In the present study, we observed that the HMGA1 IVS5-13insC variant was not associated with diabetes in the INVEST-GENES population overall or in any race/ethnic group. We observed a frequency of IVS5-13insC consistent with previously published studies in Caucasians [5, 7] and observed an increased frequency of this variant in Hispanics. Furthermore, pairwise LD in our population suggests that IVS5-13insC is not effectively tagged by rs9394200 in any race/ethnic group and in silico analysis revealed minimal evidence of putative functional consequences of the IVS5-13insC variant.
Our observations extend existing knowledge of genetic determinants of type 2 diabetes and are consistent with the findings of Marquez et al., which suggest a lack of association between IVS5-13insC and type 2 diabetes . Marquez et al. also found that the IVS5-13insC variant is not functional, which is supported by our lack of observed association in multiple race/ethnic groups and our in silico functional results which predicted no important consequences from this variability. Our findings are inconsistent with two studies that observed a significant association between IVS5-13insC and type 2 diabetes [5, 6]. This discrepancy may be due to differences in race/ethnicity of our population compared with other studies. However, a functional IVS5-13insC variant would be expected to have similar effects across multiple race/ethnicity groups.
Although our association analysis is consistent with the findings of Marquez et al. , our observation of low LD between IVS5-13insC and rs9394200 in all race/ethnic groups suggests that rs9394200 is not an appropriate tag SNP for IVS5-13insC. In Hapmap, the MAF of the T allele for rs9394200 is similar to the MAF for IVS5-13insC in Caucasians (0.03), but is 0.45 in Yorubans, further suggesting that rs9394200 does not adequately tag IVS5-13insC, especially in non-Caucasian populations. Furthermore, rs9394200 is 5000 base pairs downstream from the of HMGA1 3′ end and is represented on arrays utilized by type 2 diabetes GWAS [3, 20]. Therefore, if rs9394200 were a significant contributor to risk for diabetes, it likely would have been identified in the GWAS analyses.
Our observation of a lack of association in Hispanics and African Americans suggests that IVS5-13insC may be associated with diabetes only in Caucasian individuals. While we did not observe even a trend towards association in our Caucasian population, we acknowledge that our power to observe an association in Caucasians is lower than in the previously published studies. Although we had 84 percent power to detect an association in our overall population, we had inadequate power to definitively conclude an association in by race/ethnicity analyses.
Our in silico analysis did not reveal any direct mechanism of the IVS5-13insC variant’s effects on mRNA expression or amino acid sequence, suggesting that the variant is not likely to be functional. Although the results of Chiefari et al. suggested a functional effect of the IVS5-13insC variant on HMGA1 and INSR mRNA and protein expression in monocytes , Marquez et al. found no effect of the variant on mRNA expression in adipose tissue. The differences in observed effect on mRNA expression may potentially be explained by differential effects of the variant on transcription in monocytes and adipose tissue. The apparent functional effect of IVS5-13insC observed by Chiefari et al. may also be confounded by the lack of evaluation of IVS5-13insC variant carriers versus non-carriers among non-diabetic controls  or potential treatment with glucose-lowering medications in diabetic patients from whom monocytes were collected . Finally, IVS5-13insC may have an unknown functional mechanism not identified by in silico tools used in this study. Our study has several limitations worthy of mention. We recognize the potential for false negative results in our analyses, especially in African Americans, considering the low frequency of the variant and the limited power to detect associations within each race/ethic group. Our observations require replication in independent populations of similar race/ethnic makeup. In INVEST, incident and prevalent diabetes diagnosis was based on investigator reports, but the diabetes phenotype is well described in a previous publication , the accuracy of such reporting has been verified by others , and has been used in other trials [23–25]. INVEST investigator-reported diabetes phenotypes have also been used in a large scale gene-centric meta-analysis with HumanCVD BeadChip data , suggesting validity with regard to the diabetes phenotype and genetic association analysis. Although our HMGA1 association analysis may be confounded if control patients eventually develop diabetes after study follow-up, the high mean age and lower mean BMI suggests that our control patients are less likely to develop type 2 diabetes. In addition, although concordance within the Taqman® genotyping platform was high, our concordance between Taqman® and Pyrosequencing was 95%, suggesting some disagreement between platforms. However, Hardy Weinberg tests did not indicate genotype error and genotype error was minimized by utilization of Sanger sequencing for discrepancy confirmation.