Human Validation of Genes Associated With a Murine Atherosclerotic PhenotypeHighlights
Objective—The genetically modified mouse is the most commonly used animal model for studying the pathogenesis of atherosclerotic disease. We aimed to assess if mice atherosclerosis-related genes could be validated in human disease through examination of results from genome-wide association studies.
Approach and Results—We performed a systematic review to identify atherosclerosis-causing genes in mice and carried out gene-based association tests of their human orthologs for an association with human coronary artery disease and human large artery ischemic stroke. Moreover, we investigated the association of these genes with human atherosclerotic plaque characteristics. In addition, we assessed the presence of tissue-specific cis-acting expression quantitative trait loci for these genes in humans. Finally, using pathway analyses we show that the putative atherosclerosis-causing genes revealed few associations with human coronary artery disease, large artery ischemic stroke, or atherosclerotic plaque characteristics, despite the fact that the majority of these genes have cis-acting expression quantitative trait loci.
Conclusions—A role for genes that has been observed in mice for atherosclerotic lesion development could scarcely be confirmed by studying associations of disease development with common human genetic variants. The value of murine atherosclerotic models for selection of therapeutic targets in human disease remains unclear.
Atherosclerosis is a multifactorial process that develops over decades, underlying the majority of cardiovascular diseases. Because of its slow progression, studying the natural history of atherosclerosis requires serial examinations, thus complicating the design of studies in humans. Most research on biological mechanisms of atherosclerosis has been performed in genetically modified mice, eliminating these challenges faced in human studies. Thus, genetically modified mice elegantly allow the study of atherosclerosis in, arguably, the best-controlled model system possible. The most commonly used atherosclerotic murine models are apolipoprotein E (ApoE) or low-density lipoprotein receptor (LDLR) gene knockouts. These mice clearly display an accelerated atherosclerotic phenotype with human-like vascular lesions.1,2 Experimental modifications of these murine models may accelerate vascular plaque development resulting in advanced lesions within several weeks.3,4 Consequently, such models have been crucial in understanding the murine molecular and cellular basis of atherosclerosis. Yet, the relevance in human atherosclerotic disease remains elusive for several reasons. Primarily, the morphology of atherosclerotic plaque in mice differs from that of humans and acute events because of luminal thrombosis and evident plaque rupture are rarely observed. Second, clinical disease manifestations such as coronary artery disease (CAD) or ischemic stroke in mice are rare or lacking. Third, it is arguable whether complete knockouts in mice correspond with expression-changing mutations in humans. Finally, despite the knockout of individual genes, genetic redundancy on the pathway level further complicates the interpretation of results from animal models.5
In recent years, millions of common single-nucleotide polymorphisms (SNPs) in the human genome were identified, and our understanding of these variants with respect to the genomic architecture has increased significantly.6 This has opened up the possibility to agnostically assess the effects of genome-wide variation on human traits and disease.7 Indeed, meta-analyses of human genome-wide association studies (GWAS) have identified many risk loci for CAD8 and large artery ischemic stroke (LAS).9 These GWAS provide the unique opportunity to validate the putative disease-causing genes identified through murine models in humans.
We performed a systematic review to identify atherosclerosis-causing genes in mice and carried out gene-based association tests of their human orthologs for CAD and LAS. Moreover, we investigated the association of these genes with human atherosclerotic plaque characteristics. Furthermore, we assessed whether there are tissue-specific cis-acting expression quantitative trait loci (eQTLs) for the genes in humans. We report that putative atherosclerosis-causing genes reveal little association with human CAD, LAS, or atherosclerotic plaque characteristics, despite the fact that the majority of these genes have cis-acting eQTLs.
Materials and Methods
Materials and Methods are available in the online-only Data Supplement.
Of the 659 murine genes (Table I in the online-only Data Supplement), a total of 486 genes (73.75%) were studied in knockout mice, 57 genes (8.65%) were studied in transgenic mice, and 116 genes (17.60%) were targeted by specific compounds (Table I in the online-only Data Supplement). For 185 genes (28.07%), there are (preclinical) drugs available (Table I in the online-only Data Supplement).
