Identification of genes associated with sudden cardiac death: a network- and pathway-based approach.

BACKGROUND: Sudden cardiac death (SCD) accounts for a large proportion of the total deaths across different age groups. Although numerous candidate genes related to SCD have been identified by genetic association studies and genome wide association studies (GWAS), the molecular mechanisms underlying SCD are still unclear, and the biological functions and interactions of these genes remain obscure. To clarify this issue, we performed a comprehensive and systematic analysis of SCD-related genes by a network and pathway-based approach.
METHODS: By screening the publications deposited in the PubMed and Gene-Cloud Biotechnology Information (GCBI) databases, we collected the genes genetically associated with SCD, which were referred to as the SCD-related gene set (SCDgset). To analyze the biological processes and biochemical pathways of the SCD-related genes, functional analysis was performed. To explore interlinks and interactions of the enriched pathways, pathway crosstalk analysis was implemented. To construct SCD-specific molecular networks, Markov cluster algorithm and Steiner minimal tree algorithm were employed.
RESULTS: We collected 257 genes that were reported to be associated with SCD and summarized them in the SCDgset. Most of the biological processes and biochemical pathways were related to heart diseases, while some of the biological functions may be noncardiac causes of SCD. The enriched pathways could be roughly grouped into two modules. One module was related to calcium signaling pathway and the other was related to MAPK pathway. Moreover, two different SCD-specific molecular networks were inferred, and 23 novel genes potentially associated with SCD were also identified.
CONCLUSIONS: In summary, by means of a network and pathway-based methodology, we explored the pathogenetic mechanism underlying SCD. Our results provide valuable information in understanding the pathogenesis of SCD and include novel biomarkers for diagnosing potential patients with heart diseases; these may help in reducing the corresponding risks and even aid in preventing SCD. 2021 Journal of Thoracic Disease. All rights reserved.

Introduction

Sudden cardiac death (SCD) can either be caused by cardiac-related or unknown factors, both of which ultimately lead to a sudden, nonviolent, natural death, occurring within 1 hour of the onset of acute symptoms (1,2). In Western countries, SCD is one of the most common causes of death. At present, approximately 250,000–350,000 people die of SCD every year in America, accounting for 63% of all deaths due to heart disease (3). The majority of these events occur in older adults, and are often concurrent with coronary heart disease (4,5). Children, teenagers, and young adults—but not infants—are the other major groups that suffer from SCD (6-8). In China, SCD results in more than 544,000 deaths each year (9). Over the past decades, there have been some progressive declines in other heart diseases due to major advances in treatment and preventive measures; in contrast, the SCD rate has only seen a slight reduction. Although a few factors, including atherosclerotic cardiovascular disease (CVD), may account for a large proportion risk for SCD, the pathogenesis and especially the molecular mechanisms underlying SCD are still not fully understood. Therefore, in the last decade, in order to allow for an early diagnosis and prevention of SCD for a significant percentage of young individuals, scientists have paid considerable attention to the molecular analysis of cardiac channelopathies and cardiomyopathies (10). Numerous studies have helped us to understand the pathological and molecular mechanisms of SCD. For example, it was found that the ATP-binding cassette B1 (ABCB1) gene encodes P-glycoprotein, which plays a vital role in digoxin bioavailability. The researchers found that, in users who had all three loci mutations of the ABCB1 gene, digoxin was much more highly related to SCD than in people who had no or only 1 T allele (11). Several other genes that are associated with SCD also have been found, such as genes encoding calcium-handling regulating proteins (RyR2, CASQ2, ATP2A2, NOS1AP, TRDN and CALM1) (12-17), long QT syndrome (KCNJ5, KCNJ2, KCNE2, AKAP9, SNTA1, and AKAP9), idiopathic ventricular fibrillation (DPP6 and KCNJ8), dilated cardiomyopathy (DCM) (NEBL), and hypertrophic cardiomyopathy (HCM) (NEXN) (18,19).

In our study, we comprehensively collected the SCD-related genes from published genetic association studies. Then, to identify the significant functional themes within these genetic factors, we performed biological enrichment analysis; to analyze the interactions among the enriched biochemical pathways, we conducted pathway crosstalk analysis. In order to further study the pathogenesis of the SCD in a more specific manner, an SCD-specific molecular network was constructed and evaluated within the frame of a human protein–protein interaction network. Two SCD-specific protein networks were inferred: one based on the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database via the Markov cluster algorithm, and the other based on Pathway Commons via the Steiner minimal tree algorithm. This research should give us valuable insights into the molecular mechanisms of SCD, and are likely to yield potential biomarkers for useful diagnostic and therapeutic strategies in SCD prevention.

Methods

Collection of SCD-associated genes

The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). We searched SCD-associated candidate genes by retrieving the human genetic association studies deposited in the PubMed (https://www.ncbi.nlm.nih.gov/pubmed/) and Gene-Cloud Biotechnology Information (GCBI) databases (https://www.gcbi.com.cn). We searched for reports linked to SCD with the following terms: (sudden cardiac death [MeSH]) and (genotype [MeSH] or alleles [MeSH]) or (polymorphism [MeSH]). By December 31, 2017, a total of 2,076 publications were retrieved from PubMed. Furthermore, we searched the publications that clearly mentioned 1 or more genes significantly related to SCD in the GCBI database, and 1,094 publications were retrieved. We reviewed all the abstracts of these publications, and chose the genetic studies of SCD. To remove the false-positive findings, the publications reporting negative or insignificant association were excluded. To ensure the conclusions were supported by the content, we reviewed the full articles of the remaining publications. From these studies, genes that were reported to be significantly related to SCD were chosen for our study. In some studies, several genes were reported have high associations with SCD, while the actual effects were found to be moderate; however, based on our inclusion criteria of gene collection, these genes were included in our study. Moreover, genes from genome-wide association studies (GWAS) that had a demonstrated genetic association with SCD at a significant genome-wide level or that were frequently mentioned in the original publications were included. Finally, we collected 257 genes that were related to SCD, and referred to these as the SCD-related gene set (SCDgset), for further study. To check the mRNA expression of each gene in normal human heart, three different databases: Genotype-Tissue Expression (GTEx) database (https://gtexportal.org/home/topExpressedGenePage), BioGPS (http://biogps.org/#goto=welcome) and Serial Analysis of Gene Expression (SAGE, https://www.genecards.org/) were applied.

