Introduction

Coronary heart disease (CHD) remains a leading cause of mortality and morbidity worldwide, claiming approximately 7 million deaths each year1, and Acute coronary syndrome (ACS) reflects the most urgent and severest clinical condition of CHD2. The pathophysiology of ACS remains to be fully understood but chronic inflammation has been widely considered as a potential contributor3. Several inflammatory biomarkers have been reported to be associated with ACS risk, and most of them were selected based on their presumed pathophysiological roles in ACS. However, a large degree of uncertainty remains regarding early detection and risk discrimination of ACS, and new biomarkers for better ACS risk prediction are urgently needed.

Cytokines are pleiotropic proteins mainly released from immune cells, which can act in concert with specific receptors or inhibitors to regulate inflammation4,5. Previous studies6,7 have identified several cytokines like interleukin-6 and C-reactive protein (CRP) as potential biomarkers for atherosclerosis, myocardial infarction, or main biological pathways involved in those cardiovascular conditions. However, few studies have used high-throughput methods to screen for a large number of cytokines as potential biomarkers for ACS.

In this multi-stage study, using a protein microarray which facilitate standardized comparison of the observed associations across cytokines, we aimed to identify novel cytokines associated with ACS, and to examine the predictive value of identified cytokines beyond traditional ACS risk factors.

Results

Basic characteristics of the study population

Table 1 showed the demographic characteristics of ACS patients and controls in each study set. Compared with controls, ACS patients were more likely to have hypertension, diabetes, and to take anti-hypertensive and lipid-lowing medication across all study sets (all P < 0.05).

Table 1 Demographic and Clinical Characteristics of the Study Populations.
Full size table

Unexpectedly, we also observed higher LDL levels in controls in comparison with cases in discovery set and validation set 1 (P < 0.05), probably due to higher prevalence of statin use in the cases. In the nested case-control study, compared with controls, ACS cases had slightly higher glucose and lower high density lipoprotein levels at baseline (P < 0.05).

Association between plasma cytokines and ACS risk

In the discovery stage, we identified 15 different expression cytokines (all q-value < 0.05 and above the detection limit of protein chip) in ACS cases as compared with healthy controls, among which nine cytokines were up-regulated and six cytokines were down-regulated, with MSP as the most regulated cytokine (fold change = 4.65; Table S1 and Fig. S1). Five cytokines, namely osteopontin, CCL23, MSP, BDNF and CRP, met the selection criteria listed in the Material and Method section were selected for next-stage validation (Fig. S1).

In the validation stage, we pooled results from the two case-control studies as similar associations were observed between selected cytokines and ACS in both studies (Table 2). After adjustment for potential covariates, the pooled ORs (95% CI) of ACS for each SD increase of log-transformed osteopontin, CCL23, BDNF, CRP and MSP were 4.64 (3.16–6.79), 3.60 (2.45–5.28), 0.66 (0.49–0.90), 2.67 (1.88–3.79), and 1.33 (0.92–1.91), respectively.

Table 2 Adjusted Odds Ratio for Risk of ACS According to Five Selected Plasma Cytokines in Two Case-control Validation Sets.
Full size table

We further investigated whether osteopontin, CCL23, BDNF and CRP were associated with incident ACS in a nested case-control study after a median follow- up of 1.6 years. With adjustment for potential confounding, we observed positive associations of osteopontin and CRP with incident ACS (OR 1.29; 95% CI 1.06–1.57 for osteopontin; OR 1.30; 95% CI 1.02–1.66 for CRP; Table 3). Osteopontin showed a moderate positive correlation with CRP in health controls across all datasets (all P < 0.05; Fig. S2). Therefore, we conducted a sensitivity analysis to include both osteopontin and CRP in a single regression model, and the results did not materially change (OR 1.35; 95% CI 1.02–1.78 for osteopontin; OR 1.42; 95% CI 1.06–1.91 for CRP).

Table 3 Adjusted Odds Ratio for Risk of ACS According to Four Replicated Plasma Cytokines in Nested Case-control Study.
Full size table

After stratification for onset time windows, we observed a lower association for osteopontin in ≥1 year time period compared with <0.5 year time period and 0.5–1 year time period (P > 0.05; Table S3). In stratification analysis, we did not observe evidence for effect modification by baseline covariates (all P > 0.05 for interaction; Table S4).

