Multiple Metabolic Genetic Risk Scores and Type 2 Diabetes Risk in Three Racial/Ethnic Groups

Abstract

Context/Rationale:

Meta-analyses of genome-wide association studies have identified many single-nucleotide polymorphisms associated with various metabolic and cardiovascular traits, offering us the opportunity to learn about and capitalize on the links between cardiometabolic traits and type 2 diabetes (T2D).

Design:

In multiple datasets comprising over 30 000 individuals and 3 ethnic/racial groups, we calculated 17 genetic risk scores (GRSs) for glycemic, anthropometric, lipid, hemodynamic, and other traits, based on the results of recent trait-specific meta-analyses of genome-wide association studies, and examined associations with T2D risk. Using a training-testing procedure, we evaluated whether additional GRSs could contribute to risk prediction.

Results:

In European Americans, we find that GRSs for T2D, fasting glucose, fasting insulin, and body mass index are associated with T2D risk. In African Americans, GRSs for T2D, fasting insulin, and waist-to-hip ratio are associated with T2D. In Hispanic Americans, GRSs for T2D and body mass index are associated with T2D. We observed a trend among European Americans suggesting that genetic risk for hyperlipidemia is inversely associated with T2D risk. The use of additional GRSs resulted in only small changes in prediction accuracy in multiple independent validation datasets.

Conclusions:

The analysis of multiple GRSs can shed light on T2D etiology and how it varies across ethnic/racial groups. Our findings using multiple GRSs are consistent with what is known about the differences in T2D pathogenesis across racial/ethnic groups. However, further work is needed to understand the putative inverse correlation of genetic risk for hyperlipidemia and T2D risk and to develop ethnic-specific GRSs.

Clinical risk prediction models for type 2 diabetes (T2D) (1) often include information about anthropometric, glycemic, lipid, and other heritable traits. However, the causal relationship between these traits and T2D, particularly in terms of genetic susceptibility, is unclear. If such causal relationships exist, then genetic factors that influence these traits are also likely to influence T2D risk.

Genome-wide association studies (GWASs) and subsequent meta-analyses have improved our understanding of the genetic basis of T2D (2) and related traits (3–5), making it possible to genetically profile individuals for many traits to explore the genetic etiology of T2D as well as racial/ethnic differences in T2D etiology (6). Previous work has shown that the inclusion of genetic markers associated with lipid, body mass index (BMI), and glycemic traits improves prediction accuracy of T2D (7) and that genetic risk for obesity (8) and some glycemic traits (5) are associated with higher T2D risk. However, the association of genetic risk for dyslipidemia (9, 10), adiponectin (11, 12), and other traits with T2D remains unclear or unknown.

Here, we conduct a comprehensive analysis of the associations between genetic risk to a wide range of metabolic-related traits and T2D in multiple datasets of African American (AA), Hispanic American (HA) and European American (EA) subjects to gain insight into T2D etiology in different racial/ethnic groups, and to determine whether these genetic profiles can improve T2D prediction accuracy.

Materials and Methods

Studies examined

We used data from the Framingham Heart Study (FRAM; n = 8831 EA), the Atherosclerosis Risk in Communities study (ARIC; n = 9157 EA and 2924 AA), the Multi-Ethnic Study of Atherosclerosis (MESA; n = 2210 EA, 1300 AA, and 1180 HA), the Starr County Health Studies' Genetics of Diabetes Study (STARR; n = 1890 HA), and the Northwestern NUgene Project (NW; n = 1300 EA and 1601 AA). Datasets were obtained from dbGaP (database of Genotypes and Phenotypes). Further details and references are provided in Table 1 and in Supplemental Methods. All analyses were conducted separately by study and by self-identified race/ethnicity. Institutional review board approval was obtained from the University of Arizona.

Table 1.

Number of Cases and Controls and Mean Age for Each Dataset Used

	Cases, n	Controls, n	Mean Age, y
EAs
ARIC (n = 9157)	1671	7486	54.4
FHS (n = 8831)	878	7953	37.9
NW (n = 1300)	649	651	54.7
MESA (n = 2210)	225	1985	67.0
AAs
ARIC (n = 2924)	1066	1858	53.3
NW (n = 1601)	755	846	50.5
MESA (n = 1300)	290	1010	66.3
HAs
STARR (n = 1890)	931	959	47.6
MESA (n = 1180)	272	908	65.7

T2D case definition

T2D cases were defined as having a casual glucose level ≥200 mg/dL, a fasting glucose ≥126 mg/dL, a report of taking diabetes medication, or a self-reported physician diagnosis of T2D at any examination or visit. Details about case and control definition can be found in Ref. 13 for NW and Ref. 14 for STARR.