Functional enrichment analysis of genes related to SCD

To analyze the biological themes of the SCD-related genes, we used the functional annotation tool DAVID (Database for Annotation, Visualization and Integrated Discovery; https://david.ncifcrf.gov/summary.jsp) with the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (https://www.kegg.jp/) and overrepresentation analysis (ORA), functional class scoring (FCS), and pathway topology (PT)-based approaches. DAVID is able to organize and condense a wide range of heterogeneous annotation content, such as Gene Ontology (GO) terms, protein domains, pathways etc., into term or gene classes (20,21). However, this database is updated relatively slowly. To gain insight into the underlying biology of differentially expressed genes and proteins, pathway analysis should be the first choice. For obtaining the most recent data and pathway enrichment, we combined the data from KEGG with that from DAVID and analyzed the enrichment by ORA, FCS, and PT-based approaches that provided us better specificity, sensitivity, and relevance in pathway analysis (22,23). It has been reported that the higher hierarchical level of GO terms in the tree structure is, the more explicit the biological function demonstrated (24). Therefore, only the leaf GO terms of biological processes with a false discovery rate (FDR) value less than 0.05 were kept as the remarkably enriched terms. We extracted all the pathways when 1 or more genes overlapped with candidate genes, and Fisher’s exact test for P value assignment was used to denote the significant overlap between the pathway and the input genes. The pathways with an FDR value less than 0.05 and which contained at least 5 genes were considered to be significantly enriched. Since there were numerous of pathways involved in cancer, we excluded these pathways.

Analysis of pathway crosstalk

To investigate interlinks and interactions of the enriched pathways, we further performed pathway crosstalk analysis. For the measurement of the overlap between any given pathway pair, we calculated the overlap coefficient (OC) () and the Jaccard coefficient (JC) () by the corresponding formulas, in which A and B are the gene lists contained in the two examined pathways. To build the pathway crosstalk, we executed the following procedure:

We only chose the pathways with an FDR value less than 0.05 and which contained more than five candidate genes that would give us sufficient biological information;

We counted the common candidate genes of each pathway pair which had more than two overlapped genes. Furthermore, the pathway crosstalk needed to match with KEGG;

We calculated the overlap in every pathway pair and ranked them according to their JC and OC values;

We constructed pathway crosstalk via Cytoscape software (25).

Establishment of the human protein-protein interaction network

To deduce the interaction and correlation between the SCD-related genes, we constructed a relatively comprehensive and dependable human interactome and inferred and analyzed the potential topological characteristics of the SCDgset molecular network. By pooling and curating the nonredundant physical interaction data from the STRING database (26), we selected the interactions with a score more than 0.6 and which were approved with reported experimental data or database annotation. Then, to cluster the interaction network and recover clusters of associated interactions, Markov cluster algorithm, which is an algorithm for clustering graph flow simulation, was used at an inflation value of 3.0 (27). Finally, we established a relatively full-scale human physical interactome, which contained 196 genes and 653 edges.

Construction of the SCD-specific network

A subnetwork specific to certain diseases could guide us in clarifying the interaction mechanisms of the disease-related molecules. A network parsimony principle in the context of biological processes has been previously proposed (28), and the Steiner minimal tree algorithm meets this principle (29). We extracted the SCD-specific potential pathological network from Pathway Commons (http://www.pathwaycommons.org) (30), which currently contains data from nine databases with over 1,400 pathways and 687,000 interactions, including biochemical reactions, complex assembly, transport and catalysis events, and physical interactions involving proteins, DNA, RNA, small molecules, and complexes. We used the minimum spanning tree-based approximation (KB) method to search for the Steiner tree from the SteinerNet package (31). Here, we mainly focused on the human database and built the directed networks based on analyzing genes that the regulate state change of the SCDgset, expression of the SCDgset, transport of the SCDgset, and phosphorylation of the SCDgset, which yielded 17,901 genes/proteins and 385,553 interactions.

More specifically, 255 genes associated with SCD were input into the Pathway Commons. Except for C1ORF185, which had no annotation, all the genes in the SCDgset turned out to be involved in the interaction networks (undirected networks). During the process of building the Steiner minimal tree, the reciprocal degree of each node was used as its weight to ease the effect of the genes with extremely high interactions in the network structure. The Erdos-Renyi model in R software “igraph” package was used to obtain the nonrandomness of the constructed network. We then generated 1,000 random networks with the same number of vertices and interactions as the SCD-specific network (32). Next, we calculated the arithmetic average values of the shortest-path distance and clustering coefficient. We evaluated the significance level of nonrandomness according to the number of random networks with an average shortest-path distance (ND) less than that of the SCD-specific networks, and the number of random networks with an average clustering coefficient (NC) more than the observed clustering coefficient. Lastly, the empirical P values using ND/1,000 and NC/1,000 were separately calculated.

Statistical analysis

Functional enrichment of KEGG pathways and GO terms were done based on Fisher’s exact test which is performed by David online system. FDR in DAVID requests adaptive linear step-up adjusted P values for approximate control of the FDR. FDR <0.05 was considered statistically significant.

Results

Identification of SCD-related genes

We reviewed publications on the genetic association studies related to SCD by searching the PubMed and GCBI databases. Only those reported genes that were significantly associated with the SCD were collected, and we discarded those reported as having a negative or insignificant association by the authors of the original literature. In total, from 1,094 publications, we selected 257 genes that were reported to be associated with SCD, and the detailed list of all genes (referred to as the SCDgset) is provided in online table (available at https://cdn.amegroups.cn/static/public/jtd-21-361-1.pdf). Besides, we searched three different databases (GTEx, BioGPS and SAGE) to check all these gene expression patterns. Unsurprisingly, all genes are expressed in normal human heart.

