Genotype-by-Sex Interaction on Fasting Insulin Concentration: The HyperGEN Study

HyperGEN methods have previously been reported (17). In brief, hypertensive sibships and their offspring and/or parents were recruited from five field centers. The present report includes 1,505 European-American and 1,616 African-American participants without type 2 diabetes. Additionally, a random sample of age-matched individuals from the same base populations (182 Caucasians and 198 African-American subjects) was included to estimate population parameters. The sample of examined individuals included 2,433 relative pairs: 360 parent-offspring pairs, 1,213 sibling pairs, 534 avuncular pairs (aunt/uncle/niece/nephew), and 152 first-cousin pairs (online appendix Table 1 [available at http://diabetes.diabetesjournals.org]).

Venipuncture was performed on fasting participants following standard protocol (17). Anthropometric measurements were obtained as previously reported (17). Serum concentration of insulin was measured using an automated immunoassay instrument. All other variables were collected through interviews performed by trained interviewers. Informed consent was obtained from all participants, and this project was approved by the institutional review boards of all participating institutions.

Genotyping was provided by the National Heart, Blood, and Lung Institute Mammalian Genotyping Service (Marshfield, WI). Details on gel preparation and PCR conditions are available from the Mammalian Genotyping Service website (http://research.marshfieldclinic.org/genetics/home/index.asp). The Cooperative Human Linkage Center screening set 8 was used, which included 391 microsatellite markers equally spaced (∼9 cM distance) throughout the genome. The average marker heterozygosity was 77.7%. Analyses and assignment of the marker alleles were done using computerized algorithms. Relationship status among the purportedly full sibs was tested using ASPEX, a likelihood-based method (18), and MERLIN (19). We used the University of California Santa Cruz (http://genome.uscs.edu/) and Online Mendelian Inheritance in Man (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=OMIM) to determine the cytogenic location of markers and to search for candidate genes.

Variance component linkage analysis (SOLAR version 2.1.2) (20) was used to detect and localize quantitative trait loci (QTL) influencing fasting insulin levels among the combined sample of European-American and African-American participants and stratified by race and sex. We computed exact conditional probabilities using the Lander-Green algorithm as implemented in MERLIN (19). Allele frequencies from founders were calculated separately in African-American and European-American groups.

Genotype-by-sex interaction on fasting insulin levels was explored using a three-step strategy. We initially tested for the evidence of additive genotype-by-sex interaction, by accounting for the genetic covariance among female and male relative pairs. In these analyses, the likelihood of a model including genotype-by-sex interaction is compared with the likelihood of restricted models in which such interactions are excluded. We tested for differential additive genetic effects among female and male subjects [genetic correlation (rhoG) ≠ 1]; for differences in the magnitude of the genetic effects among female and male subjects [genetic variance (σ_g) ≠ among female and male subjects]; and for differences in residual environmental interaction with sex status [environment variances (σ_e) ≠ among female and male subjects]. Second, we performed linkage analysis for male and female subjects (sex-stratified subsets) and compared the model results with the results of the combined sample to restrict the number of regions considered in the QTL-specific genotype-by-sex interaction analysis. Finally, we examined the evidence for a QTL-specific genotype-by-sex interaction at regions identified in the linkage analysis. The likelihood of the model including QTL-specific genotype-by-sex interaction was compared with the likelihood of the restricted model in which such interaction was excluded using a likelihood-ratio test (21).

We excluded 10 individuals who were part of monozygotic twin pairs. In addition, six European-American individuals were excluded from Alabama because a sample size of six was too small to consider as a separate race-center grouping. The insulin variable was logarithmically transformed, and outliers (any observation >4 SDs from the mean and at least 1 SD from the nearest data point) were also excluded (number excluded varies by analysis, less than five). We screened covariates for statistical significance using backward- and forward-stepwise linear regression implemented in SAS 8.02 (Cary, NC). All covariates whose effects were significant at the P ≤ 0.10 were retained in subsequent analysis. Residuals were generated for models and used in all genetic analyses. We considered two different models of covariate adjustment. In model 1, fasting insulin phenotype was stratified by sex and race and adjusted for the effects of age, study center, and age-by-sex interactions. In model 2, we adjusted for additional covariates such as ever smoker, ever drinker, self-reported physical activity (number of hours of exercise per week), and hypertension status in male subjects and for ever smoker, ever drinker, self-reported physical activity, hypertension status, and estrogen use in female subjects. We did not adjust for BMI and other obesity-related traits because these traits are highly genetically correlated with the insulin trait (rhoG = 0.64, P < 0.001).

