Beyond GWASs: The usefulness of mouse genetics in understanding the complex etiology of atherosclerosis

Abstract

The development of population-based genome-wide association studies (GWASs) has led to the rapid identification of large numbers of genetic variants associated with coronary artery disease (CAD) and related traits. Together with large-scale gene-centric studies, at least 35 loci associated with CAD per se have been identified with replication. The majority of these associations are with common single nucleotide polymorphisms (SNPs) exhibiting modest effects on relative risk. The modest nature of the effects, coupled with ethical/practical constraints associated with human sampling, makes it difficult to answer important questions beyond gene/locus localization and allele frequency via human genetic studies. Questions related to gene function, disease-causing mechanism(s), and effective interventions will likely require studies in model organisms. The use of the mouse model for further detailed studies of GWAS-identified CAD-associated loci is highlighted herein.

Introduction

The development of population-based genome-wide association studies (GWASs) has led to the rapid identification of large numbers of genetic variants associated with coronary artery disease (CAD)/ myocardial infarction (MI) and related traits such as plasma LDL-cholesterol, HDL-cholesterol, triglycerides, obesity and hypertension (reviewed in^1–7). The majority of these associations are with common single nucleotide polymorphisms (SNPs) exhibiting modest effects on relative risk. The use of combined analyses, or meta-analyses, increases the power to detect modest associations.^{3, 8, 9} Currently, at least 35 loci associated with CAD per se have been identified and, importantly, replicated in at least one independent study (Table 1). While some of the loci are associated with traditional risk factors, many of the loci likely affect atherogenesis via non-traditional mechanisms.

Table 1.

Human-mouse genomic homologies for CAD-associated loci: availability of mutant mouse models.