Validation Within GWAS of CAD and LAS
We obtained summary statistics from GWAS on CAD8 and large artery stroke9 for SNPs ±50 kb from the 5′ and 3′ gene borders of the 659 studied murine genes (online-only Data Supplement). We used these as input for a gene-based analysis using versatile gene-based association study (VEGAS), which assigned an empirical P value (after permutations) to each of the 659 genes based on the P value of the SNPs in and ±50 kb around the genes of interest.10 Thus, each gene was given a P value of association to CAD or LAS based on the GWAS results while taking into account the correlation between SNPs that may exist. Of the 659 studied genes that have been shown to affect atherosclerotic phenotypes in mice, 11 (1.7%) genes were associated with CAD after correction for multiple testing (P≤0.05/659≤7.59×10−5, Table III in the online-only Data Supplement). In contrast, none of the genes were associated with LAS after correction for multiple testing. The top 10 most significant genes for CAD and LAS are shown in Table 1. A total of 84 (12.7%) and 41 (6.2%) of the genes were associated with CAD and LAS, respectively, at a nominal P≤0.05. The overlap of associated genes between LAS and CAD is limited; only the locus at 9p21 (containing CDKN2A/B), significantly associated with CAD, was also nominally associated with LAS (P<0.0062). We did not observe any further overlap between CAD and LAS top-associated genes. When looking at model groups to which the genes were assigned (knockout, transgenic, or compound), we did not observe significant differences between groups for the 11 significant genes (P=0.513 using a χ2 test).
Validation Using Human Atherosclerotic Plaque
Subsequently, we conducted a similar gene-based analysis using VEGAS on 7 plaque characteristics in the Athero-Express Biobank Study.11 These human plaque characteristics have previously been associated with clinical presentation12 and have been examined in many atherosclerotic murine models. Overall, of the 657 genes a low number of genes nominally associated with a human plaque characteristic and followed the expectation under the null (range, 4.1%–6.1%, Table V in the online-only Data Supplement). Only 2 genes were significantly associated with human intraplaque macrophages, F10 on chromosome 13 (P=1.00×10−6), and TNFAIP8L2 on chromosome 1 (P=5.80×10−5) after correction of multiple testing. All gene-based association results for the 659 target genes with plaque characteristics are provided in Table VI in the online-only Data Supplement.
Validation Using Pathway Analyses
The genes that reached nominal significance in the gene-based analysis of CAD and LAS were further analyzed using ingenuity to identify canonical pathways associated with CAD and LAS (Table IV in the online-only Data Supplement). Table 2 provides the 25 most significant canonical pathways based on the 659 murine target genes and the translation of these genes to human CAD and LAS. The liver X receptor/retinoid X receptor (LXR/RXR) activation pathway (involved in lipid metabolism) that has been extensively studied in atherosclerotic mice was found to be significantly enriched for genes associated with CAD and LAS. In contrast, the nuclear factor κB (NFκB) inflammatory signaling pathway that has been extensively studied in murine models, revealed less genes that associated with CAD or LAS (Table 2). Another example is the T-lymphocyte differentiation pathway that has been extensively studied in mice and associated with murine atherosclerosis, but for which we found little supportive evidence in our gene-based analysis associating with human CAD or LAS.
eQTL Analysis of the Murine Genes
In murine models, the 659 genes are clearly affected through knockout, transgenic techniques, or targeted compound treatment. A close human analog of such an effect would be cis-acting common genetic variants, affecting gene expression through reducing or upregulating expression. Such variants are known as eQTLs. We conducted cis-eQTL analyses using the Stockholm Atherosclerosis Gene Expression (STAGE) study and queried 3 online public resources (Table VII in the online-only Data Supplement) to identify common variants modulating the expression of the 659 genes in humans. Across the 4 data sets and 11 cell/tissue types, we found eQTLs (P≤7.31×10−7 after correction for 68 402 variants in and around these genes) for 411 of the 659 genes (these genes are marked orange in Table VIII in the online-only Data Supplement). For all genes, we found SNPs in cis-affecting expression (P<0.05) in any of the 4 data sets queried (Table VIII in the online-only Data Supplement). In a representative example of a canonical pathway (NFκB), we show that each target gene has a cis-eQTL, that is, a common variant (SNP) in or around the gene that has a significant effect on tissue-specific gene expression in humans.
