Genetic Predisposition to Coronary Artery Disease in Type 2 Diabetes Mellitus

Supplemental Digital Content is available in the text.


Joint analysis and combination of evidence
We had access to full summary statistics for the discovery analysis and requested summary statistics for variants selected for replication from replication cohorts. Thus, we performed a joint analysis between the estimates for individual variants from the discovery analyses and summary statistics for a subset of variants selected for replication. Analyses were performed in each study to test the following comparisons: CAD in subjects with T2D and CAD in subjects without diabetes. Association was tested with CAD status in a regression model, adjusted for age, sex and study specific covariates such as principal components to account for population structure, where applicable. Age was defined as age of event for CAD cases and age at sampling for controls.
We used the additive model to generate association summary statistics and combined these statistics in a fixed-effect inverse variance-weighted meta-analysis using GWAMA v2.1. 3 We used a fixed effects model to estimate the allelic effects in individual strata, under the assumptions that any differences in allelic effect between strata were due to type 2 diabetes background. This method did not account for between study variation in allelic effects. To test for heterogeneous allelic effects by T2D status, we used the method outlined by Magi et al., 2010. 3 We double genomic control (GC) corrected association summary statistics both at the study level and in the overall discovery meta-analysis.
Variants were excluded from the discovery meta-analysis if the effective sample size < 4000.
We estimated the interaction effects based on comparing the summary allelic effect on CAD for each variant between subjects with and without T2D. This approach allowed us to include more samples in the meta-analyses as studies that did not contain both subjects with and without T2D could be included in the stratified analyses but would have been excluded from a meta-analysis of the interaction term. By adopting this approach, we were able to increase the sample size but were unable to account for between study variance in allelic effects.

Signal declaration criteria
We selected loci for replication based on a p value of association for the lead variant ≤1×10 -4 in the T2D only, non-diabetic only and interaction analyses. The genotype characteristics of the replication cohorts are given in Supplementary Table 4. We estimated the combined effect sizes for variants in a joint analysis based on a fixed-effect inverse variance-weighted meta-analysis using GWAMA v2. 1. 3 Novel loci were declared at p≤5×10 -8 . For declaring interaction signals, we required that the directions of effect for individual strata were consistent across the discovery and replication analyses and achieved a pinteraction<0.05/175 (the number of variants selected for replication) in the replication analysis. Suggestive interaction signals were identified as those that showed directional consistency across the discovery and replication analyses, and when combined in joint analysis achieved combined pinteraction < discovery pinteraction.

Stratified and overall analyses
Power calculations were conducted in R statistics using the gap package. 4 For the purpose of the power calculations effective population size was used (4/ (1/Ncases+1/Nctrls) and a disease prevalence of 5%. The calculation also considered allele frequency and effect size. We used an α≤5×10 -8 for novel loci.

Interaction analysis
Statistical interaction was calculated by testing the difference between two estimates of allelic effect on CAD. The allelic effects were estimated in subjects with diabetes and without diabetes separated and were compared using GWAMA v2.1 to calculate a pinteraction. 3 The power to detect an interaction depends on how accurately the allelic effect can be estimated in each stratum. We assessed the power to detect an interaction effect of a CAD-risk variant with T2D in three allelic effect scenarios: a) an effect on CAD in subjects with T2D only (i.e. OR is 1 in subjects without diabetes, but varies between 1 and 1.2 in subjects with T2D); b) an effect on CAD in subjects with T2D and without diabetes in the same direction but of differing magnitude (i.e. OR is 1.10 in subjects without diabetes, but varies between 1 and 1.2 in subjects with T2D); and c) an effect on CAD in subjects with T2D and without diabetes but in opposite directions (i.e. OR is 0.90 in subjects without diabetes, but varies between 1 and 1.2 in subjects with T2D). For each scenario, we evaluated a range of risk allele frequencies: 10%, 20% and 50%. Power was calculated for α≤1 ×10 -4 in discovery (using discovery sample sizes) based on the threshold for replication.

Genetic correlation with related risk factors
We assessed the genetic correlation of CAD by diabetes subgroup with related risk factors using LDHub. 5 Genetic correlation was calculated by taking the slope of the regression of the product of trait 1 z scores on trait 2 z scores on the LD score for a SNP. Z scores were derived from the allelic effects and standard error for that trait. We restricted the analysis to 106 traits available in LDHub that are known risk factors for T2D and CAD.

Genetic risk score analysis
SNPs associated with Waist-Hip-ratio (adjusted for body mass index [BMI]), number of SNPs HOMA-IR(NSNPs=15) 17 and insulin resistance (NSNPs=10) 18 at genome-wide significance (p≤5×10 -8 ) were included in a genetic risk score (GRS) for each trait respectively. To account for the pleiotropic effects amongst the lipid associated loci a multivariable model was employed from in R using the TwoSampleMR package. 19 For all other GRS the inverse variance weighted method was used to associate each of the GRS with CAD summary statistics. 20 Figure 1: Odds ratios (OR) for 160 known coronary artery disease (CAD) loci from this study compared to the published OR in the combined analysis of CAD, CAD in subjects with T2D and for CAD in subjects without diabetes. The blue colour indicates a p<5x10-3 in at least one of the analyses. For these SNPs the odds ratios >/= 1.00 for the published risk allele. artery disease (CAD) is specific to subjects with type 2 diabetes (T2D) (i.e. OR is 1.0 in diabetes free subjects, but varies between 1.0 and 1.2 in subjects with T2D); B) effect on CAD is heterogeneous, but is in the same direction irrespective of diabetes status (i.e. OR is 1.10 in diabetes free subjects, but varies between 1.0 and 1.2 in subjects with T2D); and C) effect on CAD is heterogeneous, and is in the opposite direction in subjects with T2D and diabetes free subjects (i.e. OR is 0.9 in diabetes free subjects, but varies between 1.0 and 1.2 in subjects with T2D). The continuous lines represent the power to detect an interaction p<1×10 -4 in discovery and the dashed line a p<0.05 in the replication. Figure 3: Locuszoom plots of association statistics in the region of ZNF648 for a published variant rs10911021 with coronary artery disease (A) and the interaction p value with type 2 diabetes (B). This variant has been previously reported to interact with T2D to modify the risk of coronary artery disease. Figure 4: Genetic correlation of coronary artery disease stratified by type 2 diabetes with known risk factors for CAD. Asterisks indicate a p value < 4.1x10 -4 (0.05/121) for accuracy in the estimation of genetic correlation. Where the error bars do not cross zero-line p<0.05. Genetic correlations are calculated from all variants rather than those reaching genome-wide significance and give a broader picture of the overall genetic overlap. A different colour point is assigned to each trait.

Supplementary Figure 5:
The heat map of genetic risk scores for known coronary artery disease (CAD) risk factors does not show any significant differences between CAD in subjects with and without diabetes. Genetic risk scores (GRS) were for known CAD risk factors were constructed and associated with CAD in all individuals, CAD in subjects without T2D and with CAD in subjects with T2D. This was to determine if there was a different effect of risk factors on CAD by T2D background. Similar colours indicate a similar strength of association between the GRS for the known CAD risk factor and CAD in the different contexts examined.

Supplementary Table 1:
The discovery meta-analysis included 24 studies and the sample characteristics of those studies are provided in the accompanying excel file.