Human chr	SNP	Gene(s) in region	Related phenotype (human)	Murine homolog(s)	Mouse chr	Mutant models*	Lesion phenotype	Related phenotype (mouse)
1	rs1120651041	PCSK9	LDL, TC3, 42, 43	Pcsk9	4	KO, Tg (liver)		LDL, TC44, 45
1	rs171140368	PPAP2B		Ppap2b	4	KO (embryonic lethal)
1	rs59983941, 46	SORT1	LDL, TC3, 42	Sort1	3	KO, Tg		LDL, TC20, 47
1	rs1746563741, 46	MIA3		Mia3	1	KO (perinatal lethal)
2	rs429937617	ABCG8	LDL, TC,3, 48 sitosterolemia48	Abcg8	17	KO, Tg	↓ B6-Ldlr−/−, Abcg8 Tg, WTD49 -- B6-Ldlr−/− or -Apoe−/−, Abcg8 liver-specTg, chow50 ↓ B6-Ldlr−/−, Abcg8 liver-specTg, WTD+ezetimibe51	LDL, TC, sitosterolemia49, 52
2	rs672588741	WDR12		Wdr12	1	KO (ES cell)
3	rs230637453	MRAS		Mras	9	KO
5	rs270639917	IL5		Il5	11	KO, Tg
6	Rs690395631	C6orf105		9530008L14Rik	13
6	rs1252645341	PHACTR1		Phactr1	13	KO (ES cell)
6	rs176099408	ANKS1A		Anks1	17	KO (perinatal lethal)
6	rs121902878	TCF21		Tcf21	10	KO
6	rs379822054 rs1045587254	LPA	LDL, TC3 Lp(a)54	___	___	huTg	↑ B6xSJL huLPA Tg, HFD55 ↑ B6 huAPOB, LPA Tg, WTD56 -- B6-Ldlr−/−, huAPOB, LPA Tg, chow57	LDL, TC, Lp(a)58
7	rs109535419	BCAP29 PRKAR2B HBP1 COG5 GPR22 DUS4L		Bcap29 Prkar2b Hbp1 Cog5 Gpr22 Dus4l	12	KO (ES cell) KO KO KO (ES cell) KO KO (ES cell)		body weight59, 60 insulin resistance60 heart failure61
7	rs115569248	ZC3HC1		Zc3hc1	6	KO
8	rs1732151517	TRIB1	TG, LDL, TC, HDL3, 42	Trib1	15	KO		LDL, TC, TG21
9	rs497757441, 46, 62–64	ANRIL CDKN2A (p16INK4a, p14ARF) CDKN2B(p15INK4b) MTAP		no contiguous segment) Cdkn2a (p16Ink4a, p19Arf) Cdkn2b (p15Ink4b) Mtap	4	locus deletion KO, Tg KO, Tg,(promoter), spont mutation	-- 129Sv locus deletion, HFD30 ↑ B6-Ldlr−/−, p16Ink4a, p19Arf BM-specKO, WTD24 ↑ B6-Apoe−/−, p19Arf KO, HFDb65
9	rs5794598, 22	ABO	LDL, TC3, 66 thrombosis67 inflammation68–71	Abo	2	KO (ES cell)
10	rs25050839	KIAA1462		9430020K01Rik	18	KO (ES cell)
10	rs174604841, 46	CXCL12		Cxcl12	6	KO (embryonic lethal) Tg (general, EGFP, endothelial-specific)
10	rs22469429	LIPA		Lipa	19	KO		lipase deficiency
10	rs124134098	CYP17A1 NT5C2	blood pressure72	Cyp17a1 Nt5c2	19	KO (embryonic lethal) KO (ES cell)
11	rs9748199	PDGFD		Pdgfd	9	KO
11	rs9641848	APOA1/C3/A4/A5	TG, HDL3, 42	Apoa1/c3/a4/a5	9	KO, Tg (single and compound alleles)	↓ B6-huAPOA1 Tg, HFD73 ↓ B6-Apoe−/−, huAPOA1 Tg, chow74 ↑ B6,mixed-huAPOBTg, Apoa1−/−, WTD75 ↓ B6-Apoe2 knock-in, huApoa5 Tg, WTD76 -- 129/OlaxB6-Apoa1/c3/a4−/− KO, WTD77	HDL, LDL, TC, TG73, 77–79
12	rs318450480	SH2B3	blood pressure81	Sh2b3	5	KO
13	rs47731448	COL4A1/A2		Col4a1/a2	8	KO, chem/rad-induced
14	rs28958118	HHIPL1		Hhipl1	12	KO (ES cell)
15	rs38258078, 9, 22	ADAMTS7		Adamtsl7	9	KO (ES cell)
17	rs2161728	SMG6 SRR		Smg6 Srr	11	KO KO, Tg (EGFP), chem-induced
17	rs129365878	PEMT		Pemt	11	KO, spont mutation	↓ B6-Ldlr−/−, Pemt−/−, HF/HC82 ↓ B6-Apoe−/−, Pemt−/−, chow83	TG, LDL, VLDL, TC,82, 83 intrahepatic cholestasis84
17	rs46522	GIP SNF8 UBE2Z ATP5G1		Gip Snf8 Ube2z Atp5g1	11	KO (ES cell) KO KO (ES cell) KO (ES cell)
19	rs112260841	LDLR	LDL, TC, FH3, 85	Ldlr	9	KO, Tg, chem.-induced	↑ B6x129Sv-Ldlr−/−, HFD86	LDL, TC, xanthomatosis87
19	rs207565017	APOE	LDL, TC, HDL3, 42, 88	Apoe	7	KO, Tg, rad-induced, spont mutation	↑ B6x129Sv-Apoe−/−, chow89 ↑ B6x129Sv-Apoe−/−, chow and WTD90, 91	LDL, TC, HDL, xanthomatosis89, 90
21	rs998260141	MRPS6		Mrps6	16	KO (cell line)

Chr, chromosome; SNP, lead disease-associated single nucleotide polymorphism; KO, knock-out allele(s); Tg, transgenic allele(s); EGFP, enhanced green fluorescent protein; chem-induced, chemically-induced mutant allele(s); rad-induced, radiation-induced mutant allele(s); spont mutation, spontaneous mutant allele(s); FH, familial hypercholesterolemia.

Diets: WTD, Western-type diet (0.15% cholesterol, 21% fat); HFD, high-fat diet (1.25% cholesterol/15% fat/0.5% sodium cholate); HFDb, high-fat diet (10.8% total fat, 0.75% cholesterol without sodium cholate); HF/HC, high-fat/ high cholesterol diet (1.25% cholesterol, 20% olive oil.)

^*Mutant alleles listed at www.informatics.jax.org, www.sanger.ac.uk, www.mmrrc.org, and www.fimre.org.

