Identification of key genes and pathways affected in epicardial adipose tissue from patients with coronary artery disease by integrated bioinformatics analysis

Liao Tan; Qian Xu; Qianchen Wang; Ruizheng Shi; Guogang Zhang

doi:10.7717/peerj.8763

. 2020 Mar 25;8:e8763. doi: 10.7717/peerj.8763

Identification of key genes and pathways affected in epicardial adipose tissue from patients with coronary artery disease by integrated bioinformatics analysis

Liao Tan ^1,², Qian Xu ^2,³, Qianchen Wang ^1,², Ruizheng Shi ^1,^2,^✉, Guogang Zhang ^1,^2,^4,^✉

Editor: Burcu Bakir-Gungor

PMCID: PMC7102503 PMID: 32257639

Abstract

Background

Coronary artery disease (CAD) is a common disease with high cost and mortality. Here, we studied the differentially expressed genes (DEGs) between epicardial adipose tissue (EAT) and subcutaneous adipose tissue (SAT) from patients with CAD to explore the possible pathways and mechanisms through which EAT participates in the CAD pathological process.

Methods

Microarray data for EAT and SAT were obtained from the Gene Expression Omnibus database, including three separate expression datasets: GSE24425, GSE64554 and GSE120774. The DEGs between EAT samples and SAT control samples were screened out using the limma package in the R language. Next, we conducted bioinformatic analysis of gene ontology terms and Kyoto Encyclopedia of Genes and Genomes pathways to discover the enriched gene sets and pathways associated with DEGs. Simultaneously, gene set enrichment analysis was carried out to discover enriched gene functions and pathways from all expression data rather than DEGs. The PPI network was constructed to reveal the possible protein interactions consistent with CAD. Mcode and Cytohubba in Cytoscape revealed the possible key CAD genes. In the next step, the corresponding predicted microRNAs (miRNAs) were analysed using miRNA Data Integration Portal. RT-PCR was used to validate the bioinformatic results.

Results

The three datasets had a total of 89 DEGs (FC log2 > 1 and P value < 0.05). By comparing EAT and SAT, ten common key genes (HOXA5, HOXB5, HOXC6, HOXC8, HOXB7, COL1A1, CCND1, CCL2, HP and TWIST1) were identified. In enrichment analysis, pro-inflammatory and immunological genes and pathways were up-regulated. This could help elucidate the molecular expression mechanism underlying the involvement of EAT in CAD development. Several miRNAs were predicted to regulate these DEGs. In particular, hsa-miR-196a-5p and hsa-miR-196b-5p may be more reliably associated with CAD. Finally, RT-PCR validated the significant difference of OXA5, HOXC6, HOXC8, HOXB7, COL1A1, CCL2 between EAT and SAT (P value < 0.05).

Conclusions

Between EAT and SAT in CAD patients, a total of 89 DEGs, and 10 key genes, including HOXA5, HOXB5, HOXC6, HOXC8, HOXB7, COL1A1, CCND1, CCL2, HP and TWIST1, and miRNAs hsa-miR-196a-5p and hsa-miR-196b-5p were predicted to play essential roles in CAD pathogenesis. Pro-inflammatory and immunological pathways could act as key EAT regulators by participating in the CAD pathological process.

Keywords: Bioinformatics analysis, Epicardial adipose tissue, Gene expression omnibus, Coronary artery disease, Subcutaneous adipose tissue

Introduction

Atherosclerosis is a chronic artery disease that is the major cause of coronary artery disease (CAD) and stroke. It is a leading cause of death worldwide (Herrington et al., 2016; Jia et al., 2014; Zhang, Lai & Jia, 2015). In recent years, visceral adipose tissue (VAT) has been identified as affecting atherosclerosis and CAD pathogenesis persistently releasing pro-inflammatory and decrease anti-inflammatory adipokines into the circulation via a paracrine or endocrine pathway (Alexopoulos, Katritsis & Raggi, 2014; Wu & Zhao, 2006). Epicardial adipose tissue (EAT) is one of the most important VAT components. EAT, due to its position close to the myocardium, which shares the same microcirculation as the coronary artery, has a unique effect on CAD (Nakanishi et al., 2014; Patel et al., 2016).

