Two cohorts recruited through clinical care were included in our study. Cohort 1 was a subcohort of the Peking University Health Science Center and the University of Michigan Medical School study of Myocardial Infarction (PUUMA-MI) project designed to analyze the effects of SNPs of APOB, LDLR, and PCSK9 on the risk of PMI in the Chinese Han population and was used to develop a predictive model of PMI risk. Cohort 2 was an external cohort for validating the effects of the SNPs on the risk of PMI in the Chinese Han population and the predictive performance of the established model. Premature myocardial infarction was defined according to Adult Treatment Panel III (ATP III) as initial onset age ≤55 in males and ≤65 in females.

Cohort 1: A total of 9,823 cases in the PUUMA-MI project (8) were screened for qualified subjects according to the following criteria and were further divided into a PMI group and a control group. Criteria for inclusion included (Ma et al., 2020): Age ≥18, and Age ≤55 for males and ≤65 for females; (Egred et al., 2005); Chinese Han population (McManus et al., 2011); no previous history of taking any types of lipid-lowering therapies (such as statins, cholesterol absorption inhibitors, probucol, cholic acid chelators, fibrates, nicotinic acids, and high purity fish oil) at enrollment (Echols and O’Connor, 2010); no kinship between enrolled subjects. The exclusion criteria were as follows (Ma et al., 2020): abnormal hepatic function (alanine aminotransferase/aspartate aminotransferase ≥ upper limit of reference value) (Egred et al., 2005); abnormal renal function (serum creatinine ≥ upper limit of reference value) (McManus et al., 2011); thyroid diseases (such as hyperthyroidism, hypothyroidism, and thyroiditis) (Echols and O’Connor, 2010); cancer or cachexia (Singh et al., 2018); cerebrovascular diseases (including stroke and TIA) (Nanchen et al., 2015); peripheral vascular diseases; and (Cohen et al., 2005) incomplete clinical information.

Enrolled subjects were further divided into a PMI group and a control group according to the following criteria. For the PMI group, patients with initial AMI type I confirmed by symptoms, signs, electrocardiograms, myocardial injury markers, and coronary angiography (CAG) were included. Myocardial infarction and AMI type I were defined according to the Third Universal Definition of Myocardial Infarction (Thygesen et al., 2012). For the control group, a 1:3 age-matched PMI group met the following criteria: 1) No stenosis ≥50% in any main coronary arteries, including the left main artery, left anterior descending artery, left circumflex artery, and right coronary artery, as confirmed by CAG or coronary CT scan, and exclusion of individuals with MI or myocardial ischemia as determined by clinical symptoms, electrocardiograms, and myocardial injury markers; 2) no medical history of cardiovascular diseases.

Cohort 2: Subjects hospitalized in the Department of Cardiology, Peking University Third Hospital from January 2013 to December 2018 who underwent CAG and met the same criteria as those from the PUUMA-MI study, were consecutively included and divided into the PMI group and the control group according to the criteria described in Cohort 1.

This study was approved by the Ethics Committee of Peking University (Approval No. IRB00001052-11068). Informed consent was obtained from each patient included in the study.

Samples from Cohort 1 were collected by the Joint Institute of the PUUMA-MI. In total, 1,652 samples from 4 hospitals were collected: 157 from Peking University First Hospital, 535 from Peking University Third Hospital, 747 from Beijing Shijingshan Hospital, and 213 from Asia Heart Disease Hospital of Wuhan. Medical records were obtained as previously described (Tang et al., 2015). Plasma lipid levels were measured after overnight fasting, total cholesterol (TC) and triglycerides (TGs) were measured using an enzymatic method, and high-density lipoprotein cholesterol (HDL-C) and low-density lipoprotein cholesterol (LDL-C) were measured using a liquid selective detergent method (Tang et al., 2015).

Medical records of subjects from Cohort 2 were obtained from Peking University Third Hospital. General information, previous medical history, family history, medications, blood pressure at admission, height, and weight were collected through the electronic medical record system. Body mass index (BMI) was calculated for each patient. Blood samples were collected from the median cubital vein after fasting (at least 8 h) into an EDTA vacuum anticoagulant tube and a procoagulant tube for DNA extraction and biochemical tests, respectively. The blood samples for DNA extraction were centrifuged, numbered, and stored at –80°C for later use. Biochemical markers, including fasting blood glucose, TC, TGs, HDL-C, and LDL-C, were assessed in the clinical laboratory of Peking University Third Hospital.

