We tested 19 common genetic risk markers that were discovered in European populations. We found that association with all 19 of these SNPs trended in the same direction in this large multiethnic study, and the majority of these variants were nominally significant in their association with diabetes risk. A risk score comprised of these alleles was significantly associated with diabetes risk in all five racial/ethnic groups, with the only significant heterogeneity being larger effect sizes in Japanese Americans. However, in comparing the distribution of risk conferred by these alleles between populations we found that they explain little, if any, of known differences in the prevalence of diabetes between these populations.
These observations indicate that most, if not all, of these alleles show directionally similar association to T2D across many populations. Such a pattern indicates that the causal alleles at these validated risk loci (which have yet to be found) likely predate the migrations that separated these populations now residing in Europe, Africa, East Asia, the Pacific Islands and the Americas. We note that this pattern is unexpected under the recently described “common SNP, rare mutation” model of Goldstein that suggests that GWAS signals with common alleles for T2D and other diseases may be “synthetic associations” created by one or more rare alleles 
. Under the Goldstein Hypothesis the consistent associations that we noted at these loci across populations would only be observed if, in each population, one or more distinct rare alleles arose at each locus, and they happened to arise each time on the same haplotype background. Although possible, this scenario seems unlikely, and a more parsimonious explanation would be the “synthetic association” hypothesis of Goldstein does not apply to a majority of these T2D SNPs.
The modest number of cases and controls in this study (as compared to the initial discovery studies) likely underlies the lack of statistically significant associations in some groups. Weaker associations in some racial/ethnic groups may also be due to differences in allele frequencies, linkage disequilibrium, and environmental and genetic modifiers. In two cases (WFS1 and CDKAL1), significant heterogeneity by race/ethnicity reflected a lack of association in African Americans, perhaps because of lower linkage disequilibrium between the marker and the biologically relevant allele.
It is interesting that the odds ratios observed for these marker SNPs were larger in Japanese Americans than in the original discovery cohorts, and in the other ethnic groups in our study. A meta-analysis of 7 association studies in Japanese populations replicated associations from studies in European populations for 7 loci under study (TCF7L2
, and HHEX
. A recent GWAS in Japanese observed significant associations in KCNQ1
as well as these same 7 loci and, similar to our observations, noted magnitudes of effect that were generally stronger than previously observed in European populations 
. Additional studies in other Asian populations have replicated associations with many of these loci as well 
In the Multiethnic Cohort, we have found the prevalence of T2D to be at least 2-fold higher in African Americans, Latinos, Japanese and Native Hawaiians compared to European Americans, with these differences being independent of body weight 
. We examined the extent to which the known genetic risk alleles for diabetes could explain these disparities by quantifying and comparing the relative risk distributions between populations. Compared to European Americans, we did not observe evidence of greater genetic risk in any population. Our findings therefore indicate that these risk markers explain little, if any, of racial/ethnic disparities in T2D prevalence. It remains possible that the actual causal alleles in these regions may be more common in frequency and/or have larger effects than the index signals in non-European populations. As seen with KCNQ1 
, GWAS in non-European populations are effective in discovering risk loci that are important in multiple populations but difficult to identify in European populations where the alleles are rare.
This study had a number of limitations. First, a self-report of diabetes and use of medication for diabetes was used to define cases and controls. We observed that approximately 1% of a random sample of the controls in this study had HbA1C levels above 7.0%, which suggests that only a small portion of controls had undiagnosed diabetes (see Materials and Methods
). Also, our case definition did not differentiate between T1D and T2D, however we expect this misclassification to be minor as <3% of T2D cases had a previous diagnosis of T1D based on other sources (see Materials and Methods
). The highly consistent findings of this study, as compared to the discovery GWAS reports, argue that our phenotypic characterization is adequate to observe the association to T2D.
Some caution should also be given to the interpretation of the risk modeling conducted in each ethnic group, as the genetic markers included are unlikely to be the causal alleles. Future fine-mapping and sequencing studies to identify the functional variants (common and/or rare) and large-scale testing of each allele will be required to more precisely model risk as well as assess differences in the distribution of genetic risk across populations.
Another limitation is that we did not account for the potential confounding effects of population stratification. However, odds ratios were essentially unchanged after adjusting for global European ancestry in a subset of African Americans (336 cases 397 controls) for whom ancestry markers were available, suggesting that effects due to population substructure were not substantial, at least in this group. We also noted that controlling for education, a proxy for SES which has been shown to be significantly associated with Native American ancestry in Latinos 
, had little effect on the associations with these risk alleles. Furthermore, the risk alleles were not generally more frequent in Latinos than in European Americans which would be likely if these alleles were proxies for more general ancestry differences. While population stratification is unlikely to fully explain these findings, it remains possible that at some loci, the causal alleles may be more correlated with ancestry than the index SNPs.
In summary, our data provide strong support for common genetic variation contributing to T2D risk in multiple populations. Our findings in T2D do not support the theory that GWAS signals are due to rare alleles. Nonetheless, GWAS and sequencing studies in these and other racial/ethnic populations are needed to reveal a more complete spectrum of risk alleles that are important globally as well as those that may contribute to risk disparities.