An evaluation of inflammatory gene polymorphisms in sibships discordant for premature coronary artery disease: the GRACE-IMMUNE study

The study was approved by the Multicentre (MREC) and Local Research Ethics Committees (LRECs) throughout the UK. The details of the recruitment process can be found in the Additional File 1. Premature CAD was defined as a validated MI, percutaneous transluminal coronary angioplasty (PTCA), coronary artery bypass surgery (CABG) or angina (exercise test positive or angiogram showing at least one lesion >50%) before the age of 66. Hypercholesterolaemia was considered as either being on lipid lowering therapy or having total cholesterol greater than 4.9 mmol.l^-1 for individuals with a previous MI or greater than 5.2 mmol/l for individuals with other affected phenotypes (PTCA, CABG or angina) and unaffected individuals. Hypertension was either being on therapy or a blood pressure greater than 150/90 mmHg.

Fifty SNPs and one insertion/deletion, mainly located in coding regions of 35 inflammatory genes, were genotyped (48 polymorphisms from coding regions and three from the introns). The primer mix and all other necessary components for the SNP genotyping were provided by Roche Molecular Systems (Pleasanton, USA). Characteristics of the polymorphisms are shown in the Additional File 2. DNA extraction and storage was performed using the PUREGENE^® DNA extraction kit (available from Gentra Systems, MN, USA). A pooled polymerase chain reaction (PCR) amplified the chosen targets from 50 ng of genomic DNA to produce biotin labelled products. These products were then hybridized with sequence specific oligonucleotide probes immobilized in a linear array. A series of development stages enabled detection of hybridization between probe and PCR product thus allowing identification of genotypes (homozygous wild, homozygous mutant, heterozygote). The processed strip was then scanned and genotypes were assigned by StripScan software (Roche Molecular Systems) [3] providing a semi-automated reading of the data assigning genotypes.

Family relationships were validated using the Graphical Representation of Relationships (GRR) and RELATIVE programs [4, 5] based on the 51 polymorphisms from this study plus 65 additional SNPs (analysis details in the Additional File 3). Allele frequencies were estimated using the program MENDEL which accounts for family relationships [6] and Hardy-Weinberg equilibrium was tested for each SNP using a goodness-of-fit test. The primary analysis was a test of association in presence of linkage between CAD and each individual SNP using the family-based association test (FBAT) [7]. We used the version of this test that employs empirical variance to account for residual familial correlation [8] (program FBAT 2.0.2C, option -e). An additive mode of inheritance was assumed. The analysis was repeated in two subsets of the data: (1) all cases affected with MI and their siblings and (2) all cases of CAD before the age of 50 and their siblings (see details in the Additional File 3). For genes or regions where two or more polymorphisms were in linkage disequilibrium (LD), haplotype analysis was also carried out in FBAT.

The haplotypes showing the highest evidence of association were further analysed using simple conditional logistic regression (CLR) in STATA [9], subsequently adding in clinical covariates. Robust variance was used to account for correlation between family members in CLR [10, 11]. Haplotypes were inferred using the program HAPLORE [12]. As parents were not genotyped, there was some uncertainty in haplotype assignment. This uncertainty was accounted for in CLR by using a weighting approach originally designed for unrelated cases and controls [13] and adapted here to family data (see details in the Additional File 3). After identifying a haplotype associated with CAD, we used unconditional logistic regression to assess whether the same haplotype predicts hypercholesterolaemia, the most predominant clinical risk factor for CAD in our dataset.

The initial data set consisted of 2870 individuals from 930 families. As a result of relationship checking, a total of 41 families were discarded - mainly because reported siblings were found to be more likely half-siblings. In addition, two families were each split into two, leaving a final sample size of 2699 persons from 891 families. The structure of the families included in this study is detailed in Table 1. The median age of disease onset was 50 years (range 21-66) and 69% of affected individuals had a MI as the index event. The median age of unaffected individuals was 57 years (range 35-85) and intra-family comparisons of ages revealed that over 80% of unaffected siblings had surpassed the age at which their affected siblings had their events. Clinical characteristics of the study sample are shown in Table 2.

Considered individually, each of the traditional risk factors of CAD [male gender, diabetes, hypertension, hypercholesterolaemia, cigarette smoking and body mass index (BMI)] has a highly significant effect on disease risk. When all these factors are included in the model jointly, male gender, hypertension, cigarette smoking and hypercholesterolaemia still show a highly significant effect on CAD, whilst diabetes mellitus has a borderline effect and BMI was not associated with CAD (Table 2).

