A Systems Biology Framework Identifies Molecular Underpinnings of Coronary Heart Disease

Abstract

Objective

Genetic approaches have identified numerous loci associated with coronary heart disease (CHD). The molecular mechanisms underlying CHD gene-disease associations, however, remain unclear. We hypothesized that genetic variants with both strong and subtle effects drive gene subnetworks that in turn affect CHD.

Approach and Results

We surveyed CHD-associated molecular interactions by constructing coexpression networks using whole blood gene expression profiles from 188 CHD cases and 188 age- and sex-matched controls. 24 coexpression modules were identified including one case-specific and one control-specific differential module (DM). The DMs were enriched for genes involved in B-cell activation, immune response, and ion transport. By integrating the DMs with altered gene expression associated SNPs (eSNPs) and with results of GWAS of CHD and its risk factors, the control-specific DM was implicated as CHD-causal based on its significant enrichment for both CHD and lipid eSNPs. This causal DM was further integrated with tissue-specific Bayesian networks and protein-protein interaction networks to identify regulatory key driver (KD) genes. Multi-tissue KDs (SPIB and TNFRSF13C) and tissue-specific KDs (e.g. EBF1) were identified.

Conclusions

Our network-driven integrative analysis not only identified CHD-related genes, but also defined network structure that sheds light on the molecular interactions of genes associated with CHD risk.

Introduction

Atherosclerotic coronary heart disease (CHD) is a multi-factorial disease with a prominent inflammatory vascular component that remains the leading cause of death in most developed countries and will soon become the leading cause of death in developing countries ¹. Atherosclerosis involves an interaction between modified lipoproteins, monocyte-derived macrophages, T cells, and the normal cellular elements of the arterial wall, resulting in the formation of atherosclerotic plaques. Plaque rupture and thrombosis can result in myocardial infarction (MI) and stroke ². Despite rapid advances in the genetic and genomic analysis of CHD in recent years, the molecular elements and the mechanisms involved in CHD pathogenesis remain unclear. Understanding the molecular basis of this disease can help to identify biomarkers for more accurate clinical diagnosis and point to candidate drug targets for more effective therapy.

At the genetic level, recent genome-wide association studies (GWAS) have identified 27 loci associated with CHD and myocardial infarction (MI), with many representing novel risk loci ^{3, 4}. Identifying the casual genes and mechanisms underlying GWAS signals, however, remains a challenging task. In addition, even large scale GWAS involving tens of thousands of cases and controls are unlikely to explain more than 15–20% of the heritability of CHD and MI ⁵. Therefore, further experiments and integrative analyses are needed to elucidate functional mechanisms underlying CHD risk loci and to identify the missing heritability.

At the gene expression level, peripheral blood has been investigated widely due to its ease of collection and its pathological relevance to CHD, as multiple blood constituents appear to play a role in plaque accumulation, rupture, and thrombus formation ⁶. Gene expression changes in whole blood of CHD patients may reflect the interaction of genetic predisposition, disease activity, immunity, thrombosis, metabolism, and environmental modifiers underling disease mechanisms. In recent studies ^7-9, mRNA profiling of tens to more than one hundred CHD cases has yielded lists of differentially expressed genes in an effort to identify effective markers of disease severity. Most of these studies, however, identified relatively few significantly differentially expressed genes between cases and controls at a stringent statistical cutoff and there was little overlap in the gene signatures of different CHD expression profiling studies. For instance, there is no common CHD gene signature among 160 genes reported by Sinnaeve et al ⁷, 50 genes reported by Wingrove et al ⁸, and 23 genes identified by Rosenberg et al ⁹. This failure to replicate may be due to the small sample sizes, differences in phenotypes or experimental platforms, and the low levels of differential expression of single genes.