So far, clinical studies have shown that EAT expansion is an independent CAD risk factor (Herrington et al., 2016; Mahabadi et al., 2013; Nakanishi et al., 2014; Liu et al., 2019). Some EAT functions were reported in cell and animal studies. EAT releases some factors (apelin (Yao et al., 2015), adiponectin (Ouchi & Walsh, 2008), MCP-1 (Niu & Kolattukudy, 2009)) that effect the myocardium and coronary arteries via paracrine and vasocrine pathways. Furthermore, exosomes released from EAT could carry molecules (proteins, RNA and lipids) that establish cross-talk between the pathological EAT and the coronary artery (Thomou et al., 2017). The NALP3/inflammasome pathway is activated by microbial colonisation in EAT in CAD patients (Guauque-Olarte et al., 2011). The expression of proteins involved in oxidative stress, metabolism regulation, gene transcription regulation, and angiogenesis were significantly increased in CAD patient EAT (Vacca et al., 2016).

However, it is not completely understood how EAT participates in CAD. One of the major limitations in studying EAT function is that only patients undergoing cardiac surgery are studied. Collecting EAT from healthy subjects is not possible for obvious ethical reasons. Therefore, subcutaneous adipose tissue (SAT) was always used as the second-best control in previous studies (Gruzdeva et al., 2017; Salgado-Somoza et al., 2010; Hirata et al., 2011).

We attempted to verify the pre-existing function and predict a new effect of EAT in pathological CAD processes. Here, microarray analysis of EAT and SAT was conducted based on three independent expression array databases. These gene expression features could help discover new CAD biomarkers and pioneer therapeutic strategies.

Materials and Methods

Data source

The microarray expression data sets (GSE24425, GSE64554 and GSE120774) were downloaded from the Gene Expression Omnibus (GEO) database. A total of 28 EAT samples and 27 SAT control samples were collected from the same CAD patients undergoing coronary artery bypass grafting surgeries (CABG). GSE24425 used the Illumina HumanWG-6 V3.0 expression beadchip containing six paired EAT and SAT samples. GSE64554 used the Illumina HumanHT-12 V3.0 expression beadchip. GSE120774 used the HuGene-1_0-st Affymetrix Human Gene 1.0 ST Array, including 13 EAT samples with EAT controls and nine EAT samples with eight SAT controls, respectively. The three datasets were all quantile normalised and log2-transformed.

Data pre-processing

Data pre-processing consisted of three units: transition from gene probes to gene symbols, data consolidation, and batch normalisation. First, the three series matrix files were annotated with an official gene symbol using the data table of the microarray platform, and the gene expression matrix files were obtained. The three gene expression matrix files were merged into one file using a Perl script. Gene probes without gene symbols or genes with more than one probe were eliminated or averaged, respectively. To ensure the integrity and comparability of the datasets, the batch normalisation of merged data was pre-processed by sva package (Leek et al., 2012) using the R language (R Development Core Team, 2018). Batch effects are the most widely recognised potential latent variable in genomic experiments. The sva package was used to eliminate the latent variables or unwanted heterogeneity in the high-throughput data. The batch normalisation including two steps: identification of potential impact factors and elimination of batch effects using the ComBat function.

Differentially expressed gene analysis

The differentially expressed genes (DEGs) between EAT samples and SAT samples were determined using the limma package in R. The thresholds were log2 (fold change) > 1 and P value < 0.05. DEG visualisation was done using a volcano map and heatmap using gglot2 (Ginestet, 2011) and the pheatmap package (Kolde, 2015).

Functional enrichment analysis of DEGs

The Cluster Profiler R package (Yu et al., 2012) and the database for annotation, visualisation, and integrated discovery (DAVID 6.8, http://david.ncifcrf.gov) (Huang et al., 2007) were used to functionally analyse and analyse the enriched pathways of the key DEGs in gene ontology (GO) terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways. The P value was corrected using the Benjamini method or false discovery rate (FDR) for multiple testing calibrations. The threshold was P < 0.05.

Gene set enrichment analysis

Gene set enrichment analysis (GSEA) (Subramanian et al., 2005) was performed using GSEA software. Gene sets used here were downloaded from the Molecular Signatures Database. Enrichment results satisfying a nominal P-value cut-off of < 0.05 with a FDR > 0.25 were considered statistically significant. MSigDB were download from http://software.broadinstitute.org/gsea/index.jsp (Liberzon et al., 2011).

PPI network construction and analysis

To explore the interacting genes, the search tool (STRING 10.5; http://string-db.org) (Szklarczyk et al., 2015) was employed to establish a DEG PPI network, which was drawn using Cytoscape (Kohl, Wiese & Warscheid, 2011). Interaction with a combined score > 0.4 was set as the cut-off point. The most important module in the PPI network was identified using the plug-in Molecular Complex Detection (MCODE) (Bader & Hogue, 2003) of Cytoscape, an application to cluster a given network by topology, to find densely connected regions. The criteria for selection were as follows: MCODE scores > 5, degree cut-off = 2, node score cut-off = 0.2, max depth = 100, and k-score = 2. Subsequently, the maximal clique centrality (MCC) algorithm of CytoHubba (Chin et al., 2014) was used to explore the PPI network hub genes.

