Novel risk genes identified in a genome-wide association study for coronary artery disease in patients with type 1 diabetes

Type 1 diabetes (T1D) is a chronic disease characterized by an increase in blood glucose due to a lack of insulin production. Diabetes is a major health concern globally with a prevalence ranging between 4 and 7.8% in industrialized countries [1, 2]; among persons with diabetes, it is estimated that 5–10% are affected with T1D [3]. Recent large-scale epidemiological studies suggest that T1D is associated with a reduced lifespan of approximately 12 years [4] due to at least two-fold increased risk for death due to cardiovascular (CV) events.

Of note, in a nation-based registry of type 1 diabetes, coronary heart disease was 4 times more prevalent in patients with type 1 diabetes compared to sex- and age-matched controls [5].

However, studies exploring cardiovascular disease (CVD) risk variants in T1D patients are lacking. Traditional risk factors for CVD including age, circulating lipid levels, and smoking are well established in the general population [6] and in patients with type 1 diabetes [7]. Coronary artery disease (CAD) could be strongly influenced by genetic factors [8]. However, to the best of our knowledge, the relationship between genetic factors and CAD in type 1 diabetes was largely examined with a candidate gene approach [9‐12]. Another important question is whether genetic markers of CAD established in the general population, known to roughly explain 10% of the heritability [13, 14], also play a role in individuals with T1D.

In an attempt to unravel the genetic determinants of CAD among T1D patients, we examined the association of genome-wide genotype array data with CAD in multiple T1D cohorts of European descent (The British Isles, Denmark and France).

A total of 6,728,637 imputed SNPs were tested for association with CAD in 3557 T1D patients made of 434 with CAD cases and 3123 controls in the discovery dataset. The meta-analysis results of the discovery GWAS have been summarized in the Manhattan and QQ plots shown in Additional file 2: Figure S1, Additional file 3: Figure S2, respectively.

One locus at chromosome 5q13.2 reached genome-wide significance (p < 5 × 10⁻⁸) with the lead SNP, rs115829748, located upstream of the MAP1B gene. The T allele of this low frequency SNP (MAF ~ 0.04), was associated with an Odds Ratio (OR) of 3.16 [95% confidence interval (CI) 2.18–4.59] (p = 1.36 × 10⁻⁹). No other SNP demonstrated suggestive association with CAD at this locus (Additional file 4: Figure S3).

At the p < ~ 1.0 × 10⁻⁵ threshold, 20 additional loci demonstrated evidence for suggestive association with CAD with little heterogeneity across cohorts (Table 1). Imputation metrics of the top SNPs are provided in the Additional file 5: Table S2. Therefore, we sought to replicate the top 21 signals in five independent T1D cohorts totaling 585 CAD cases and 2612 controls. Replication was feasible for 17 SNPs while four SNPs (rs34319244, rs373009901, rs143723948, rs571622299) were not properly imputed in the replication stage (Table 1). While none of the 17 SNPs reached the pre-specified Bonferroni threshold of 3.0 × 10⁻³ for positive statistical replication, three were however nominally (p < 0.05) associated with CAD in the replication stage, with genetic effects consistent between the discovery and replication studies (Table 1). Of note, no trend for association was observed (p = 0.279) with the MAP1B rs115829748 that came out first in the discovery GWAS and that showed similar allele frequencies in the discovery and replication studies.

The strongest association was observed at the CDK18 locus where, in the replication stage, the rs138760780-T allele, with frequency 0.02, was associated with an increased odds ratio (OR) for disease of 1.88 [1.07–3.31] (p = 0.014). This value has to compare with 3.48 [2.00–6.04] observed in the discovery cohorts. In the combined discovery and replication cohorts, the meta-analyzed OR associated with the rs138760780 T allele was 2.60 [1.75–3.85] (p = 2.02 × 10⁻⁶) with no heterogeneity across the discovery and replication stage (p = 0.545).

The second suggestive association holds at the FAM189A2 locus. The rs12344245-G allele found associated with an increased OR of 2.52 [1.68–3.81] in the discovery cohorts also demonstrated a trend for association with the disease in the replication stage, OR = 1.45 [1.01–2.08] (p = 0.022). Combining the discovery and replication study led to a meta-analyzed OR for disease of 1.85 [1.41–2.43] (p = 8.52 × 10⁻⁶) with no significant evidence for heterogeneity across stage (p = 0.426).

The third suggestive association was observed for the PKD1 locus. The rs116092985 minor G associated with a 1.85 [1.408–2.44] and a 1.29 [1.00–1.67] increased risk of CAD in the discovery and replication cohorts, respectively. Altogether, the combined statistical evidence for association of the rs116092985-G allele reached p = 1.01 × 10⁻⁵ (OR = 1.53 [1.27–1.85]) (p = 0.220 for heterogeneity across studies).

The present work was aimed at identifying susceptibility alleles for CAD risk in patient population of T1D using a GWAS approach with a two-step framework (discovery + replication). Albeit we identified one locus (MAP1B) reaching genome-wide significance in the discovery stage, it did not replicate with similar effects. Nevertheless, in the combined T1D dataset of 1019 cases and 5735 controls, we observed strong statistical evidence for association with CAD at three biological candidate genes, CDK18, PKD1 and FAM189A2.

We consider our study as very original as it is the first one, to the best of our knowledge, to use a GWAS approach for CAD, in patients with type 1 diabetes. We have thus pooled the largest collection of type 1 diabetes patients with available genetic data.