In order to address the limitations of previous studies, we designed a systems biology framework to explore the molecular underpinnings of CHD via integration of whole blood gene expression profiles of RNA collected from Framingham Heart Study (FHS) participants with network approaches, GWAS, and genetics of gene expression (Figure 1). The FHS case-control study includes 188 pairs of prevalent CHD cases and age- and sex-matched controls. In our companion paper, we identified 35 differentially expressed genes between CHD cases and controls at FDR<50% ¹⁰. To extend upon result of analysis of differential gene expression, in this investigation, we first constructed coexpression networks and compared the robust coexpression patterns of cases versus controls to identify differential coexpression modules (DMs) that demonstrate gene co-regulation in cases but not in controls or vice versa. The CHD DMs were then integrated with SNPs associated with altered gene expression (eSNPs or eQTLs) from CHD-relevant tissues as well as CHD and CHD risk factor GWAS association results to look for enrichment of CHD risk eSNPs among the DM genes. This integration allowed us to differentiate causal (i.e. upstream) from reactive (i.e. downstream) DMs. Last, using a key driver analysis (KDA) that leverages DMs and graphical networks – including external tissue-specific Bayesian networks (BNs) and protein-protein interaction (PPI) networks – we identified putative regulators of the DMs and derived a CHD subnetwork.

Figure 1. Systems Biology Analysis Flow Chart Depicting the Process of Identifying CHD Causal Modules and Key Drivers (KDs).

The overall analysis includes 3 steps. First, coexpression networks are constructed from CHD cases and controls separately. CHD differential modules (DMs) are then identified by comparing the network structures between cases and controls. Step 2, the DMs are then integrated with GWAS of CHD and related traits using a SNP Set Enrichment Analysis (SSEA) to identify causal DMs. Third, key regulatory genes or KDs are identified for the causal DMs based on directional networks derived from independent studies using a Key Driver Analysis (KDA) algorithm.

Materials and Methods

Materials and Methods are available in the online-only Supplement.

Results

Clinical characteristics of the study sample

The case control study included 188 CHD cases and 188 matched controls. Demographic data and the clinical and laboratory characteristics of the study participants are summarized in Table 1. CHD cases had a higher prevalence of hypertension than controls (p=0.019).

Table 1.

Clinical Characteristics of the Case-control Study Groups

	CHD Cases N=188	CHD Controls N=188
Age (yr )	71±7.9	71±8.0
Sex, male (female)	140(48)	140(48)
Smoking, n (%)	15 (8)	2 (1)
Alcohol, n (%)	31(16)	34(18)
Glucose (mg/dl)	115±33	111±26
Diabetes, n (%)	51(27)	37(20)
Hypertension, n (%)	158 (84)	130(69)
BMI (kg/m2)	28.9±4.7	28.7±4.8
Systolic BP (mm Hg)	127 ±18	129±17
Diastolic BP (mm Hg)	68±10	72±10
Total cholesterol (mg/dl)	151±33	174±31
Triglycerides (mg/dl)	126±85	116±70
HDL cholesterol (mg/dl)	48±14	53±15
Hypertension treatment, n (%)	154(82)	109(58)
Diabetes treatment, n (%)	41(22)	35(19)
Lipid treatment, n (%)	175(93)	125(66)

Values are mean± SD, sample number or percentage.

BMI= body mass index

BP=blood pressure

CHD=coronary heart disease

Construction of coexpression networks for CHD and identification of DMs

Using differential expression analysis, we identified 35 genes that were differentially expressed between CHD cases and controls at a false discovery rate <50%; these results are summarized in our companion report ¹⁰. We hypothesized that alterations in the gene regulatory network architecture may better reflect the underlying molecular differences between groups. Therefore, we compared coexpression networks between CHD cases and controls to identify differential coexpression modules (DMs).