The results of SNP by SNP analysis are shown in Table 3. Results are reported only for SNPs that were in Hardy-Weinberg equilibrium and for which we had at least 50 informative families (the average number of informative families was 345). Associations with CAD significant at 5% level were found for polymorphisms in interleukin 9 [(IL9) C4244T, P = 0.01], nitric oxide synthase 3 [(NOS3) A498G, P = 0.05], complement component 5 [(C5) A2416G, P = 0.04], interleukin 4 receptor [(IL4R T1682C, P = 0.01 and A1902G, P = 0.04] and eotaxin [(SCYA11) G361A, P = 0.04]. In the subsets of MI cases and those with younger CAD onset (<50 years) the associations with C5 (A2416G) remained (MI only P = 0.006; <50 years P = 0.04) and additional significant associations were observed with interleukin 1 alpha (IL1α) C549T (MI only P = 0.02; <50 years P = 0.02).

More than one SNP were genotyped within 13 genes and one gene cluster (IL1α-IL1β). Table 4 shows the genes or gene cluster where any haplotype was significantly associated with CAD, CAD before 50 years or MI at the nominal 5% level using FBAT. The LD (measured with D') between adjacent SNPs is also shown along with their physical position from dbSNP build 128. The most significantly associated haplotype was found in the IL1 cluster (CCC haplotype, frequency of 41% in the total sample), which was found at increased frequency in all CAD cases (P = 0.006), MI cases (P = 0.01) and young CAD cases (P = 0.0002), although the overall haplotype effect only reached the 5% significance level in early-onset CAD cases (P = 0.02, with 7 degrees of freedom). In NOS3 gene, there was a significant protective effect of the AT haplotype (frequency 11%) in overall CAD (single haplotype P = 0.009 and overall test P = 0.04) and in MI (single haplotype P = 0.005, overall test P = 0.03). However, there was no evidence of haplotype association in this gene in younger disease onset. Other haplotypes with nominally significant effects were found in IL4R and SCYA11 (Table 4).

Using an additive model with no covariate adjustment in weighted CLR, each copy of IL1-CCC haplotype confers an increase in risk of CAD with estimated odds ratio (OR) of 1.21 [95% confidence interval (95 CI) = 1.04-1.40, P = 0.01] in the total sample. For young-onset cases, the estimated OR was 1.50 (95 CI = 1.22 - 1.86, P = 0.0001) and for MI the OR was 1.30 (95 CI = 1.09 - 1.56, P = 0.003). The overall significance level of all haplotypes (7df) was 0.28 in the whole sample, 0.18 for MI only and 0.02 in the affecteds before age 50. When tested jointly with four clinical covariates (hypercholesterolaemia, hypertension, smoking and sex), the effect of haplotype IL1-CCC on CAD was no longer significant in the whole sample (OR = 1.07, P = 0.57) nor in those affected by MI (OR = 1.26, P = 0.07), but it was nominally significant in younger cases (<50 years OR = 1.39, P = 0.05). No significant interaction was found between IL1-CCC haplotype and any of the covariates.

The most potent predictor of increased risk of disease in our population was hypercholesterolaemia (see Table 2). When it was excluded from the multivariable analysis, the effect of haplotype IL1-CCC in the whole sample was significant (OR = 1.22, P = 0.05), and was more so in those with younger onset disease (OR = 1.53, P = 0.002, see Table 5). This indicates a possible relationship between the haplotype and hypercholesterolaemia itself. In order to test this, a simple unconditional logistic regression analysis was carried out on the whole study population and stratifying by CAD, using hypercholesterolaemia as the trait of interest with IL1-CCC haplotype as predictor. In the whole dataset, the estimated OR of having hypercholesterolaemia was 1.15 (95 CI = 1.03 - 1.29, unadjusted P = 0.01) per copy of haplotype IL1-CCC. This was comparable to the OR of developing CAD per copy the same haplotype using unconditional logistic regression (OR = 1.21, 95 CI = 1.02 - 1.43, unadjusted P = 0.02). The haplotype effect on hypercholesterolaemia was of the same magnitude in subjects affected and unaffected by CAD, although not significant in these smaller subsets (OR = 1.10, 95 CI = 0.88-1.37 and OR = 1.13, 95 CI = 0.96-1.33, respectively). This result suggests that the haplotype IL1-CCC may increase the risk of CAD through increasing the risk of hypercholesterolaemia, although the high correlation between hypercholesterolaemia and CAD in our study makes this difficult to evaluate.