Among these genes, there were 15 genes related to potassium sodium-activated channel (HERG, KCNA5, KCND2, KCND3, KCNE1, KCNE2, KCNE3, KCNE5, KCNJ11, KCNJ2, KCNJ5, KCNJ8, KCNQ1, KCNT1, and KCTD1), six genes encoding sodium voltage-gated channel alpha subunit (SCN10A, SCN1B, SCN2B, SCN3B, SCN4B, and SCN5A), three genes encoding genes that were involved in calcium voltage-gated channel (CACNA1C, CACNA2D1, and CACNB2), four genes belonging to the solute carrier family (SLC12A9, SLC19A2, SLC25A26, and SLC4A4), two transforming growth factor beta genes (TGFB2 and TGFB3), two transforming growth factor beta receptor genes (TGFBR1 and TGFBR2), and three fibroblast growth factors (FGF12, FGF13, and FGF23). A further four tropomyosin family members were collected in the SCDgset (TPM1, TPM2, TPM3, and TPM4). Apparently, the significant SCD-related genes had a diversity of functions but were mostly reported to be involved in cardiac function.

Biological function enrichment analysis of the SCDgset

One more specific biological function array of these SCD-related genes was analyzed by the DAVID website, the results of which are provided in online table (available at https://cdn.amegroups.cn/static/public/jtd-21-361-2.pdf). Among the 383 GO terms overrepresented in the SCDgset, we unsurprisingly found 67 biological functions clearly related to heart diseases, including cardiac muscle contraction, heart development, muscle contraction, cardiac conduction, regulation of heart rate by cardiac conduction, regulation of heart rate, muscle filament sliding, action potential of myocardial cells participating in contraction, regulation of cardiac muscle contraction, histomorphogenesis of ventricular muscle, heart contraction, regulation of atrial cardiac muscle cell membrane depolarization, and regulation of cardiac conduction. Obviously, any of these biological dysfunctions could lead to cardiac diseases, including SCD. In this study, we also found results containing several biological functions that could lead to noncardiac causes of SCD, such as positive/negative regulation of RNA polymerase II promoter transcription and neurologic disorders (negative regulation of neuronal apoptosis, neuron fate commitment, regulation of calcium ion-dependent exocytosis of neurotransmitter, neuron projection morphogenesis, and neurotransmitter receptor metabolic process). Moreover, regulation of cell proliferation, cell adhesion, and apoptotic process were highly enriched in GO terms, while 13 and 22 GO terms related to cell membrane ion channels and transporters were enriched in the results, respectively. As one previous study reported, the cardiac action potential is made possible by active and passive processes that maintain highly regulated electrochemical gradients for sodium, potassium, and calcium ions through cell membrane ion channels and transporters (33). These results demonstrated that the collected candidate genes were relatively dependable for our further study.

Analysis of pathway enrichment in the SCSgset

In order to acquire useful information concerning the pathogenic molecular mechanism underlying SCD, it was necessary to identify the biochemical pathways enriched in the candidate genes. Pathway enrichment of the SCDgset was conducted through using KEGG and DAVID, and 47 significant enrichment pathways for SCD were identified (). Among these, HCM (ranked first in ), DCM (ranked second in ), and adrenergic signaling in cardiomyocytes (ranked third in ) were highly enriched in the SCDgset. Consistent with previous studies (34,35), HCM is a leading cause of SCD in young adults, and DCM is another major cause of SCD (36,37). Arrhythmogenic right ventricular cardiomyopathy (ARVC) is a genetic form of cardiomyopathy which is usually transmitted with an autosomal dominant pattern, and it accounts for 11–22% of cases of SCD in the young athlete population (38). In this study, we found that ARVC was enriched in the SCDgset at the fourth position. Furthermore, various signaling pathways that are involved in cell metabolism and cell behavior were highly enriched in our results, including calcium signaling pathway, oxytocin signaling pathway, cyclic GMP (cGMP)-protein kinase G (PKG) signaling pathway, cyclic AMP (cAMP) signaling pathway, HIF-1 signaling pathway (responding to hypoxic stress), mitogen-activated protein kinase (MAPK) signaling pathway, and others. Moreover, pathways in cancer, proteoglycans in cancer, pancreatic cancer, and AGE-RAGE signaling pathway in diabetic complications were highly enriched, which is consistent with previous studies which suggest these pathways have important functions in SCD (39,40).