The definitions of the risk factors of coronary heart disease were identical to previous description (Tang et al., 2015), including age, gender, body mass index, current smoker, hypertension, dyslipidaemia, diabetes history, and family history of premature coronary heart disease. Dyslipidaemia is defined as having one or more of the following criteria (Ma et al., 2020): TG ≥ 1.7 mmol/L (Egred et al., 2005); HDL-C < 1.04 in males and <1.29 mmol/L in females (McManus et al., 2011); LDL ≥3.4 mmol/L; and (Echols and O’Connor, 2010) already on lipid-lowering drugs.

A QIAamp DNA Blood Mini Kit (QIAGEN, Germany) was used to extract DNA from blood samples. A full-wavelength nucleic acid protein analyzer was used to determine DNA quality. Samples that met the concentration and purity requirements were sent to Beijing Genomics Institute (BGI) for analysis. The SNPs were detected using a HumanExome BeadChip (Illumina, United States). Quality of SNPs for Cohort 1 was determined as previously described (Tang et al., 2015). The genotypes of SNPs in Cohort 2 were determined using mass spectrometry. Sample quality control and SNP detection was performed by BGI. For all tested samples in Cohort 2, a 5% random sample was reciprocally tested, and the reproducibility was 100%.

Matched subjects for the control group in Cohort 1 were selected by propensity score (matching tolerance, 0.18) using SPSS 22.0 software (IBM, United States). Subsequent statistical analyses were performed using SPSS 22.0 software and R software.

Categorical data are presented as percentages (%) or ratios. Pearson’s chi-squared test or Fisher’s exact test was used for comparisons between different groups. Normally distributed continuous data are presented as the mean ± standard deviation. Non-normally distributed data are shown as the median and IQR. Independent t tests or Mann–Whitney U tests were used for comparisons between two groups. The odds ratio (OR) was determined using logistic regression, and the 95% confidence interval (CI) was calculated. Univariate and multivariate logistic regression models were used to identify risk factors related to PMI. First, risk factors for CHD and SNP loci that were associated with risk of PMI were included in the univariate logistic regression model analysis. Risk factors with p < 0.05 in the univariate analysis were then included in a multivariate logistic regression model and analyzed by the enter method. A predictive model of PMI risks was developed based on the weight of each factor. A receiver operating characteristic (ROC) curve was used to analyze the predictive performance of the model for evaluation of PMI risks in different cohorts. The diagnostic cutoff value, sensitivity, and specificity of the model were obtained by the Youden index. Receiver operating characteristic curves were generated using GraphPad Prism 5.0 software. The synergy index (SI) was calculated to determine the gene–environment interactions between the selected SNP loci and traditional CHD risk factors. In this study, the dominant model was used. p values were adjusted for false discovery rate in multiple comparison testing. p < 0.05 was considered statistically significant.

Using Plink 1.07 software, the goodness of fit chi-squared test was performed to compare whether the SNP genotypes in each cohort conformed to the law of Hardy–Weinberg equilibrium (HWE). p < 0.05 represented failure on the HWE test. To guarantee the power of statistical analysis, only variants with MAF >1% that passed the HWE test were analyzed in this study.

A total of 1,652 subjects were enrolled in Cohort 1, including 413 in the PMI group and 1,239 in the control group. A total of 405 (98.1%) subjects in the PMI group had at least one risk factor for CHD, and total of 853 subjects (68.8%) in the control group had at least one risk factor for CHD. The proportions of male patients, and patients with hypertension, diabetes, dyslipidemia, history of smoking, and positive family history of premature CHD were higher in the PMI group than those in the control group (p < 0.001). Body Mass Index, TC, TG, and LDL-C were higher in PMI group than those in the control group (p < 0.05). In contrast, systolic blood pressure (SBP) and HDL-C were lower in the PMI group than those in the control group (p < 0.001). A study flow diagram is presented in Figure 1.

A total of 847 subjects were enrolled in Cohort 2, including 577 in the PMI group and 270 in the control group. A total of 568 patients (98.4%) in the PMI group had at least one risk factor for CHD, and a total of 206 subjects (76.3%) in the control group had at least one risk factor for CHD. The proportions of male patients, patients with a history of smoking, positive family history of premature CHD were higher in the PMI group than those in the control group. Furthermore, BMI, TC, TG, and LDL-C were higher in the PMI group than those in the control group (p < 0.05). The levels of HDL-C in the PMI group were lower than those in the control group (p < 0.001). Details are shown in Table 1. A study flow diagram is shown in Figure 2.

