ORIGINAL RESEARCH article

Front. Bioeng. Biotechnol., 12 March 2020

Sec. Computational Genomics

Volume 8 - 2020 | https://doi.org/10.3389/fbioe.2020.00187

The Joint Analysis of Multi-Omics Data Revealed the Methylation-Expression Regulations in Atrial Fibrillation

  • 1. Department of Cardiology, Shanghai Tenth People’s Hospital, Tongji University School of Medicine, Shanghai, China

  • 2. Department of Pediatric Cardiovascular, Xin Hua Hospital, School of Medicine, Shanghai Jiao Tong University, Shanghai, China

  • 3. Department of Cardiology, Shanghai Tongji Hospital, Tongji University School of Medicine, Shanghai, China

  • 4. Shanghai East Hospital of Clinical Medical College, Nanjing Medical University, Shanghai, China

  • 5. Department of Cardiovascular Surgery, Shanghai East Hospital, Tongji University School of Medicine, Shanghai, China

  • 6. Department of Cardiovascular Surgery, National Clinical Research Center for Geriatric, West China Hospital, Sichuan University, Chengdu, China

  • 7. Department of Cardiovascular Surgery, Zhongshan Hospital, Fudan University, Shanghai, China

Abstract

Atrial fibrillation (AF) is one of the most prevalent heart rhythm disorder. The causes of AF include age, male sex, diabetes, hypertension, valve disease, and systolic/diastolic dysfunction. But on molecular level, its mechanisms are largely unknown. In this study, we collected 10 patients with persistent atrial fibrillation, 10 patients with paroxymal atrial fibrillation and 10 healthy individuals and did Methylation EPICBead Chip and RNA sequencing. By analyzing the methylation and gene expression data using machine learning based feature selection method Boruta, we identified the key genes that were strongly associated with AF and found their interconnections. The results suggested that the methylation of KIF15 may regulate the expression of PSMC3, TINAG, and NUDT6. The identified AF associated methylation-expression regulations may help understand the molecular mechanisms of AF from a multi-omics perspective.

Introduction

Atrial fibrillation (AF), one of the most prevalent heart rhythm disorders, is a potential cause of heart failure and ischemic stroke with high morbidity and mortality (; ). The cause of AF is multifactorial which include age, male sex, diabetes, hypertension, valve disease, and systolic/diastolic dysfunction (; ; ). Depends on how often atrial fibrillation occurs and how it responds to treatment, AF is roughly divided into two major subtypes-paroxysmal atrial fibrillation (PAF) and persistent atrial fibrillation (PeAF). In the treatment of AF, drugs were the first choice, non-drug therapies were used only when drug therapy failed or patients could not tolerate the medication. In contrast to the extensive knowledge of etiology, the underlying mechanism of AF remains elusive. Further study of the potential mechanisms of AF could provide novel strategies for the treatment and management to increase quality of life and reduce economic burden of social ().

With the development of next generation sequencing (NGS) technologies, growing evidence have demonstrated that AF is a disease with a significant genetic contribution. Previous studies have filtered novel genetic variants and candidate genes including transcriptional factor genes (PITX2, PRRX1, ZFHX3, NKX2.5, TBX5), ion channel genes (KCNN3, HCN4, CACNA1C, SCN5A, KCNQ1, KCNH2), and caveolin genes (CAV1 and CAV2) (; ; ; ; ). However, these genes explain only a small fraction of the biology and genetic underpinnings of AF.

Epidemiological studies have demonstrated that genetic, environmental, behavioral, and clinical factors contribute to AF pathogenic mechanism (). Emelia J. B performed genome-wide methylation using whole blood samples from 183 prevalent AF and 220 incident AF cases. They examined the association between DNA methylation and GWAS loci, suggesting DNA methylation might be a possible mechanism through which AF-specific genetic variations affect gene regulation (). To date, only a few studies have investigated differential DNA methylation as a predictor biomarker at specific candidate loci that were previously associated with AF.

Therefore, we applied DNA methylation profiling study to identify the likely rare damaging variants and putative candidate genes from 10 patients with persistent atrial fibrillation (PeAF), 10 patients with paroxymal atrial fibrillation (PAF) and 10 healthy individuals. Interestingly, we identified top 10 genes (KIF15, ABCA3, FOXG1, VGF, PDE4D, EIF3C, CNTNAP5, SHOX2, VGF, TRIM59) as functional candidate genes and the expression level are significantly increased in PeAF and PAF patients than control. Given the importance of DNA methylation to gene expression, we investigated the gene expression of the same participants using RNA sequencing. We also defined top 10 genes (EPN3, EMD, SMCO4, F2RL2, TMED1, PSMC3, PDZD11, NUDT6, TINAG, GALNT5) in gene expression data and the expression pattern of these genes was significantly different between PeAF and PAF. These results have improved our understanding of the underlying mechanism and offer new insights into the potential pathway of AF, which could provide novel therapeutic option for this disease.