We observed elevated levels of osteopontin and CRP in 82 ACS patients with different ACS subtypes, stenotic vessels and onset time windows after ACS onset (Table S5). Association of osteopontin and CRP with severity and onset time window of ACS.

In the nested case-control study, we further identified the association of osteopontin with severity and onset time window of ACS. Among ACS cases, higher plasma osteopontin levels were observed with increasing number of stenotic vessels: the median (25th–75th percentile) levels of plasma osteopontin were 47.12 (37.62–55.09) ng/ml in 160 patients with 1-vessel disease, 59.66 (43.78–71.76) ng/ml in 85 patients with 2-vessels disease, and 79.08 (69.51–98.99) ng/ml in 73 patients with 3-vessels disease, respectively (P < 0.001, Fig. 1A). There was a positive correlation between plasma osteopontin and Gensini score among ACS patients (r = 0.42; P < 0.001; Fig. S3). We also observed a stepwise increase in plasma osteopontin levels with increasing tertile of Gensini score: the median (25th–75th percentile) levels of plasma osteopontin were 46.97 (36.06–55.45), 56.16 (45.57–75.73) and 65.43 (50.13–91.83) ng/ml, given Gensini scores of ≤11, 12–31, and >31, respectively; Fig. 1B). Similar trends were observed in ACS patients with shorter onset time windows such that plasma osteopontin were 61.42 (46.83–81.74), 54.27 (43.65–67.52) and 51.87 (41.16–65.97) ng/ml in ACS patients who had their blood collection <0.5 year (49 patients), 0.5–1 years (124 patients), and ≥1 years (145 patients) before ACS onset, respectively (Fig. 1C). Moreover, higher osteopontin levels were observed in patients with ST-segment elevation myocardial infarction (n = 11) compared with patients with UAP (n = 291) (P < 0.05; Fig. S4). However, no similar trends were observed for CRP (Figs 1D–F and S4).

Figure 1
figure 1

Association of Plasma Osteopontin and CRP Levels with ACS Severity and Onset Time Windows among ACS Cases in the Nested Case-control Study. Higher levels of circulating osteopontin were correlated with higher severity of ACS (A and B), and earlier ACS onset time (from the time of measurement to the ACS onset; C). *represents P < 0.05 and **represents P < 0.001.

Full size image

Risk discrimination and reclassification

Figure 2 summarized the results of C-index, NRI, and IDI analysis in the nested case-control study. The C-index for incident ACS at 2 years of follow-up was increased from 0.69 to 0.73 and to 0.71, respectively, with the addition of osteopontin and CRP. Moreover, the continuous NRI and IDI metrics were also modestly improved with the addition of osteopontin and CRP separately. However, the largest reclassification was observed for the combination of osteopontin and CRP into the model with an improved area under the curve from 0.69 to 0.74, an improved NRI of 26.7%, and an IDI of 0.034 (all P < 0.05).

Figure 2
figure 2

ROC Curve, NRI, IDI Evaluating ACS Cases and Controls in the Nested Case-control Study. The area under the curve (AUC) is shown with its 95% CIs. Traditional risk factors including age, sex, BMI, smoking status, total cholesterol, high-density lipoprotein cholesterol, systolic pressure, status of diabetes and hypertension.

Full size image

Discussion

Through a multi-stage association study, we first identified the prospective associations of osteopontin and CRP with incident ACS. To our best knowledge, this is the first multi-stage study to identify osteopontin as a biomarker for risk and severity of ACS.

In the case-control studies, we identified BDNF and CCL23 in association with ACS. Consistent with our findings, Kaess et al.8 found an inverse association of BDNF and CVD, supporting the potential role of BDNF in CVD pathogenesis. CCL23 was also reported to participate in inflammatory responses and tube formation, both of which play critical roles in the progression of ACS9,10. Our failure to replicate these findings may be due to the elderly population and, relatively modest sample size of the validation study or changes of cytokines levels during disease progression. In addition, we further confirmed the association of higher ACS risk with elevated levels of CRP. In line with our results, Kaptoge et al.6 found a 37% higher CHD risk per 1-SD elevated log-transferred CRP levels in an updated meta-analysis. In another study from Kaptoge et al.11, the addition of CRP to the Framingham risk score only increased the C-index by 0.0039, and yielded a NRI of 1.52%, which is similar with our finding that CRP might be a biomarker and only added limited predictive value beyond established risk factors for ACS.