The low frequency CDK18 rs138760780-T allele (frequency ~ 0.02) was found associated with ~ 2.5 fold increased risk of CAD. According to public database (e.g. Haploreg [27]), this SNP does not show strong linkage disequilibrium (LD) (pairwise r² > 0.80) with other SNPs at this locus, consistent with the regional association plot that does not suggest any evidence of disease associated SNPs (Additional file 7: Figure S4). Interrogating the functional status of this SNP through HaploReg tool [27] suggested that this SNP may be involved in some epigenetic regulatory mechanisms. CDK18 encodes for a cyclin-dependent kinase, suggesting a role in cell cycle. This predicted protein is also related to CDK1, which is involved in the G2/M transition in eukaryotic cells [28]. Although cell cycle is a very broad pathway, CDK1 has also been associated with T1D [29], but at this point little is known about a potential involvement of CDK18 in the pathophysiology of T1D or its complications.

We also observed some evidence that the low frequency FAM189A2 rs12344245 G allele, (frequency ~ 0.04), associated with a ~ 1.8 fold-increased risk of CAD. We did not find any evidence suggesting that this intronic SNP, or any other SNPs in LD (Additional file 8: Figure S5: regional association plot) with it, could be functional. Nevertheless, even though not much is known about the role of the encoded protein, this locus is a good candidate. Indeed, genetic variations at this locus have been found associated with albumin to creatinine ratio [30]. More interestingly, two FAM189A2 SNPs (rs10780297 and rs10120442) have been reported to moderately associate (p = 9.3 × 10⁻⁴) in a large GWAS for CAD in ~ 63,000 non-diabetic populations [14], suggesting that this locus could be a CAD locus in some specific at-risk groups of diabetic patients. The latter two SNPs are in moderate LD (D’ = 1 but r² = 0.05) with our lead rs12344245 SNP, indicating that a fine mapping analysis of this locus would warrant further investigations. Of interest, it was not identified as a common gene in both type 2 diabetes and CAD. It can thus be speculated this gene is an important gene in high-glucose environment rather than a gene leading to high-glucose.

Finally, we observed an association of the non-synonymous PKD1 rs116092985 (Trp1399Arg) with CAD among T1D patients where the Arg1399 minor allele (frequency ~ 0.10), was associated with an increased CAD risk (OR ~ 1.5). The regional plot (Additional file 9: Figure S6) shows that there are several SNPs in LD with this PKD1 top SNP that associate with CAD. PKD1 encodes for the Polycystin 1, Transient Receptor Potential Channel Interacting protein, a member of the polycystin protein family. Recent reports have suggested a role of PKD1 not only in renal tubular function and structure [31] but rare mutations in this gene as the main also cause underlying polycystic kidney disease [32], highlighting its importance in kidney complications. One important question is whether PKD1 risk allele is involved in a common genetic background linking DN and CAD. This question was not duly analyzed due to power issue. However, no clear association was established with DN in previous GWAS focusing on this question [18].

Our study also enabled us to assess in patients with T1D the impact of common SNPs that have been found associated with CAD in large GWAS performed in unselected individuals. Beyond the observation that most of the previously reported SNPs showed consistent association with CAD in our T1D population, this look-up identified a few CAD loci (ANKS1A, COL4A2, TOMM40/APOE) where the reported CAD associated SNP could have a stronger effect in T1D patients. However, this hypothesis would require further investigation.

Some limitations must be acknowledged. We did not consider differently T1D patients with and without DN, and all of the analyses were not stratified on DN status in order to keep all CAD patients in the analysis. However, this should have limited impact on our main results as none of the SNPs identified here were positive considering previously reported GWAS with regard to DN as primary endpoint [18]. Another limitation pertains to limited power of our sample size required, particular to overcome the harsh genome-wide statistical significance threshold. Indeed, our discovery GWAS was not well powered to identify common SNPs associated with moderate genetic effects as those frequently encountered in a GWAS context. For instance, our discovery study had no power to detect at the genome-wide statistical threshold the genetic effect of a variant with an associated allelic OR less than 1.40. It was only well powered (> 80%) to detect OR greater than 1.6 as soon as the allele frequency of the disease allele is greater than 0.28 and well powered to detect OR greater than ~ 2 for allele frequency greater than 0.05. In particular, we had no power to detect the well-established association of the 9p21 locus at the 5 × 10⁻⁸ threshold while we had a chance of 60% to detect it would the liberal threshold of 0.05 had been used. Similarly, we acknowledge the low power of our replication studies where none of the tested associations achieved the Bonferroni threshold of 3 × 10⁻³. We only had a power of 53, 38 and 26% to detect at this threshold a significant association at the CDK18 rs13876070, PKD1 rs116092985 and the FAM189A2 loci, respectively.

Despite these limitations, we have assembled the largest cohort available and conducted novel analyses to discover novel candidate loci for CAD in T1D patients that need to be further studied with additional epidemiological data and functional work to confirm our findings. Of interest, our negative study could suggest a role of epigenetics beyond genomics. In this regard, insulin promotes the expression of DNA methyltransferases leading to methylation resulting in atherosclerosis [33], broadening the research field of CAD in type 1 diabetes.

We identified three new candidate loci for CAD in individuals with T1D, these loci weren’t previously found associated with CAD in the general population. Three other loci previously reported in the general population were found associated with CAD in our setting, namely ANKS1A, COL4A2 and TOMM40/APOE. Although this work needs further investigation, studying the function of these loci could lead to a better understanding of the physiological pathways involved in the development of CAD as a complication of T1D.