Genotypes and population stratification

Genotypes were obtained from genome-wide arrays typed in each study. After standard quality-control procedures, whole-genome imputation was performed. Population stratification in AAs and EAs was accounted for through a principal-components approach. Further details on the above methods can be found in Supplemental Methods.

Genetic risk scores

Seventeen weighted genetic risk scores (GRSs) were calculated based on the single-nucleotide polymorphisms (SNPs) and effect sizes identified in a recent GWAS meta-analysis for each respective trait: T2D, fasting glucose (FG), fasting insulin (FI), 2-hour glucose (2HG), proinsulin (PRO), hemoglobin A_1c (HbA_1c), BMI, waist-to-hip ratio (WHR), total cholesterol (TC), triglycerides (TGs), high-density lipoprotein (HDL), low-density lipoprotein (LDL), adiposity (FAT), adiponectin (ADPN), blood pressure (BP), C-reactive protein (CRP), and serum urate (URA). A list of SNPs with corresponding effect sizes can be found in Supplemental Table 1, and further details concerning the GWAS meta-analyses used to identify SNPs and effect sizes as well as how the GRSs were calculated can be found in Supplemental Methods.

Statistical analysis

Logistic regression was used to examine the odds ratio (OR) of each of the 17 GRSs on T2D, controlling for age, sex, and the T2D GRS. The latter was modeled to account for the correlation between the T2D GRS and other GRSs. In FRAM, we also included cohort as a discrete covariate to account for the long-running nature of this study. Association estimates were combined across studies within each ethnic group using a random-effects meta-analysis implemented in the metafor package (15) in R (16). To account for the testing of 17 GRSs, we considered statistical significance based on a Bonferroni correction for multiple testing (α = 0.05/17 = 0.0029), as well as nominal statistical significance. We note that the Bonferroni correction in this case is quite conservative, given the correlation among some of the GRSs, and the strong prior evidence that these SNPs are implicated in metabolic traits.

Prediction accuracy was assessed using a training-testing procedure in each ethnic/racial group, where 1 or 2 datasets comprised the training set, and the remaining independent datasets comprised the testing set. Details concerning the procedure for training and testing can be found in the Supplemental Methods.

To identify genomic regions with potential pleiotropic effects, we plotted the genomic location and corresponding phenotype for all SNPs included in GRSs that were found to have consistent evidence of association with T2D. We used a web-based tool, Idiographica (17), to construct this plot.

Results

Estimates of pairwise correlation between GRSs are shown in Supplemental Table 2. The highest correlation was between the LDL and TC GRSs (r = 0.87), likely due to the high correlation between the traits and hence the inclusion of common SNPs in the respective GRS. The T2D GRS was most strongly correlated with the FG and FI GRSs (r = 0.27 and 0.19, respectively).

The associations between the GRSs and T2D risk in EA are shown in Figure 1. The T2D GRS was positively associated with T2D (OR = 1.4 [1.35–1.46], P < .0001) in all EA datasets. Among the other GRSs, the FG (OR = 1.15 [1.11–1.20], P < .0001) and BMI (OR = 1.08 [1.04–1.12], P = .0001) GRSs were positively and significantly associated with T2D, whereas the FI (OR = 1.05 [1.01–1.09], P = .027) was nominally associated with T2D. Interestingly, we observed a trend suggesting that the TC (OR = 0.96 [0.91–1.01], P = .14), TG (OR = 0.98 [0.94–1.02], P = .30), and LDL (OR = 0.97 [0.93–1.01], P = .14) GRSs were negatively associated with T2D risk (see Figure 1). Upon further inspection, it appears that this trend of negative associations (with the LDL and TC GRSs) is due to an excess of SNPs that are directionally inconsistent with T2D risk, as opposed to a single lipid-associated SNP being strongly negatively associated with T2D (Supplemental Figure 1). In addition, based on a query of the DIAGRAM v3 meta-analysis (n > 65 000) summary statistics (http://diagram-consortium.org/downloads.html), only 25% of LDL and 33% of TC SNP effects are directionally consistent with T2D risk (Supplemental Figure 2). In these summary statistics, 2 SNPs stood out with directionally inconsistent effects on T2D and LDL/TC: rs4420638 in APOE and rs10401969 in CILP2, for which the alleles associated with decreased LDL and TC are associated with higher T2D risk (P = 3.2 × 10⁻⁷ and P = 5.4 × 10⁻³, respectively). Among the other GRSs, we observed weak to null associations with T2D, all in the expected direction, except for the BP GRS.

Figure 1.

Meta-analysis of association of each GRS with T2D risk in 4 datasets of EAs. ORs for T2D are shown on the x-axis. Abbreviation: RE, random effects.

