in controls and DCM hearts, but overall CpG methylation was increased in DCM. LADs were associated with increased CpG methylation and suppressed gene expression. Integrated analysis identified genes whose expressions were regulated by LADs or CpG methylation, or by both, the latter pertained to genes involved in cell death, cell cycle, and metabolic regulation.

Conclusion:

LADs encompass ~20% of the genome in human cardiac myocytes comprised of several hundred coding and non-coding genes. LADs are redistributed in LMNA-associated DCM in association with markedly altered CpG methylation and gene expression. Thus, LADs through genomic alterations contribute to the pathogenesis of DCM in laminopathies.

INTRODUCTION

Mutations in the LMNA gene, encoding Lamin A and its isoforms, cause a variety of distinct phenotypes, collectively referred to as laminopathies. ^{1, 2} Laminopathies involve multiple cell types and organs, which is in accord with ubiquitous expression of the LMNA gene in almost all differentiated cells, including cardiac myocytes. ^{1, 3, 4} Cardiac involvement is common and manifests with dilated cardiomyopathy (DCM), arrhythmogenic cardiomyopathy, arrhythmias, conduction defects, and sudden cardiac death ^5–8. DCM is the major cause of morbidity and mortality in laminopathies. ^{5, 6, 9–12}

LMNA is a nuclear inner membrane protein and essential for various biological processes in the nucleus such as chromatin organization, DNA repair and gene expression. ^{3, 13} In mammals, LMNA interacts with hundreds of large chromatin domains, which are referred to as Lamin-associated domains (LADs). ³ LADs are distributed throughout the genome and are involved in genomic organization, recruitment of epigenetic regulators, and consequently, gene expression. ^{3, 14, 15} LADs are considered either constitutive or facultative. ³ Constitutive LADs comprise a subset that is cell type invariant, highly conserved across species, and low in gene density. ¹⁶ In contrast, facultative LADs show cell type-specific genomic distributions, are less conserved, and exhibit dynamic positioning during biological processes. ^{16, 17} LADs are predominantly located at heterochromatic gene-desert regions and are enriched in repressive histone marks, such as H3K9me3 and H3K27me3. ^{16, 18–22} Consequently, genes located at LADs are generally silenced or expressed at low levels. ^{14, 17}

The molecular pathogenesis of DCM caused by LMNA mutations is not well understood. A number of effector molecular pathways, including signaling pathways, such as mitogen-activated protein kinases, and changes in the linker of nucleoskeleton and cytoskeleton complex (LINC) and consequent mechanotransduction to the nucleus have been implicated in laminopathies ^{1, 23–26} In view of the extensive contact of LMNA with chromatin, dysregulated gene expression has emerged as a plausible mechanism to explain phenotypic diversity of laminopathies, including DCM. ²⁷ However, LADs and their roles in regulation of gene expression in cardiac myocytes in normal and pathological states have not been defined. We posit that DCM in laminopathies results from dysregulated gene expression as a consequence of reorganization of LADs in cardiac myocytes. To test this hypothesis, we isolated cardiac myocyte nuclei from control human hearts and hearts with DCM associated with pathogenic variants in the LMNA gene. We performed chromatin immunoprecipitation, using an anti-LMNA antibody, followed by deep sequencing of the precipitated DNA fragments (ChIP-Seq). We also delineated methylation states of the CpG sites by reduced representative bisulfite sequencing (RRBS) and determined genome-wide transcript levels by RNA-sequencing in the same heart samples. We identified LADs, analyzed their distributions, and determined their associations with CpG methylation states and transcript levels in control and DCM hearts.

METHODS

Adherence to TOP Guidelines.

Large data sets are already available to other investigators through GEO (RNA-Seq: GSE120836 and ChIP-Seq GSE120837). Detailed information about material and methods are available in Online Supplementary Material. All other data and material are available from the corresponding author upon request.

Human myocardial tissues.

Human heart tissue samples were obtained from the University of Colorado at Denver tissue Biobank. Institutional review board approved use of the tissues. Left ventricular samples from explanted hearts of 5 patients with an established diagnosis of DCM who underwent cardiac transplantation were used for isolation of cardiac myocyte nuclei and RNA and DNA extractions. Similarly, left ventricular tissues from 5 donor hearts, which were not used for cardiac transplantation, were included as controls.

Immunofluorescence.

Pericentriolar material 1 (PCM1) was used as a marker to identify and isolate cardiac myocyte nuclei. ^28–30 Thin fresh frozen myocardial sections were stained with anti-PCM1 (Rabbit polyclonal Sigma Cat# HPA023370) anti α-actinin (Sigma Cat# 7811) antibodies, incubated in 4′, 6-Diamidino-2-phenylindole di-hydrochloride, the latter to stain the nuclei (Sigma, cat# D8417), and examined under epifluorescence microscopy, as described previously. ²⁶

Isolation of cardiac myocyte nuclei.

Aliquots of 250 – 300 mg of human ventricular tissues were used to isolate cardiac myocyte nuclei per a published protocol. ^28–30 Briefly, frozen tissues were homogenized in a cold lysis buffer containing protease and phosphatase inhibitors (Roche Cat# 11 836 153 001, Cat# 04 906 845 001). The homogenates were passed through 100 μm and 70 μm nylon filters (VWR Cat#732–2759, Falcon Cat# 352350) successively, centrifuged at 1000g, and layered onto a sucrose gradient buffer. The nuclei were centrifuged at 30,000g and the pellets were resuspended in a sorting buffer. The nuclei were incubated with 2 μg of either a rabbit IgG (Millipore, Cat# 12–370) or an anti PCM1 antibody (Sigma, Cat# HPA023370), pre-labelled with Alexa 488 Goat anti Rabbit IgG secondary antibody (Invitrogen, Cat#A21206). Nuclei were also stained with 5μM DRAQ5 (Abcam Cat#108410) and subjected to fluorescence activated cell sorting (FACS) using ARIA III (BD Bioscience). Flow cytometry data were processed using FlowJo v10 software.

Chromatin immunoprecipitation-Sequencing (ChIP-Seq).

Sorted PCM1 positive cardiac myocyte nuclei were crosslinked using 1% formaldehyde (Sigma, Cat# F8775) ³⁰. The nuclei were disrupted in a lysis buffer containing protease and phosphatase inhibitors and the lysates were sonicated using a Bioruptor Pico (Diagenode) to achieve fragment sizes ranging from 200bp-500bp. Lysates were precleared with Protein G Sepharose (GE Healthcare Cat# 17–0618-01) and incubated with a 2 µg aliquot of an anti-LMNA antibody (Abcam Cat# ab26300), followed by pulldown with BSA coated Protein G-Sepharose beads. The pulldown complexes were washed with a low and high salt wash buffers, consecutively, and the protein/DNA complexes were extracted in an elution buffer, and reverse crosslinked. DNA was extracted after proteinase K digestion and phase separation using phenol/chloroform/isoamyl alcohol (25:24:1).