Regulating miRNA prediction

The online prediction tool microRNA Data Integration Portal (mirDIP) (http://ophid.utoronto.ca/mirDIP) (Tokar et al., 2018) was used to predict potential microRNA (miRNA) targeting. Ten top hub genes were submitted, and the top five predicted miRNAs of every gene were chosen and listed.

Assessment of the mRNA expression of hub genes using qRT-PCR

Eight EAT tissues and SAT tissues were collected from Xiangya Hospital. This study was approved by the Ethical Committee of Xiangya Hospital and conducted in accordance with the Declaration of Helsinki. In addition, each patient volunteered written informed consent. All tissues were immediately frozen in liquid nitrogen after resection and stored in liquid nitrogen. The clinical characteristics of included CAD patients were listed in Table 1.

Table 1. Clinical features of included CAD patients.

Clinical features of included patients	n = 8
Age	66.36 ± 8.33
Sex
Male	6
Female	2
Heart function grade of NYHA
I–II	6
III	2
IV	0
BMI (kg/m²)	23.01 ± 3.75
Hypertention	5
Diabetes	0
Hyperlipidemia	3
Smoking	4
Triple-vessel disease	8

Open in a new tab

Total RNA was extracted from tissues using TRIzol (Takara, Japan) according to the manufacturer’s instructions. Total RNA was reverse transcribed into cDNA using a cDNA Synthesis Kit (Takara, China). mRNA levels were tested using Nanodrop one (Thermo Fisher Scientific, Waltham, MA, USA). All reactions were performed on the Eppendorf Mastercycler ep realplex (2S; Eppendorf, Hamburg, Germany) using following cycling parameters: 95 °C for 30 s, followed by 40 cycles of 95 °C for 15 s, 60 °C for 34 s. RT-qPCR was performed using the FastStart Universal SYBR^® Green Master (ROX) (Takara, China). The primer sequences were listed in Data S1.

The relative expression level for each target gene was normalised by the Ct value of β-actin (internal control) using a 2^−ΔΔCt relative quantification method. A meaningful analysis between the two groups was performed by a paired t-test, and a P value < 0.05 was considered statistically significant.

Results

DEG analysis

Here, 28 EAT samples and 27 SAT samples of CAD patients from the GSE24425, GSE64554 and GSE120774 datasets were analysed. After pre-processing, the raw data were merged and normalised (Fig. S1). Based on the cut-off criteria (adjusted P value < 0.05 and |log2 foldchange (FC)| > 1), a total of 89 DEGs were identified, including 43 up-regulated and 46 down-regulated DEGs. A DEG expression heat map and volcano map were shown in Figs. 1 and 2.

Each column represents a adipose tissue sample, and each row represents a DEG. The gradual colour change from blue to red indicates the changing process from downregulation to upregulation. DEGs, differentially expressed genes; EAT, epicardial adipose tissue; SAT, subcutaneous adipose tissue.

Blue dot represent DEGs with fold change <1; red dot represent DEGs with fold change >1. P value < 0.05.

Gene ontology enrichment analysis

Functional enrichment analysis of DEGs was performed by using clusterProfiler package in the R language. The top 12 GO analysis results (Fig. 3) showed that genes were mainly enriched in G protein-coupled receptor binding, chemokine activity, chemokine receptor binding, retinoid binding, isoprenoid binding, CCR chemokine receptor binding, retinal binding, cytokine binding, cytokine activity, haptoglobin binding, molecular carrier activity, and phospholipase activator activity. All results were listed in Table S1.

The y-axis labels represent clustered GO terms; x-axis represent the number of gene enriched in GO clusters. GO, gene ontology.

Pathway enrichment analysis

Significantly enriched pathways of DEGs are shown in Table 2 based on the DAVID online tool that was used to perform KEGG pathway analysis. Genes were mainly enriched in cytokine–cytokine receptor interaction, chemokine signalling pathway, prion diseases, malaria, and arginine and proline metabolism. Enriched pathways identified from DEGs were fewer but more reliable due to relatively limited DEGs from the three data sources.

Table 2. The significant KEGG pathways enriched by DEGs. P value < 0.05. KEGG: Kyoto Encyclopedia of Genes and Genomes.

Category term	Count	P value
hsa04060: Cytokine–cytokine receptor interaction	6	0.003647442
hsa04062: Chemokine signalling pathway	5	0.00818195
hsa05020: Prion diseases	3	0.009459285
hsa05144: Malaria	3	0.019044109
hsa00330: Arginine and proline metabolism	3	0.019784372

Open in a new tab