Abstract
There is a significant unmet need in the diagnosis and prediction of coronary artery disease (CAD) and subsequent major adverse cardiovascular events including myocardial infarction. Recently, a number of genomic approaches have been taken to address these problems, including genetic, transcriptomic, and proteomic methodologies. Most of these efforts focused on the discovery of putative markers and initial construction of multimarker classifiers for these clinical endpoints and were limited by incomplete clinical validation and comparison to clinical factor models. However, one such classifier for obstructive CAD has been validated in independent multicenter validation cohorts. Initial forays into combining different types of classifiers (genetic, proteomic, etc.) using so-called systems biology approaches is in its infancy but holds considerable promise to further improve the outlook for patients at risk for these serious and prevalent conditions.
Keywords
Coronary artery disease, genomics, genetics, gene expression, proteomics, myocardial infarction
Chapter Outline
Introduction 83
What’s New? 84
Clinical Challenges 84
Genetics 86
Transcriptomics 88
Clinically Validated Gene Expression Tests for Obstructive CAD 90
Gene Discovery for MI Signatures 93
Multivariate Proteomic Predictors of CAD and MACE 93
Multimodality or Systems Biology Approaches 96
Other Approaches 96
Conclusion 96
References
Introduction
Diseases of the cardiovascular system encompass a wide diversity of etiologies, severity, and unmet medical needs. Coronary artery disease (CAD) and its sequelae, including acute coronary syndromes, myocardial infarction (MI), heart failure, and cardiac arrhythmias are the leading causes of morbidity and mortality in the developed and developing world. Substantial progress has been made in elucidating genetic and environmental risk factors for these diseases, and the availability of 3-hydroxy-3-methylglutaryl-coenzyme A reductase inhibitors (statins), beta-blockers, aspirin, and angiotensin converting enzyme inhibitors/Adrenergic receptor antagonists, as well as new devices and imaging modalities such as drug eluting stents, implantable cardiac defibrillators, ventricular assist devices, and coronary computed tomography angiography have made substantial inroads on reducing the prevalence and severity of these diseases. These advances have created a challenge to individualize and focus treatment for those who will most benefit and conversely to spare those at low risk expensive and invasive diagnostic procedures, thus creating an opportunity for the development of new diagnostics, as the clinical presentation for these diseases, especially in their early stages, are variable and complex.
What’s New?
A number of new technologies have become available over the last decade which have had significant impact on our understanding of the development, diagnosis, natural history, and treatment of CAD. For genetic methods, these include the large-scale application of genome-wide association studies (GWAS) using single nucleotide polymorphisms (SNPs) such as in the CARDIOgram consortium , exome sequencing , and the utilization of SNPs to test causal roles for specific genes by Mendelian randomization methods . Measurement of the “expressed genome” by microarrays and RNA sequencing for the transcriptome and targeted and unbiased approaches to the proteome and metabolome are technically more challenging but have very significant promise. Noninvasive imaging improvements, especially for coronary CT-angiography, have resulted in increasingly detailed pictures of in vivo atherosclerosis, including both hydrodynamic functional characterization by coronary CT-fractional flow reserve measurements, as well as determinations of plaque composition, all while reducing radiation dosing . These advances have occurred in the setting of several seminal clinical studies which have shown that optimal medical therapy is comparable to percutaneous coronary intervention (PCI) in the stable obstructive CAD setting , and that the post-PCI natural history of atherosclerosis in terms of MI, often occurs at nonculprit, nonobstructive lesions . All of this has occurred in a setting where overall disease prevalence has decreased over the last few decades .
Clinical Challenges
Despite these advances, key clinical challenges remain. The diagnosis of obstructive or significant CAD in stable symptomatic patients without known disease has been complex due to variability in symptoms and presentation, coupled with the decrease in disease prevalence. In addition, assessment of major adverse cardiovascular event (MACE) likelihood, especially MI, in both primary and secondary prevention settings, remains difficult. Finally, the triage of patients who present in the emergency clinic with chest pain symptoms suggestive of acute coronary syndromes or MI also remains challenging. The landscape in these areas has shifted recently with the recognition that overall disease burden may be a more important factor for event prognosis than vessel obstruction or ischemia . For CAD diagnosis in symptomatic patients without known disease, a typical workup may include a family history, risk factor assessment, stress echocardiogram, and then if clinically indicated, noninvasive imaging such as myocardial perfusion, stress echocardiography, or coronary CT-angiography. Positive imaging results will then often lead to invasive coronary angiography (ICA) to further define anatomy and potentially lead to revascularization. The inefficiency of this process was demonstrated in the report from Patel et al. showing that for approximately 400,000 patients referred for ICA, the fraction of those with obstructive disease was only 38% , despite >80% having had noninvasive imaging performed ( Fig. 6.1 ). These observations have been confirmed in additional studies and raise concerns about over testing and unnecessary imaging and concomitant radiation to the patient . An additional issue is referral bias in estimation of the performance characteristics of noninvasive imaging which likely reduces true sensitivity and increases specificity of imaging methods for obstructive disease, as patients with negative imaging tests are rarely referred for ICA.