Not all genes will exert their effect on clinical outcome via gene expression, rather gene function. Thus, we compared the gene-based association results for CAD and LAS for the genes with a valid eQTL (PeQTL<7.31×10−7) and without (PeQTL≥7.31×10−7), but found no significant difference in enrichment for disease association in either CAD (P=0.882) or LAS (P=0.634). Similarly, we compared the gene-based association results between the model groups for the 84 and 42 genes that associated with CAD and LAS (knockout, transgenic, or compounds) and found a statistical difference between groups for CAD (P=0.0013, χ2=13.291). A pairwise comparison of the 3 model groups revealed that knockout and compounds do not have statistically different gene-based results (P=0.718), whereas the results between knockouts and transgenic mice (P=0.00049, χ2=12.146) or transgenic mice and compounds do (P=0.0029, χ2=8.852). We found no statistical difference for LAS (P=0.637) when comparing the 3 model groups. We also stratified our pathway analyses based on these 3 model groups, but this revealed no additional significant pathways.
This study shows that putative atherosclerosis-causing genes identified in murine atherosclerosis models reveal little association with human CAD, LAS, or atherosclerotic plaque characteristics, despite the fact that the majority of these genes have cis-acting eQTLs.
Overall, the majority of genes associated with an atherosclerotic phenotype in mice do not carry variants that associate with human CAD, LAS, or advanced plaque characteristics. In contrast, murine genes involved in lipid metabolism significantly associated with human CAD, which is consistent with the known role of lipids in CVD risk. Indeed, lipid-lowering drugs, such as statins, act through HMGCR (chromosome 5q13.3) to lower circulating lipids.13 A recent GWAS showing that variants in the HMGCR locus are associated with an increase of 2.84 mg/dL total cholesterol,14 effectively confirmed this drug action retrospectively. However, for most murine genes and pathways of innate and adaptive immunity, there was no association with human CAD, LAS, or plaque characteristics.
ApoE−/− and LDLR−/− models are widely used to study the initiation and progression of atherosclerosis. To study the effect on plaque development15 or therapeutic strategies,16 additional (double) knockout or transgenic atherosclerotic models have been developed. Yet, they lack plaque rupture thrombosis and subsequent cardiac or cerebral ischemia. Although therapeutic strategies have been developed based on animal models, failures in clinical utilization underscores the need for human verification and translation before initiating targeted drug development programs.17,18 Indeed, a post hoc analysis by deCODE genetics and Amgen assessed the validity of results from human genetic studies as positive predictors of successful clinical trials.19 Essentially all failed clinical trials targeting a gene (locus) lack any evidence from genetic association studies.19
There are several potential explanations for the observed discrepancies of genes involved in atherosclerotic murine studies and human cardiovascular disease. First, they may be explained by differences in effect sizes. Common genetic variants associated with human disease often have a modest effect, in contrast to experimental gene manipulation (ie, knockout) in animal models to study atherosclerotic disease. These genetic modifications in mice limit inferences about dose-dependent effects, which is relevant for predicting drug effects in human disease. Furthermore, the combined effect of human population history and selection may have yielded little functional genetic variation and thus no association with CAD or LAS, even in a large sample. The low number of murine genes associated with plaque characteristics in the Athero-Express study may be explained by limited sample size and, therefore should be interpreted with caution. However, for our gene-based analysis of CAD and LAS, we had access to GWAS results based on large mete-analyses, providing substantial statistical power to detect genes associated with disease. In addition, we studied the enrichment of murine-derived atherosclerosis-related gene sets within canonical pathways. We then tested whether their human orthologs (that were nominally associated with disease based on a gene-based test) were also observed in these pathways. For the majority of pathways, weak or even absent evidence was found in humans. Of note is the LXR pathway which is proven to be linked with human cardiovascular disease, and for which we found ample evidence using our gene-based and our pathway analysis. Furthermore, in atherosclerotic mouse models, plasma lipid levels are the main determinant of lesion development. The associations between mice and human GWAS studies may improve when human individuals are studied with (genetic) susceptibility to abnormal lipid metabolism. Such interactions have not been explored in this study. However, such an association would imply that the atherosclerotic mouse models cannot represent disease development in the general population.