Because a nonnormal trait distribution increases type I error in the variance component method, we calculated the distribution of logarithm of odds (LOD) scores under the assumption of multivariate normality, using 10,000 replicates and simulation methods (“-lodadj ” command in SOLAR). We determined the robust LOD score by multiplying the observed LOD score by a correction coefficient, calculated by regressing the expected LOD scores on the observed simulated LOD scores (22). Only robust LOD scores are reported in this study. In addition, we determined the 1-LOD unit drop support interval for all linkage results with a LOD score ≥1.8 (23).

Overall, ∼75% of subjects had hypertension (Table 1). Female and male subjects had similar age and physical activity levels within race, but female subjects had lower alcohol consumption and tobacco exposure than males. African-American female subjects had the highest fasting insulin levels among all groups.

The heritability of fasting insulin concentration was ∼0.48 ± 0.09 for female subjects and 0.40 ± 0.13 for males, after adjusting for covariates. The additive genetic correlation (rhoG) between female and male subjects in the combined sample was 0.65 ± 0.21 (P = 0.07), and the additive genetic variance of males was 4.05 ± 0.66 and of female subjects 4.10 ± 0.40 (P = 0.96). In analyses stratified by race, the rhoG was 0.49 (P = 0.02) for European-American subjects and 0.64 (P = 0.19) for African-American subjects, with no differences in the additive genetic variance of male and female subjects in both racial groups.

As hypothesis testing for genotype-by-sex interaction is notably underpowered and we found significant evidence for aggregate interaction in European-American subjects, we proceeded with sex-specific linkage analysis and QTL-specific interaction hypothesis. QTL-specific interaction hypothesis testing was performed only in those regions where suggestive evidence for linkage was observed in the stratified linkage analyses in either male or female subjects. The results of these analyses are reported in Table 2 (for LOD scores ≥1.8), and an extended table with all sex-specific and combined sample LOD scores can be found in online appendix Table 2. A QTL influencing variance of fasting insulin concentration on chromosome 2 at 95 cM (between markers GATA69E12 and GATA71G04) was detected in female subjects (adjusted LOD = 3.0, model 1) but not in males (LOD = 0.0) (interaction P value = 0.0071) (Table 2 and Fig. 1). The 1-LOD support interval spanned from 78 to 101 cM. This sex-specific signal at 2p13.2 was detected in both European-American (LOD = 2.0) and African-American (LOD = 1.2) female subjects. The LOD was greatly reduced in the combined model of female and male subjects (LOD = 0.7) and, therefore, would not be detected without the gene-by-sex interaction analysis.

We also detected suggestive evidence of linkage (23) on chromosomes 1q, 4q, 7q, 9q, 13q, 17q, and 19p and a second linkage peak on chromosome 2p (Table 2). Some of the linkages were detected only in male subjects (i.e., 1p32.2). The QTL-specific interaction P values for these analyses are displayed in online appendix Table 2. Supporting evidence of linkage to the regions identified in our study is summarized in Table 3. Supplemental figures display linkage results for all chromosomes (supplemental online Fig. 1).

We detected a QTL-specific genotype-by-sex interaction influencing variation of fasting insulin concentration in female subjects (LOD = 3.4) on chromosome 2p. This sex-specific linkage signal was detected in both African-American and European-American families and is substantially higher than the linkage signal from the total sample (LOD = 0.7). Interestingly, this region of the genome would not be detected without considering gene-by-sex interactions, stressing the importance of considering complex context-dependent interactions in the search for insulin-related genes.

Sex-specific genetic effects have been previously described for fat deposition and distribution (15,16) and blood pressure (24), traits associated with insulin resistance. It is possible that the regulation of genes influencing insulin is modulated by environment factors such as the hormonal environment in men and women. In addition, these sex-specific effects may be due to gene-gene interaction including interactions with genes in sex chromosomes.

The region of 2p overlaps with positive findings for several insulin-related traits and has been detected in multiple ethnic populations (8,11,13,25,26) (Table 3). For example, suggestive linkage on chromosome 2 at 88 cM (LOD = 2.3) for the homeostatis model assessment of insulin resistance was described for Japanese and Chinese descents enrolled in the Stanford Asian Pacific Program in Hypertension and Insulin Resistance study (26). Among participants from four different ethnic groups (Hispanics, African Americans, European Americans, and Asians) in the Family Blood Pressure Program project, a meta-analysis of fasting glucose and insulin concentration identified suggestive linkage to chromosome 2p at 99 cM (11). Studies including only subjects of European ancestry (the Framingham Offspring Study and the Finland-U.S. Investigation of Non-Insulin-Dependent Diabetes Mellitus Genetics Study) have also identified suggestive evidence for linkage on 2p13 (13,25). Except for our study, none of these previous studies have accounted for gene-by-sex interactions in their analysis, and, therefore, one wonders whether their ability to detect a genome-wide linkage was hampered by not considering complex gene-environment interactions.

