Coronary artery disease (CAD) is estimated to affect >200 million people worldwide (1). The currently accepted main risk factors for CAD remain those of the Framingham set, i.e. male sex, family history of premature CAD, age, smoking, diabetes, hypertension, hypercholesterolemia and HDL-cholesterol (2). Additional markers (e.g. troponin-T, C-reactive protein, and N-terminal pro-B-type natriuretic peptide) have proven useful in patients with acute myocardial infarction (MI), heart failure or myocardial hypertrophy, but have been shown to carry little incremental predictive value for chronic CAD (3). Ideally, new markers should also constitute potential treatment targets, but in recent years no novel molecule was identified in the context of CAD.

Cardiac ischemic events may develop when CAD reaches a critical threshold. However, there are patients with severe CAD who do not develop ischemic events; conversely, individuals with minimal CAD may do so. Similar exceptions to the accepted paradigm include patients with diffuse CAD with a low cardiovascular risk factor (RF) profile and others with multiple RFs who develop only mild or no CAD. While such “outliers” represent only 15–20% of patients experiencing an acute coronary event (4), they nevertheless offer a unique model to search for unknown mechanisms predisposing to CAD beyond traditional RFs; conversely, individuals witout significant CAD but many RFs may have unknown protective factors that impede or delay CAD.

The Coronary Atherosclerosis in outlier subjects: Protective and novel Individual Risk factors Evaluation (CAPIRE) study was designed to explore this opportunity by prospectively segregating RFs and CAD (assessed by coronary computed tomography angiography, CCTA) into two outlier groups of subjects, whose event rates are being followed in time (5). The rapidly evolving technology of proteomics provides a refined tool for developing new diagnostic and therapeutic algorithms (6). High-performance platforms can screen thousands of proteins that may serve as biomarkers, eventually to be included in multi-parametric models of risk assessment, and/or serve as indicators of disease pathways or of therapeutic effectiveness (7).

Here we report the findings obtained by applying large-scale proteomics to the entire cohort of subjects in the CAPIRE study, with a special focus on the two outlier groups, RF⁻/CAD⁺ and RF⁺/CAD.

The demographic, clinical and metabolic characteristics of the study cohort were generally consistent with the predefined grouping (Tables 1, 2). Of note, a higher prevalence of male sex, older age, and higher body mass index (BMI) characterized both CAD⁺ groups as compared to the CAD⁻ groups. Most differences between the 4 groups (except for BMI, waist girth, current blood pressure, HbA_1c, and eGFR) were also present between the RF⁺/CAD⁻ and RF⁻/CAD⁺ outlier groups, reflecting the a priori criteria used to define them. Thus, RF⁻/CAD⁺ outliers had a significantly lower prevalence of hypertension, hypercholesterolemia, diabetes, smoking and family history of ischemic heart disease. High density lipoprotein (HDL)-cholesterol was higher, and hsTnT lower, in the RF⁺/CAD⁻ compared to the RF⁻/CAD⁺ group, as previously described (5).

The association of proteins with the clinical data was explored in a stepwise manner. First, the PLS analysis was restricted to the two outlier groups. Here, 20 proteins were found to significantly differ between them, i.e., they achieved a VIP score ≥3 (Figure 1). Because gender and age differed considerably between these groups (Table 1), the corresponding between-group differences in individual protein levels were adjusted for gender and age in separate ANCOVA models.

Consequently, only four proteins retained statistical significance: myosin regulatory light chain 2 atrial isoform (MYO) and C-C motif chemokine 22 (C-C22) were lower (4530 ± 104 vs. 5169 ± 92, p = 8.5e⁻⁴ and 17271 ± 624 vs. 21784 ± 780, p = 4.4e⁻², respectively), and protein shisa-3 homolog (PS-3) and platelet-activating factor acetyl-hydrolase (PAF-AH) were higher (1345 ± 70 vs. 1043 ± 30, p = 2.1e⁻² and 23065 ± 635 vs. 19484 ± 559, p = 4.4e⁻²), in RF⁻/CAD⁺ as compared to RF⁺/CAD⁻ (Figure 1).