Second, one might argue that the SNP to gene mapping done by VEGAS is conservative and consequently excludes (regulatory) variants of greater effect that may lie as far as 1 Mb. However, most variants significantly affecting gene expression are found within 50 kb of the gene body,20 and regulatory elements are usually found in intergenic regions. Nevertheless, it is unlikely that erroneous annotations alone can explain the strong discrepancy between murine genes and their human orthologs within pathways.
Third, undoubtedly the Ingenuity Knowledge Base is a comprehensive summary of the current knowledge from literature on gene networks and pathways. Although it is constantly updated and manually curated, our ingenuity-based pathway analysis is biased. Indeed, genes (and thereby networks and pathways) that are not studied (in humans or any model system) would simply not exist in ingenuity, thus partly explaining an apparent discrepancy between associated pathways in humans and mice. In addition, pathways are often cell and organ specific and a selective approach toward cell types that are considered to play a dominant role in atherogenesis may affect the readout of our pathway analyses.
Fourth, the annotation of candidate genes to GWAS loci is usually based on literature and proximity to the genome-wide significant SNP, but whether such a gene actually influences disease is still unknown. For our gene-based approach, we mapped the mouse genes to their human ortholog and tested their association with CAD or LAS. Therefore, the genes that have been annotated to GWAS loci may not seem in our results even when the SNPs in their proximity meet the genome-wide significance threshold.
Finally, a transgenic or compound-treated mouse may not be comparable with a knockout model, thus explaining the lack of association of these genes with human disease. Given the statistically different gene-based results, we stratified our pathway analyses based on these 3 different model types and found no difference. Hence, it is unlikely that the type of murine intervention model explains the lack of association with human disease.
Previous studies have raised doubts on the validity of translating murine models to human pathophysiology in studying immune responses,21–23 although conflicting results have been obtained.24 Our study adds to the debate on the relevance of murine models for human atherosclerotic disease and supports the view that the direction of the scientific process matters. The lack of association of murine atherosclerotic genes with human CAD, LAS, or human plaque characteristics may be because of diverged gene expression patterns among mice and humans resulting in different phenotypic effects. Indeed, results from the recently published Mouse ENCODE (Encyclopedia of DNA elements) project show that gene expression patterns diverge between mice and humans.25,26 Moreover, phenotypic effects of orthologous genes frequently differ between species. This suggests that a sensible approach would be to group genes based on mouse-human orthology to improve the translational power of putative murine genes. GWAS agnostically provide human evidence for the involvement of genetic loci in the underlying mechanisms of atherosclerotic disease, including CAD and LAS. Therefore, from the outset, the rationale to initiate an examination of the role of genes in these loci in murine models of atherosclerosis could be supported by human data.
Our study may have several limitations. Individuals of non-European descent are underrepresented in the GWAS we examined. It remains to be investigated if results will differ for GWAS in non-European cohorts.
The genetic association study on human plaque characteristics was executed in the Athero-Express study and included 1443 patients with significant atherosclerosis. Although this study represents the largest collection of histologically investigated plaques, this study may have limited power when examining the genetics of plaque characteristics.
We studied all published papers that applied the ApoE−/− and LDLR−/− as a model for atherosclerosis. However, alternative murine models have been studied in atherosclerotic disease that we did not include in our search and we cannot exclude that these reveal better associations with the human genetic outcome studies.
In conclusion, a role for genes that has been observed in mice for atherosclerotic lesion development could scarcely be confirmed studying associations of disease development with common human genetic variants indicating that knockout models of atherosclerosis are not a good reflection of the variation that underlies common forms of atherosclerosis. The value of murine atherosclerotic models for selection of therapeutic targets in human disease remains unclear.
We acknowledge the support from the Netherlands CardioVascular Research Initiative from the Dutch Heart Foundation, Dutch Federation of University Medical Centres, the Netherlands Organisation for Health Research, and Development and the Royal Netherlands Academy of Sciences. We thank Aisha Gohar for her critical review of the article.