Among AAs, the T2D GRS was significantly associated with T2D (OR = 1.17 [1.08–1.26], P = .0002). However, the magnitude of the association was less than that observed in EAs (P = .001). The FI (OR = 1.09 [1.03–1.15], P = .0035) and WHR GRS (OR = 1.07 [1.01–1.13], P = .020) were nominally associated with T2D (see Supplemental Table 3).

Among HAs, the T2D GRS was significantly associated with T2D (OR = 1.4 [1.16–1.68], P = .0004). The only other GRS that demonstrated a nominally significant association with T2D was BMI (OR = 1.13 [1.03–1.24], P = .0074) (see Supplemental Table 4).

The stepwise selection procedure performed in the FRAM and ARIC EA datasets resulted in the selection of 5 GRSs (T2D, FG, 2HG, BMI, and LDL) in both datasets. Only the LDL GRS was negatively associated with T2D (see Supplemental Table 3). Prediction analyses performed on testing sets (MESA and NW) using the multiple-trait GRS resulted in a small and not significant increase in prediction accuracy in both testing sets over the models that considered only the T2D GRS (see Supplemental Table 4). Among AAs, 4 GRSs (T2D, FI, TG, and LDL) were selected in the ARIC training set. All were positively associated with T2D, except the TG GRS (Supplemental Table 3). The use of the multiple-trait GRS did not significantly increase prediction accuracy in either of the 2 testing sets (see Supplemental Table 4). Finally, among HAs, the stepwise selection procedure resulted in the selection of T2D, FG, 2HG, BMI, WHR, CRP, and BP, where all GRSs were positively associated with T2D, except for the CRP and BP GRSs (Supplemental Table 3). The use of the multiple-trait GRS did not result in improved prediction accuracy in the testing set (see Supplemental Table 4).

The genomic region that stands out most prominently with pleiotropic effects is a 780-kb region on chromosome 19 in which GIPR, APOE, and other genes are located (Supplemental Figure 5). This region is linked with variation in T2D, FG, 2HG, BMI, and LDL. Furthermore, the 4 SNPs located in the 44-kb GIPR region associated with T2D, FG, 2HG, and BMI are not in strong linkage disequilibrium (r² between 0.03 and 0.67). A 1.1-Mb region on chromosome 3 also stands out, containing IGF2BP2, ETV5, and ST64GAL1, which are implicated in T2D, 2HG, FG, and BMI.

Discussion

We observed a significant association of the T2D GRS with T2D in all ethnic/racial groups and found that this association is stronger in EAs than AAs. The glycemic GRSs generally exhibited trends of positive association with T2D among EA. The finding that the FI and WHR GRSs, but not the FG GRS, are nominally associated with T2D in AAs is consistent with findings of greater insulin resistance and secretion among AAs (18, 19), suggesting that this racial difference has genetic underpinnings. However, caution is warranted in making comparisons among racial/ethnic groups, because SNPs were identified in European-descent individuals and because associations were more likely to be identified among EAs due to the larger sample sizes.

Among EAs, with the exception of lipid traits (other than HDL) and BP (discussed below), most of the other GRSs exhibit the expected directional relationship with T2D risk, consistent with prospective studies, suggesting that the corresponding phenotypes precede T2D. Most GRSs appear to have very small effects (in proportion to their respective phenotypic association with T2D) that we are likely still underpowered to observe. Additionally, GRSs based on large-sample GWAS meta-analyses may explain more of the respective trait variation than those based on smaller samples.

One of our most interesting findings is the trend suggesting that overall genetic risk for hyperlipidemia is associated with lower T2D risk. This is somewhat unexpected given the finding in prospective studies that hyperlipidemia is predictive of T2D incidence (1). However, it is consistent with the finding that T2D protective alleles at the CILP2 and GIPR genes are associated with hyperlipidemia (2) and with the recent finding of Li et al (20) that genetic risk for dyslipidemia is associated with lower fasting glucose and HbA_1c. These findings should prompt an in-depth examination of the causal link, if any, between lipid traits and T2D.

The observed relationship of the BP GRS with T2D, although not statistically significant, does suggest that the relationship, if any, could be negative, which would be inconsistent with current T2D clinical risk scores. The trend is negative in both HA cohorts, 2 of the 3 AA cohorts, and 3 of the 4 EA cohorts.

Although several GRSs are associated with T2D, they do not substantially improve prediction accuracy compared with the T2D GRS alone. Each non-T2D GRS likely explains a small proportion of its corresponding phenotype and hence an even smaller proportion of T2D.

In conclusion, our findings highlight the importance of identifying genetic risk factors for T2D within different ethnic/racial groups, understanding the differing etiologies of T2D across these groups, and better understanding how hyperlipidemia plays a role in T2D.