Notably, the CAD loci identified by GWAS thus far have been estimated to explain only ~10% of the additive genetic variance of human CAD.⁸ Several human genetic approaches towards detecting loci representing the unexplained variance have been discussed. These include approaches for detecting rare SNPs^{10, 11} or copy number variants¹² associated with disease. However, as the list of new disease loci grows, it will be important to establish the clinical or public health importance of the identified loci.

Model organisms provide useful tools for obtaining data related to clinical relevance. Due to the modest size effects of SNP variants and ethical/practical constraints associated with human sampling, questions regarding gene function and disease causing mechanism(s) can be assessed more definitively in model organisms. Further, effective interventions and potential modifiers of SNP-disease associations can be tested in model organisms prior to the design of clinical trials in humans. The mouse has become the model of choice due to small size, breeding efficiency, availability of genetic manipulation technologies, and high degree of genome similarity. Human-mouse genomic homologies have been identified for most SNP association loci identified to date (Table 1). Although some CAD-associated variants have been found in gene-poor regions or regions of unknown biological relevance, the candidate causal variants in these cases may function by regulating expression of neighboring genes. Thus, genetic studies of mice exhibiting altered expression or function of CAD SNP-residing or neighboring genes may be relevant. Mutant models – including transgenic, knock-outs derived by gene-targeting or gene trap technologies, chemical- or radiation induced mutagenesis, sub-chromosomal locus deletion, and spontaneous mutation – are available for many of the human CAD-associated loci identified to date but few have been queried for atherosclerosis susceptibility/resistance (Table 1). Further, more than 9000 conditional targeted alleles in mouse embryonic stem cells have recently become available.¹³ Lastly, random genetic variation among different inbred strains of mice can lead to the identification of novel genes underlying atherosclerosis.

Mendelian disease genes exhibiting common associations

Some of the CAD loci underlying common susceptibility to disease were previously identified in relationship to rare Mendelian forms of hypercholesterolemia/premature CAD. These include LDLR, APOE, PCSK9, ABCG8, and LIPA. Importantly, gene-specific mutant mouse models displayed effects on atherogenesis and related traits (i.e. plasma cholesterol levels, plant sterol levels, xanthomatosis, lipase deficiency) similar to that observed in humans (Table 1). These provide “positive controls” indicating that the use of mutant mouse models can be relevant to the study of human CAD loci.

The disease causing mechanism for Mendelian disease genes is at least partly established. While clearance of circulating lipoproteins is a common mechanism underlying disease pathogenesis for LDLR, APOE, and PCSK9, new studies in mouse models are revealing additional roles in inflammation (Apoe),^{14, 15} and apoB secretion (Pcsk9).¹⁶ Common variants of ABCG8 are associated with both plasma phytosterol as well as LDL levels.¹⁷ Detailed studies in mice have outlined the role of Abcg8 in dietary cholesterol absorption¹⁸ and intestinal cholesterol excretion¹⁹ but the role of plant sterols in atherogenesis remains unresolved. LIPA encodes a lysosomal acid lipase involved in the breakdown of cholesterol esters and triglycerides. The LIPA risk allele is associated with increased lipase expression but not altered lipid levels, suggesting a novel pathogenic mechanism. Thus, even the Mendelian-associated genes have the potential to reveal new pathogenic mechanisms!

New genes/loci associated with traditional risk factors

A few of the newly identified loci are associated with known risk factors for CAD, suggesting disease-causing mechanisms. For example, SORT1 and TRIB1 are associated with lipoprotein levels in human association studies. Functional validation of these associations was obtained via gene-specific over- and under-expression of these genes in genetically uniform mouse models of atherosclerosis.^{20, 21} Furthermore, Sort1 and Trib1 were shown to modulate hepatic VLDL secretion and production, respectively, from primary hepatocytes in mice.^{20, 21} These data suggest that modulation of two novel regulatory pathways for lipoprotein metabolism may alter susceptibility to CAD/MI in humans. The ABO gene is associated with multiple CAD-related traits: LDL levels, thrombosis, inflammatory gene expression and plant sterol levels.^{8, 22, 23} Further studies are required to delineate the relative role of each pathway in the pathogenesis of CAD.

