This two-stage case-control study, involving 2803 CAD patients and 2840 age- and sex-matched controls, contained two data sets: the discovery set including 1266 cases and 1296 controls from Wuhan Asia Heart Hospital between March 2011 and February 2017, and the replication set with 1537 cases and 1544 controls from Zhongnan Hospital of Wuhan University and The First Affiliated Hospital of Zhengzhou University (between July 2013 and August 2017). The diagnosis of CAD was based on ≥50% of lumen obstruction in at least one major coronary artery or their main branches by coronary angiography. CAD severity was evaluated by the presence of left main coronary artery (LMCA) lesions, vessel scores, and modified Gensini scores (Montorsi et al., 2006; Weintraub et al., 2011; Bing et al., 2015; Wang et al., 2016; Supplementary Materials and Methods). CAD patients were divided into three groups according to the following clinical presentation: (1) stable angina pectoris (SAP); (2) unstable angina pectoris (UAP); (3) MI (Wang et al., 2016; Supplementary Materials and Methods). The controls were subjects without atherosclerotic lesions and history of CAD, as confirmed by angiography, electrocardiographic test, and physical examination. We collected the following demographics for each participant: (1) traditional CAD risk factors including smoking habit, alcohol drinking habit, and histories of hyperlipidemia, type 2 diabetes mellitus (T2DM), and hypertension; (2) clinical data such as lipid levels, fasting plasma glucose (FPG) levels, blood pressure, and body mass index (BMI) (Supplementary Materials and Methods). We excluded subjects with the following diseases: (1) cardiac diseases such as valvular or congenital heart diseases, myocardial bridge, and coronary spasm; (2) systemic diseases including cancers, hepatic or renal diseases, and autoimmune diseases. This study complied with the Declaration of Helsinki, and was approved by local Ethics Committees. All subjects signed written informed consents accordingly.

Genomic DNA of peripheral blood leukocytes was extracted by a phenol-chloroform method. Genotyping for SNP rs1136410 was performed on a LightScanner 96 system (Idaho Technology, Salt Lake City, UT, USA) using high-resolution melting (HRM) analyses (Figure S1; Li et al., 2014). PCR amplification for HRM was conducted with an annealing temperature of 56°C, in 10 μL of PCR mixtures containing 1.5 mM of Mg²⁺, 200 μM of each dNTP, 0.5 μM of each primer (forward primer: 5′-GGG GGC TTT CTT TTG CTC-3′; reverse primer: 5′-TGT CCA GCA GGT TGT CAAG-3′), 1 U of Taq DNA polymerase, 1 μL of LC Green, and 50 ng of genomic DNA. Genotyping results of HRM were verified by direct sequencing and repeated assays in 10% of the whole samples.

In PBMCs, PARP activities, combined with two well-known biomarkers for oxidative inflammatory response, 8-OHdG and IL-6, were measured with competitive ELISA kits, as described in Supplementary Materials and Methods (Cui et al., 2017). CV values for inter- and intra- assays were 6.4 and 5.5% for PARP activities, 7.2 and 5.1% for 8-OHdG levels, and 4.5 and 6.7% for IL-6 levels, respectively.

To evaluate the high-order gene-environment interactions between SNP rs1136410 and traditional CAD risk factors, MDR and CART analyses were conducted by MDR 2.0 (UPenn, Philadelphia, PA, USA) and Clementine 12.0 (SPSS Inc., Chicago, IL, USA) programs, respectively. Briefly, in MDR analyses (Velez et al., 2007), the predictive accuracy for all combinations of included variables was analyzed by 1,000-time permutation and 100-time cross-validation tests. The best interaction model in predicting CAD risk was identified with the maximum cross-validation consistency (CVC) and the optimal testing accuracy.

According to the relative significance of included variables, CART analyses hierarchically subdivided data to produce a binary recursive-partitioning tree, which identified the best interaction model to predict CAD risk using logistic regression analyses (Barnholtz-Sloan et al., 2011). In CART analyses, each split was based on the Gini index, and required a minimal node size of 50 (Gu et al., 2006; Barnholtz-Sloan et al., 2011).