Using the previously developed Weighted Gene Coexpression Network Analysis (WGCNA) method ¹¹, we constructed gene coexpression networks for CHD cases and their corresponding controls, separately. A total of 12 coexpression network modules were identified for CHD cases and 12 for controls (Figure SIII). We tested whether modules were conserved between case and control networks by examining the overlap in gene memberships using Fisher's exact test. Modules that overlapped at p <2.0×10^-3 (corresponding to Bonferroni-corrected p<0.05) were defined as conserved between networks. If a module in one network did not have a corresponding conserved module in the other network, it was considered a DM. Two DMs – the CHD_tan module and the control_tan module containing 38 and 79 genes, respectively – differed between CHD cases and controls (Figure SIV and Table SIII). The genes in these two CHD DMs did not show any overlap with the 35 differential expression signatures ¹⁰, suggesting that the altered gene coexpression pattern and differential expression of individual genes captured different biological signals.

Functional annotation of the DMs

In order to understand the biological pathways and functional categories of the DMs, we conducted functional enrichment analysis. As shown in Table 2, the top enriched GO biology process term in the control_tan module is B cell activation, with 9 out of 51 module genes (existing in the GO database) overlapping with 103 genes in this pathway (19.8 fold enrichment, Bonferroni corrected enrichment p=5.6×10^-10). These B cell-related genes demonstrated strong coexpression in controls, but not in CHD cases, indicating that B cell dysregulation may be a major factor in atherosclerotic CHD. In contrast to the control_tan module, the CHD_tan module was identified in CHD cases but not in controls. The CHD_tan module is enriched for genes involved in ion transport (GO category; Bonferroni-corrected enrichment p=0.025). These results suggest a re-organization and shift of immune response and ion transporter pathways, with the former disrupted and the later enhanced in CHD.

Table 2.

Gene Ontology Enrichment of CHD Differential Modules.

Ontology category	Overlap	Fold enrichment	P value	Corrected P*
--Control_tan module
B cell activation	9	19.8	5.65×10-10	5.65×10-7
B cell differentiation	7	23.6	1.53×10-8	1.53×10-5
B cell receptor signaling pathway	6	22.6	2.30×10-7	2.30×10-4
immune response	12	6.3	2.86×10-6	2.86×10-3
lipid transport	5	9.6	1.66×10-4	0.166
hemopoiesis	9	4.2	2.23×10-4	0.229
--CHD_tan module
ion transport	10	4.7	2.53×10-5	0.025
metal ion transport	7	6.0	1.21×10-4	0.121
cell motility	7	3.0	0.007	1

^*Bonferroni correction for testing of multiple pathways and functional categories in the GO biology process and KEGG pathway databases.

Differential modules are enriched for CHD risk eSNPs

By virtue of the case-control design of this study, blood samples from the CHD cases used for gene expression profiling were drawn after the CHD events, thus some of the DMs we identified are potentially downstream of CHD rather than being causal. In order to identify causal DMs, we integrated the CHD DMs with CHD-related GWAS databases to look for enrichment of risk SNPs among DMs, with the rationale that genes harboring disease-risk SNPs are putatively causal for the disease of interest ^{12, 13}, because genotype does not change in response to disease occurrence.

We used SNP set enrichment analysis (SSEA; see description in the Methods section) to test if the DMs for CHD are likely to play causal roles in disease development by evaluating the overall association p value distribution of putative functional SNPs among DM genes in comparison to the null distribution. All 5 CHD gene sets from different resources (details in Materials and Methods) were highly enriched with low p value CHD association eSNPs in the CARDIoGRAM CHD GWAS ³ (Table SIV).These results confirm the sensitivity of SSEA to identify gene sets suspected of being causal of CHD.