New genes/loci underlying novel pathogenic mechanisms for CAD/MI

The majority of the loci listed in Table 1 have some degree of known protein function but no known role in CAD/MI pathogenesis. For these loci, basic knowledge of directional effects and tissue relevance can be sorted out in mouse models. Directional effects (i.e. for regulatory variants, whether increased or decreased gene expression is associated with disease) can be confirmed/established using general knockout or transgenic models crossed onto an Apoe^−/− or Ldlr^−/− proatherogenic background. In some cases, existing congenic,²⁴ spontaneous, chemically-, or radiation-induced mutants may be queried (Table 1). Tissue relevance (i.e. the specific tissue type affecting disease pathogenesis) can be assessed using bone marrow (BM) transplantation or tissue-specific knockouts. Reciprocal BM experiments utilizing a chromosome 4 congenic model harboring the 9p21 region of homology, and exhibiting decreased expression of macrophage Cdkn2a, indicated that BM-derived cells, but not resident vascular cells, were sufficient to confer the pro-atherosclerotic phenotype of the congenic mouse.^{24, 25} Direct testing of the candidate gene showed that BM-derived cells from Cdkn2a^+/− mice were sufficient to confer accelerated atherogenesis in the Ldlr^−/− background.²⁴ Of note, tissue macrophages and mixed monocyte/macrophage populations, but not circulating monocytes, were implicated in the study.²⁴ This study suggests that macrophage deficiency of CDKN2A may partly explain the association of 9p21 with CAD/MI in humans. The data are consistent with human studies reporting lack of association of the 9p21 risk allele with CDKN2A expression in circulating monocytes²⁶ or resident vascular cells,²⁷ but significant association with decreased levels in T lymphocytes.²⁸ Studies in human macrophages have not been reported.

Genetic variants of Anril have been implicated at the 9p21 human CAD/MI locus. ANRIL is a non-coding RNA implicated in both long-range cis-acting as well as trans-acting transcriptional control of syntenic tumor suppressor genes, CDKN2A (encoding p16^INK4a, p14^ARF) and CDKN2B (encoding p15^INK4b). Multiple ANRIL splice variants are present in human tissues, complicating genetic association studies of the structural gene.²⁹ A murine deletion mutant covering the homologous region exhibited decreased expression of the neighboring tumor suppressor genes, supporting the hypothesis of an ANRIL regulatory variant underlying the 9p21 locus.³⁰ A potential effect of the variant on atherosclerosis was not observed in the highly athero-resistant model tested. However, testing of the deletion in a more athero-susceptible model has not been carried out as yet.

Two of the CAD-associated loci listed in Table 1 were identified as open reading frames or cDNA clones via annotation efforts of the Human Genome Project (c6orf105) or the RIKEN Genome Science Lab (KIAA1462) but have no known biological function. C6orf105 exhibits ethnic-specific CAD association among Chinese Han populations³¹ but KIAA1462 exhibits association in both European and Chinese populations, with similar allele frequencies and size effects.⁹ Both loci have homologous DNA sequences in the mouse genome and, thus, targeted deletion or transcriptional disruption may shed light on the biological functions of novel proteins.

Additional mechanistic data including stage of lesion development, genetic background effects, potential effects on lesion regression, and overlapping roles of CAD loci in multiple diseases can be assessed using unique strains and experimental designs. Stage of lesion development can be tested in dietary time course studies of gene-specific knock-out or transgenic mice bred onto standard mouse models of atherosclerosis, or with conditional knock-outs induced before or after lesion development. Genetic background effects can be tested using different inbred strains of mice (exhibiting differences in susceptibility to atherosclerosis), different engineered models of atherosclerosis, or mutant mice carrying mutations in more than one CAD locus. The Reversa mouse^{32, 33} is a model of atherosclerosis regression which may be useful for differentiating hyperlipidemia from other genetic effects on lesion regression. Finally, introduction of CAD-associated mutations into mouse models of diabetes, hypertension, obesity and metabolic syndrome may shed light on shared points of regulation among multiple disease phenotypes.