Sources of Funding
S.W. van der Laan is funded through grants from the Netherlands CardioVascular Research Initiative (GENIUS, CVON2011-19), the Interuniversity Cardiology Institute of the Netherlands (09.001). S. Haitjema and S.W. van der Laan are supported by the FP7 EU project CVgenes@target (HEALTH-F2-2013–601456). The UCL (University College London) Hospitals NIHR (National Institute of Health Research) Biomedical Research Centre supports F.W. Asselbergs. The work for STAGE (Stockholm Atherosclerosis Gene Expression study) was supported by PROCARDIS in the 6th EU-framework program (LSHM-CT-2007–037273), the Swedish Heart-Lung Foundation (J.L.M. Björkegren), the King Gustaf V and Queen Victoria’s Foundation of Freemasons (J.L.M. Björkegren), the Swedish Society of Medicine (J.L.M. Björkegren), the Swedish Heart Lung Foundation and Research Council (H.F. Asl and J.L.M. Björkegren). This work was also supported by grant from University of Tartu (SP1GVARENG, J.L.M. Björkegren), the Estonian Research Council (ETIS, J.L.M. Björkegren), the Roslin Institute Strategic Grant Funding from the BBSRC (T. Michoel) and by Clinical Gene Networks AB. Supported by the EU-funded Integrated Projects Cardiogenics (LSHM-CT-2006–037593), CVgenes@target as well as the BMBF (Bundesministerium für Bildung und Forschung)-funded German National Genome Network (NGFN-Plus) Project Atherogenomics (FKZ [Förderkennzeichen]: 01GS0831) and e:AtheroSysMed, with participation in the German Excellence Center of Cardiovascular Research (DZHK [Deutsches Zentrum für Herz-Kreislauf-Forschung]), partner site Munich Heart Alliance and the Leducq consortium CADgenomics.
J.L.M. Björkegren is founder, main shareholder, and chairman of the board for Clinical Gene Networks AB (Clinical Gene Network [CGN]) and Tom Michoel is shareholder. CGN has an invested interest in microarray data generated from the STAGE (Stockholm Atherosclerosis Gene Expression study) cohort. Cavadis B.V. financed genotyping of AEGS1. G. Pasterkamp and D.P.V. de Kleijn are founders and stockholders of Cavadis B.V.
↵* G. Pasterkamp, S.W. van der Laan, and S. Haitjema are shared first authors.
↵† H.M. den Ruijter and F.W. Asselbergs are shared last authors.
The online-only Data Supplement is available with this article at http://atvb.ahajournals.org/lookup/suppl/doi:10.1161/ATVBAHA.115.306958/-/DC1.
- Nonstandard Abbreviations and Acronyms
- coronary artery disease
- expression quantitative trait loci
- large artery ischemic stroke
- genome-wide association studies
- single-nucleotide polymorphism
- Received August 27, 2015.
- Accepted February 17, 2016.
- © 2016 American Heart Association, Inc.
- von der Thüsen JH,
- van Berkel TJ,
- Biessen EA
- Cheng C,
- Tempel D,
- van Haperen R,
- van der Baan A,
- Grosveld F,
- Daemen MJ,
- Krams R,
- de Crom R
- Traylor M,
- Farrall M,
- Holliday EG,
- et al
- Verhoeven BAN,
- Velema E,
- Schoneveld AH,
- de Vries J-PPM,
- de Bruin P,
- Seldenrijk CA,
- de Kleijn DPV,
- Busser E,
- van der Graaf Y,
- Moll F,
- Pasterkamp G
- Hellings WE,
- Peeters W,
- Moll FL,
- Piers SR,
- van Setten J,
- Van der Spek PJ,
- de Vries JP,
- Seldenrijk KA,
- De Bruin PC,
- Vink A,
- Velema E,
- de Kleijn DP,
- Pasterkamp G
- Plump AS,
- Masucci-Magoulas L,
- Bruce C,
- Bisgaier CL,
- Breslow JL,
- Tall AR
- Seok J,
- Warren HS,
- Cuenca AG,
- et al
- Takao K,
- Miyakawa T
Although many putatively atherosclerosis-causing genes have been identified through the genetically modified mouse model, the translation to human atherosclerotic disease has been notoriously challenging.
Here, we systematically reviewed literature to identify putative atherosclerosis-causing genes in mice.
We identified their human orthologs and performed gene-based association tests on human coronary artery disease, ischemic stroke, and plaque characteristics.
Of the 659 identified genes, 11 were associated with coronary artery disease, none with stroke, and 2 with intraplaque macrophages.
Pathway analyses confirmed the limited association with human disease, despite the fact that many genes have a cis-expression quantitative trait loci.
Our study underlines the need for human validation of murine atherosclerosis-causing genes.