The control-specific DM, control_tan, was not only enriched for risk eSNPs for CHD in the CARDIoGRAM CHD GWAS (1.6-fold enrichment; p=0.04 and 1.5×10^-5 from Kolmogorov-Smirnov (KS) and Fisher's tests, respectively), but also is enriched for low p value associations for lipid traits including HDL cholesterol (KS p=4.8 ×10^-12 and Fisher's p=7.8×10^-9), LDL cholesterol (KS p=9.0×10^-19 and Fisher's p=1.9×10^-26), and total cholesterol (KS p= 4.2×10^-21and Fisher's p=7.5×10^-28) in the GLGC lipid GWAS ¹⁴, suggesting that this DM may play a role in CHD pathogenesis via effects on atherogenic lipids (Table SIV SSEA results for CHD related gene sets, and Table SV CHD risk eSNPs for the control_tan module). The CHD_tan DM, on the other hand, did not show enrichment for CHD risk SNPs (enrichment p=0.49 and 0.85 from Fisher's and KS tests, respectively), and thus may be reactive rather than causal. The differential gene expression signatures identified at FDR<50% also failed to show any enrichment for CHD risk SNPs.

Identification of key drivers of the CHD causal DM and their associated subnetworks via network models derived from orthogonal studies

We speculated that the CHD causal DM we identified (control_tan) was driven by key regulatory genes with broad impact on this network module when perturbed. Such key regulatory genes or key drivers (KDs) can serve as candidate genes for therapeutic intervention. To identify KDs, we projected the genes within the CHD causal DM onto the pre-compiled directional networks including tissue-specific Bayesian networks (BNs) (Table SII) ^{15, 16} and non-directional PPI networks ¹⁷ from previous studies. At a Bonferroni-corrected p<0.05, we identified 59 unique KDs for the CHD causal module in the 6 tissue-specific BNs (59 KDs for blood, 0 for liver, 1 for fat, 1 for kidney, 0 for heart, and 0 for muscle) and one PPI network (0 KD). The human blood BN gave rise to the largest number of KDs (n=59) in contrast to the limited number of KDs in other tissue-specific networks. This is not surprising given that the DMs were identified from human blood expression profiles, confirming the tightly co-regulated relationships among the DM genes in blood rather than in other tissues. Furthermore, as the DMs were from the Framingham Heart Study and the blood BN used for KD identification was constructed from an independent Icelandic cohort ¹⁵, the consistency between the two networks cross-validated our results. Among the 59 KDs, 27 KDs contained eSNPs showing association with CHD or its risk factors at p <0.05; 7 contained eSNPs at p <1.0×10^-3, and 3 contained eSNPS at p <1.0×10^-4 in GWAS (Table SVI). Two of the KDs (CD79B and SPIB) overlapped with the 23 gene signature of CHD reported by Rosenberg et al ⁹ (enrichment p=5.5×10^-5).

We further prioritized the 59 KDs based on 4 elements: 1) consistency of the KD across multiple directional networks, i.e., a higher priority is given to KDs driving the CHD causal DM in more than one tissue-specific/PPI network; 2) whether the KD is a DM gene itself; 3) if the KD contains eSNPs showing evidence of association with CHD and/or its risk factors at p <1.0×10^-3 (the liberal GWAS p value cutoff was used here because the GWAS information was not considered alone but integrated with several levels of additional data to prioritize genes and reduce false discoveries); and 4) the p value of the KD based on the KDA algorithm (Table VI). Among the top 20 KDs (Table 3), TNFRSF13C and SPIB appeared in 2 networks – TNFRSF13C in adipose and blood and SPIB in kidney and blood – and were considered as multi-tissue KDs; the remaining were blood-specific KDs. In fact, Sage et al. reported that in Ldlr knockout mice, deletion of the BAFF receptor (BAFFR, encoded by the multi-tissue KD TNFRSF13C) reduced aortic root atherosclerosis significantly ^{18, 19}. In addition, Fretz et al. observed that knockout of another top KD, EBF1, in mice induced defects in lipid metabolism, adipose tissue deposition, and impaired glucose mobilization ²⁰. Therefore, some of the top KDs we identified are supported by existing experimental evidence for their relevance to metabolism, atherosclerosis, or other CHD-related phenotypes.

Table 3.

Top 20 Key Drivers Derived from the CHD Causal Module from Tissue-specific Bayesian and Protein-protein Interaction Networks.