The differences in demographics between cases and controls were compared by the Student's t-test (for quantitative variables) and the Pearson χ² test (for qualitative variables). Hardy-Weinberg equilibrium (HWE) for SNP rs1136410 was evaluated by the Pearson χ² test. Allelic and genotypic (additive, recessive, dominant) associations of SNP rs1136410 with CAD risk were tested by logistic regression analyses before and after adjusting for age, sex, smoking status, alcohol drinking status, BMI, and histories of hyperlipidemia, T2DM, and hypertension. The 100,000-time Monte-Carlo permutation test was used to control for multiple testing (Jiang et al., 2014). The homogeneity of ORs between two sets was analyzed by the Breslow-Day test. In subgroup analyses, we used the multiplicative likelihood ratio test to examine the potential gene-environment interactions in CAD risk. The effect of SNP rs1136410 on CAD severity was appraised by the linear-by-linear association χ² test. Multivariable linear regression analyses were conducted to test the associations of SNP rs1136410 with lipid levels, IL-6 levels, 8-OHdG levels, and PARP activities, as well as the effects of IL-6 levels, 8-OHdG levels, or PARP activities on CAD risk and vessel scores. We used the Pearson (for normal distributed data) or Spearman (for skewed data) correlation test to assess the correlations between IL-6 levels, 8-OHdG levels, PARP activities, and modified Gensini scores. All these analyses set P < 0.05 (two-tailed) as a significant level in SPSS 17.0 (SPSS Inc., Chicago, IL, USA) and PLINK programs. The statistical power was estimated by PS 3.0 software (Vanderbilt University, Nashville, TN, USA).

In both discovery and replication sets (Table S1), CAD patients had higher frequencies of smoking habit, alcohol drinking habit, hyperlipidemia, T2DM, and hypertension, higher levels of BMI, blood pressure, FPG, triglyceride (TG), total cholesterol (TC), and low-density lipoprotein cholesterol (LDL-c), and lower levels of high-density lipoprotein cholesterol (HDL-c) than controls. The genotype distributions of SNP rs1136410 fulfilled expectations of HWE in merged controls (P_HWE = 0.228).

In the discovery set, the minor allele G of SNP rs1136410 conferred an 11% reduced risk of CAD [odds ratio (OR) = 0.89, P = 0.040, Table 1]. This allelic association was successfully verified in the replication set, with an OR of 0.87 and a P-value of 0.006 (Table 1). Since the allelic ORs between discovery and replication sets were homogenous (P = 0.733), we performed a meta-analysis of these two sets, which identified a more significant allelic association between SNP rs1136410 and CAD risk, with an OR of 0.88 and a P-value of 6.29 × 10⁻⁴ (Table 1).

In both discovery and replication sets, genotypic association analyses consistently found significant effects of SNP rs1136410 on CAD risk under both additive and recessive models (Table 1). In a meta-analysis of these two sets, the recessive model was identified as the best-fitting model, with the smallest OR of 0.75 and the most significant P-value of 6.53 × 10–5 (Table 1). Assuming a recessive OR of 0.75 and a type I error of 0.05, the entire population could offer a statistical power of 97.5% to address the association.

In subtype analyses based on a recessive model (Table S2), the variant genotype GG of SNP rs1136410 was significantly associated with the decreased risk of UAP and MI, with ORs of 0.71 (P = 0.002) and 0.61 (P = 6.07 × 10–6), respectively. No significant association was found between SNP rs1136410 and SAP.

All significant results remained unchanged after the permutation test for multiple testing and adjustment for covariates (Table 1 and Table S2).

We first used subgroup analyses to seek two-way gene-environment interactions in CAD risk. Assuming a recessive model (Table 2), the variant genotype GG of SNP rs1136410 constantly conferred a reduced risk of CAD in all subgroups, except for subjects with smoking habit, T2DM, and hyperlipidemia. The multiplicative likelihood ratio test further found the two-way interactions of SNP rs1136410 with smoking status (P_inter = 0.031), history of T2DM (P_inter = 0.025), and history of hyperlipidemia (P_inter = 0.010) in decreasing CAD risk.