Table 1

Pathways enriched in the SCDgseta

Pathway ID	Pathways	No. of genesb	Genes among SCD included in the specific pathway	P valuec	FDRd
hsa05410	Hypertrophic cardiomyopathy (HCM)	32	LMNA, DMD, ACTC1, MYH7, CACNA1C, TPM3, ACE, TNNT2, MYL3, ITGB3, CACNB2, RYR2, ITGA2, TNNI3, TNNC1, SGCD, TPM4, MYH6, TPM1, TGFB3, PRKAG2, MYBPC3, TPM2, ATP2A2, TTN, CACNA2D1, AGT, EMD, EDN1, TGFB2, MYL2, NCX1	4.83E−32	8.89E−30
a04261	Adrenergic signaling in cardiomyocytes	32	KCNE1, ACTC1, MYH7, CACNA1C, TPM3, TNNT2, CALM1, MYL3, CACNB2, SCN1B, SCN5A, ADRB1, KCNQ1, RYR2, TNNI3, ATP1B1, TNNC1, SCN4B, TPM4, MYH6, TPM1, AGTR1,	9.08E−24	5.57E−22
a05414	Dilated cardiomyopathy (DCM)	31	LMNA, DMD, ACTC1, MYH7, CACNA1C, TPM3, TNNT2, MYL3, ITGB3, CACNB2, ADRB1, RYR2, ITGA2, TNNI3, TNNC1, SGCD, TPM4, MYH6, TPM1, TGFB3, MYBPC3, PLN, TPM2, ATP2A2, TTN, CACNA2D1, AGT, EMD, TGFB2, MYL2, NCX1	1.84E−29	1.69E−27
hsa05200	Pathways in cancer	30	VEGFB, JUP, CREBBP, TGFBR2, VEGFA, TPM3, TGFBR1, CALM1, PPARG, RELA, CDKN1A, NOTCH1, ITGA2, GNB4, ERBB2, TGFB3, AGTR1, PLEKHG5, PRKCA, SMAD3, WNT11, GNG11, AGT, ROCK2, EDN1, CALM2, TGFB2, FGF23, KNG1, ESR1	1.35E−06	2.49E−05
hsa05412	Arrhythmogenic right ventricular cardiomyopathy (ARVC)	19	LMNA, DMD, JUP, CACNA1C, DSG2, DSC2, ACTN2, ITGB3, CACNB2, RYR2, ITGA2, SGCD, PKP2, GJA1, ATP2A2, CACNA2D1, EMD, DSP, NCX1	5.34E−16	2.46E−14
hsa04260	Cardiac muscle contraction	19	ACTC1, MYH7, CACNA1C, TPM3, TNNT2, MYL3, CACNB2, RYR2, TNNI3, ATP1B1, TNNC1, TPM4, MYH6, TPM1, TPM2, ATP2A2, CACNA2D1, MYL2, NCX1	2.72E−15	1.00E−13
hsa05205	Proteoglycans in cancer	17	VEGFA, CAV3, ITGB3, CDKN1A, ITGA2, ANK2, ERBB2, FLNC, PRKCA, FLNA, CAV1, WNT11, ROCK2, DCN, TGFB2, TLR4, ESR1	1.86E−06	3.12E−05
hsa04022	cGMP-PKG signaling pathway	16	MYH7, CACNA1C, MYLK2, CALM1, ADRB1, ATP1B1, AGTR1, PLN, ATP2A2, NPPA, ROCK2, ADRA2B, CALM2, KCNJ8, KNG1, NCX1	6.67E−07	1.53E−05
hsa04020	Calcium signaling pathway	16	TNI, CACNA1C, MYLK2, RYR1, CALM1, ADRB1, PTK2B, RYR2, ERBB2, TNNC1, AGTR1, PRKCA, PLN, ATP2A2, CALM2, NCX1	3.50E−06	4.96E−05
hsa04010	MAPK signaling pathway	16	VEGFB, TGFBR2, VEGFA, CACNA1C, TGFBR1, CACNB2, RELA, NGF, ERBB2, FLNC, TGFB3, PRKCA, FLNA, CACNA2D1, TGFB2, FGF23	0.00077	0.004863
hsa04933	AGE-RAGE signaling pathway in diabetic complications	15	VEGFB, TGFBR2, VEGFA, TGFBR1, RELA, THBD, SERPINE1, COL3A1, TGFB3, AGTR1, PRKCA, SMAD3, AGT, EDN1, TGFB2	3.85E−09	1.18E−07
hsa04371	Apelin signaling pathway	15	MYLK2, RYR1, TGFBR1, CALM1, MYL3, SERPINE1, RYR2, GNB4, PRKAG2, AGTR1, SMAD3, GNG11, CALM2, MYL2, NCX1	2.90E−07	7.63E−06
hsa04921	Oxytocin signaling pathway	15	CACNA1C, MYLK2, RYR1, CALM1, CACNB2, KCNJ2, CDKN1A, RYR2, PRKAG2, PRKCA, NPPA, CACNA2D1, ROCK2, CALM2, KCNJ5	1.22E−06	2.49E−05
hsa04510	Focal adhesion	15	VEGFB, VEGFA, CAV3, MYLK2, ITGB3, ITGA2, ERBB2, ACTN4, FLNC, PRKCA, FLNA, CAV1, ROCK2, ILK, MYL2	3.08E−05	0.000378
hsa04024	cAMP signaling pathway	15	CREBBP, CACNA1C, CALM1, HCN4, RELA, ADRB1, RYR2, TNNI3, ATP1B1, PLN, ATP2A2, NPPA, ROCK2, EDN1, CALM2	6.41E−05	0.000635
hsa04218	Cellular senescence	12	TGFBR2, IGFBP3, TGFBR1, CALM1, RELA, CDKN1A, SERPINE1, TGFB3, SMAD3, TRPM7, CALM2, TGFB2	0.00021	0.001642
hsa04925	Aldosterone synthesis and secretion	11	LDLR, CACNA1C, CYP11B2, CALM1, ATP1B1, AGTR1, PRKCA, NPPA, AGT, CALM2, KCNJ5	9.30E−06	0.000122
hsa04926	Relaxin signaling pathway	11	VEGFB, TGFBR2, VEGFA, TGFBR1, RELA, COL3A1, GNB4, PRKCA, SMAD3, GNG11, EDN1	0.00013	0.001096
hsa05226	Gastric cancer	11	JUP, TGFBR2, TGFBR1, CDKN1A, ERBB2, ABCB1, TGFB3, SMAD3, WNT11, TGFB2, FGF23	0.00043	0.003066
hsa05161	Hepatitis B	11	CREBBP, TGFBR2, TGFBR1, RELA, PTK2B, CDKN1A, TGFB3, PRKCA, SMAD3, TGFB2, TLR4	0.00092	0.005301
hsa05163	Human cytomegalovirus infection	11	VEGFA, CALM1, ITGB3, RELA, PTK2B, CDKN1A, GNB4, PRKCA, GNG11, ROCK2, CALM2	0.01089	0.0455
hsa04924	Renin secretion	10	CACNA1C, ACE, CALM1, KCNJ2, ADRB1, AGTR1, NPPA, AGT, EDN1, CALM2	2.32E−06	3.56E−05
hsa04713	Circadian entrainment	10	CACNA1C, RYR1, CALM1, NOS1AP, RYR2, GNB4, PRKCA, GNG11, CALM2, KCNJ5	5.04E−05	0.000546
hsa04066	HIF-1 signaling pathway	10	CREBBP, VEGFA, RELA, CDKN1A, SERPINE1, ERBB2, PRKCA, NPPA, EDN1, TLR4	6.56E−05	0.000635
hsa05142	Chagas disease (American trypanosomiasis)	10	TGFBR2, ACE, TGFBR1, RELA, SERPINE1, TGFB3, SMAD3, TGFB2, TLR4, KNG1	8.44E−05	0.000777
hsa04919	Thyroid hormone signaling pathway	10	CREBBP, ITGB3, NOTCH1, NCOA2, ATP1B1, MYH6, PRKCA, PLN, ATP2A2, ESR1	0.00023	0.001681
hsa04270	Vascular smooth muscle contraction	10	CACNA1C, MYLK2, CALM1, AGTR1, PRKCA, NPPA, AGT, ROCK2, EDN1, CALM2	0.00064	0.004395
hsa05152	Tuberculosis	10	CREBBP, CALM1, RELA, IL-18, TGFB3, CTSS, LAMP2, CALM2, TGFB2, TLR4	0.00614	0.027546
hsa05212	Pancreatic cancer	9	TGFBR2, VEGFA, TGFBR1, RELA, CDKN1A, ERBB2, TGFB3, SMAD3, TGFB2	3.61E−05	0.000415
hsa04068	FoxO signaling pathway	9	CREBBP, TGFBR2, TGFBR1, CDKN1A, TGFB3, PRKAG2, SMAD3, TGFB2, STK11	0.0025	0.012756
hsa05418	Fluid shear stress and atherosclerosis	9	VEGFA, CAV3, CALM1, ITGB3, RELA, THBD, CAV1, EDN1, CALM2	0.00354	0.017121
hsa05225	Hepatocellular carcinoma	9	TGFBR2, TGFBR1, CDKN1A, TGFB3, PRKCA, SMAD3, WNT11, BRG1, TGFB2	0.01187	0.0455
hsa04971	Gastric acid secretion	8	MYLK2, CALM1, KCNJ2, KCNQ1, ATP1B1, PRKCA, CALM2, KCNE2	0.00023	0.001681
hsa04974	Protein digestion and absorption	8	ACE2, KCNQ1, DPP4, COL3A1, ATP1B1, ELN, KCNE3, NCX1	0.00079	0.004863
hsa04350	TGF-beta signaling pathway	8	CREBBP, TGFBR2, TGFBR1, SMAD6, TGFB3, SMAD3, DCN, TGFB2	0.00092	0.005301
hsa05220	Chronic myeloid leukemia	7	TGFBR2, TGFBR1, RELA, CDKN1A, TGFB3, SMAD3, TGFB2	0.00137	0.007399
hsa05146	Amoebiasis	7	RELA, COL3A1, ACTN4, TGFB3, PRKCA, TGFB2, TLR4	0.00516	0.02375
hsa04725	Cholinergic synapse	7	CACNA1C, KCNJ2, KCNQ1, GNB4, PRKCA, GNG11, ACHE	0.01172	0.0455
hsa05144	Malaria	6	SELP, IL-18, TGFB3, HBA1, TGFB2, TLR4	0.00067	0.004419
hsa05416	Viral myocarditis	6	DMD, MYH7, SGCD, MYH6, CAV1, CXADR	0.00181	0.009501
hsa05321	Inflammatory bowel disease (IBD)	6	RELA, IL-18, TGFB3, SMAD3, TGFB2, TLR4	0.00297	0.014779
hsa04520	Adherens junction	6	CREBBP, TGFBR2, TGFBR1, ERBB2, ACTN4, SMAD3	0.00495	0.023373
hsa04610	Complement and coagulation cascades	6	CD46, TFPI, THBD, FACTORV, SERPINE1, KNG1	0.00777	0.034039
hsa05132	Salmonella infection	6	RELA, IL-18, FLNC, FLNA, ROCK2, TLR4	0.01159	0.0455
hsa05210	Colorectal cancer	6	TGFBR2, TGFBR1, CDKN1A, TGFB3, SMAD3, TGFB2	0.01159	0.0455
hsa04614	Renin-angiotensin system	5	ACE2, ACE, AGTR1, AGT, AGTR2	0.00012	0.001053
hsa05143	African trypanosomiasis	5	IL-18, PRKCA, HBA1, NPPA, KNG1	0.00121	0.00676