For MACE prognosis, there are a number of clinical settings where more accurate information would be useful: primary prevention in asymptomatic individuals as 30% of CAD patients present with an MI as their first symptoms, known CAD patients to tailor treatment to risk, and secondary prevention in patients who have had an MI or revascularization. In these settings, clinicians typically utilize risk factor algorithms as exemplified by the Framingham risk score, and also societal guidelines for making treatment decisions. Recent work on the genetics of MI has suggested that a great deal of the heritability of MI is through uncharacterized mechanisms .
Genetics
Large-scale GWAS including up to 200,000 patients have been applied to understand the underlying genetic basis of CAD and MACE. In the latest studies from the CARDioGRAM-G4D consortium, a total of 46 loci have been identified at nominal genome-wide significance although all have effect sizes <2-fold . In addition, a number of GWAS studies have been completed examining CAD risk factors, especially lipids. A number of groups have taken a variety of statistical approaches to combine significant SNPs from these GWAS studies into so-called genetic risk scores (GRS) and have asked whether they improve the ability to identify patients at risk for CAD or MI and cardiovascular death . Although statistically significant improvements in some parameters have been seen over clinical risk factor algorithms, the effects have overall been quite modest ( Table 6.1 ). It should be noted that the currently known genome-wide significant SNPs only account for about 10% of heritable CAD risk, which is estimated to be 40%–50%, suggesting there is room for improvement . In addition, it appears comorbidities such as diabetes may affect the significance of some genetic variants for CAD .
| Study | Population | N | Endpoints | SNP Source ( N ) | Results |
|---|---|---|---|---|---|
| Paynter, 2010 | Women’s Genome Health Study | 19,313 | MI/MACE ( N =199) | Literature (101) | Not significant |
| Ripatti, 2010 | Case:control and Prospective Finnish, Swedish | 70,000 | First MI/revasc., unstable angina, CV death ( N =1200) | Literature (13) | 1.7-Fold increase top to bottom quintile |
| Brautbar, 2012 | ARIC, Rotterdam, Framingham offspring | Total 12,900 | MI, CV Death, revasc. | Literature (13) not associated with intermediate phenotype; weighted and unweighted GRS | Improvement in ARIC but not other studies |
| Patel, 2012 | Emory and Cleveland Clinic Cath Lab cohorts | Total 5000 | Earlier prevalent MI versus older disease free | Literature (11) genome-wide significant prior to 2009; weighted GRS | Significant for earlier onset prevalent MI, but not incident MI |
| Thanassoulis, 2012 | Framingham | Total 3000 | MI, CV death, or calcium score | Literature (29) for MI/CAD and 89 for CHD-related phenotypes | Significant but not for discrimination of MI/CAD; better for CAC |
| Isaacs, 2013 | Erasmus Rotterdam | Total 10,000 | Lipid levels, cIMT, carotid plaque, incident MI | Lipid-related SNPs from Literature—4 GRS calculated—up to 26 SNPs | Significant but not better than FRS by AUC and only marginally additive to FRS |
| Smith, 2014 | CHARGE Consortium | Total 6900 | Aortic calcium and valve disease | Lipid-related SNPS from Literature—LDL (57), TG (39), HDL (70) weighted GRS calculated | Significant but did not correct for clinical factors; Mendelian randomization design |
| Bjornsson, 2015 | Icelandic Emory | Total 10,400 | Angiographic disease severity | Literature SNPs related to CAD (50), restricted GRS (32) | 1.7-fold increase in multivessel disease top to bottom quintile |
One hypothesis for the lack of larger effects with the GRS is that the SNP-based technology, which only interrogates common polymorphisms, may be missing rare variants with much larger effect sizes. A recent study examining this hypothesis used exome sequencing to compare an early onset MI population with older MI-free controls, searching for enrichment of rare nonsynonomous coding sequence variants .
Although both large-scale GWAS and exome-sequencing methods are very powerful, the former has often identified loci of unknown function, and the latter is by definition focused on a specific genomic subset. By utilizing the randomization of parental alleles which occurs at meiosis, a randomized genetic experiment (so-called Mendelian randomization) can be done on a population basis, analogous to a randomized clinical trial of a therapeutic drug. The prerequisite is identification of SNPs which cleanly and selectively affect the gene activity and phenotype of interest. These usually correspond to SNPs in cis to the target gene which affect either the expression of the gene or its activity. Recent work has used these methods with significant success both to confirm causal relationships with known risk factors, validate candidates [Lp(a) , triglycerides and low-density lipoprotein (LDL) , NPCIL1 ] and call into question others (High-density lipoprotein (HDL), sPLA-2, C-reactive protein (CRP) ).
