This prospective study shows that three genetic risk variants for T2D and CVD on 9p21 are associated with total mortality in patients with T2D. We found that being homozygous for the diabetes risk allele rs10811661 was protective for total mortality. We confirmed the previously reported association between two cardiovascular risk alleles of the rs10757278 and rs2383206 variants with an increased mortality risk. We also confirmed a possible synergistic effect of glycemic control with the ‘CVD’ variant rs2383206 on mortality. However, this effect was present in the patients with good glycemic control, not in individuals with a poor glycemic control as was shown previously.
All three SNPs rs10811661, rs10757278 and rs2383206 are located in adjacent LD blocks of the 9p21 locus and represent independent genetic markers for susceptibility to T2D (rs10811661) and CVD (rs10757278 and rs2383206, in high LD with each other (r2=0.86)). The closest genes in proximity of this region are CDKN2A and CDKN2B, both of which are highly expressed in pancreatic islet cells; CDKN2A plays a role in pancreatic islet regenerative capacity. To date, no evidence of an association between the ‘T2D‘variant, rs10811661, and CDKN2A and CDKN2B gene expression in three relevant human tissues (i.e. the colon, liver and pancreas) was found. In contrast, the rs10757278 from the ‘CVD’ risk interval on 9p21 has been shown to regulate cardiac CDKN2A/B expression. Recently, Harismendy et al. reported that the 9p21 locus is enriched in enhancers including one that interacts with the risk allele of rs10757278, emphasizing the important role of rs10757278. The risk allele disrupts a binding site for STAT1, the signal transducer for interferon-γ and cytokines. And thus suggesting a direct mechanistic link between the 9p21 region and atherosclerosis risk via alteration of vascular cell proliferation[19–21].
Association of the ‘T2D’9p21 variant with mortality
We found that having two ‘T2D‘risk T-alleles of the rs10811661 variant was associated with a decreased total mortality risk. One possible explanation of the observed relationship between the genetic risk to T2D and lower mortality could be that subjects who are genetically prone to develop diabetes could develop this disease in the relative absence of lifestyle related diabetes risk factors. These lifestyle related factors, for example, increased general and central adiposity are also associated with chronic diseases and mortality[22, 23]. Although our study included only a limited number of patients with newly diagnosed diabetes, there were no significant differences in baseline variables, for example in BMI, between patient’s homozygous vs. non-homozygous for the T2D risk allele.
Another explanation could be that the observed relationship between the ‘T2D’ risk allele of rs10811661 and low mortality in our study was in fact caused by its protective effect on other diseases. Interestingly, Beekman et al. have recently found a positive relationship, although non-significant, between the two ‘T2D’ risk alleles of rs10811661 and longevity (OR 1.11 (95%CI 0.96-1.23)), which is in agreement with our results. This hypothesis may also be supported by recent studies reporting an association between increased T2D genetic susceptibility and reduced prostate cancer risk[25, 26]. In our study, we found no significant relationship with the ‘T2D’-risk allele and cancer mortality, although our sample size was probably too small to detect this association.
Finally, it is important to note that no underlying causative mechanisms linking the genetic marker rs10811661 with the pathophysiology of T2D were revealed yet. Further studies are needed to confirm this relationship and should also focus on the discovery of potential molecular and physiological pathways explaining T2D development.
Interaction between the ‘CVD’9p21 variant and glycemic control on mortality
We observed an association between the homozygote CVD-risk genotypes for rs10757278 and rs2383206 and an increased risk of total mortality. Although there was an increased risk for macrovascular complications at baseline in patients with the ‘CVD’-risk genotype of rs10757278, the association with the 9p21 variant and cardiovascular mortality after follow-up was not significant. The most likely explanation is that our study had a lack of power to detect a difference in cardiovascular mortality. Another explanation could be that in the patients with T2D, the CVD-risk genotype influences the initiation of atherosclerosis but not progression of atherosclerosis.
In our study, the effect of the rs2383206 variant on total mortality in diabetic patients was further influenced by the degree of glycemic control. The effect of the risk allele disappeared in individuals within the highest HbA1c tertile. These results possibly suggest that for the patients with low levels of HbA1c compared to the patients with poor glycemic control, total mortality is determined to a larger extent by genetic factors. Indeed, in a previous ZODIAC study, we showed that HbA1c is an important factor with regard to mortality risk, but only in the patients with higher HbA1c level. In contrast with this, Doria et al. reported that cardiovascular and total mortality associated with the 9p21 variant rs2383206 were higher in patients with poor glycemic control. Although we used the same statistical model as Doria et al., we could not confirm the synergism between rs2383206 and poor glycemic control on mortality, neither did we observe an interaction between HbA1c level and rs2383206 with cardiovascular mortality. The difference in these two observations can be explained by various factors. First, HbA1c at baseline was substantially lower in our cohort compared to the HbA1c level in the prospective sample reported by Doria et al. (7.3±1.1 versus 8.3±1.4, respectively). Secondly, the study by Doria et al. used multiple HbA1c measurements before the study recruitment. Unfortunately, these data were not available in the ZODIAC study. Finally, it is possible that the strength of the association reported by Doria et al. may have been overestimated due to selective forwarding of especially severe CAD cases to the medical centers from which participants were recruited in the above mentioned study. Thus, it remains unclear to what extent differences between the study populations, for example, in the baseline HbA1c level or in the inclusion criteria, influenced the relationships between glycemic control and mortality in the carriers of two risk alleles of rs2383206. Hence, additional replication studies are necessary to further investigate the interaction between the 9p21 locus and glycemic control.
Strengths and limitations
The unique strengths of our study include its prospective study design using a cohort of T2D patients with a long follow-up for incident mortality, and the availability of the baseline covariates to adjust for in the survival analysis. Also, we selected three 9p21 genetic variants based on the original GWA scans and the replication studies in Europeans[5, 6, 8, 12]. Recently a comprehensive fine-mapping of the 9p21 region using the targeted sequencing and the reference panel from 1000 Genome project has been performed. Although an additional set of common associated variants was identified, no stronger associations than the original GWAS signals have been revealed.
It needs to be noted that the a priori selection process of the SNPs could be the subject of debate. For example, other ‘CVD’ associated variants in the 9p21 region such as rs1333040 or 1333049 were also reported as independently associated with cardiovascular disease. However, rs1333040 and 1333049 are in high LD with rs10757878 (r2=0.57 and r2=0.93 for rs1333040 and 1333049, respectively). Moreover, it has recently been shown that the CVD risk allele of rs10757278 is located in one of the enhancers and disrupts a binding site for the transcription factor STAT1, thus, highlighting the potential functional importance of rs10757278. Another limitation of our study is the number of events that may have been insufficient to draw valid conclusions on cause specific mortality, i.e. cardiovascular mortality. A third potential limitation concerns that only the recessive genetic model was tested in our analysis. To alleviate the multiple testing problems in the data by reducing the number of statistical tests we have exclusively performed our analysis according to the recessive model as a predefined model. This model was selected a priori and based on the reports from all previous prospective studies[12, 14, 15, 17], although we acknowledge that some previous cross-sectional studies have also used the additive model in their analyses.