a, SCD-related gene set; b, number of genes among the SCDgset; only the pathways containing 5 or more genes among the SCDgset are shown; c, P values were calculated with a hypergeometric test; d, FDRs were adjusted by the Benjamini-Hochberg method. SCD, sudden cardiac death; SCDgset, sudden cardiac death-related gene set; FDR, false discovery rate.

Crosstalk among significantly enriched pathways

Besides identifying lists of significantly enriched pathways, we also conducted pathway analysis among the 47 enriched pathways to explore the correlations between the pathways and understand how they interact with each other. There were 53 pathways including 5 or more members in the SCDgset. If two pathways shared a proportion in the SCDgset, we assumed that there was a degree of crosstalk between them (41). Only if one pathway shared at least two genes with 1 or more other pathway(s), we considered it to meet the criterion for crosstalk analysis, and 34 pathways met this criterion. To construct the pathway crosstalk network, we used all the pathway pairs (edges). Based on the average scores of the JC and OC, we calculated the overlap significance of each pathway pair.

We roughly divided the pathways into two major modules according to their crosstalk. Each module has more interactions than other pathways, and all the genes from one module might participate in the same/similar biological process (). One module primarily is connected by calcium signaling pathway while the other is related to MAPK signaling pathway. GWAS studies have determined that calcium signaling pathway and MAPK signaling pathway are the common pathways between type 2 diabetes (T2D) and coronary artery disease (CAD) (42). The two modules were not independent; instead, the two main pathways were connected directly with each other and indirectly via a few other pathways.

Figure 1

Pathway crosstalk among the SCDgset-enriched pathways. Nodes represent pathways, and edges represent crosstalk between pathways. The edge width is directly proportionate to the crosstalk level of a given pathway pair. SCDgset, sudden cardiac death-related gene set.