To further identify the high-order gene-environment interactions in CAD risk, data on SNP rs1136410 and eight traditional CAD risk factors (age >60, male, smoking habit, alcohol drinking habit, BMI >25, hypertension, T2DM, and hyperlipidemia) were included in the MDR analyses. As presented in Table 3, the four-factor model containing “smoking status,” “history of T2DM,” “history of hyperlipidemia,” and “SNP rs1136410” variables were selected as the best interaction model for predicting CAD risk, with the optimal testing accuracy of 0.6773, the maximal CVC of 100/100, and the smallest P < 0.0001 in the permutation test. Subsequent CART analyses including these four variables showed that individuals with the variant genotype GG but without smoking habit, T2DM, and hyperlipidemia, were associated with an 84% (OR = 0.16, P < 0.001) decreased risk of CAD, compared with the reference group at the greatest risk (Figure 1). Taken together, we suggested the high-order gene-environment interactions among SNP rs1136410, smoking status, history of T2DM, and history of hyperlipidemia in decreasing CAD risk.

First, in CAD patients from discovery (P = 0.044), replication (P = 0.023) and merged sets (P = 0.003, Figure 2 and Table S3), the variant genotype GG of SNP rs1136410 was significantly associated with decreased frequencies of LMCA lesions under a recessive model. Then, in replication (P = 0.015) and merged sets (P = 0.003, Figure 2 and Table S3), the frequencies of the genotype GG showed a decreasing trend from the one-vessel CAD group to the two-vessel CAD group, and ultimately to the three-vessel CAD group. Finally, along with the increasing quartiles (Q1-Q4) of modified Gensini scores, the frequencies of the genotype GG were gradually decreased in patients from all sets (Figure 2 and Table S3).

In controls from the replication set, linear regression analyses using a recessive model observed a significant association between SNP rs1136410 and TC levels, with a 0.139 mmol/L (P = 0.032, Table 4) decrease in TC levels per minor allele. This correlation became more significant in merged controls, with a P-value of 0.005 and a β value of −0.132 (Table 4). Furthermore, the genotype GG was consistently associated with increased HDL-c levels in both cases and controls from all sets (Table 4). In merged controls, the variances of TC and HDL-c levels explained by SNP rs1136410 were 0.40 and 0.82%, respectively.

In two sets of our study, 120 CAD patients and 120 controls were randomly selected to detect PARP activities, 8-OHdG levels, and IL-6 levels in PBMCs (Table S4). As presented in Figures 3A,B, PARP activities and 8-OHdG levels in CAD patients were significantly higher than those in controls (P < 0.001). A positive correlation was also found between PARP activities and 8-OHdG levels (r = 0.244, P < 0.001, Figure 3C). In subtype analyses (Table S5), patients with UAP and MI had increased PARP activities compared with controls; 8-OHdG levels in all CAD subtypes were significantly higher than those in controls.

In CAD patients, along with the increasing levels of 8-OHdG, vessels scores (P = 0.024, Figure 3D) and modified Gensini scores (r = 0.228, P = 0.012, Figure 3E) were gradually increased; there was a positive correlation between PARP activities and modified Gensini scores (r = 0.234, P = 0.010, Figure 3F).

In PBMCs stimulated by lipopolysaccharide (LPS) (Qin et al., 2016a), IL-6 levels were significantly correlated with CAD risk (P < 0.001), 8-OHdG levels (r = 0.318, P < 0.001), PARP activities (r = 0.516, P < 0.001), and modified Gensini scores (r = 0.226, P = 0.013), but not with vessel scores and SNP rs1136410 (Figure S2).

As presented in Figure 4 and Table S5, PARP activities and 8-OHdG levels were gradually decreased from the genotype AA to the heterozygous variant genotype AG and ultimately to the homozygous variant genotype GG in both cases and controls.

The 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.