ChIP was performed using either an anti-LMNA antibody or a rabbit IgG isotype control using 50,000 cardiac myocyte nuclei. To assess efficient immunoprecipitation of LMNA, a subset of samples were analyzed by immunoblotting protein/DNA complexes. The samples were resuspended in a Laemmli sample buffer (Bio-Rad Cat#161–0737), loaded onto an SDS-polyacrylamide gel, subjected to electrophoresis, and transferred to a nitrocellulose membrane. The membranes, after blocking in nonfat dry milk, were incubated with an anti-LMNA antibody (Abcam Cat# ab26300), followed by incubation with an HRP-conjugated secondary antibody. Signals were detected using the ECL western blotting detection kit (Amersham Cat# RPN2106).

Immunoprecipitated DNA and input samples were used for library preparation and sequencing. ChIP-Seq libraries were prepared using the NEBNext Ultra II DNA Library Prep Kit (New England BioLabs, Inc. NEB#E7645) following the manufacturer’s protocol. Briefly, approximately 1 ng of fragmented DNA was used to repair the ends before ligation to NEBNext Adaptors. The adaptor-ligated DNA was enriched by PCR and purified using AMPure XP beads (Beckman Coulter). The libraries were run on a 2200 Tape Station (Agilent Technologies) for quality control. A KAPA Library Quantification Kit (KAPA Biosystems) was used to quantify the libraries. A final concentration of 1.5 nM was loaded onto an Illumina cBOT for cluster generation before sequencing on an Illumina HiSeq3000.

Paired-end, 100 bp sequencing reads were obtained from matched input and anti-LMNA ChIP-seq libraries. Sequencing reads were aligned to the hg19 release of the human genome by Bowtie v2.1 ³¹. Only uniquely mapping reads were selected for further analysis. Each pair of input and immunoprecipitated samples was adjusted to achieve the same read depth using the DownsampleSam feature of Picard software (www.broadinstitute.org/gatk/events/2038/GATKwh0-BP-1-Map_and_Dedup.pdf). LADs were identified using the Enriched Domain Detector (EDD) software with parameters set at bin size of 11 kbp, gap penalty of 9 and a false discovery rate (FDR) of 0.05. ³² For each sample, any predicted LADs >= 20 Mbp were excluded. LADs were annotated against NCBI RefSeq entries, NONCODE, and the Broad ChromHMM feature database using the BEDTOOLS software and annotation files obtained from RefSeq, NONCODE, or in the case of the Broad ChromHMM features, a custom annotation table obtained via the UCSC genome browser. ³³ The log₂ ratio of signal in ChIP vs. Input comparisons were computed using NGS-Deep tools from the Galaxy server (https://usegalaxy.org/) and signal tracks were visualized using Integrated Genomic Viewer (IGV). ³⁴ Common LADs and differentially regulated LADs in both control and DCM cardiac myocytes were identified using the multiple intersect function in BEDTOOLS and the Jaccard bed function was used to compute the relative distance between LADs. ³³ ChIP-Seq data have been submitted to GEO (GSE120837).

ChIP-quantitative polymerase chain reaction (ChIP-qPCR).

Approximately 100,000 PCM1 positive nuclei were isolated from each heart by FACS and ChIP was performed using an anti-LMNA antibody, as described. Specific primers were designed using IGV genome browser and primer blast program to detect LADs that were present in at least 3 control or DCM samples and qPCR was performed as published (Online Table I). ³⁵

Immunoblotting.

Immunoblotting was performed on approximately 50 mg of protein extract from each human left ventricular tissue, as published. ^{26, 36}

Reduced representative bisulfite sequencing (RRBS).

Genomic DNA, extracted from the same control DCM hearts used for ChIP-Seq, was analyzed for CpG methylation by RRBS. The RRBS libraries were made following a published protocol except for the addition of a column purification step ³⁷. Briefly, purified DNA was digested by overnight incubation with MspI and the digested DNA fragments were subjected to end repair and A-tailing, followed by linker ligation. Bisulfite conversion was done using the EZ DNA Methylation-Gold Kit (Cat. # D5030, Zymo Research, Irvine, California, USA) according to the manufacturer’s instructions. Barcoding of the RRBS libraries was performed using adapters from Bioo Scientific (NEXTflex™ DNA Barcodes). RRBS libraries were sequenced on a HiSeq 2500 (Illumina, San Diego, California, USA) using 36 bp sequencing. The adaptors and low-quality bases were trimmed with Trim Galore (Version 0.2.5) prior to alignment of sequencing reads. The quality scores and bp distribution of the sequencing reads was assessed using FastQC (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/). A minimum of 30 million reads / sample were aligned to human reference genome assembly hg19 by Bismark (version 0.13.0) with default parameters. Differentially methylated CpGs were obtained using R package methylkit based on logistic regression and only CpGs with a coverage of at least 10 reads were used for further analysis. ^{38, 39}. Differentially methylated regions were defined as genomic regions with mean differences in methylation compared to their matched normal, at an FDR-adjusted p-value < 0.05. Data were visualized using the UCSC genome browser, the IGV browser, EaSeq software using default bin size settings (http://easeq.net) and Circos software packages. ^{40, 41}

Identification of differentially methylated TF binding sites (TFBS).

MEME Suite was used to predict TFBS in hypermethylated and hypomethylated CpG sites. ⁴² Motif discovery MEME-ChIP was performed followed by enrichment analysis using CentriMo. ^43,44 DNA sequences 500 bp upstream and downstream from each differentially methylated sites were analyzed. CentriMo enrichment tool was used to obtain only motifs that were centrally located and therefore, were most proximal to the CpG sites. MEME-ChIP and CentriMo were run using the default parameter. Likewise, the default setting of the MEME software used for the background sequences, which were determined from the primary input sequence and accounted for dinucleotide bias. An alternative background sequence, as permitted by MEME, was also included in the analysis to reduce potential fortuitous effects of the background selection. To account for dinucleotide biases like GC content, the first-order background model was used, which uses first-order Markov Model over a input sequence centered at the base pair being scored and corrects for the underlying dinucleotide composition, resulting in a better signal/noise. ⁴⁵ To further asses the reproducibility of enrichment, our dataset was trained to a higher order (2 and 3rd) background models, which identified identical enrichment values for TFBS.

RNA-sequencing.

Total RNA was extracted from human left ventricular myocardial tissues using Qiagen miRNeasy Mini Kit (cat # 217004). Integrity of the RNA extracts was analyzed on an Agilent Bioanalyser RNA chip and samples with an RNA Integrity Number (RIN) read out of >8 were used. Strand-specific sequencing libraries were generated using the TruSeq stranded total RNA library preparation kit, including depletion of ribosomal RNA (Illumina Inc. cat # 20020596) and sequenced on an Illumina HiSeq 4000 instrument to give 100 bp, paired-end reads. ^{26, 35}

RNA sequencing reads were aligned to the human reference genome build hg19 using TopHat2. ⁴⁶ The aligned read pairs were annotated using either Ensembl- or UCSC-supplied gtf annotation files, or with the addition of a gtf from the NONCODE database of noncoding RNA (http://www.noncode.org/). The numbers of aligned read pairs that uniquely mapped to any coding or noncoding RNA entry were compiled into a read count matrix, from which RNA abundance ratios (‘fold-changes’), p-values, and FDRs were calculated in a between-group fashion using EdgeR statistical package in an R/Bioconductor framework. ^{47, 48} Reads were quantified using the featureCounts software. Data was normalized using the Remove Unwanted Variation method, as implemented in the R scientific analysis platform. ⁴⁹ Significance level was set at a Benjamini-Hochberg FDR-adjusted p value (i.e., q value) of <0.05. RNA-Seq data has been submitted to GEO (GSE120836).