In order to unravel the possible pathological protein network for the SCDgset, 1 subnetwork for SCD from the STRING database via the Markov cluster algorithm was built. The protein network of SCD comprised 196 nodes and 653 edges (interactions; ). In , the proteins labeled in red mainly play important roles in the regulation of heart contraction and ion transport. For example, most of the proteins, such as KCNE1, KCNE1L, KCNE2, KCNE3, and KCNJ5, are involved in the potassium sodium-activated channel, the proteins that belong to the sodium voltage-gated channel alpha subunit, SCN10A, SCN1B, SCN2B, SCN3B, SCN4B, and SCN5A, and the members of the calcium voltage-gated channel, CACNA1C, CACNA2D1, and CACNB2, are all circled in the red part. SCN5A-encoded cardiac sodium channel is the basis of cardiac excitability, and glycerol phosphate dehydrogenase 1-like protein (GPD1L) can be found on the two endpoints of 1 line, meaning they can interact with each other to induce certain functions. It has been reported that dysfunction of sodium current (INa) may lead to fatal ventricular arrhythmia in maladies, including long QT syndrome, sudden infant death syndrome (SIDS), and Brugada syndrome (BrS), and that mutations in GPD1L are also related to SIDS, BrS, and a decrease in INa. Valdivia et al. clarified the relationship between GPD1L and SCN5A: PKC-dependent phosphorylation of SCN5A leads to the connection of redox state and cardiac excitability through GPD1L (43). Here, we depicted the relationship between GPD1L and SCN5A via a protein network in a more intuitive way. The light green-labeled proteins mostly act in the modulation of muscle structure development (). Several sarcomeric proteins, such as alpha myosin heavy chain, play crucial roles in cardiac development. According to the reports, mutations in these genes result in congenital heart defects (CHDs), which occur in approximately 1 in 145 live births. Therefore, identifying these potentially pathogenic genes is vital. Tropomyosin 1 (TPM1), an essential sarcomeric component, has multiple roles in the developing heart and in the formation of CHDs (44). Consistent with the previous study, our results showed that TPM1, and other TPM family members, TPM2, TPM3 and TPM4, were all in the light green circle suggesting their roles in cardiac development. Interestingly, we found that junctophilin-2 (JPH2) directly interacts with the proteins (red circles). JPH2 contributes to the construction of skeletal muscle triad junctions and is integral to proper intracellular Ca2+ signaling in cardiac myocytes via its involvement in ryanodine receptor–mediated calcium ion release. Jiang et al. suggested that JPH2 directly or indirectly interacts with Cai-handling proteins, Cav1.2 and KCNQ1 (45). In our results, we demonstrated that the interaction between JPH2 and KCNQ1 is indirect, because ANK2 or/and SCN5A is/are the bridge(s) that connect these two proteins. The remaining proteins outside of the red or light green circles are involved in various other pathways.

Figure 2

The SCD-specific network was constructed via Markov cluster algorithm, with 653 edges and 196 nodes. Red nodes are genes mainly involved in regulation of heart contraction and regulation of ion transport, while light green nodes are genes related to muscle structure development. SCD, sudden cardiac death.

We also extracted the specific network for SCD from Pathway Commons via the Steiner minimal tree algorithm. The SCD protein network comprised 272 edges (interactions) and 273 nodes (). To examine the nonrandomness of the extracted network, we created 1,000 random networks by using the Erdos-Renyi model. The average clustering coefficient and shortest-path distance were compared with the corresponding values of the SCD-specific network. The mean shortest-path distance of these random subnetworks was 5.23, while that of the SCD-related network was 3.30. The random networks had an average clustering coefficient (NC) of 0.008, while the SCD distinctive network had an NC of 0. Therefore, we extracted a nonrandom SCD-specific network.

Figure 3

The SCD-specific network was constructed using a node-weighted Steiner minimal tree algorithm, with 272 edges and 273 nodes. The green nodes are genes of the SCDgset, while the red nodes are nonoriginal/extended genes. SCD, sudden cardiac death; SCDgset, sudden cardiac death-related gene set.

Except for C1ORF185, which had no annotation, all genes in the SCDgset were inside of the SCD-specific network. They accounted for 93.0% of the 273 genes in the network, suggesting a very high coverage rate of the SCDgset in the subnetwork. In the SCD-specific molecular network, we found 23 genes that were enrolled outside of the SCDgset (). Interestingly, there were 3 big clusters in our directed SCD-specific network (). Angiotensinogen (AGT), a gene in the SCDgset, was the jumping-off point of all 3 of these clusters and interacted with noggin (NOG) to extend 1 cluster and with insulin (INS) to spread the third cluster. Surprisingly, NOG and INS were not included in the SCDgset. It is known that AGT plays crucial roles in heart diseases, such as hypertrophy of cardiac myocytes and coronary heart disease (46-48). The bone morphogenetic protein antagonist, NOG, has been shown to be a regulator in mammalian cardiac development in several previous studies (49,50), while INS has been shown to be related to cardiovascular events (51,52). However, no research has demonstrated that AGT can directly interact with NOG or INS. Considering the close interaction between other intermediate genes and known SCD-related genes, they may also participate in the pathological process of disease phenotypes. Interestingly, a number of the genes that occupy the jumping-off points of small clusters, such as presenilin 1 (PSEN1), ATP-binding cassette subfamily B member 1 (ABCB1), and vascular endothelial growth factor A (VEGFA), have been shown to be related to SCD (11,53,54). Although no reports have suggested that any of these 23 genes directly participate in the pathophysiological process of SCD, genes interacting with them or their family have been indicated to have functions in such processes. These include paired box 4 (PAX4), a node in the SCD-specific network that is not part of the SCDgset, was reported as an additional promising candidate of circadian transcriptional regulator, and Asian-specific missense variant in PAX4 associated with T2D in Chinese individuals. However, the genes linking to it were all included in the SCDgset and have been shown relate to heart diseases. Therefore, among these 23 genes, there are several that may figure prominently in SCD susceptibility and may be considered as novel targets for further study.