Transcriptomics
Since coronary atherosclerosis is characterized by a long-term maladaptive inflammatory immune response, it is possible that biological variables which integrate both genetics and environment factors, such as RNA and protein expression in circulating blood cells or plasma, might provide a fruitful avenue for discovery and development of signatures for the diagnosis or prognosis of CAD and MACE. Since circulating cell-based RNA signatures may reflect both the development of and response to atherosclerotic plaque, their utility is particularly likely given the systemic nature of atherosclerosis . Early work showed that the dysregulation of messenger RNA expression in circulating cells appears to correlate with the burden of atherosclerosis . In addition, the emerging areas of micro-RNAs and long noncoding RNAs may also be of interest . A role for circulating micro and long noncoding RNAs packaged in vesicles in CAD needs further validation, but cardiac derived noncoding RNAs in MACE diagnosis and prognosis has shown interesting results . Recent analysis of the relative global importance of transcriptional and translational regulation of protein expression has suggested that the former has been underestimated and that transcriptional regulation accounts for a large majority of variation in the expressed genome, reinforcing the importance of examining RNA levels .
Methodology
The discovery of informative transcripts, development of a multigene classifier, analytical validation, and independent clinical validation of such a classifier are required before the clinical utility of such a diagnostic or prognostic test can be evaluated. There are significantly greater experimental challenges with reproducibly measuring RNA profiles as compared to DNA, whether by microarray, RNA sequencing, or Real-time-quantitative polymerase chain reaction (RT-qPCR), including sample preservation, RNA purification, and data normalization and analysis .
Gene Discovery for CAD
In the case of gene discovery, there have been studies from four groups with significant numbers of patients, where investigators used microarrays for gene discovery and then confirmed results with RT-qPCR either in the same (technical replication) or an independent (biological replication) cohort . These studies are summarized in Table 6.2 and differ significantly in their clinical phenotypes, especially by the focus on either stable CAD or the inclusion of patients with a history of MI or revascularization, as well as definitions of CAD, allowance for comorbidities such as diabetes, microarray platforms, size of cohorts, and statistical analysis methods. Given these differences and the significant correlation structure of whole blood gene expression data, largely driven by cell-type specific expression, it is not surprising that the sets of genes identified as significantly associated with CAD or MI do not show substantial overlap . However, a few similarities are worth noting: in the Wingrove and Elashoff studies, a significant upregulation of neutrophil genes and decrease in lymphocyte genes is observed, similar to that seen in the Kim study . This reflects, at least in part, the known correlation of neutrophil/lymphocyte ratio with noncalcified plaque and MI prognosis . In addition, the key-driver systems biology analysis of the Framingham study identifies a set of genes of which 2 (CD79B and SPIB) are found in the classifier derived by Elashoff and another 3 (CD200, BACH2, and TSPAN13) are found in the 655 genes found as significant in both the larger Elashoff microarray studies. Given the differences found between diabetics and nondiabetics with respect to CAD gene expression, and the inclusion of both sets in the Joehanes work , a nondiabetic subset analysis may have shown even more overlap.
| Study | Microarray Platform, Sample type | Population | N | Case Definition | Genes ( N ) | Biological Pathways |
|---|---|---|---|---|---|---|
| Wingrove | Agilent PBMC and PAXgene | Cath Lab, includes unstable angina. Controls 0% stenosis. Replication by RT-PCR excluded MI, revasc, known CAD | 41 95, 107 | Obstructive CAD70% stenosis or 50% in two vessels | 526, 1.3-fold P <0.05 Included literature | Neutrophil activation |
| Elashoff | Agilent PAXgene | Cath Lab, only stable patients; no known prior CAD/MI/revasc. First cohort (36% diabetic) Second cohort (nondiabetics only) Replication by RT-PCR for 110 genes | 195 198 640 | Obstructive CAD 50% stenosis by clinical read QCA | N =2438 N =5935 655 in both | Neutrophil apoptosis, neutrophil to lymphocyte ratio B, NK-cell activation CD8 T cells in diabetics |
| Sinnaeve | Affymetrix PAXgene | Cath Lab, nondiabetics, no MI within a month for cases but old MI allowed (25% of cases). Controls had 0% stenosis | 222 | Obstructive CAD 50% stenosis CAD index | 160, correlation-based analysis | |
| Joehanes | Affymetrix Exon arrays PAXgene | Framingham, 25% diabetics Controls age and sex matched | 188 | MI and revasc. versus age, sex matched | 269 P <0.01 | NK/CD8 T cell; erythroid |
| Kim | Illumina Bead arrays PAXgene | Cath Lab, 35% diabetics Replication | 175 163 | 4 Groups, no CAD, CAD (50% stenosis, MI, prior MI) | Neutrophil to lymphocyte ratio |
Clinically Validated Gene Expression Tests for Obstructive CAD