Pathway analysis.

Gene set enrichment analysis (GSEA) was performed on ranked gene lists, created based on the expression levels of differentially expressed genes (DEGs) in DCM as compared to control samples (q<0.05). ⁵⁰ Significance was assessed by analyzing signal-to-noise ratio and gene permutations based on 1,000 cycles of permutations. Molecular signature database (MSigDB) 3.0 curated gene sets for hallmark and canonical pathways were used for the analysis. ⁵¹ Gene sets were also curated using the compute overlap, a hypergeometric distribution function of GSEA. The data were presented as enrichment score and an FDR cutoff of 0.05 was used to identify significant gene sets.

Upstream regulators Analysis.

To identify likely transcriptional regulators of dysregulated genes, the Upstream Regulator Analysis module of Ingenuity Pathway Analysis software (IPA®, QIAGEN Redwood City) was used. The IPA predicted upstream regulators with Z score greater than +2 or less than −2 are reported.

Statistical analysis.

Normality of data distribution was tested by Wilk-Shapiro normality test. Normally distributed data, presented as mean ± SD, were compared by t-test and one-way ANOVA. Data deviating from a Gaussian distribution were presented as the median values and analyzed by Kruskal-Wallis test or Welch’s approximation test, as were the categorical data using Graph pad Prism 7 (www.graphpad.com/scientific-software/prism/) or Stata Statistics/Data Analysis software version 10.1 (College Station, TX). Similarity among the genomic regions of LADs were calculated using the Jaccard BEDTOOLS from Galaxy tool shed (https://toolshed.g2.bx.psu.edu/).

RESULTS

DCM and control human heart specimens.

Myocardial tissues from five explanted hearts of patients (4 males and 1 female) with primary DCM who had undergone heart transplantation were used (Online Table II). Secondary causes of heart failure, including coronary artery disease, were excluded. Only DCM hearts that carried pathogenic variants in the LMNA gene were included (Online Table II). As a control group, myocardial tissues from 4 male and 1 female donor hearts, which were not utilized for heart transplantation, were used. The mean age of the control group was similar to that of the DCM cases (controls: 35.6 ± 16.1 vs. DCM; 50.8 ± 10.7, p=0.117 by Kruskal-Wallis rank test).

LMNA expression in control and DCM hearts.

Western blot analysis of whole heart protein extracts from 5 controls and 5 LMNA-associated DCM showed a mild reduction in the LMNA protein levels in DCM hearts (Online Figure I). This finding is in accord with the effects of the LMNA mutations on transcription and translation, including the effects of the heterozygous premature stop codon mutation, which is not expected to lead to expression of stable protein from the mutant allele. (reviewed in ⁵²).

LADs in cardiac myocytes.

Cardiac myocyte nuclei were identified upon staining with an antibody against PCM1, which has been shown to tag cardiac myocytes in the heart. ^28–30 Expression of PCM1 in human cardiac myocytes was analyzed upon staining of thin myocardial sections with antibodies against PCM1 and myocyte marker ACTN2 (Figure 1, A and B). Approximately, 28.0±1.9% of nuclei in the myocardial sections stained positive for PCM1, which is in accord with the expected percentage of cardiac myocytes in the myocardium (Figure 1C). ^{53, 54}

Figure 1. Characteristics of lamin-associated domains (LADs) in human cardiac myocytes.

A. Immunofluorescence panel showing staining of human thin myocardial sections with an anti PCM1 and ACTN2 antibodies. The former specifically marks cardiac myocyte nuclei in the heart and the latter cardiac myocytes.

B. A higher magnification of immunofluorescence panels stained as in panel A.

C. Quantitative data showing percent of myocardial cells stained positive for PCM1 in control human hearts

D. FACS data showing isolation of PCM1 stained cardiac myocyte in control and DCM human hearts and the corresponding IgG panel, the latter a control for gating.

E. Graph depicting PCM1 stained nuclei as a percentage of total nuclei in FACS data

F. Graph depicting the number of PCM1 stained nuclei per milligram of myocardial tissue

G. Blot showing immunoprecipitation of lamin A/C (LMNA) protein with an anti-LMNA antibody, and the corresponding input and anti-IgG antibody as controls.

H. A genome browser map of Log of ChIP-Input signal ratio identifying a lamin-associated domain (LAD) and the surrounding non-LAD regions in 3 control and 3 DCM samples. Chromosome ideogram and list of genes located at the region are depicted.

I. Panel depicts the number of LADs identified in each control and DCM genome (N=5 per group)

J. Panel depicting whisker and bar showing mean size of LADs in each heart sample. Median, 25% and 75% quartiles and range values are shown.

K. Validation of selected LADs identified in control hearts by an independent set of ChIP and qPCR experiments, showing enrichment of the LADs by up to 50-fold, as compared to non-LAD regions. First lane represents the average value of 5 non-LAD (NL) regions arbitrary set at 1 for normalization purpose. Only LADs with a q value of <0.05 are shown.

L. Validation of selected LADs identified in LMNA-associated DCM hearts by an independent set of ChIP and qPCR experiments, showing enrichment of the LADs by up to 60-fold, as compared to non-LAD regions. First lane represents the average value of 6 non-LAD (NL) regions arbitrary set at 1 for normalization purpose. Only LADs with a q value of <0.05 are shown.

M. Number of genomic regions encompassing LADs unique to control and DCM hearts or shared in both groups.

Whole heart nuclei were isolated from control and DCM hearts, stained with an anti PCM1 antibody to tag myocyte nuclei, and then subjected to FACS (Figure 1D). PCM1 stained nuclei comprised 23.1 ± 6.8 % and 10.9 ± 2.8 % of all DRAQ5 stained nuclei in the control and DCM groups, respectively (p=0.005 by Welch’s approximation test, Figure 1E). The mean number of nuclei expressing PCM1 per mg of myocardial tissue was lower by about 2-fold in the DCM as compared to the control hearts (474 ± 201 vs. 191 ± 101, p=0.0132 by Welch’s approximation test, Figure 1F).

To determine whether anti LMNA antibody effectively precipitated LMNA, nuclear proteins, extracted from the PCM1 stained nuclei, were immunoprecipitated (IP) with an anti-LMNA antibody and analyzed by immunoblotting. LMNA was enriched in the precipitates with the anti-LMNA antibody, whereas LMNA protein was not detected in the corresponding IP with IgG alone (Figure 1G).