In the next step, bivariate PLS was used on the entire database, with RFs and CAD as the dichotomic response variables. Using again the restrictive criterion of a VIP value ≥3, 13 proteins topped the list (Table 3). Following adjustment for gender, age and multiplicity, four proteins (polymeric immunoglobulin receptor (PIR), neurocan core protein (NCP), vesicular overexpressed-in-cancer pro-survival protein (Ves), and insulin-like growth factor binding protein 3 (IGFBP3) remained significantly associated with the presence of RFs only, 1 [brevican core protein (BCP)] with the presence of CAD only, and 3 proteins [coiled-coil domain-containing protein 126 (C-C126), extracellular superoxide dismutase [Cu-Zn] (SOD), and renin] with the presence of both RFs and CAD. In addition, in this analysis the four proteins previously identified from the direct comparison of the outlier groups retained statistical significance, three of them (PS-3, PAF-AH, and C-C22) in association with RFs only, and one (MYO) in association with CAD alone. The direction and strength of association of each of the 17 proteins in Table 3 were tested by principal component analysis and are shown in Figure 2 as loadings on the two dichotomic factors, RFs and CAD. Thus, by restricting statistical significance to p values <0.001, 12 out of ~5,000 proteins were associated with the 2 pre-set discriminants of the four study groups, namely RFs and CAD, independently of the main anthropometric determinants (gender and age). The strength of this panel of proteins was tested in a logistic model where the response variable was the study group, and the 12 proteins were the predictors. As shown in Figure 3, the AUC_ROC ranged from 0.72 for the RF⁺/CAD⁻ group to 0.81 for the RF⁺/CAD⁺ group, with an overall r value of 0.47 and a p value of 1.0e⁻³⁸.

To quantify the unique association of each protein with individual clinical parameters independently of the other proteins and of gender, age and BMI, we ran multivariate regression models with each the 12 proteins as the dependent factor and all clinical parameters as independent variables in addition to gender, age and BMI. As shown in Table 4, serum HDL-cholesterol was a strong positive correlate of 6 proteins (C-C126, NCP, Ves, BCP, IGFBP3, and SOD, which were all interrelated), while renin was strongly associated with higher HbA_1c and lower eGFR. Of note, PAF-AH was strongly associated with serum non-HDL-cholesterol.

The next step was to test the independent association of each of the 12 proteins with the presence of RFs and CAD (Supplementary Table 1). A higher BMI was associated with lower levels of the proteins that were directly related to HDL-cholesterol. Gender also had a strong impact on most of the 12 proteins, as did smoking. Plasma renin was the strongest correlate of CAD. MYO and PS-3 (involved in vascular repair or adverse remodeling) retained independence from BMI, cholesterolemia, hsCRP or HbA_1c levels (Table 4, Contribution to the field). By way of example, MYO was reduced not only in the group with CAD and low RFs (RF⁻/CAD⁺) but also in the whole cohort in association with a Leaman Score of ≥5. Conversely, PS-3 was not only reduced in the group with high RFs and no CAD (RF⁺/CAD⁻) but also in association with one of the major RFs, i.e., smoking (Figure 4).

Finally, each of the 12 proteins was tested against the percentage of patients on treatment with the main classes of drugs (while adjusting for gender, age, and BMI). As shown in Supplementary Table 2, drug therapies generally had weak or null associations with protein levels except for some expected link (e.g., higher renin levels in patients using ACEi or ARB). For completeness, Supplementary Tables 3, 4 list the top proteins associated with the extreme groups (CAD⁻/RF⁻ and CAD⁺/RF⁺, respectively).

In the present study we searched for new potential biomarkers by applying high performance proteomics in four different phenotypes, defined according to the presence of traditional cardiovascular RFs and CCTA-assessed CAD. Additionally, linked the resulting “hit” proteins with clinical and biochemical characteristics as well as other risk markers.

