Serum protein profiles predict coronary artery disease in symptomatic patients referred for coronary angiography

Coronary heart disease is the most prevalent chronic disease and the leading cause of death in the US, with more than half a million newly diagnosed coronary artery disease (CAD) patients annually [1, 2]. Cardiac catheterization and coronary angiography are often necessary for definitive evaluation of coronary artery anatomy, the presence of coronary atherosclerosis and to determine the need for interventional therapy. Despite the high prevalence of CAD, approximately half of patients undergoing invasive cardiac catheterization either have no significant coronary lesions or do not require any mechanical or surgical form of revascularization [3‐5]. Thus, the procedure could be eliminated in many cases if alternative, non-invasive tools were available to assess the presence or absence of significant CAD and confirm the need for angiography.

Clinical assessment of CAD represents a significant medical and economic challenge comprising more than a million coronary angiograms annually in the US alone with demographics of aging and obesity forecasting growing demand [2‐5]. The risk and expense of cardiac catheterization (ionizing radiation, contrast media, morbidity) and the large number of patients with normal coronary arteries or 'non-significant' CAD undergoing invasive angiography warrant development of alternative tests for CAD without cardiac catheterization [5]. While progress has been made using non-invasive computed tomography (CT) particularly for its negative predictive value, CT incorporates significant exposure to ionizing radiation with considerably lower resolution than catheterization-based angiography [6].

Efforts to identify circulating biomarkers for CAD have shown promise by interrogating transcriptional profiles of patient blood cells and plasma for unique mRNA and microRNA signatures [7, 8]. Since extracellular RNA undergoes rapid degradation, prospective mRNA signatures were derived predominantly from nucleated blood cells while the miRNAs identified in plasma were likely protected in circulating vesicles or bound to protective protein complexes [9]. Consequently, the utility of RNA as an indicator of CAD is constrained by its selective cell source in the bloodstream, the friability of the ribonucleotide targets and the arduous process of RNA capture, purification, amplification and analysis. In contrast, circulating proteins are more stable than RNA in blood and serum with several individual markers identified previously as prospective biomarkers for the presence of atherosclerosis, myocardial infarction, heart failure, or markers of pathways involved in these cardiac conditions such as inflammation, thrombosis, plaque stability, and oxidative stress, for example, troponin C, pro-brain natriuretic peptide (BNP) and C-reactive protein (CRP) [10, 11]. However, circulating biomarkers have proven to be of limited value in clinical tests to diagnose coronary artery disease antecedent to a cardiac event, primarily because most studies have focused on single or at most a few markers to make the diagnosis [12]. The difficulty in identifying predictive factors of CAD in blood or serum is compounded by the multifactorial etiology of coronary artery disease which makes early diagnosis by a single, endpoint marker unlikely prior to activation of a common ischemic pathway or until significant myocardial compromise has occurred.

The hypothesis underlying the current study was that coronary artery disease status can be assessed via individual and/or combinatorial protein changes in serum that assess multiple pathways of atherosclerosis as a low-risk, non-invasive approach for screening of symptomatic patients, that is, patients referred for cardiac catheterization. The study targeted patients who were referred for a clinically-indicated cardiac catheterization from either the emergency room or outpatient cardiac clinic in a major metropolitan center who presented with symptoms of heart disease. All patients had blood drawn prior to coronary angiography and revascularization. By analyzing a compendium of 41 circulating protein targets associated with atherogenesis, inflammation, thrombosis, and plaque vulnerability, we discovered 12 diverse proteins expressed across a broad dynamic range that were significantly different concomitant with the need for these patients to undergo therapeutic intervention including stent placement, angioplasty, or coronary artery bypass graft surgery (CABG). We also tested multiplex biomarker signatures for the potential to discriminate patients lacking significant coronary artery disease from patients with CAD requiring corrective interventional therapy. In particular, the ability to rapidly and decisively rule out clinically significant coronary artery disease using a potentially low cost, low risk blood test in even a small percentage of patients with normal coronary arteries could be highly beneficial.