Table 2

Genes inside of the SCD-specific network but outside of the SCDgseta

Gene ID	Gene symbol	Gene name	Alias	Edges	mRNA expression in normal human heart fromb
					GTExc	BioGPSd	SAGEe
9241	NOG	Noggin	SYM1/SYNS1/SYNS1A	71	√
3630	INS	Insulin	IDDM/IDDM1/IDDM2/ILPR/IRDN/MODY10	54	√	√
5465	PPARA	Peroxisome proliferator activated receptor alpha	NR1C1/PPAR/PPARalpha/hPPAR	17	√	√	√
5078	PAX4	Paired box 4	KPD/MODY9	15	√	√
6667	SP1	Sp1 transcription factor	–	7	√	√	√
3725	JUN	Jun proto-oncogene, AP-1 transcription factor subunit	AP-1/AP1/c-Jun	6	√	√	√
5566	PRKACA	Protein kinase camp-activated catalytic subunit alpha	PKACA/PPNAD4	4	√	√
4783	NFIL3	Nuclear factor, interleukin 3 regulated	E4BP4/IL3BP1/NF-IL3A/NFIL3A	3	√	√	√
6777	STAT5B	Signal transducer and activator of transcription 5B	STAT5	2	√	√
6513	SLC2A1	Solute carrier family 2 member 1	CSE/DYT17/DYT18/DYT9/EIG12/GLUT/GLUT-1/GLUT1DS/HTLVR/PED/SDCHCN	2	√	√	√
4303	FOXO4	Forkhead box O4	AFX/AFX1/MLLT7	2	√	√
4654	MYOD1	Myogenic differentiation 1	MYF3/MYOD/PUM/bHLHc1	2	√	√	√
29803	REPIN1	Replication initiator 1	AP4/RIP60/ZNF464/Zfp464	2	√	√	√
4609	MYC	MYC proto-oncogene, bhlh transcription factor	MRTL/MYCC/bHLHe39/c-Myc	2	√	√	√
5444	PON1	Paraoxonase 1	ESA/MVCD5/PON	2	√	√
1385	CREB1	Camp responsive element binding protein 1	CREB/CREB-1	2	√	√	√
324	APC	APC, WNT signaling pathway regulator	BTPS2/DP2/DP2.5/DP3/GS/PPP1R46	2	√	√
6714	SRC	SRC proto-oncogene, non-receptor tyrosine kinase	ASV/SRC1/THC6/c-SRC/p60-Src	2	√	√
7003	TEAD1	TEA domain transcription factor 1	AA/NTEF-1/REF 1 /TCF-13/TCF13/TEAD-1/TEF-1	2	√	√	√
2353	FOS	Fos proto-oncogene, AP-1 transcription factor subunit	AP-1/C-FOS/p55	2	√	√	√
387	RHOA	Ras homolog family member A	ARH12/ARHA/RHO12/RHOH12	2	√	√	√
4217	MAP3K5	Mitogen-activated protein kinase kinase kinase 5	ASK1/MAPKKK5/MEKK5	2	√	√	√
5270	SERPINE2	Serpin family E member 2	GDN/GDNPF/PI-7/PI7/PN-1/PN1/PNI	2	√	√	√

a, SCD-related genes gene set; b, mRNA expression of each gene in normal human heart was searched from three different database; c, GTEx (Genotype-Tissue Expression) database: https://gtexportal.org/home/topExpressedGenePage; d, BioGPS: http://biogps.org/#goto=welcome; e, SAGE (Serial Analysis of Gene Expression): https://www.genecards.org/. SCD, sudden cardiac death; SCDgset, sudden cardiac death-related gene set.

Discussion

In the past decades, substantial progress has been made to clarify the molecular mechanisms underlying SCD. With the development of high-throughput technology, the elements related to this disease have been identified on a much larger scale. GWAS, in particular, represents a significant step forward in investigating biologic pathways of disease causation and towards identifying a huge number of genes that are involved in SCD (10,37). Even though the numbers of reports regarding the genes and proteins potentially involved in this disease are growing, a thorough understanding of the molecular biological process related to SCD pathogenesis is still far from complete. Therefore, a systematic analysis of SCD-related genes is needed to decode the potential pathogenesis of SCD. In the current study, we collected genetically SCD-associated genes, and explored the interaction of these genes via a pathway-based and network-based analytical framework, with SCD-related biochemical processes and their interactions also being analyzed.

Although some knowledge concerning the factors involving SCD have been obtained from the candidate gene(s)-based genetic association and biochemical studies, our comprehensive analysis on SCD-related has significant advantages. First, we implemented an extensive compilation and curation of genetically SCD-associated human genes, which provided us a valuable gene data source for further study. Moreover, since many diseases are not caused by a single gene and instead are the result of multiple genes working together, each gene can be said to only play a small to medium role in the emergence of disease (4). Considering this, we retrieved genes jointly showing significant genetic association with SCD to select a high coverage of related genes. Furthermore, by paying attention to the biological correlation of genes, pathway and network analysis can resist the effect of false-positive gene results more effectively and may offer a more comprehensive glimpse into the pathological mechanisms of SCD.

The results of biological function enrichment analysis demonstrated that genes in the SCDgset participated in various biological processes and that the dysfunction of these genes might lead to heart diseases. For instance, terms that were directly related to cardiac functions, such as cardiac muscle contraction, heart development, regulation of heart rate, regulation of heart rate by cardiac conduction, and cardiac conduction, were highly enriched in SCD-related genes, implying these activities in the pathologic process of SCD are significant; furthermore, all of these terms were concurrent with the priori biological findings for SCD (4,18,39). Of further note, the GO biological processes that might lead to noncardiac causes of SCD, such as response to hypoxia, neurologic disorders, and terms related to cell membrane ion channels and transporters, were also enriched in the results. It was reported that central nervous system disorders can directly (or indirectly via cardiac interaction) result in SCD (41,55). Surprisingly, the gene numbers related to the terms like transcriptional regulation of RNA polymerase II promoter were in the top rank, while 31 genes functioned as positive regulators and 28 genes functioned as negative regulators. However, few previous reports have examined the relationship between transcriptional regulation of RNA polymerase II promoter and SCD, therefore, the current findings may give us new insights into the mechanisms and causes of SCD.