Although a number of discovery transcriptional studies have been performed as enumerated above, the only gene-expression-based test that has been clinically validated in multiple studies and is in clinical use appears to be the Corus CAD test. The Corus CAD test algorithm consists of a linear combination of sex-specific age-dependent risk functions and the gene expression levels determined by RT-qPCR of 23 genes from whole blood cell RNA, which are transformed to a 1–40 scale for clinical reporting. Algorithm development focused solely on nondiabetic patients, based on the observation that diabetic and nondiabetic signatures for obstructive CAD were significantly different . A candidate set of 655 genes was identified from the intersection of the two independent microarray experiments and using a variety of statistical techniques, a set of 113 RT-qPCR gene expression assays, comprising obstructive CAD classifiers, surrogates for specific cell-types, and normalization genes, were derived ( Table 6.2 ). These assays were then run on 640 independent nondiabetic patient samples from the PREDICT study ( www.clinicaltrials.gov , NCT 00500617) that comprised the algorithm development data set. Based on the high degree of correlation structure largely driven by cell-type specific gene expression, metagene terms were constructed, as well as sex-specific age-dependent risk functions. A combination of penalized logistic regression methods was then used to construct two overlapping algorithms for men and women containing a total of 23 gene expression values and the sex-specific age-dependent risk functions.
Corus CAD Analytical Validation
The analytic validity of the Corus CAD test has been evaluated in two published studies . First, more than 800 control blood samples were tested to assess intra and interbatch variability and reproducibility across a variety of experimental parameters. In this study, the overall process standard deviation was approximately 1 unit on the transformed 1–40 Corus CAD score scale, representing a small 1.7% change in estimated obstructive CAD likelihood. A second study examined the stability of the process with stored samples and the change in patient scores over time, along with an analysis of an expanded cohort from the PREDICT study, with a focus on non-Caucasian patients. The results showed that there was no significant change in the area under the Receiver-operating characteristics curve (ROC AUC) with an expanded cohort ( N =1502) versus the original validation study ( N =526, AUC=0.70) with samples stored over a 5-year period. In the expanded cohort, a subset analysis of non-Caucasians showed an AUC of 0.72, which was not significantly different from the AUC of 0.70 for the entire cohort.
Corus CAD Clinical Validation
The Corus CAD test performance has been assessed in two independent multicenter validation studies in nondiabetic patients without known CAD, examining two different disease prevalence populations of relevance to clinical decision-making. It should be noted that patients with preexisting CAD, chronic autoimmune or inflammatory disorders, and treatment with steroids or chemotherapy were excluded. The PREDICT trial evaluated test performance in a patient population ( N =526) referred for ICA, the gold standard for obstructive disease evaluation . Disease prevalence was 37%, as measured by core laboratory quantitative coronary angiography, and very similar to that observed by Patel et al. .
The COMPASS study ( www.clinicaltrials.gov , NCT01117506) evaluated test performance in symptomatic patients ( N =431) referred for myocardial perfusion imaging (MPI), the primary gateway to ICA in the United States . The gold standard was either ICA, coronary CT-angiography, or a combination of both, determined in core laboratories, so that all patients independent of their MPI results had gold standard data on their coronary anatomy. Obstructive CAD prevalence was only 15%, lower than seen in PREDICT. Positive MPI scans were seen in 11% of patients, very similar to that seen in a recent study reporting the changes in MPI positivity over the last two decades .
Results of the two Corus CAD validation studies representing 58 centers in the United States were very consistent ( Table 6.3 ). In addition, in the subset of PREDICT patients who had MPI (70%) and in the entire COMPASS study, Corus CAD showed superior diagnostic performance to MPI driven by greater sensitivity and diagnostic accuracy. The test showed superior performance to the Diamond–Forrester classification in COMPASS and significantly added to the classification in PREDICT. In both studies, increasing Corus CAD score was significantly associated with increasing maximum percent stenosis. Finally, clinical follow-up for subsequent revascularization and MACEs was performed in PREDICT for 1-year postindex catheterization and showed a very significant association of Corus CAD score and the composite revascularization and event endpoint , whereas in COMPASS, 6-month follow-up also showed significantly fewer revascularization and events with low (≤15) Corus CAD scores . The test is commercially available in the United States; samples are processed at a College of American Pathologists (CAP) and Clinical Laboratory Improvement Amendments (CLIA) accredited clinical lab, with the scores being resulted within a few days.