Several of the proteins identified in this study have been investigated in human disease and their role in the atherosclerotic process has been assessed; others have only been studied in animal models, with no clear disease correlates. From a general point of view, the twelve hit proteins take part in immune system responses, fibrosis, oxidation, or proliferating/healing signaling pathways, reflecting the complexity of the human proteome applied to CAD.

After adjustment for gender and age, two proteins remained associated with the absence of CCTA-proven CAD despite the presence of multiple RFs, i.e., myosin regulatory light chain 2 atrial isoform (MYO or MLC-2a) and C-C motif chemokine 22 (C-C22). Therefore, for these two proteins a protective role against CAD may be hypothesized. Sarcomeric proteins have recently drawn considerable attention as therapeutic targets to improve myocardial function. The recently published results of the Global Approach to Lowering Adverse Cardiac Outcomes through Improving Contractility in Heart Failure (GALACTIC-HF) trial showed that omecamtiv mecarbil, a cardiac myosin activator, reduced the incidence of heart failure hospitalization and cardiovascular death in patients with systolic heart failure compared with placebo (16). MYO is a sarcomeric protein that enhances migration, contractility and inotropism in vascular smooth muscle cells (VSMC) in a calcium-independent fashion (17). Its reduction in the intima of coronary plaques has been shown to wound the healing potential of VSMC (17, 18). In our dataset, the biological strength of this association is confirmed by the reduced MYC levels in individuals with a high Leaman score, i.e., a high burden of atherosclerotic involvement in their coronaries. C-C22 is an immunosuppressant chemotactic peptide, predominantly investigated in cancer biology, which also seems to have a protective role against atherogenesis (19).

In the outlier counterpart (presence of CAD with low levels of RFs), protein Shisa 3 homolog (PS-3) and platelet-activating factor acetyl hydrolase (PAF-AH) were identified as markers of CAD. PS-3 is a scaffold protein mediating the Wnt/beta-catenin signaling pathway (20). Recent data identify PS-3 involvement in pathological adverse vascular remodeling (21), suggesting a detrimental role of PS-3 in CAD development. A large number of studies have identified PAF-AH (also known as lipoprotein-associated phospholipase A2, Lp-PLA₂) as a biomarker of vascular inflammation and atherosclerosis (22). PAF-AH serves the biologic function of degrading products within oxidized LDL, causing the instability of the atherosclerotic plaque; accordingly, in the current data, PAF-AH was strongly associated with non-HDL-cholesterol (23). Numerous studies have shown its correlation with long-term cardiovascular events in patients with both stable and unstable CAD (22, 24). However, two large randomized, placebo-controlled trials of the PAF-AH inhibitor, darapladib, failed to prove a cardioprotective effect (25, 26), therefore its value as a therapeutic target is uncertain.

Among the four proteins associated with RF⁺/CAD⁻, insulin-like growth factor binding protein 3 (IGFBP3) is the best known (27). This protein binds ~90% of insulin-like growth factors and is upregulated by hypoxia; recently, its suppression by salvianolic acid B has been suggested to improve myocardial function in diabetes-associated cardiac fibrosis (28). Moreover, growth hormone, the stimulus for IGF production and the key regulator of IGFBP3, has been shown to be associated with coronary atherosclerosis, independently of other RFs. Hence, it is not unlikely that IGFBP3 clusters with other cardiovascular RFs, although its role remains disputed (29).