Osteopontin is a multifunctional protein which was thought to play a critical role in atherosclerosis, and there is convincing evidence linking osteopontin to the onset of ACS. Osteopontin is abundantly present in atherosclerotic plaques12, it was also reported to interact with integrins13 to participate in numerous physiological and pathological events including macrophage chemotaxis, inflammation, and cell survival14. Previous studies have found that osteopontin could increase endothelial cell migration via αvβ3 ligand, thus increasing the risk of atherosclerosis15. Additionally, osteopontin was considered to be a macrophage-chemotactic stimulant and participated in the recruitment of monocytes-macrophages16,17. The potential role of osteopontin in promoting retention of macrophages at sites of chronic inflammation16 indicated a possible mechanism linking to ACS onset and progression. Besides, osteopontin was found to be associated with accumulation of calcium in tissues of CHD patients and may, therefore, serve as a surrogate biological marker of coronary arteries calcification18. Several case-control studies18,19,20,21,22,23 have examined the associations between osteopontin and CHD, however, with inconsistent findings. Ohmori et al.19 and Abdel-Azeez et al.20 found the association of osteopontin with risk and severity of coronary artery disease. Similar to our findings, compared with healthy controls, Tousoulis et al.21 found higher osteopontin levels among patients with 3-vessels CHD. Higher osteopontin levels were also observed among CHD patients complicated with diabetes22. In a recent study, Mohamadpour et al.23 found an association between osteopontin and CHD, but failed to observe differences in osteopontin levels among CHD subgroups with different narrowed vessels. This discrepancy may be attributed to the small, selected groups of CHD patients recruited after disease onset. However, the association we found in the nested case-control study was much weaker in comparison with that in case-control studies, indicating that osteopontin levels rapidly increased after ACS onset. The dynamic changes before and after ACS onset were further confirmed by the measurement of osteopontin levels in the longitudinal study. Despite an independent association, we found in the present study that osteopontin was modestly correlated with CRP levels, suggesting that osteopontin, as an important inflammatory cytokine, may activate the low-intensive inflammation associated with CRP elevation and other conventional risk factors for ACS, such as hypertension and obesity18,24. Although our finding does not establish causality, the comprehensive evaluation of the association with ACS could be useful, given the emerging literature on cytokines as potential targets for drug development25.

The finding of osteopontin alone or on top of CRP added to the predictive value for ACS is speculative but attractive. Our study suggested that osteopontin might reflect certain stages of ACS, thus being a useful biomarker in clinically discrimination of the high risk population for ACS. Nevertheless, since this was an observation study, we cannot quantify the clinical benefits associated with the improvement in early diagnostic accuracy, intervention study was still warranted to provide this information.

Our study has several strengths. First, we used the protein microarray to measure 280 cytokines simultaneously in the discovery stage while most previous studies examined only a few selected cytokines. This method allowed direct comparison of circulating plasma cytokines levels, which broadened our abilities to screen cytokines as potential biomarkers for ACS. Moreover, the validation studies were conducted in three independent populations, therefore minimizing the chance of false positive findings. Third, we included a nested case-control study to prospective investigate the associations of cytokines with incident ACS. Last, we observed the association of osteopontin with both early risk and severity of ACS. Risk prediction performance measures further confirmed that adding osteopontin in combination with CRP modestly improved the ability to predict ACS risk.

Some limitations of our study merit consideration. First, our nested case-control participants were only followed up for a median of 1.6 years. Despite that, the rapid increase in the osteopontin levels related to the severity and onset time of ACS has potential implications in clinical practice, suggesting that close monitoring of osteopontin levels in high risk individuals may help clinicians make decisions to reduce disease risk and prevent disease onset. Second, the prediction calculation may be overestimated because plasma cytokines were measured in the nested case-control study instead of the entire cohort. Therefore, our findings could only be interpreted as potential biomarkers, large-scale prospective studies are still warranted. Third, evaluation of ACS severity was based on number of stenosis vessels or segments and the degree of luminal narrowing, further studies were expected to validate the association using more precise scores such as SYNTAX score or calcium score. Last, it was difficult to exclude the possibility of subclinical CHD in controls. Nevertheless, we collected detailed information on symptoms, hospital records, clinical examinations, laboratory tests of blood and urine, and electrocardiogram results to minimize undiagnosed CHD in controls.