In Cohort 1, 111,716 SNP loci passed SNP quality control, 243 of which were located in PCSK9, APOB, and LDLR. There were 32 SNPs in PCSK9, 4 of which had MAF >1%: rs11583680, rs505151, rs151193009, and rs2479409. There were 167 SNPs in APOB, 6 of which had MAF >1%: rs1042034, rs13306198, rs1367117, rs676210, rs679899, and rs13306194. There were 46 SNPs in LDLR, 3 of which had MAF >1%: rs6511721, rs2738446, and rs2738459. Therefore, 13 SNPs were subjected to the HWE test, of which 3 failed (p < 0.05). These 3 SNPs were excluded from further analysis. Details of the distributions of genotypes of the SNPs of PCSK9, APOB, or LDLR are shown in Supplementary Table S1. The workflow of loci screening is shown in Figure 1.

In Cohort 1, the distribution of the T allele at the PCSK9 R93C variant (rs151193009, C > T) was lower in the PMI group than that in the control group (PMI vs. Control: 0.8% vs. 2.3%, P _adjust < 0.05), and the distribution of the A allele at the APOB R532W variant (rs13306194, G > A) was also lower in the PMI group than that in the control group (PMI vs. Control: 9.3% vs. 12.8%, P _adjust < 0.05). However, the distribution of alleles of other SNPs in the two groups did not differ significantly. Details are shown in Table 2.

Logistic regression analysis of Cohort 1 showed that the C > T variant (adjusted OR 0.354, 95% CI 0.139–0.900, p = 0.029) of PCSK9 R93C was significantly associated with reduced risk of PMI in Chinese Han individuals not taking lipid-lowering medications. Other SNPs showed no association with the risk of PMI in the Chinese Han population after adjusting for confounding factors. Details are shown in Table 2.

As shown in Table 3, PCSK9 R93C passed the HWE test (p > 0.05) in the PMI and control groups in Cohort 2. We also found that the distribution of the T allele at the PCSK9 R93C was lower in the PMI group than that in the control group in Cohort 2 (PMI vs. Control: 1.0% vs. 2.4%, P _adjust < 0.05), and the C > T variant of PCSK9 R93C was significantly associated with reduced risk of PMI, with a decrease of ∼60% (OR 0.420, 95% CI 0.189–0.933, p = 0.033; adjusted OR 0.394, 95% CI 0.157–0.987, p = 0.047) in the Chinese Han population, as determined using logistic regression. This result was significant regardless of whether the data were adjusted for confounding factors. These results were consistent with those from Cohort 1. Details are shown in Table 4.

Gene–environment interactions between the PCSK9 R93C variant and traditional risk factors for CHD were analyzed. There was a trend toward a synergistic interaction between the variant and diabetes in PMI occurrence (SI = 2.769), and an antagonistic interaction between hypertension/smoking and PMI occurrence (with SI of 0.413 and 0.120, respectively) in the Chinese Han population. However, none of the SIs was statistically significant. Details are shown in Supplementary Table S2.

Univariate logistic regression analysis of Cohort 1 showed that six risk factors for CHD apart from age (sex, hypertension, diabetes, hyperlipidemia, smoking, and family history of premature CHD) were closely related to the prevalence of PMI in the Chinese Han population. Multivariate logistic regression analysis of these six variables and PCSK9 R93C was used to develop a 7-variable predictive model of PMI risks in the Chinese Han population. The weight of each variable in the model is shown in Table 5.

Receiver operator curves were generated to test the predictive ability of the model, as illustrated in Figure 3A. The area under the curve (AUC) of the model was 0.839 (95% CI 0.815–0.862, p < 0.001), which suggested that the model showed good predictive performance for PMI risks in the Chinese Han population. The cutoff value of the model was 2.862, and the Youden index was 0.550 (sensitivity 66.6%, specificity 88.4%).

The performance of the prediction model was validated using Cohort 2 as an external cohort. The ROC curve of the predictive model applied to Cohort 2 is shown in Figure 3B. The AUC of the model was 0.840 (95% CI 0.810–0.871, p < 0.001), which suggested that the model also had good predictive power for PMI risks in the Chinese Han population in an external cohort.