To identify LADs in cardiac myocytes, LMNA ChIP was performed using ~ 50,000 PCM1-stained nuclei per heart and sequencing libraries were prepared from equal amounts of the precipitated DNA fragments as well as from the input fragments, to account for potential differences in the LMNA protein levels. The average number of sequencing reads and the percent mapped did not differ between control and DCM hearts (about 70 million reads in each sample of which about 40 to 66% were uniquely mapped reads, Online Table III). LADs were detected and called by the EDD that efficiently identifies broad domains with extensive but low-amplitude enrichment between ChIP and input reads, characteristic of LADs. ³² A representative genome browser profile of LADs and non-LAD regions in the control and DCM samples is depicted in Figure 1H, which illustrates LADs as contiguous regions exhibiting increased sequence read density relative to the genomic background (presented as log₂ of ChIP/Input signal intensity). Genome-wide a total of 331 ± 77 LADs in the control and 248 ± 40 LADs in the DCM groups were identified (Figure 1I and Online Table IV, p=0.047 by Kruskal-Wallis rank test). The mean sizes of LADs were 2.1 ± 1.5 and 3.5 ± 0.91 Mbp in control and DCM groups, respectively (Figure 1J, p=0.117 by Kruskal-Wallis rank test). Collectively, LADs covered 640 ± 361 and 862 ± 232 Mbp of the genome in the control and DCM groups, respectively (p=0.295 by Kruskal-Wallis rank test).

Validation of LADs by LMNA ChIP.

To test for validation of the LAD identified by LMNA ChIP-Seq, an independent set of LMNA ChIP experiments was performed in the same heart samples and under identical experimental conditions. Fifty-two and 46 LADs, which were present in at least three heart samples, in the control and DCM groups, respectively, were tested for validation. As shown in Figure 1, panels K and L, 46/52 (88%) and 41/46 (89%) of the LAD in the control and DCM groups, respectively, showed evidence of significant enrichment in the ChIP reaction, as compared to the non-LAD regions, validating the ChIP-Seq findings. In addition, 2 and 3 LADs in the control and DCM groups, respectively, showed a trend toward significant enrichment in the validation studies.

Redistribution of LADs in DCM.

To discern common and phenotype-specific LADs, genomic regions that were present in at least 3 samples in each group were compared. A total of 221 genomic regions, comprising 251 Mbp were common between control and DCM hearts, whereas 149 genomic regions comprising 95 Mbp were exclusive to controls and 520 regions encompassing 539 Mbp were distinct to DCM hearts (Figure 1M, Jaccard similarity coefficient: 0.28). To assess distribution of LADs according to functional chromatin states, LADs in the control group were depicted against the 15-state epigenome partitions computed using ChromHMM 15 on NIH Epigenome Roadmap. ⁵⁵ Functional chromatin states differed significantly in the LAD regions as compared to the overall epigenome map, referred by low Jaccard similarity coefficient, suggesting a non-random binding of LMNA to chromatin (Online Figure II). LADs showed a preponderance toward heterochromatin regions and away from promoter and transcriptionally active domains, as compared to the epigenome map. The LADs in the control and DCM hearts differed significantly in genomic localization, indicated by a low Jaccard similarity coefficient, showing a redistribution of the LADs in DCM. The shift in DCM was notable for a distribution of LADs toward the genome-wide epigenome pattern, suggesting partial loss of specific functional chromatin binding in DCM, albeit the similarity coefficients to the genome still remained low (Online Figure II).

Differential gene expression in DCM.

To determine association of LADs with transcriptomic changes in DCM, ribosome-depleted RNA preparations from the control and DCM hearts were analyzed by RNA-Seq. Principal component analysis illustrated distinct separation of the control and DCM transcriptomes, with genotypes accounting for 76.4% of the variance (Figure 2A). Transcript levels of 2,514 coding genes were increased whereas those of 2,083 were decreased in DCM as compared to control hearts at an FDR of <0.05 (GSE120838 , Figure 2B). A heat map depicting the differentially expressed coding genes (DEGs) and showing intra-group clustering but inter-group dissimilarity in the transcript levels, is presented in Figure 2C. A partial list of the top dysregulated genes is provided alongside the heat map. To identify putative transcriptional regulators responsible for DEGs in DCM, the Upstream regulator prediction module of Ingenuity Pathway Analysis (IPA) was used, which showed TP53, NUPR1, and KDM5B as the transcriptional regulators most likely to have increased activity and NRF family of transcription factor (TF) NFE2L2 as most likely to have decreased activity (Figure 2D).

Figure 2. Transcriptomics in human hearts with DCM associated with defined pathogenic variants in the LMNA gene.

A. Principal component analysis of the whole heart RNA-Seq data showing distinct clustering of control and DCM samples.

B. A pie chart illustrating the number of differentially upregulated (red), downregulated (blue), and unchanged (grey) genes in DCM hearts (N=5 per group, FDR<0.05).

C. A heat map of differentially expressed genes and the corresponding top upregulated and downregulated genes in human hearts from patients with DCM.

D. The graph illustrated top dysregulated TFs, obtained from IPA upstream regulator module, in the heart in DCM (red: activated, blue: suppressed) and the corresponding Z scores, deduced from differential expression of their known target genes. Size of the nodes indicates number of dysregulated genes contributing to the node. Only upstream regulator with Z score more than 2 or less than −2 along with the p<0.05 for overlaps are shown.

E. A heat map of differentially expressed long non-coding RNAs (lncRNAs) and the corresponding list of top differentially expressed lncRNAs in the DCM hearts.

F. A Pearson correlation matrix between differentially expressed lncRNAs and the transcript levels of the corresponding genes in cis (100 kbp). Adjusted r² and the corresponding p value for all transcripts are shown.

G. Selected examples of correlation between transcript levels of non-coding RNA and their corresponding coding RNAs (in cis) in controls and DCM. An anti-sense lncRNA showing an inverse correlation is also shown.

H. Venn diagrams showing number of coding and non-coding genes located in LAD genomic regions shared between control and DCM hearts or exclusive to each group.

I. Violin plots depicting median, 25% and 75% quartiles, and range of the transcript levels as CPM in LAD and non-LAD genomic regions in controls and DCM, showing lower levels of the transcripts in the LAD genomic regions in both controls and DCM hearts. Overall ANOVA p value and all pairwise p values for the transcript levels were <0.005.

J. Representative sort maps showing transcript levels plotted with reference to the RefSeq genes in LAD and non-LAD regions in control and DCM hearts, both showing reduced CPM levels and lower number of coding genes in the LAD genomic regions.

To identify dysregulated RNA transcripts other than the coding RNAs in DCM and thus, to test the hypothesis that altered lamin A binding affects transcription beyond coding mRNA regions, annotated RNA sequence reads were aligned to the NONCODE database, which contains 96,308 human long non-coding RNA (lncRNA) genes (www.noncode.org/). Of those listed in the NONCODE database, 26,924 lncRNAs were expressed, defined as count per million (CPM) value of ≥1, in the human heart (control and DCM), of which 367 were upregulated and 439 downregulated in the DCM hearts (q <0.05). A heat map of the differentially expressed lncRNAs illustrating similar intra-group clustering and inter-group distinction, along with the list of top differentially expressed ncRNAs is presented in Figure 2E.