Materials and Methods

Atrial Fibrillation Patients

Ten patients with paroxymal atrial fibrillation (g1), 10 patients with persistent atrial fibrillation (continuous atrial fibrillation lasting more than 12 months) (g2) and 10 healthy individuals (g3) were enrolled in this study (Table 1). All patients were subjected to detailed medical evaluation, which included medical history, physical examination, electrocardiography (ECG), and echocardiography. Patients with chronic heart failure, coronary heart disease, cardiomyopathy, hyperthyroidism or chronic pulmonary heart disease were excluded.

TABLE 1

Coronary
AgeWeightHeightDiabetesangiographyLVEFLeft atrialDuration of
No.(years)Gender(Kg)(cm)SmokingHypertensionmellitusor CTA(%)diameter (mm)AF (years)
169Male76169NoYesNoNegative7040–
263Male64170NoNoNoNegative5946–
363Male70170NoNoNoNegative6639–
469Male67173NoYesNoNegative6746–
569Male75165NoNoNoNegative7036–
661Male76176NoYesYesNegative6042–
764Male52168YesNoNoNegative6440–
864Male71181NoYesYesNegative6339–
961Male87167YesYesNoNegative6237–
1066Male82173NoYesNoNegative6342–
1163Male86176NoYesNoNegative57462.5
1263Male80178NoNoNoNegative68553
1364Male70170NoNoNoNegative67414
1464Male84164NoYesNoNegative55482
1565Male73169NoYesNoNegative69553.5
1666Male66168NoYesNoNegative64454
1767Male80175NoYesYesNegative59472.5
1867Male73165YesYesNoNegative59473
1963Male61164NoNoNoNegative73492
2067Male90178NoYesNoNegative70582.5

Demographic characteristics of AF patients.

LVEF, left ventricular ejection fraction; AF, atrial fibrillation; CTA, CT angiography.

The study was conducted in accordance with the Declaration of Helsinki, and the protocol used to collect human heart tissue was approved by the Ethics Committee of Shanghai East Hospital (DI: 0402017).

Written informed consents to participate in this study were provided by all the enrolled patients before operation of fibrillation ablation. The left atrial appendage tissues which were abandoned during isolated surgical ablation were collected. Normal left atrial appendages were collected from healthy male donors.

The Methylation Profiles

The DNA methylation status of 850K probes in the 30 samples was measured using Methylation EPICBead Chip. The raw data was quality controlled and preprocessed using R/Bioconductor package minfi1 (). The beta value ranged from 0 to 1 was calculated to represent how each position was methylated. 1 meant high methylation and 0 meant low methylation.

The RNA Sequencing Profiles

The total RNAs were extracted using RNeasy Mini Kit (Cat#74106, Qiagen) and the RNA integrity was checked using Agilent Bioanalyzer 2100 (Agilent technologies, Santa Clara, CA, United States). Qualified total RNA was further purified by RNAClean XP Kit (Cat A63987, Beckman Coulter Inc., Kraemer Boulevard, Brea, CA, United States) and RNase-Free DNase Set (Cat#79254, QIAGEN, GmBH, Germany). Pair-end sequencing reads were generated using Illumina data collection software. First, the reads were mapped onto human reference genome GRCh38 using Hisat2 (version:2.0.42) (). Then, Stringtie (version:1.3.03) () was used to calculate the FPKM (Fragments Per Kilobase of exon model per Million mapped reads).

Feature Selection Algorithm

There were 866,091 methylation probes and 50,868 RNAs. The number of features were extremely large. It was difficult to select key features using traditional statistical methods. Therefore, we adopted the latest machine learning based feature selection method Boruta to get the key methylation probes and RNAs.

Boruta () is a feature selection method based on random forest. It can select sample group relevant features effectively. First, it will shuffle the features to create many permuted datasets. Then, it will evaluate the importance score of each feature in the original actual dataset and permuted datasets. Then, it will compare the actual importance score with permuted scores and find the features with significantly higher actual importance scores than permuted scores. After multiple iterations, it will select all the sample group relevant features. The python code from https://github.com/scikit-learn-contrib/boruta_py was used to apply the Boruta feature selection algorithm.

Classification Predictor

To evaluate how well the selected features can classify the samples, we built an SVM (Support Vector Machine) classifier using the methylation data and another RNA-Seq data based SVM classifier. The svm function in R package e101714 () was used to apply the SVM classification algorithm.

LOOCV (leave-one-out cross validation) was used to objectively evaluate the classification performance. Each time, one sample was treated as test sample while all the other samples were used to train the model. After 30 rounds, all samples had been tested once and the overall accuracy was calculated based on the confusion matrix. In confusion matrix, the actual sample groups were compared with predicted sample groups.