Diagnostic coronary angiography revealed that 209 of the patients in this study exhibited significant coronary artery disease requiring therapeutic intervention while 150 patients did not exhibit clinically significant coronary artery disease despite symptoms or other findings that led to referral for cardiac catheterization. These two distinct outcome groups were otherwise identical upon admission regarding clinical symptoms and physical characteristics including gender, diabetic status, smoking history, body surface area, basal metabolic rates, cholesterol, LDL and creatinine values (see Table 1). Among continuous variables, there were small albeit significant differences in age, HDL levels and ejection fraction between groups; but the differences were of minimal diagnostic value and all patients required coronary angiography. Regarding categorical variables, there were no significant differences in gender or diabetes between the two groups; however, the number of patients with hypertension was significantly higher in the CAD group.

All serum samples were collected, processed, stored and analyzed in an identical manner to limit the effect of preanalytic variability including differential protein degradation among specimens. Significant differences were detected in 12 serum proteins (Q value = 0.01; P < 0.01) between patients diagnosed as having CAD requiring intervention and those with non-significant CAD after diagnostic coronary angiography. The differences detected in the stage 1 study (n = 239) were reinforced by the additional samples from the validation study (n = 120) (see Table 2). APO-A1 and APO-B100 were among the highest expressed proteins overall averaging approximately 300 μg/ml of serum (Figure 1). APO-A1 fell significantly in patients with significant CAD versus non-significant CAD while APO-B100 was significantly increased. Within the same concentration range, fibrinogen was present at levels typically exceeding 1 μg/ml with values fivefold higher in patients with significant CAD (Figure 1). At serum concentrations from 10 ng to 1 μg/ml serum, five proteins were significantly higher in CAD patients. Specifically, CRP, VCAM-1, MPO, resistin and osteopontin were 1.2 to 3.1-fold higher than patients with non-significant CAD (Figure 2). Four analytes, IL-6, IL-1β, IL-10 and NT-pBNP were significantly higher in the CAD group among analytes detected in a range from 1 pg/ml to 1 ng/ml (Figure 3). There were no significant statistical correlations between any of these 12 analytes and age, ejection fraction or hypertension status which were significantly but incrementally different between the patient outcome groups. No other analytes among the 41 interrogated targets were significantly altered between the 2 groups of patient samples using either bead-based or planar platforms.

We identified 14 multiplex signatures of 2 to 5 proteins each with the highest acuity to detect patients without significant CAD (22.6% to 58.4% SP) while detecting 95% of the significant CAD group (95% SN) in the stage 1 study (see Additional file 2 Table S1). A total of 11 distinct proteins were shared among the 14 signatures with osteopontin (14 of 14), and resistin (10 of 14) most frequently represented. There was a trend for protein signatures with increased numbers of analytes to detect more normal patients at a fixed sensitivity for CAD patients (95%) (two proteins = 39.3% ± 0.3% vs five proteins = 50.0% ± 0.01% of normal patients). However, a performance plateau was reached at five biomarkers based on crossvalidated classifier performance and the frequency of appearance of 'artificial' markers in test signatures exceeding five proteins. Receiver operating characteristics analysis indicated that these signatures were effective in discerning patients without significant CAD. The area under the curve (AUC) for the top signatures ranged from a low of 0.839 ± 0.028 (mean ± SD) for a two-protein signature (OPN, resistin) to a maximum AUC of 0.845 using three biomarkers (OPN, resistin, APO-B100) (Figure 4). These ROC curves were compared to those generated by the Bayesian compound covariate predictor algorithm for the same data set. The area under the curve using the scoring function algorithm exceeded that obtained by the Bayesian predictor in every case. A clinical validation test of 120 additional serum samples (49 normal, 71 patients requiring intervention) was performed to test the performance of the scoring function algorithm. In two separate studies, the best performing multiplex signatures contained five proteins (OPN, resistin, MMP7, IFNγ with either CRP or ACRP-30) and were able to correctly classify 88% and 92% of patients requiring percutaneous intervention while delineating 33% and 36% of the patients with normal coronary arteriograms.