To analyze association of lncRNAs with the transcript levels of coding genes located in cis, genomic juxtaposition and co-expression patterns of differentially expressed lncRNA-mRNA within 100 Kbp were correlated. LncRNAs that mapped within the coding RNA open reading frames and were in the sense orientation were excluded from analyses. A total of 435 lncRNA-mRNA pairs were located in cis. A Pearson correlation matrix of lncRNA-mRNA cis pairs is shown in Figure 2F and Online Figure IIIA, which illustrate a strong correlation in the expression patterns of lncRNA-mRNA pairs (293 LncRNA-mRNA pairs with correlation coefficients ≥ 0.9 and 63 lncRNA-mRNA pairs with ≤ −0.9). Specific examples of correlation of the differentially expressed lncRNAs and mRNAs, including an example of lncRNA expressed in anti-sense direction, are shown in Figure 2G.

LADs and gene expression.

To determine whether LADs were associated with gene expression, genomic regions that were present in at least 3 samples were analyzed for their associations with transcript levels of genes located in the LAD regions. Shared genomic regions encompassed 1,042 coding and 9,940 non-coding genes in the control group, whereas the corresponding numbers in the DCM group were 2,160 coding and 19,461 non-coding genes (Figure 2H). Control and DCM groups shared 455 coding genes in the LAD genomic regions, whereas 1,705 were unique to DCM and 587 to controls (Figure 2H). Likewise, 5,847 non-coding genes were common to both, 13,614 unique to DCM, and 4,093 were found only in controls. Genome-wide transcript levels were compared between the LAD and non-LAD regions. Globally, transcript levels of coding genes were lower in the LAD as compared to non-LAD regions both in the control and in DCM groups (Figure 2I). Representative sort maps of the transcripts in the LADs and non-LADs genomic regions in control and DCM groups, showing lower CPM levels in the LAD regions, as compared to non-LAD regions, are shown in Figure 2J. Likewise, analysis of transcript levels of non-coding genes that were expressed (CPM ≥ 1) showed lower expression levels in the LAD as compared to non-LAD regions both in the control as well as in DCM hearts (Online Figure IIIB).

Finally, the number of coding genes whose transcripts were not detected was higher in the LAD regions than in the non-LAD regions (Online Figure IV). Approximately, half of the coding genes in the LAD regions were not expressed, whereas a third of the coding genes in the non-LAD region was not expressed in the control and DCM groups (Online Figure IV).

Differential CpG methylation in DCM.

Because LADs have been associated with CpG methylation in cancer cells and senescent fibroblasts, methylation states of the genome-wide CpG sites in the control and DCM hearts were determined by RRBS, which led to identification of 906,812 individual CpG sites at a minimum sequencing depth of 10 reads. ^{56, 57} The genome-wide CpG methylation pattern showed a strong propensity towards hypomethylation, as approximately 81.5% of the CpG sites were either unmethylated or were less than 20% methylated in both DCM and control hearts (Figure 3A and Online Figure VA). This was particularly the state for the CpG sites at the promoter regions, which encompassed 70.5% of all CpG sites analyzed. Hypermethylated CpG sites, defined as >80% methylation, were the second preponderant sites in the genome comprising 9.5% of the total CpG sites. Comparing the genome-wide percent CpG methylation in DCM and control samples showed a strong correlation (Figure 3A, Pearson r² = 0.983). CpG methylation at the promoter, gene body, and intergenic regions between controls and DCM hearts also showed strong correlations (Figure 3A).

Figure 3. CpG methylation and transcript levels.

A. Smoothed scatter plots showing correlations between percent methylation of all CpG sites in control and DCM hearts throughout the genome, promoter, genic (gene body), and intergenic regions

B. Plots, as in panel A, except shown only for differentially methylated CpG sites (FDR<0.05) showing correlations in % CpG methylation in the promoter and gene body regions between control and DCM hearts. Intergenic CpG sites were most differentially methylated and showed least correlation between control and DCM samples. Pearson correlation coefficient (r) values are indicated in each panel.

C. Enrichment of specific TF binding motifs in hypomethylated CpG sites located in the promoter regions. Top 4 centrally located motifs are shown along with the corresponding E-values of enrichment.

D. As in C except the enrichment is shown in hypermethylated CpG sites

E. Graph showing correlation between differential CpG methylation (FDR < 5%) and differential gene expression (FDR < 5%) in DCM samples. All combinations are depicted. The overall Pearson correlation r value and p value, reflecting all four quadrants is shown.

F. Pathway analysis of genes in Panel E predicted enrichment of TP53 targets (Z score = 2.23). Corresponding CpG methylation states are indicated in red (down) and blue shades (up) and differential gene expressions are indicated in parenthesis.

G. Density plots of CpG methylation distribution in LAD and non-LAD genomic regions for genome-wide, promoter, gene body, and intergenic regions are shown. Overall ANOVA p-value for each panel and pairwise comparisons of the genomic regions were < 0.001, except for CpG methylation density in gene body regions located in LADs in the control as compared to DCM hearts.

A total of 19,291 CpG sites (~ 2% of all sites) were differentially methylated between controls and DCM hearts (FDR < 0.05), which included 10,606 and 8,685 hypermethylated and hypomethylated sites, respectively, in the DCM hearts as compared to controls (Figure 3B and Online Figure VB). Genome-wide distribution of the differentially methylated CpG sites (DMS) showed predominant localization to the promoter (8,900 sites, 46% of DMS) and the intergenic regions (9,988 sites, 51%), whereas only 403 DMS (2%) were located in gene body regions. Overall, CpG methylation at the promoter, gene body, and intergenic regions showed strong correlations between controls and DCM samples (Figure 3B), albeit gene body and intergenic regions showed increased CpG methylation state in DCM as compared to controls (Online Figure VB).

Because CpG methylation is known to affect TF binding at the promoter regions, enrichment of TF sequence motifs in the differentially expressed CpG regions located at the promoter regions (0 to −2 Kbp from the ATG site) were analyzed using the de novo TF binding site discovery and enrichment tool MEME (http://meme-suite.org/) ^{58, 59}. Over a dozen TF binding motifs were enriched in differentially methylated CpG sites located in the promoter region. Top predicted TFs whose binding sites were enriched in the hypomethylated and hypermethylated CpG sites, respectively, are depicted in Figure 3, C and D. The analysis included methylation states of all CpG sites within the gene promoter regions. In some cases, both hypomethylated and hypermethylated CpG sites within the same promoter region were juxtaposed to each other. Consequently, some of the enriched motifs were similar between hyper- and hypo-methylated CpG sites. A complete list of the enriched TF binding sites in the DMS is shown in Online Table V.

CpG methylation and transcript levels.

To detect association of methylation states of CpG sites located at the promoter regions with the transcript levels of the corresponding genes, levels of CpG methylation at the DMS (FDR <5%) were correlated with the transcript levels of the DEGs (FDR <5%). Transcript levels of the majority of the genes were increased in DCM, regardless of whether the CpG sites at their promoter regions were hypo- or hypermethylated (Figure 3E). Methylation states of all differentially methylated CpG sites and the transcript levels of their corresponding genes (FDR: < 5%, differential methylation by >5%) were weakly correlated (Figure 3E). This finding is in accord with the known modest effects of CpG methylation on transcription in pathological cardiac remodeling. ⁶⁰ The observed association between the CpG methylation states in the promoter regions and the transcript levels of the corresponding gene reflects the sum effects of methylation states of all CpG sites, which might be different from the effects of the methylation state of each CpG site on the transcript levels. About half of the DEGs (734/1,561) had concordant anticipated changes in the CpG methylation states at their promoter regions, i.e., reduced or increased CpG methylation were associated with increased or reduced transcript levels, respectively (Figure 3E, r=0.75, p<2.2*10⁻²⁶). Pathway analysis of this subset of DEGs displaying concordant changes in CpG methylation showed enrichment of the TP53 targets (Figure 3F, Z score: 2.2, p=0.04). The majority of the TP53 target genes were associated with hypomethylation, predicting activation of the TP53 pathway in DCM.