New discovery of atherosclerosis susceptibility genes

As mentioned above, the CAD loci identified by human GWAS thus far have been estimated to explain only a fraction of the genetic variance of human CAD.⁸ Mouse genetic/mapping approaches provide a means of identifying new genes, perhaps untractable by human genetic approaches because of modest effect. While mouse linkage studies pinpoint disease susceptibility loci to relatively large genomic intervals containing large numbers of genes, several techniques have been applied to narrow the list of disease causal candidate genes. Refined mapping of loci can be obtained through the generation of interval-specific congenic strains. A cross between B6-Apoe^−/− and the more athero-resistant strain FVB-Apoe^−/− revealed two intervals contributing to atherosclerosis susceptibility; one locus was narrowed to 7 genes, the other to 21 genes.³⁴ Similarly, the congenic mapping efforts in a cross between B6-Ldlr^−/− and a wild-derived MOLF strain revealed two atherosclerosis loci on chromosome 4.²⁴ Cdkn2a was identified as a disease-causing gene in one of the intervals but mapping of the distal locus is still underway. Combining a congenic mapping approach with gene expression profiling, Lusis and colleagues identified Zhx2³⁵ as a novel regulator of plasma lipid metabolism.³⁵ Copy number variants can also be applied to mouse mapping studies. In a cross between B6 and C3H, gene expression levels and several metabolic traits mapped to three unique copy number variants, suggesting novel loci involved in regulation of plasma lipoprotein levels, glucose and body weight.³⁶ Recently, a hybrid mouse panel was developed for high-resolution association studies in mice. This approach aims to provide refined mapping and increased sensitivity compared to linkage studies. Together, these studies have the potential to reveal new genes and pathways underlying atherosclerosis susceptibility.

Limitations of the mouse model

Many of the loci listed in Table 1 were discovered based on case-control studies of MI. While atherogenesis precedes MI, not all cases of atherosclerosis lead to acute complications. This suggests that different pathologies underlie these clinical phenotypes. Furthermore, the ABO gene was specifically associated with MI in the presence of coronary atherosclerosis.²² Currently available mouse models are susceptible to atherosclerosis but resistant to acute complications. Thus, studies of genes affecting plaque rupture may be limited in the mouse. ABO is associated with thrombosis,²² and at least one spontaneous mutation in mice leads to atherothrombosis.³⁷ In addition, murine plaques exhibit features of human vulnerable plaques, a precursor to plaque rupture and infarction. These can be assessed by qualitative changes in plaque morphology.^{38, 39} Thus, although plaque rupture/MI may not be amenable to study per se in murine models, pathogenic mechanisms leading to clinical consequences may be queried.

The identification of phenotypically causal variants underlying CAD susceptibility is important for the delineation of biological genotype-phenotype relationships as well as discovering potential predictors of disease. Lead association SNPs may represent causal variants or may be associated by circumstance alone (ie. exhibiting strong linkage disequilibrium with the lead SNP). In most cases, disease-causing variants will not be the same in human and mouse. In particular, regulatory variants in non-coding regions may not be conserved. For example, the putative causal variant at SORT1 is human specific. The causal allele creates a binding site for the CEBP family of transcription factors that does not exist in the mouse.²⁰ However, data supporting the role of a regulatory variant can be gained from studies in mice. For example, the lead SNP identified for TRIB1, a triglyceride-, LDL- and CAD-associated locus, is located downstream of the coding sequence and suggested a regulatory effect on gene expression.^{3, 17, 40} Subsequent studies in Trib1-overexpressing and -deficient mice showed decreased and increased plasma triglyceride levels, respectively.²¹ Demonstration of regulatory effects stemming from an allele-specific mutant construct will be necessary to solidify the genotype-phenotype relationship.

Most disease-associated SNPs exhibit modest effects on relative risk. Thus, the relevance of complete gene knockout and highly-expressing transgenic mice comes into question. Several genetic methods exist for testing modest effects. Mice carrying heterozygous deficiency of a particular gene will likely demonstrate differences in gene expression more closely mimicking the situation in humans.²⁴ Secondly, BAC transgenic mice generally express only 1–3 copies of a transgene. Thirdly, spontaneous, radiation- and chemically-induced mutants usually harbor point mutations. Some of these models are available for the human CAD-associated loci (Table 1) and can be tested for differences in atherosclerosis susceptibility or plaque morphology.

Summary/Conclusions

Although recent human genetic studies have met with remarkable success in terms of identifying CAD/MI-associated loci, many details regarding the underlying genes/mechanisms remain unanswered. The high degree of genomic similarity between humans and mice, along with the wide array of genetic tools available, indicate that much can be learned from parallel studies of mice and human.