Pathway analysis showed that cardiac -related pathways were enriched in the SCDgset, and included HCM, DCM, adrenergic signaling in cardiomyocytes, ARVC, and cardiac muscle contraction (), which further highlights the dysfunctions of cardiac pathway as the key precipitators of SCD (35,37,38). Meanwhile, cGMP-PKG pathway was found to be enriched in the SCDgset at a high rank, which is consistent with its indispensable role in the pathogenesis and pathological development of SCD. The cGMP-PKG pathway plays a major role in the regulation of cardiac function, and is involved in heart failure, cardiac hypertrophy, and ischemic cardiomyopathy (56,57). The next highest-ranking pathway was calcium signaling pathway; interestingly, previous studies have demonstrated that crosstalk occurs between Ca2+ and cGMP signaling during cardiac hypertrophy (58). Calcium directly mediates cellular depolarization, diastolic current flow, early ischemic arrhythmias, and calcium homeostasis dysfunction to induce various cardiac diseases (59,60). MAPKs are prominent players that have been the focus of extensive investigations in the past decades. As MAPKs are essential to cardiac development, cardiac hypertrophy, ischemia /reperfusion injury, and pathological remodeling, they have been considered to be viable targets in therapeutic development (61). Our study showed that the MAPK pathway was one of the enriched pathways of SCD-related genes. However, within the pathway, the 4 best characterized MAPK subfamilies, ERK1/2, JNK, p38, and ERK5 were not included (), which might give clues that these genes participated SCD via noncanonical MAPK pathways. Consequently, these results indicated that the molecular mechanisms underlying SCD are fairly complex and require further study.

Much more interestingly, we identified two main modules via by pathway crosstalk analysis. One module was mostly composed of pathways related to calcium modulation and the other module was related to MAPK pathway. As previous studies have shown, calcium signaling and MAPK signaling are the furthest upstream nodes in heart development, and they are involved in induction of cardiac mesoderm, outflow tract formation, trabeculation, cardiac cushion formation, and valve formation. Furthermore, we observed that calcium signaling pathway and MAPK pathway were interconnected and shared most of the hormone-related pathways that were enriched in the SCDgset (i.e., apelin signaling pathway, oxytocin signaling pathway, and FoxO signaling pathway), suggesting that the two pathways share the same molecular mechanisms that regulate SCD. However, as it pertains to the MAPK module, it showed a greater degree and strength of crosstalk. The module emphasizes the notion that there might be a connection between virus infection pathway (e.g., tuberculosis, human cytomegalovirus infection, hepatitis B) and SCD via MAPK signaling pathway. Due to the limited number of relevant studies attesting to the close relationship between virus infection and SCD, our study may offer only preliminary evidence for the regulatory aspects of SCD, and this requires further examination. Hence, by combining the analytical results from biochemical pathway and pathway crosstalk with the a priori biological knowledge base, the major pathways associated with SCD can be identified.

Further, we inferred the SCD-specific protein network on the basis of the human reference interactome network. It is worth noting that 23 novel genes from the SCDgset that also appear in the human interaction group were ones credibly reported to be related to SCD. Interestingly, in the SCD-specific protein network, NOG has 71 edges that interact with the greatest number of proteins, and INS has 54 interactions, which is far more than that of other protein-protein interaction networks. By interacting with AGT and NOG, INS became the jumping-off point of two large clusters. Moreover, PPARA is a famous gene in studying the mechanisms of overweight or obese patients, and is potentially one of the causes of SCD (62). Even though we did not collect PPARA in the SCDgest, pathway crosstalk analysis identified it as a potential target gene in SCD. As for PAX4, it showed no definitive linkage to SCD. However, our results found it to be one of the important elements in the SCDgset because it interacted with 15 SCD-related proteins. SP1 and JUN mainly acted by means of MAPK pathway to contribute to the emergence of SCD (63,64).

We further extracted the specific protein network for SCD through two methods (Markov cluster algorithm and Steiner minimal tree algorithm) with different databases (STRING database and Pathway Commons), and the subnetworks also showed different patterns. In the first approach, the genes had different roles in SCD that could easily be identified; the genes labeled in red dots are known as regulators in heart contraction and ion transport, while the light green genes are modulators in muscle structure development (). AGT was identified as the most central point in , and it links to all other small clusters characterized by NOG, INS, PPARA, PSEN1, PAX4, SP1, ABCB1, and VEGFA. Surprisingly, even though AGT, PSEN1, ABCB1, and VEGFA were included in the SCDgset and were the jumping-off points to their corresponding clusters, they were just as normal as other factors in the subnetwork in . However, other jumping-off points, such as NOG, INS, PAX4, PPARA, JUN, SP1 and PPKACA were not included in the SCDgset but do play important roles in the SCD-specific protein network, and reports do indicate that they are directly linked with SCD (46-48,62-64). Here, we strongly suggest that further effort should be put into their functional study. According to the results, our predictive approach could not only produce an important predictive subnetwork of the SCDgset for SCD but could potentially help detect related genes.

In current study, we compiled SCD-associated genes from selective literature deposited in the PubMed and GCBI databases and adopted a systems biology framework to conduct a comprehensive and systematic biological function- and network-based analysis of SCD. We focused only on those genes showing a positive association with SCD and that were proven with sufficiently solid evidence by original authors. Via integrating the analysis results from GO, pathway, and pathway crosstalk, we demonstrated that the basic cardiac-related pathways, including HCM, DCM, adrenergic signaling in cardiomyocytes, ARVC, and cardiac muscle contraction were enriched in the SCDgset. Furthermore, calcium signaling pathway and MAPK pathway were enriched in the SCDgset and were interrelated. Moreover, the SCD-specific pathological molecular network indicated some potential genes associated with SCD (NOG, INS, PAX4, PPARA, JUN, SP1, and PPKACA) which should be noted for further study. To confirm the association of these novel genes with any inherited cardiac disease, we will check the expression patterns in both mRNA and protein levels, build the overexpressing or/and knock-out mutants in cell and animal levels to perform functional study, and the molecular mechanisms will also be studied.

Therefore, our systematic and comprehensive exploration of the SCD-associated genes may not only offer broad insights into understanding the contribution of genetic factors and other related environmental factors in the pathogenesis of SCD but can also aid in exploring the molecular mechanisms underlying SCD. The SCDgset should provide an informative source and be considered a useful dataset for SCD inquiry. Finally, several potential biomarkers were identified that may be valuable for the prevention of SCD.

The article’s supplementary files as