Rank	KD	Tissue	Is DM gene	Adjusted KD p value by subnet size	GWAS source where KD eSNPs show association at p <1×10-3	Literature evidence for CHD association
1	SPIB	Blood / Kidney	0	Blood p=1.76×10-25 / Kidney p= 0.045		One of the CHD biomarkers of the 23 gene signatures9
2	TNFRSF13C	Blood / Adipose	0	Blood p=1.94 ×10-24/ Adipose p=0.041		Deletion of BAFF Receptor Reduced Atherosclerosis in ldlr Knockout Mice18, 19
3	PPAPDC1B	Blood	1	2.1 ×10-30	ICBP_SBP
4	EBF1	Blood	1	3.2 ×10-18	ICBP_DBP	EBF1-/- mice resulted in lipid abnormalities20
5	RALGPS2	Blood	1	1.8 ×10-16	GLGC_HDL
6	CD72	Blood	1	4.6 ×10-25
7	RASGRP3	Blood	1	1.7 ×10-22
8	CD200	Blood	1	3.2 ×10-22
9	TNFRSF17	Blood	1	5.8 ×10-22
10	BACH2	Blood	1	1.0 ×10-04
11	COBLL1	Blood	0	1.2 ×10-28	GLGC_HDL, TC, TG
12	CD79B	Blood	0	1.6 ×10-28	GIANT_BMI	One of the CHD biomarkers of the 23 gene signatures9
13	SAV1	Blood	0	8.8×10-22	GLGC_TC
14	FAM3C	Blood	0	2.2×10-16	CARDIoGRAM
15	LARGE	Blood	0	9.3 ×10-32
16	PKIG	Blood	0	2.6 ×10-31
17	TSPAN13	Blood	0	2.9 ×10-30
18	CELSR1	Blood	0	4.3 ×10-30
19	CTGF	Blood	0	3.0 ×10-29		Increased expression in atrial fibrillation patients and animals 39
20	BANK1	Blood	0	4.1 ×10-29

The key drivers (KDs) were ranked based on 4 elements: 1) consistency of the KD across multiple directional networks; 2) whether the KD is a DM gene itself; 3) if the KD contains eSNPs showing evidence of association with CHD and/or its risk factors at p <1.0×10^-3; and 4) the p value of the KD based on the Key driver analysis algorithm.

As shown in Figure 2, these top KDs constitute a tightly linked subnetwork with certain elements being tissue-specific and others showing cross-tissue connections. The KDs are at the center of the subnetwork and hence represent master regulators of this subnetwork. Not surprisingly, this subnetwork, which includes 152 genes depicted in Figure 2, is significantly enriched for immune response (GO functional category) and B cell receptor signaling pathways (Bonferroni-corrected p=1.9×10^-6 and p=5.1×10^-6, respectively). In addition, the subnetwork is highly enriched for genes whose eSNPs show low p value associations for both CHD in the CARDIoGRAM GWAS (enrichment p= 9.2×10^-3 and 3.8×10^-3 from Fisher's and KS tests, respectively) and CHD risk traits including SBP and DBP in the ICBP GWAS and lipid traits (HDL, LDL, and total cholesterol) in the GLGC GWAS (Table SIV). Furthermore, 15 genes were candidate cardiovascular genes curated in the Genetic Association Database (http://geneticassociationdb.nih.gov/). All these lines of evidence support the relevance of the KDs and their related subnetwork to atherosclerotic CHD.

Figure 2. Top KDs of the CHD causal module and the associated subnetwork.

The subnetwork was derived from the top 20 KDs using the tissue-specific networks from which the KDs were identified. The nodes of the largest size in the network are the top 20 KDs, and the middle-sized nodes represent other KDs and genes in the CHD causal model. Red rectangular, red circular, and green circular nodes are multi-tissue KDs, blood-specific KDs, and causal DM genes, respectively. The network graph were drawn by ProteoLens³⁸.