Results

The Key Methylation Features Identified With Boruta

We ran Boruta feature selection algorithm on the methylation data and got 10 key methylation features. These 10 key methylation features were listed in Table 2. The probes were annotated to genome positions (Genome Build 37) and genes using the official annotation file from Illumina. Sometime, one position may be associated with multiple genes. Therefore, the 10 methylation probes can be mapped onto 15 gene symbols. We checked the GO annotation of these genes and found that cg16703882 (SHOX2) was associated with GO:0007507: heart development which was closely relevant to AF.

TABLE 2

ILMNIDChromosomePositionStrandUCSC Ref gene name
cg00702638344803293RKIF15; KIAA1143
cg02331561162391081FABCA17P; ABCA3
cg029913381429236017RFOXG1
cg040841577100809049FVGF
cg05995159559325256RPDE4D
cg063576151628403195RMIR6862-2; MIR6862-1; EIF3CL; EIF3C
cg113445662124782885FCNTNAP5
cg167038823157823479RSHOX2
cg211862997100808810RVGF
cg268560803160167746RTRIM59

The 10 key methylation features identified by Boruta.

We plotted the heatmap of these 10 key methylation features in Figure 1. It can be seen that most of the 10 patients with paroxymal atrial fibrillation (g1), 10 patients with persistent atrial fibrillation (g2) and 10 healthy individuals (g3) were cluster into the right groups.

FIGURE 1

The Key Gene Expression Features Identified With Boruta

Similarly, we ran Boruta feature selection algorithm on the RNA-Seq gene expression data and got 10 key gene expression features. The 10 key gene expression features were given in Table 3. We also plotted their heatmap in Figure 2. The clusters were also largely correct.

TABLE 3

Gene IDNameDescriptionGRCh38 locus
ENSG00000049283EPN3Epsin 317:50532543-50543750
ENSG00000102119EMDEmerinX:154379197-154381523
ENSG00000166002SMCO4Single-pass membrane protein with coiled-coil domains 411:93478472-93543508
ENSG00000164220F2RL2Coagulation factor II (thrombin) receptor-like 25:76615482-76623434
ENSG00000099203TMED1Transmembrane p24 trafficking protein 119:10832438-10836318
ENSG00000165916PSMC3Proteasome 26S subunit, atpase 311:47418769-47426473
ENSG00000120509PDZD11PDZ domain containing 11X:70286595-70290514
ENSG00000170917NUDT6Nudix hydrolase 64:122888697-122922968
ENSG00000137251TINAGTubulointerstitial nephritis antigen6:54307859-54390152
ENSG00000136542GALNT5Polypeptide N-acetylgalactosaminyltransferase 52:157257598-157314211

The 10 key gene expression features identified by Boruta.

FIGURE 2

The Classification Performance of Methylation and Gene Expression

From Figures 1, 2, we can see that both methylation and gene expression features can correctly cluster most samples. But we would like to evaluate their performance objectively and quantitatively. Therefore, we applied LOOCV to test the SVM classifiers of methylation and gene expression features. The confusion matrixes of methylation features and gene expression features were listed in Tables 4, 5, respectively.

TABLE 4

Predicted g1Predicted g2Predicted g3
Actual g1820
Actual g2370
Actual g30010

The confusion matrix of key methylation features.

TABLE 5

Predicted g1Predicted g2Predicted g3
Actual g1901
Actual g2091
Actual g3109

The confusion matrix of key gene expression features.

From Table 3, we can see that the AF patients (g1 + g2) and healthy individuals (g3) were perfectly classified using the methylation features, but the methylation data did not have great performance on classifying the subtype of AF (g1 and g2). From Table 4, we can see that the two subtype of AF, paroxysmal atrial fibrillation (g1) and persistent atrial fibrillation (g2), had very different gene expression pattern. In other words, the methylation data and gene expression data complement each other. The methylation data can be used to predict the AF and the gene expression data can be used to classify the subtypes of AF.

We checked the wrongly predicted samples in Tables 3, 4. They were different. Within the 30 samples, 22 samples had the same predicted labels by expression and methylation. All these 22 samples were correctly predicted. For the 8 inconsistent samples between expression and methylation predictions, at least one of the two predictions (the expression-based prediction and the methylation-based prediction) was correct. In other words, all the samples can be corrected classified based on either expression or methylation. The expression-based prediction and the methylation-based prediction were complementary.

The Methylation-Expression Regulation Network

We mapped the genes of methylation features and gene expression features to the STRING network (Version 11.05) () and visualized the network using R package igraph ()6 to identify the potential relationship between two candidate genes sets. The methylation-expression regulation network was shown in Figure 3. In the network, the methylation and expression genes were marked in red and green. The methylation genes located in three clusters: EIF3CL-EIF3C, KIF15-TRIM59, and SHOX2-FOXG1-CNTNAP5. The three expression genes (PSMC3, TINAG, and NUDT6) were connected with methylation gene KIF15. Even EPN3 can be indirectly connected to KIF15. That made KIF15 at the center of the network. These results suggested that KIF5 may play important roles in the pathogenesis of AF.