Distribution of CpG methylation sites according to LADs.

To determine whether redistribution of LADs in DCM was altered with CpG methylation states and altered LAD enrichment at defined genomic domains, CpG methylation densities were analyzed in LAD and non-LAD genomic regions that were present in at least 3 samples. Overall, the methylation patterns followed those that were observed for genome-wide CpG methylation patterns in DCM and controls (Figure 3A), as the majority of the sites were hypomethylated (0–20%), including the promoter regions (Figure 3G). Gene promoters located in the LAD regions had a lower CpG density than those located in the non-LAD regions. Likewise, promoters in the DCM groups had a higher CpG density than in controls. DCM patients had a higher CpG density in gene body regions that were located in LADs, as compared to control hearts (Figure 3G).

CpG methylation density in gain of LADs (GoL) and Loss of LADs (LoL) in DCM.

To determine methylation states of the CpG sites located in the redistributed LAD genomic regions in DCM, CpG methylation density was analyzed in regions present in DCM but absent in controls, i.e, GoL, and absent in DCM but present in controls, i.e., LoL. GoL in DCM was associated with increased percent CpG methylation in the promoter and intergenic region (Figure 4A). Increased percent CpG methylation in DCM was even more prominent in the LoL regions, particularly in the intergenic region (Figure 4B). Genomic regions outside of the gained or lost LADs also showed a similar pattern, i.e, showing a higher percent CpG methylation in DCM as compared to control hearts (Figure 4 C and D). Comparing GoL or LoL to their corresponding non-LAD regions showed increased percent CpG methylation in the LAD as opposed to non-LAD regions (Figure 4, A-D).

Figure 4. Redistribution of LADs in DCM hearts with corresponding changes in CpG methylation and gene expression.

A. Genome-wide and domain-specific CpG methylation states depicted as density plots generated using EaSeq software with average %CpG methylation from 0 to 100% and number of sites with default bin size in each Gain of LAD (GoL) regions in DCM.

B. Loss-of-LAD (LoL) regions showing increased CpG methylation density, genome-wide, promoter region, and intergenic regions (FDR<0.01)

C. Corresponding CpG methylation maps for the non-LAD regions, defined as regions outside the GoL, showing significantly increased CpG methylation density in promoter and intergenic regions in DCM (FDR<0.001).

D. Corresponding CpG methylation maps for the non-LAD regions, defined as regions outside the LoL, showing significantly increased CpG methylation density in DCM, particularly in the intergenic region (FDR<0.001).

E. A heat map of transcripts of differentially expressed genes located in gain of LAD regions in DCM (N=304).

F. A heat map of transcripts of differentially expressed genes located in loss of LAD regions in DCM (N=120).

G. Transcription factors (TFs) motifs enriched in the differentially expressed genes located in GoL and LoL regions in DCM, as shown in panels E and F.

H. Circos map showing integration of DEGs, differentially methylated CpG (average methylation of CpG in promoter regions) and GoL and LoL in DCM. The tracks from outer to inner in the Circos plot represent the followings:

1. Track I shows transcript levels of all DEGs in DCM presented as fold change (log 2)
2. Track II depicts methylation of CpG sites in the promoter regions of differentially expressed genes (Pearson correlation p value <0.05).
3. Track III shows correlation between coding transcript levels (fold-change in log₂) and the corresponding promoter CpG methylation states (average of DMS are used)
4. Track IV illustrates genes that are located in the GoL and LoL regions and show significant changes in their expression levels (Pearson correlation p value <0.05) and
5. Track V shows correlation between transcript levels of DEGs and GoL and LoL regions.

I. Predicted TF binding motifs enriched in the DEGs that are part of GoL or LoL and were associated with changes in CpG methylation in DCM.

GoL and LoL and gene expression in DCM.

GoL in DCM was associated with suppressed gene expression, as evidenced by reduced transcript levels of 238 as opposed to increased transcript levels of 66 genes located in the GoL regions (Fisher exact test p<0.0001, Figure 4E). In contrast, LoL was associated with increased transcript levels of 98 genes, whereas transcript levels of 22 genes were reduced (Fisher exact test p<0.0001, Figure 4F). Genes whose expressions were dysregulated upon redistribution of LADs were further analyzed to predict enrichment of TF binding sites at their promoter regions. The approach led to identification of 32 TF motifs predicted to be enriched in DCM, including THAP1, NRF1, ZBT14, SP1, and KLF12 (Figure 4G and online Table VI).

Consistent with TF enrichment analysis, Gene Ontology (GO) biological pathway analyses of gene dysregulated in DCM and located in the GoL regions showed those involved in mitochondrial function, stress response, and cell cycle regulation, as the top dysregulated pathways (Online Figure VIA). In contrast, LoL affected expression of genes that were involved in cell catalytic activity, extracellular organization, and MTOR signaling, as the major suppressed pathways (Online Figure VIB).

GoL and LoL were also associated with the transcript levels of lncRNAs in a pattern similar to those of coding RNAs (Online Figure VII). Accordingly, GoL was associated with reduced transcript levels of 89 and increased levels of 10 non-coding genes in DCM, as compared to controls (p<0.0001, Online Figure VII-A). Conversely, LoL was associated with increased transcript levels of 3 and reduced transcript levels of 40 non-coding genes in DCM (p<0.0001, Figure VII-B).

To determine whether changes in the expression levels of genes located in the GoL and LoL regions were a function of LADs or heart failure per se, DEGs located in the GoL and LoL regions were compared between LMNA-associated DCM and a garden variety of heart failure. ⁶¹ There were 305 DEGs in the GoL regions in the LMNA-associated DCM and 2,933 DEGs in the garden variety heart failure. Of the 305 DEGs located in the GoL regions in LMNA-associated DCM, only 56 were shared with the garden variety of heart failure (18%), the remainder (249, 82%) were exclusive to LMNA-associated DCM (Online Figure VIII). Similarly, of the 121 DEGs located on the LoL regions in the LMNA-associated DCM, 29 (24%) were shared with the garden variety of heart failure and 92 (76%) were exclusive to LMNA-associated DCM (Online Figure VIII).

Integrated analyses of LADs, CpG methylation, and transcript levels.