Discussion

We conducted a systems biology investigation of gene expression in a case-control study of CHD using network approaches. To our knowledge, this is the largest gene expression study to date of CHD incorporating multidimensional genome-wide data from multiple sources including GWAS, gene expression, eSNPs, and molecular networks.

Extending a traditional approach that targeted differentially expressed individual genes ¹⁰, we compared the coexpression network structure in CHD cases with that in matched controls to derive differential network modules. The advantage of a network approach is that it provides a contextual framework to explain how genes interact with each other differently in CHD cases versus controls. Previous research has reported strong correlations of gene coexpression patterns with causal disease mechanisms. For example, a conserved coexpression module discovered in mouse and human liver and adipose tissues was found to be correlated with multiple metabolic traits ^{15, 21} and its causal relationship with metabolic diseases was further validated experimentally ^21-23.

Using coexpression network analyses, we identified 2 DMs that demonstrated differential coexpression patterns between cases and controls. The 79 genes in the control_tan module (Table SIII) showed a tightly co-regulated pattern in controls but a disrupted coexpression pattern in CHD cases. B cell-centered immune genes were highly enriched in this module, along with a suggestive enrichment for lipid transport genes (ABCA6, ABCA9, ABCB4, APOD, and OSBPL10). This is consistent with the known roles of immune response and lipid related pathways in atherogenesis. At a cellular level, atherosclerosis involves lipid accumulation in the artery wall, which can lead to plaque rupture, platelet aggregation, and coagulation leading to occlusive thrombus and MI. Immune responses are involved in multiple phases of atherogenesis ^{6, 24}. From the inception of atherogenesis, multiple immunocytes including macrophages, T cells, and mast cells are recruited by adhesion molecules or leukocyte receptors expressed on the surface of endothelial cells. The precise role of B cells in atherosclerosis, however, is relatively less well studied. In the traditional view, B cells are thought to play a protective role against atherosclerosis ²⁵. Kyaw et al. found that innate-like B1 cells mediate protection from atherosclerosis via the secretion of natural IgM antibodies ²⁶. The tightly co-regulated pattern of B cell-related genes in controls and the disrupted pattern in CHD cases found in our study are consistent with an athero-protective role of B cells. A recent study, however, showed reduced atherosclerosis in mice after B-cell depletion ²⁷, suggesting that B-cell subpopulations may have contrasting roles in the pathogenesis of atherosclerosis ²⁸.

It is of note that the blood samples used in this study were collected after the onset of clinical CHD events. Therefore, the differential gene signatures and DMs identified here could be a consequence of CHD or its treatment (i.e. downstream signals) rather being causal of CHD (i.e. upstream signals). GWAS has identified scores of loci associated with CHD and CHD risk factors, with about 27 loci for CHD or MI ^{3, 4}, 30 for blood pressure ^29-31, 95 for lipids ¹⁴, 28 for BMI ³², and 27 for type 2 diabetes ³³. Traditional GWAS, however, can only detect common loci yielding stringent p value associations, typically p<5×10^-8, and susceptibility loci with more subtle effects are missed. In an attempt to separate the causal and reactive effects of the DMs that differed between CHD cases and controls, we used SSEA to integrate our gene expression findings with publicly available GWAS results for CHD and its risk factors based on the reasoning that genetic variation is unaltered over time and that converging genetic associations with CHD gene expression signatures will highlight causal genes and pathways. SSEA is a complementary approach to single-point analysis, as it tests the overall association of a set of SNPs derived from a given gene set to a trait rather than a single SNP or gene association with a trait ¹², thus empowering the identification of biological pathways, functional categories, or networks. Whereas individual genes may only exert subtle effects on a trait, collectively they may contribute to disease susceptibility on a larger scale. Another advantage of SSEA lies in the fact that it utilizes eSNPs (SNPs associated with the expression levels of genes) from disease-relevant tissues to map genes to their functional SNPs. This is in contrast to distance-based gene-to-SNP mapping employed in other studies ^34-36.