FIGURE 3

Discussion

DNA methylation, a pre-transcriptional modification characterized by the addition of methyl groups to specific nucleotides, regulates the stability of gene expression states and maintains genome integrity by collaborating with proteins that modify nucleosomes (; ; ). Previous studies considered that changes in DNA methylation states contribute to the regulation of biological processes underlying AF, such as fibrosis, atrial dilatation, atrial fibroblast proliferation and differentiation from fibroblasts into myofibroblasts (). To further enhance the biological understanding of the atrial fibrillation, our study focused on DNA methylation, particularly with respect to how it relates to mRNA expression. Among our two gene sets of top 10 genes, we found PDED4, SHOX2, and EMD were the most important genes for AF which have been reported associated with AF in previous reference.

Atrial fibrillation is reported to be associated with a profound remodeling of membrane receptors and alterations in cAMP dependent regulation of Ca2+ handling. PDE4 is expressed in human atrial myocytes and accounts for approximately 15% of PDE (phosphodiesterase) activity (). PDE4D encoded protein has 3’,5’-cyclic-AMP phosphodiesterase activity and degrades cAMP, which acts as a signal transduction molecule in multiple cell types and represents the major PDE4 subtype (). The activity of PDEs decreased with age, and the relative PDED4 activity was lower in patients with permanent atrial fibrillation than in age-matched sinus rhythm controls (). Previous study provided evidence that patients with pAF were found to have a decreased PDE4 activity as compared with patients in sinus rhythm ().

Short Stature Homeobox 2 (SHOX2) is a member of the homeobox family of genes in which mutations associated with early-onset and familial AF (). SHOX2 is considered as a key regulator of sinus node development of which deficiency could lead to bradycardia in animal models (). Previous study demonstrated SHOX2 was susceptible for SND and AF by screening 98 SND patients and 450 individuals with AF. In the heart development of mouse and zebrafish, they also proved SHOX2 plays an important role, the mutation of SHOX2 could lead to severe bradycardia (; ).

Emerin (EMD) encodes a serine-rich nuclear membrane protein which located on the cytoplasmic surface of the inner nuclear membrane and related to X-linked Emery-Dreifuss muscular dystrophy (EDMD) (). Previous study found a nonsense mutation in EMD from two EDMD families which is associated with X-linked recessive inheritance, result in serious cardiac complication, including AF (). Cardiologic assessment revealed slow atrial fibrillation in a recent case of a 65-year-old male patient with a hemizygous duplication of 5 bases in exon 6 of the EMD, gene on the X chromosome (; ; ).

Statements

Data availability statement

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation, to any qualified researcher.

Ethics statement

The studies involving human participants were reviewed and approved by the Ethics Committee of Shanghai East Hospital. The patients/participants provided their written informed consent to participate in this study. Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this manuscript.

Author contributions

YZ and CG conceived and designed the experiments. BL, XS, and KD performed the experiments. ML, YQ, and SZ analyzed the data. BL, XS, and YZ wrote the manuscript. All authors read and approved the final version of the manuscript.

Funding

The present study was funded by the National Key Research and Development Program (Grant No. 2018-YFC-1312505 to YZ), the National Natural Science Foundation of China (Grant No. 81770408 to CG), and (Grant No. 81900280 to XS).

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

References

Summary

Keywords

atrial fibrillation, methylation, multi-omics, feature selection, classification

Citation

Liu B, Shi X, Ding K, Lv M, Qian Y, Zhu S, Guo C and Zhang Y (2020) The Joint Analysis of Multi-Omics Data Revealed the Methylation-Expression Regulations in Atrial Fibrillation. Front. Bioeng. Biotechnol. 8:187. doi: 10.3389/fbioe.2020.00187

Received

19 December 2019

Accepted

26 February 2020

Published

12 March 2020

Volume

8 - 2020

Edited by

Peilin Jia, The University of Texas Health Science Center at Houston (UTHealth), United States

Reviewed by

Yan Yao, Chinese Academy of Medical Sciences, China; Jian Zhong Zhou, The First Affiliated Hospital of Chongqing Medical University, China

Updates

Copyright

*Correspondence: Changfa Guo, Yangyang Zhang, ;

†These authors have contributed equally to this work

This article was submitted to Bioinformatics and Computational Biology, a section of the journal Frontiers in Bioengineering and Biotechnology

Disclaimer

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.

Outline

Figures

Cite article

Copy to clipboard


Export citation file


Share article

Article metrics