Among 172 genes present in the 1-LOD support region, 4 are involved in pathways of glucose or complex carbohydrate metabolism (glutamine-fructose-6-phosphate transaminase 1 [GFPT1], glucosidase I [GCS1], hexokinase 2 [HK2], and N-acetylglucosamine kinase [NAGK]). The GFPT1 gene encodes a rate-limiting enzyme of the hexosamine pathway and this pathway has been implicated in the pathogenesis of insulin resistance and diabetes vascular complications (27). However, studies have failed to show association of GFPT1 variants with diabetes and diabetic nephropathy (28,29).

Other candidate genes located in the 1-LOD unit support interval of the 2p12 signal are four genes from the Reg family, which are involved in regulating β-cell islets mass (30). Regenerating islet-derived 1 β gene A (REG1A) encodes a protein secreted by the exocrine pancreas and has been associated with islet cell regeneration. Another gene located at 2q13 is the Alstrom syndrome 1 (ALMS1) gene that is associated with an autosomal recessive syndrome characterized by childhood obesity, type 2 diabetes, and neurosensory degeneration (31). However, a study of coding ALMS1 variants failed to detect any significant association with type 2 diabetes (32).

We identified several other regions with suggestive evidence of linkage to fasting insulin concentration (Table 2). Most of these regions have been previously described (Table 3). Of interest is the 9q32 region, nearby the region where a putative gene for diabetes (carboxyl ester lipase [CEL] 9q34.3) has recently been identified (33). CEL product is involved in the hydrolysis and absorption of dietary lipids. The enzyme is a component of the exocrine pancreas, and the findings link β-cell dysfunction to exocrine pancreas injury (33). In our study, the highest LOD score at this region was detected in African-American female subjects, who also had the highest mean fasting insulin concentration.

We also identified suggestive linkage of fasting insulin concentration to 1q32.2, a region previously identified only for fasting glucose and homeostatis model assessment of insulin resistance traits in studies of European Americans and American Indians (7,9,25,34). Other regions of suggestive evidence of linkage were 2p21, 4q13.3, 7q31.33, 9q32, 13q12.12, and 17q25.3.

In summary, we detected a QTL-specific gene-by-sex interaction influencing variation of fasting insulin concentration at 2p13.2. The genome-wide linkage was specific to female subjects and was present across racial backgrounds. These results demonstrate the importance of considering complex context-dependent interactions in our search for insulin related genes.

This hypertension network is funded by cooperative agreements (U10) with National Heart, Blood, and Lung Institute (HL54471, HL54472, HL54473, HL54495, HL54496, HL54497, HL54509, and HL54515).

We acknowledge the participants of the HyperGEN Study. We also acknowledge the HyperGEN participating institutions and principal staff: Network Center/University of Utah Field Center: Steven C. Hunt, Roger R. Williams (deceased), Hilary Coon, Paul N. Hopkins, Janet Hood, Lily Wu, and Jan Skuppin; University of Alabama at Birmingham Field Center: Albert Oberman, Cora E. Lewis, Michael T. Weaver, Phillip Johnson, Susan Walker, and Christie Oden; Boston University/Framingham Field Center: R. Curtis Ellison, Richard H. Myers, Yuqing Zhang, Luc Djoussé, Jemma B. Wilk, and Greta Lee Splansky; University of Minnesota Field Center: Donna Arnett, Aaron R. Folsom, Mike Miller, Jim Pankow, Gregory Feitl, and Barb Lux; University of North Carolina Field Center: Gerardo Heiss, Barry I. Freedman, Kari North, Kathryn Rose, and Amy Haire; Data Coordinating Center, Washington University: D.C. Rao, Michael A. Province, Ingrid B. Borecki, Avril Adelman, Derek Morgan, Karen Schwander, David Lehner, Aldi Kraja, and Stephen Mandel; Central Biochemistry Lab, University of Minnesota: John H. Eckfeldt, Ronald C. McGlennen, Michael Y. Tsai, Catherine Leiendecker-Foster, Greg Rynders, and Jean Bucksa; Molecular Genetics Laboratory, University of Utah: Mark Leppert, Steven C. Hunt, Jean-Marc Lalouel, and Robert Weiss; and National Heart, Lung, and Blood Institute: Susan E. Old, Millicent Higgins (retired), Cashell Jaquish, Dina Paltoo, Martha Lundberg, andMariana Gerschenson. We also thank the University of Minnesota Supercomputing Institute for Digital Simulation and Advanced Computation for use of the IBM SP supercomputer.