The control_tan DM (enriched for genes involved in B cell activation/differentiation and immune function), but not the CHD_tan DM (enriched for ion transport genes) was identified as a causal module showing enrichment for eSNPs with low p value associations from CHD GWAS in CARDIoGRAM ³ and lipid traits GWAS in GLGC ¹⁴. These results indicate that B cell-related immune function may exert a causal effect on CHD and on lipids. As discussed above, prior studies support a causal role of B cells in atherogenesis. In addition, recent research by Fretz et al. revealed direct evidence that B cell dysfunction (Ebf1-/- mice) results in lipid abnormalities ²⁰. Therefore, our results not only support previous phenotypic observations linking B cells to both atherogenesis and lipid dysregulation, but also provide network and genetic support at the molecular level.

We further identified and prioritized putative key regulatory genes or KDs for the CHD causal DM based on the global networks derived from tissue-specific BNs and PPI networks. We identified 59 KDs for the control_tan module, 2 from multiple tissue networks and 57 from the blood-specific network. After prioritization of the 59 KDs, we retrieved a subnetwork derived from the top KD and found that the KDs and other control_tan DM genes form a tightly linked subnetwork. Perturbations of the multi-tissue KD, TNFRSF13C, and a tissue-specific KD, EBF1, in animal models have been shown to affect aortic root atherosclerosis and other CHD-related phenotypes ^{18, 20}. Therefore, the subnetwork structure and the KDs we uncovered provide both known CHD candidate genes with experimental support and novel candidate genes that warrant further validation.

In summary, we systematically compared the gene expression features of 188 CHD cases and 188 matched controls from the FHS via network modeling. Orthogonal studies incorporating CHD and CHD risk factor GWAS and data-driven gene regulatory networks from human and mouse models were integrated in this analysis to infer casual effects and possible pathological molecular mechanism. We found B cell-centered immune function to be related to CHD pathogenesis. We hypothesize that disruption of gene co-regulation of a B cell related DM via KDs contributes to CHD susceptibility. Our results provide further evidence supporting a critical role of adaptive immunity in atherosclerosis, especially aspects involving B cell function. Consequently, we speculate that immunomodulatory therapy may be useful to prevent or treat atherosclerotic CHD ³⁷.

Significance.

Atherosclerotic coronary heart disease (CHD) as a multi-factorial disease is influenced by both genetic and environmental factors. This study utilized a systems biology framework to identify genes and networks associated with CHD via integration of whole blood gene expression profiles with network approaches, GWAS, and genetics of gene expression. This study demonstrated a CHD differential coexpression module (DM), involving B-cell centered immune functions with a disrupted coexpression pattern in CHD cases but not in matched controls. This DM was implicated as causal based on its significant enrichment for both CHD and lipid eSNPs. This causal DM was further integrated with networks to identify regulatory key driver (KD) genes and their driving subnetwork structure. These findings provided further evidence of a critical role of adaptive immunity in atherosclerosis, especially aspects involving B cell function, and may help to identify new drug targets for the treatment or prevention of CHD.

Abbreviations

BMI

Body Mass Index

CABG

coronary artery bypass grafting

CHD

coronary heart disease

CARDIoGRAM

Coronary ARteryDIsease Genome wide Replication and Meta-analysis

CVD

cardiovascular disease

DBP

diastolic blood pressure

DIAGRAM

DIAbetes Genetics Replication and Meta-analysis

FHS

Framingham Heart Study

GIANT

Genetic Investigation of Anthropometric Traits

GLGC

Global Lipid Genetics Consortium

GO

gene ontology

GWAS

Genome-wide association studies

ICBP

International Consortium of Blood Pressure

MI

myocardial infarction

PPI

protein-protein interaction

SBP

systolic blood pressure

SSEA

SNP set enrichment analysis

PTCA

percutaneous transluminal coronary angioplasty with or without a stent

WGCNA

Weighted Gene Coexpression Network Analysis