In summary, we confirmed the association of osteopontin with incident ACS independent of conventional risk factors in four independent studies of Chinese adults. In addition, our data suggested osteopontin could be a potential biomarker for risk and severity of ACS.

Methods

Study design and study populations

We conducted a multi-stage study design in four independent Chinese populations to discover and validate cytokines associated with ACS, including three case-control studies and one prospective nested case-control study (Fig. 3).

Figure 3
figure 3

Flowchart of the Study.

Full size image

In the discovery stage, 12 ACS cases (8 AMIs and 4 UAPs) and 16 frequency matched healthy controls were recruited from Wuhan, Hubei26. Clinically confirmed ACS patients, including UAP, NSTEMI, and STEMI, were recruited from Wuhan Union Hospital and Wuhan Central Hospital from 2010 to 2013. ACS were confirmed based on clinical history, symptoms, electrocardiograph, cardiac biomarkers, coronary angiography, risk factors, and/or other clinical examinations according to World Health Organization guidelines27,28. Patients who were complicated with congenital heart disease, cardiomyopathy, severe kidney or liver dysfunctions, and cancer were excluded from the study. Healthy controls were randomly selected from Wuhan residents in the Wuhan-Zhuhai cohort during the same period29, and were frequency matched on age (±5 years), sex, and BMI (±1 kg/m2).

The first validation set recruited 107 ACS cases (76 AMIs and 31 UAPs) from the same sources as in the discovery set. The second validation set recruited 103 ACS cases (76 AMIs and 27 UAPs) from two hospitals in Guangdong, China (Shenzhen Bao’an Hospital and People’s Hospital of Zhuhai, south China) from 2010 to 2013. In the validation studies, healthy controls without cardiovascular disease (CVD) and cancer from Wuhan-Zhuhai cohort were random selected, and were 1:1 matched on age (±5 years), sex, and BMI (±1 kg/m2). In the nested case-control study, we enrolled 318 ACS (27 AMIs and 291 UAPs) case-control pairs from the Dongfeng-Tongji cohort. Details of the Dongfeng-Tongji cohort have been reported elsewhere30. Briefly, between April and November 2013, we conducted questionnaire inquiries (including major chronic diseases) and physical examinations among retired employees from the Dongfeng Motor Corporation. The company has its own affiliated hospitals and comprehensive health care system, which allowed us to track for morbidity and mortality records of all participants. Baseline CVD and cancer cases were excluded based on self-report or medical records. Incident ACS cases were identified through review of medical insurance documents, hospital records, and death certificates during the follow-up until June 2015. All the diagnostic information and medical records for participants with diagnosed ACS were carefully checked and adjudicated by a group of trained physicians who were blinded to cytokine data. After a median of 1.6 years of follow-up, 318 eligible incident ACS cases were included in the present analysis. Controls were randomly selected from participants who were free of CVD and cancer at baseline and were also CVD-free up to June 2015, and were 1:1 matched on age (±5 years), sex, BMI (±1 kg/m2) and smoking status (current, former and never) to incident ACS cases.

In the longitudinal study, we recruited 82 out of the 318 ACS cases from the nested case-control study who were admitted to the Dongfeng Central Hospital (affiliated with the Dongfeng Motor Corporation) in Shiyan City (central China) from February 2014 to June 2015. Blood samples were drawn within 24 hours of the ACS onset before any medication use. We then compared their blood cytokine levels at baseline (2013) and immediately after ACS onset (2014–2015).

The study was approved by Ethics Committee of the Tongji Medical College and all participating hospitals. All participants signed informed consents and all experiments were performed in accordance with relevant guidelines and regulations.

Biomarker measurements

For the discovery set, 280 plasma cytokines were quantitated using Quantibody Human Cytokine Antibody Array 6000 (Raybiotech Inc., Georgia, USA) according to the manufacturer’s procedure31. Five cytokines (osteopontin, chemokine ligand 23 [CCL23], macrophage stimulating protein [MSP], brain derived neurotrophic factor [BDNF], and CRP) were selected for the next stage validation. Details of the selection criteria were described under statistical analyses. In the validation stage, plasma levels of osteopontin (Cat. # SOST00, R&D, USA), BDNF (Cat. # SBD00, R&D, USA), and CRP (Cat. # SCRP00, R&D, USA), MSP (Cat. # ab100612, Abcam, USA) and CCL23 (Cat. # ab100611, Abcam, USA) were determined by high-sensitivity enzyme linked immunosorbent assay (ELISA) kits.