To determine association of LADs with CpG methylation and transcript levels, changes in distribution of LADs, CpG methylation states, and transcript levels were integrated. These analyses led to identification of genes whose transcript levels were mostly correlated with presence or absence of LADs, percent CpG methylation, a combination of both, or were independent of them (Figure 4H). Pairwise analysis showed that expression of 530 genes were affected by CpG methylation only (r² set at >0.25, Online Figure IX). Likewise, redistribution of LADs in DCM was associated with altered transcript levels of 357 genes (Online Figure IX). Intersection of the redistributed LADs, CpG methylation percentage, and transcript levels identified 50 genes whose expressions were affected by CpG methylation as well as redistributions of LADs (Online Figure IX-A). Pathways analysis showed dysregulated genes involved in cell death and survival, cell cycle, and metabolism, among others (Online Figure IX-B). Analysis of the promoter regions of genes whose expressions were regulated by CpG methylation as well as by LAD redistribution led to identification of enrichment of predicted motifs for the several TFs in DCM, including MBD2, NRF1, SP1, and CTCFL, among others (Figure 4-I, and Online Table VII).

DISCUSSION

An integrated approach utilizing cardiac myocyte-specific LMNA ChIP-Seq, whole heart transcriptome, and CpG methylation analyses in control and DCM hearts, the latter associated with defined pathogenic variants in the LMNA gene, partially defined the role of LMNA in restructuring the human genome and altering gene expression in patients with DCM. The findings are notable for the presence of extensive interactions between LMNA and the genome, encompassing about 20% of the genome in human cardiac myocytes. Approximately, one-third of the LADs, comprised of about 250 Mbp, were shared among cardiac myocytes in control and DCM groups, likely representing constitutive LADs, whereas the remainder differed among the genomes and according to the phenotype (control and DCM), likely representing facultative LADs. LADs are predominantly located in the heterochromatin and less so in the promoter regions and the actively transcribed genes. Consequently, transcript levels of genes located at the LAD regions were lower as compared to non-LAD regions in both control and DCM hearts. LADs were partially redistributed in DCM and this redistribution covered ~ 10% of the coding genes and was associated with differential expression of lncRNAs and CpG methylation in DCM hearts. Altered expression levels of lncRNAs was strongly associated with the transcript levels of coding genes located in cis, suggesting a co-expression and probable cis-regulation of the coding genes by lncRNAs in DCM. LADs were associated with increased CpG methylation, particularly in the intergenic regions, albeit differential CpG hypermethylation as a whole was increased in DCM in both LAD and non-LAD regions, indicating factors other than LADs were also determinants of CpG methylation density in DCM. CpG methylation as a whole was associated with transcript levels of ~ 500 genes, comprising of about 10% of all differentially expressed genes in DCM. Pathway analysis of differentially expressed genes in DCM predicted TFs TP53, NUPR1, and KDM5B as the top upregulated and NFE2L2 as the most downregulated pathways in DCM. A subset of genes whose expressions showed an inverse correlation with percent CpG methylation correlated with activation of the TP53 pathway in DCM. Integrated analysis of LADs, CpG methylation, and gene expression indicated the combined role of LADs and CpG methylation in regulation of expression of gene targets of over a dozen key TFs involved in cell death/survival, cell cycle, and metabolism. Collectively, the results implicate redistributed LADs, altered CpG methylation, and dysregulated gene expression in the pathogenesis of DCM in laminopathies.

Our findings showing extensive interactions of LMNA with the genome in human cardiac myocytes, while novel, are in accord with the findings in other species and mammalian cell types. ^{3, 14, 17, 62} LADs were initially identified by the DamID method, which is a proximity labeling assay whereby bacterial DNA adenine methyltransferase (Dam) is tethered to a nuclear envelope protein, such as LMNA. The tethering results in methylation of adenine nucleotides that are in contact with the nuclear envelope protein, which is then identified by visualization or sequencing. ^{14, 63} More recently, LADs have been identified by ChIP-Seq method, upon chromatin fragmentation either by sonication or by digestion with micrococcal nuclease and subsequent sequencing of the antibody precipitated DNA fragments. ¹⁸ Despite the differences, all indicate extensive interactions of LMNA with chromatin through LADs, often involving a third of the genome and ranging in size from 10 Kbp to larger than 20 Mbp.^{14, 17, 18, 21} LADs were validated in an independent set of LMNA ChIP experiments in both control as well as in DCM hearts. However, testing for replication of the findings in independent heart samples is not feasible as each LMNA mutation is rare or private. Likewise, mutation-specific analysis is not feasible because of the unique nature of the LMNA mutations or their rare population frequencies. Our findings, while novel in cardiac myocytes, are in principle in accord with the existing data on interactions of LMNA with the genome and its important role in regulation of gene expression.

Patients with DCM associated with the LMNA pathogenic variants also had advanced heart failure, which raises the question of whether the changes in gene expression were due to localization and re-distribution of LADs alone or the presence of concomitant heart failure. Comparison of the DEGs located in the GoL and LoL regions between LMNA-associated DCM and a garden variety of heart failure patients, suggested that differential expression of genes located in LADs was mainly a function of LADs and not heart failure per se (Online Figure VIII). Notwithstanding these findings, LADs are expected to be dynamic in nature and change in heart failure or other pathological states. Likewise, LMNA is known influence spatial re-organization of the chromatin ^{3, 15, 64–66}. In general, genes located in the nuclear periphery are transcriptionally less active than those located within the center of chromatin. LMNA being a nuclear envelop protein interacts with the chromatin mostly in the periphery and contributes to spatial localization of the genes within the nucleus. ^{3, 15} Consequently, structural changes in the LMNA protein, whether due to mutations or altered mechanical stress, would be expected to exert direct (direct binding) and indirect (altered spatial organization of chromatin) effects on gene expression. Analysis of the RNA-Seq data in LMNA-associated DCM showed that only a subset of about 9% (407/4,597) of the DEGs were associated with LAD and the remainder were not located in LAD regions and hence, likely were changed due to chromatin spatial re-organization in heart failure. Therefore, LMNA through LADs as well as spatial distribution of chromatin is expected to affect gene expression under various pathological conditions, including in LMNA-associated DCM and heart failure. As regards the observed changes in the CpG methylation and LADs, the study design does not enable unambiguous identification of the changes that result from the presence of the LMNA mutations from those that are the consequence of advanced heart failure, as patients with the LMNA mutations also had advanced heart failure. In the broader view, however, LADs are expected to be dynamic in nature and change in heart failure as well, as discussed above.

Our data identified TP53 as a major activated pathway in DCM associated with LMNA mutations. The findings are in part in accord with the previous data reporting elevated TP53 protein levels in hearts of patients with DCM. ^{67, 68} Likewise, TP53-responsive microRNAs, including miR-192, have been identified as potential biomarkers in heart failure post myocardial infarction. ⁶⁹ Analysis of the publicly available RNA-Seq datasets, however, did not identify activation of TP53 in patients with a garden variety form of heart failure, including heart failure due to myocardial ischemia, suggesting likely specific activation of the TP53 pathway in DCM in laminopathies. ^{70, 71} Deletion of Tp53 gene in mice, while initially protects the heart from pressure overload-induced cardiac hypertrophy, induces age-dependent cardiac hypertrophy and heart failure through dysregulation of genes involved in a diverse array of biological pathways, including apoptosis, autophagy, mitochondrial biogenesis and bioenergetics, cell cycle, and growth arrest, among others. ⁷² The findings, collectively, point to the significant role of the TP53 protein in regulating myocardial structure and function.