Finally, two proteins associated with the presence of both RFs and CAD have a convergent biologic role. The first one, extracellular superoxide dismutase (Cu-Zn SOD), is a major antioxidant enzyme whose alterations possibly reflect enhanced oxidative stress; moreover, impaired Cu-Zn SOD expression or catalytic activity has been identified in several physiological situations such as aging and age-associated diseases (30). The second one is renin, which was the strongest correlate of CAD in our cohort and whose link with cardiovascular events has a firm pathological basis and is supported by growing clinical evidence: renin is the activator of the renin-angiotensin-aldosterone system, whose pharmacological inhibition reduces morbidity and mortality of patients with CAD (31), and high levels of renin are associated with atherosclerosis, hypertensive cardiomyopathy, and impaired left ventricular function (32). The observational Multiethnic Study of Atherosclerosis reported an independent association between plasma renin activity and cardiovascular outcomes in patients without previous CV events who had angiographically proven CAD or subclinical cardiovascular disease (33).

Our qualitative principal component analysis indicated a clustering of seven of the twelve hit proteins with HDL-cholesterol. It is well established that HDL-cholesterol levels are strongly and inversely correlated with the risk of CAD; however, interventional studies raising HDL levels did not reduce cardiovascular events (34–37). In fact, it is the quality of HDL-cholesterol particles that seems to be relevant, because it affects their functionality not only in the atherosclerotic process but also in immunomodulation and inflammation (38). In line with the pathophysiological relevance of lipid metabolism, several classes of apolipoproteins have been studied as cardiovascular risk factors, and their pathophysiological role in CAD is firmly established (38). Consistently, in the current study the PAF-AH signal was significantly higher in the RF⁻/CAD⁺ group as compared to RF⁺/CAD⁻, and was reciprocally related to HDL-cholesterol. A weak negative association with HDL-cholesterol was also seen for PS-3, the other protein that was higher in the RF⁻/CAD⁺ group and that remains inversely associated with the presence of other RFs. Of all hits, reduced MYO and reduced PS-3 (involved in vascular repair and adverse remodeling) retained their strong associations with CAD⁺/RF⁻ and CAD⁻/RF⁺ phenotypes, respectively, independently of body mass, cholesterolemia, C-reactive protein or glycated hemoglobin levels.

Thanks to a complete segregation of CAD phenotypes and a reasonably wide separation of risk factor distribution, in the present CAPIRE study proteomic screening per se was able to discriminate CAD+/RF– from CAD–/RF+. Correlation analyses linking physiology to proteins, clinical risk factors and biochemical risk markers led to the extraction of MYO and PS-3 as two discriminators, independent of established risk factors or biomarkers. Of note, each of them points to opposing vascular processes (healing vs. adverse remodeling), potentially protecting or predisposing to CAD, which warrants further mechanistic investigation as well as validation in other cohorts.

The data underlying this article will be shared on reasonable request to the corresponding author.

This study, involving human participants, was reviewed and approved by Istituto Svizzero per gli Agenti Terapeutici, Comitato Etico Centro Cardiologico Monzino-IEO, Comitato Etico Aziendale dell'Azienda Ospedaliera-Universitaria, Comitato Etico Parma, Comitato Etico Modena, Comitato per la Sperimentazione Clinica dei Medicinali, AUSL 12 di Viareggio – Ufficio Ricerca, Segreteria Locale ASUR del CERM Marche. The patients provided their written informed consent to participate in this study.

EF wrote the manuscript, researched data, and is the guarantor taking responsibility for the contents of the article. MLM contributed to drafting the manuscript and carried out statistical analyses. GF contributed to drafting the manuscript and managed revision and editing. RL supervised the routine laboratory analyses. SW was responsible for the proteomics. FA, DA, RL, MM, SW, AMas, and AMag critically reviewed and edited the manuscript and contributed significantly to discussion. All authors contributed to the article and approved the submitted version.

This work was supported by Heart Care Foundation of the Italian Association of Hospital Cardiologists, Florence, Italy.

FA reports consultancy/speaker fees from Amgen, Bayer, BMS/Pfizer, and Daiichi Sankyo outside the submitted work. SW is an employee and shareholder of SomaLogic Inc., Boulder, Colorado, USA. AMag reports personal fees from Bayer, Fresenius, and Novartis outside the submitted work. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.