In the nested case-control study, overnight fasted blood samples were collected during the physical examination at baseline in 2013, before the ACS onset. Plasma biomarkers were measured from plasma samples that had been stored at −80 °C immediately after collection and have not been thawed until analysis. To avoid batch effect, all matched case-control pairs and the measurement sequence were randomized before analysis. Measurements of each case-control pair were performed in duplicate in the same plate. Intra-assay and inter-assay coefficients of variation for all the measurements were <5% and <10%, respectively.

Statistical analysis

The baseline characteristics of ACS cases and controls were compared using one-way analysis of variance (ANOVA) or Mann-Whitney U test for continuous variables and the chi-square for categorical variables. The cytokine microarray data were analyzed using Significance Analysis of Microarray (SAM) 3.00 algorithm (http://statweb.stanford.edu/~tibs/SAM/index.html). SAM assigns each cytokine a d-score based on a multi-comparison analysis of expression changes and indicates significance by fold change and q-value (q-value was defined as the false discovery rate [FDR] adjusted p-value). Five cytokines (osteopontin, CCL23, MSP, BDNF and CRP) were selected for the next stage validation based on the following three selection criteria: 1) at least a 2-fold higher (fold change ≥2) or lower (fold change ≤0.5) expression in the ACS group compared with the control group; 2) q-value <0.05 between ACS and control group; 3) above the limit of detection in each individual. Cytokine levels measured by ELISA were natural-logarithm transformed before analysis. Conditional logistic regression analysis was used to calculate the odds ratios (ORs) of cytokines with adjustment for age, sex, BMI, drinking status, smoking status, systolic blood pressure (SBP), total serum cholesterol, low density lipoprotein (LDL), triglyceride, estimated glomerular filtration rate (eGFR; in mL/min per 1.73 m2), medication of anti-hypertensive and lipid-lowing medications. Correlation coefficients between validate cytokines and blood lipid levels in controls were estimated by Spearman partial correlation coefficients with adjustment for age, sex, BMI, and smoking status.

In the nested case-control study, cytokines were divided into tertiles, from the lowest to highest levels, on the basis of the distributions among the controls. To test the linear trends of the associations between cytokines and ACS, we used the median levels of cytokines in each tertile as continuous variables. To investigate the association of cytokines with ACS severity, two coronary scoring systems were used to evaluate ACS severity: the most adopted clinical 1- to 3- vessels disease score32,33 and the Gensini score34. All ACS patients were classified according to the number of >50% stenotic vessels, and a ≥50% narrowing of the left main coronary artery was considered as 2-vessels disease, based on which ACS cases were categorized into 1-vessel disease, 2-vessels disease, or 3-vessels disease27,28. Among 291 ACS patients with sufficient information to calculate Gensini score, each segment score equals weighting factor (5 for the left main, 2.5 for the proximal circumflex [Cx] and left anterior descending [LAD], 1.5 for the mid LAD, 1 for the right coronary artery, the obtuse marginal branch of Cx, the distal Cx, the posterior descending artery, and the first diagonal branch and the distal LAD, and 0.5 for the posterolateral system and the second diagonal branch, respectively) multiplied by a severity score that represents the percentage luminal diameter reduction of the coronary artery lumen (32 for 100%, 16 for 99%, 8 for 90%, 4 for 75%, 2 for 50%, and 1 for 25% lumen diameter reduction, respectively). Scoring of all coronary angiograms was done by two investigators who were unware of clinical and laboratory data. Onset time window was defined as the time period from blood sample collection to ACS onset. We stratified ACS nested case-controls into <0.5 year, 0.5–1 year and ≥1 year groups according to onset time window of ACS cases. Stratified analysis was conducted with unconditional logistic regression models to evaluate associations between osteopontin levels and ACS risk in each stratum of traditional cardiovascular risk factors. For the nested case-control analysis, we constructed models by adding osteopontin and CRP independently or simultaneously to the Framingham risk score, and looked for the additive value of cytokines. The discrimination value of cytokines for the ACS prediction was illustrated by comparing area under the ROC curve (AUC)35, while the added predictive ability of cytokines combined with Framingham risk score was assessed by the integrated discrimination improvement (IDI) index and net reclassification index (NRI)36,37. We conducted all analysis using SAS version 9.3 (SAS institute Inc., Cary, NC) and a two-sided P-value < 0.05 was considered statistically significant.