In the present study, LADs were detected in isolated human cardiac myocyte nuclei by the LMNA ChIP-Seq method and the EDD algorithm. EDD has been shown to have superior sensitivity to the width of enriched domains and more robust detection of LADs than the other peak detectors. ³² While there is no other data in human cardiac myocytes, LADs, detected by the EDD algorithm, have been defined in human primary dermal fibroblasts and in differentiated human adipocyte progenitors. ^{32, 65} LADs common to control and DCM samples comprised 251 Mbp genomic regions, which might be constitutive LADs. Comparing the candidate constitutive LAD genomic regions with the existing databases identified 181 Mbp (72% base pair overlap) and 136 Mbp (54% base pair overlap) overlap with dermal fibroblasts and adipocyte progenitors, respectively. ^{32, 65} Likewise, comparing the CpG methylation data in the present study involving over 900,000 CpG sites and the ENCODE CpG methylation by RRBS (~ 650,000 CpG sites), showed a strong correlation between the two data sets (R²= 0.9533). ⁷³ Accordingly, methylation states of ~71.5% of the CpG sites (coverage ≥10) in the present datasets were concordant with those in the ENCODE dataset. ⁷³

A number of mechanisms have been implicated in regulation of gene expression by LADs. LMNA, directly or indirectly through association with retinoblastoma protein (RB1) and E2F TFs, recruits a number of epigenetic factors, including histone modifying enzymes, suppressor of variegation 3–9 homolog (SUV39H1), euchromatic histone lysine methyltransferase 2 (EHMT2), histone deacetylase 3 (HDAC3), and DNA methyltransferase 1 (DNMT1) to chromatin ^74–80,81. LADs are typically enriched in repressive histone marks, such as H3K9me3 and H3K27me3. ^{18–20, 22}. LADs are characterized by methylation of the CpG islands and spatially segregate with active histone marks H3K4me2/3, H3K9ac, and H3K27ac during interphase ^{21, 79, 82–85}. Finally, LADs also associated with three-dimensional organization of chromatin within the nucleus through association with pericentric heterochromatin as well as nucleolus. ^{3, 18, 21, 66, 86} Given that the majority of the DEGs were outside of the LAD regions, one may speculate that restructuring of the chromatin by the redistributed LADs is a major mechanism regulating gene expression in DCM.

The number of PCM1+ nuclei was smaller in DCM than in control hearts, which might reflect known myocyte drop out, due to apoptosis, necrosis, etc in DCM. Likewise, LMNA protein levels were modestly reduced in LMNA-associated DCM than in control hearts. To account for such confounding issues, equal amounts of ChIP libraries were used for sequencing. LMNA ChIP-Seq was performed in isolated cardiac myocyte nuclei, whereas CpG methylation and RNA-Seq were performed in whole heart DNA and RNA, respectively. Consequently, whereas the studies define extensive interactions of LMNA with the genome in cardiac myocytes, the findings with regards to CpG methylation and transcriptomic changes are likely influenced by cellular admixture of the myocardium. The limitation was in part imposed by the difficulty in isolating human adult cardiac myocytes for RNA-Seq as opposed to isolation of myocyte nuclei, which has recently become feasible due to specific expression of PCM1 in cardiac myocyte nuclei as opposed to other cardiac cells. ^28–30 Consequently, association of LADs with the transcript levels and CpG methylation are not myocyte -specific and are expected to be diluted by the CpG methylation states in other cardiac cells, particularly endothelial cells and fibroblasts, which are the other common cells in the heart. ^{53, 54} Nevertheless, despite heterogeneity of cellular origins of the transcripts and CpG methylation, significant associations between LADs and CpG methylation status and the transcript levels were observed. The modest effects of CpG methylation on gene expression in DCM, while in accord with the previous studies on the role of CpG methylation on gene expression, might be partially reflective of the cellular admixture as well the sum effects of methylation states of multiple CpG sites on each gene promoter. ⁶⁰ Nevertheless, the presence of LADs was associated with increased CpG methylation and reduced transcript levels of coding and non-coding genes, offering a mechanism for the pathogenesis of the phenotype in laminopathies in general, and in DCM in particular.

In conclusion, the findings indicate extensive interactions of LMNA with chromatin in human cardiac myocytes and re-arrangement of LMNA-chromatin interactions in DCM. Re-structuring of LADs are associated with altered CpG methylation and dysregulated expression of a large number of coding and non-coding genes, involved in cell death/survival, cell cycle, and metabolism, among others, in DCM. The findings provide novel insights into the molecular pathogenesis of DCM in laminopathies.

NOVELTY AND SIGNIFICANCE.

What Is Known?

• Mutations in the LMNA gene, encoding lamin A/C (LMNA), cause a diverse array of phenotypes, including dilated cardiomyopathy (DCM), which are collectively referred to as laminopathies.

• LMNA is a nuclear envelop protein and is known to interact with chromatin at regions referred to as Lamin-Associated Domains (LADs).

• LAD and their functional significance in human cardiac myocytes are undefined.

What New Information Does This Article Contribute?

• We identified about 330 LADs with an average size of 2 Mbp each in both control human cardiac myocytes and myocytes isolated from DCM hearts associated with the pathogenic variants in the LMNA gene.

• LADs comprised about 20% of the genome in human cardiac myocytes and were predominantly located in the heterochromatin regions.

• LADs were associated with increased CpG methylation and suppressed gene expression.

• LADs were redistributed in DCM, which was associated with altered expression of about 4,500 coding genes and 800 long non-coding RNAs (lncRNAs).

• Integrated analysis of LADs, CpG methylation, and transcript levels indicated the combined roles of LADs and CpG methylation in regulation of expression of genes involved in cell death/survival, cell cycle, and metabolism.

LMNA is a nuclear envelope protein that interacts with chromatin, influences its spatial reorganization, and affects accessibility of the chromatin to transcription. In human cardiac myocytes, LADs interact with about 20% of the genome, and influence CpG methylation and transcription of a diverse array of genes belonging to various biological pathways. LADs are redistributed in LMNA-associated DCM, which results in altered CpG methylation and gene expression, including those regulated by TP53, NUPR1, KDM5B, and NFE2L2. The findings implicate LADs, through genomic reorganization, CpG methylation, and gene expression, in the pathogenesis of DCM in laminopathies.

Nonstandard Abbreviations and Acronyms:

ChIP

Chromatin immunoprecipitation

CPM

Count per million

DAPI

4′, 6 Diamidino-2-phenylindole dihydrochloride

DCM

Dilated cardiomyopathy

DDR

DNA damage response

DEGs

Differentially expressed genes

DMS

Differentially methylated sites

E2F

E2F transcription factor

EDD

Enriched domain detector

FDR

False discovery rate

GoL

Gain of LAD

GSEA

Gene Set enrichment analysis

LAD

Lamin-associated domain

LMNA

Lamin A/C

LoL

Loss of LAD

NCOA1

Nuclear Receptor Coactivator 1

NEF2L2

Nuclear factor erythroid 2 like 2

NUPR1

Nuclear protein 1

PCR

Polymerase chain reaction

PCM1

Pericentriolar material 1

RNA-Seq

RNA-sequencing

TF

Transcription factor

TP53

Tumor protein 53