Proteins were selected for evaluation in this study based on their roles in mechanisms underlying atherogenesis, atherosclerosis and plaque instability including vascular inflammation, thrombosis, aberrant lipid regulation, metabolism hormones, and vascular smooth muscle and extracellular matrix (ECM) remodeling [14]. The 41 preliminary targets we interrogated were restricted by availability of monoclonal antibody pairs optimized for use in the commercial assay platforms. IL-1β, IL-6, IL-10 and VCAM-1, were significantly elevated in patients with CAD in the present study consistent with an injury-induced, inflammatory response [15, 16]. Elevated IL-1β and IL-6 have been associated previously with acute phase protein induction and may explain the concomitant significant increases in fibrinogen and CRP concentration we detected. CRP has long been proposed as a surrogate marker for inflammatory mediators in predicting coronary events while NT-pBNP has been used as an indicator of left ventricular dysfunction in CAD patient cohorts comparable to this study [11, 17, 18]. Both analytes were significantly elevated in the present study among patients requiring therapeutic intervention and CRP was among the best single molecule classifiers delineating 19% of normal samples while detecting 95% of the patients with significant CAD.

Significant reciprocal changes were detected in APO-A1 and APO-B100 in CAD patients consistent with reports defining aberrant lipid transport and accumulation as contributory to atherosclerosis [19]. Mutations in the APO-B100 gene cause autosomal dominant, hereditary familial hypercholesterolemia and premature coronary artery disease due to defective ligand binding [19, 20]. At the same time, elevated APO-A1 is associated with a cardioprotective effect and enhancement of APO-A1 expression has been proposed as a therapeutic strategy to inhibit atheroma formation [19, 21]. The increased APO-B100 and decreased APO-A1 levels in our patients requiring PCI versus normal controls support these previous findings. Myeloperoxidase was also significantly increased in CAD patients associated with its role as a catalyst for lipid peroxidation at inflammation sites and as a marker of plaque instability [22, 23]. Resistin levels were elevated in the PCI patients indicative of 1) metabolic shifts in lipid utilization and adipogenesis and/or 2) an inflammatory response with resistin secreted from macrophages concomitant with the release of proinflammatory cytokines [24].

Many targets traditionally associated with vascular smooth muscle and ECM remodeling were not significantly altered among these patient groups including matrix metalloproteinases 1, 2, 3, 7, 9 and tissue inhibitors of metalloproteinases 1, 2, 3 and 4. Only osteopontin, which acts as a negative regulator of calcification in bone remodeling, was elevated within this category with the rejoinder that OPN also may act as a chemokine in the cell mediated type 1 immune response associated with inflammatory cell accumulation rather than as a substrate for cell adhesion [25]. Thus, the proteins that demarcated our patient outcome groups were predominantly associated with processes of inflammation and lipid regulation rather than cellular aggregation and ECM remodeling. However, we recognize that the domain of proteins susceptible to interrogation in this study was limited to analytes for which high affinity antibody pairs precisely characterized to two different epitopes were available. The involvement of additional proteins and pathways associated with CAD will likely be reinforced and/or revealed as the inventory of immunoassays becomes more comprehensive.

Our data indicate multiplex proteomics analyses using monoclonal antibodies provide relevant information regarding circulating serum analyte concentrations when assayed at a dilution that allows direct comparison to parallel recombinant calibration standards. Advantages include small serum volumes (< 100 μl) collected by standard clinical protocols, rapid turnaround times (minutes to hours), high sensitivity (pg) and a broad dynamic range (8 logs). Disadvantages include high assay cost, limited target availability and poor concurrence of concentration measurements across dilutions and commercial platforms associated with variations in antibodies, buffers, diluents and capture structures. In the present study, 15 targets were tested at identical serum dilutions using bead-based (Luminex) and planar (Aushon) technologies in 56 identical samples, albeit with different aliquots and in serial studies. A total of 12 assays concurred in detection of statistically significant differences between the 2 patient outcome groups. These results suggest that multiplex immunochemical assays of serum may provide information of diagnostic relevance but that protocols and reagents must be optimized and standardized prior to routine clinical application.

The results of this study were somewhat surprising both for discovery of unique proteins as discriminants of CAD and for the absence of statistically significant differences in many targets with established roles in atherosclerosis. For example, osteopontin has been only indirectly associated with atherosclerosis yet exhibited the greatest statistical difference between patient groups (P = 1.75 × 10^-12). Osteopontin was first identified as a sialoprotein from mineralized bone matrix and only recently was associated with calcification of plaques in cardiac valves and vessels [25‐27]. Similarly, resistin has been linked only indirectly to CAD through a role in metabolic homeostasis and insulin sensitivity [28]. On the other hand, multiple growth factors (VEGF, leptin, ghrelin), lipoproteins (APO-A2, E, serum amyloid A: SAA), cell adhesion molecules (thrombospondin, PECAM-1, ICAM-1, selectins E, L, P) and MMP and TIMP targets associated with ECM remodeling exhibited no statistically significant differences. There are several potential explanations for the latter findings: (1) a rigorous statistical standard was utilized to avoid multiple testing errors and while MMP1, MMP7, ACRP-30, and leptin were borderline for statistical significance (P = 0.015, 0.045, 0.027, 0.027 respectively) they failed to reach the Q = 0.01 level established for significance with adjusted P values ≤ 0.01 in this study; (2) serum may not be an effective transducer of deleterious protein changes participating in structural rearrangements within the coronary vascular anatomy and the extracellular matrix; and (3) the patients comprised a diverse range of coronary obstruction and plaque vulnerability since they were selected for symptoms upon emergent presentation requiring diagnostic coronary angiography without the occurrence of a clinically obvious myocardial infarct or an 'event'. A subset of patients selected for advanced disease might reveal additional protein changes but stray from the intended focus of this study.

A scoring function algorithm was developed, tested and validated for the ability to classify patients symptomatic for heart disease consistent with the outcome of coronary angiography studies and need for interventional therapy. We minimized selection bias by testing a hypothesis driven biomarker panel and avoided overfitting by performing cross validation and follow-up testing utilizing additional serum samples from the cohort. The algorithm was designed to be 'tuned' to increase the sensitivity for capture of patients who required coronary revascularization at the expense of detecting fewer patients who did not require coronary revascularization. All serum signatures with highest classification strength from the training trial (239 samples) included osteopontin and signatures containing 4 or 5 proteins performed best during both training and validation phases. The most efficacious protein signature in validation studies comprised OPN, resistin, MMP7 and IFNγ as a four-marker panel while the addition of either CRP or ACRP-30 yielded comparable results in five protein signatures.

Further validation of the diagnostic accuracy of this approach will require extensive testing in greater numbers of patients at multiple locations as well as a prognostic cohort. It is possible that inclusion of clinical variables and risk factors in the biomarker algorithm or using the algorithm as part of a clinical scoring system will enhance both the fidelity and the efficacy of this approach for diagnostic purposes [29, 30]. In that context, we calculated 10-year Framingham Coronary Heart Disease (CHD) Risk Scores for patients where all clinical variables (gender, age, total cholesterol, HDL, systolic blood pressure, smoking and diabetes status, use of antihypertensive medication) were acquired prior to coronary angiography [31]. This represented 91 patients who subsequently required therapeutic revascularization (CAD: CHD Score = 14.9 ± 8.5) versus 63 patients who were determined to be free of significant coronary artery disease (no CAD: CHD score = 10.2 ± 6.7). The Framingham CHD scores were statistically different between groups (P < 0.001, unpaired Student's t test) but they classified only 16% of the subjects without significant CAD (10 of 63) at a 95% sensitivity for patients with CAD. In contrast, our algorithm incorporating serum values for OPN, RES, CRP, MMP7 and IFNγ identified 63% of the subjects without significant CAD (40 of 63) at 95% sensitivity for patients with CAD. Thus, our multiplex serum protein classifier correctly identified four times as many patients as the Framingham index. The strength of adding clinical variables to our scoring function remains to be determined, but the ability to exempt significant numbers of patients with normal coronary arteries or non-significant CAD from cardiac catheterization with a blood test represents a major economic and health benefit given the growing epidemic of CAD in the US and abroad.

The results of the present study indicate that a serum, multiplex biomarker assay may provide a clinically useful tool in combination with other standardized clinical tests to facilitate the decision-making process for performing cardiac catheterization in symptomatic patients. The development of highly sensitive monoclonal antibodies to additional pertinent targets along with the formulation of novel predictive algorithms will likely improve the efficacy of this approach. The long-term potential benefits include reduced patient exposure to ionizing radiation and minimization of the rapidly escalating healthcare costs associated with the use of invasive angiography to rule out coronary artery disease.