Abstract
Background Addressing gene-gene interactions is essential in defining a trait implicating complex disease-related mechanisms. In this study, we aimed to explore both main effects of single-locus and multi-locus interactions to test the hypothesis that the ectonucleotide pyrophosphatase/phosphodiesterase 1 (ENPP1) and perilipin (PLIN) genes may contribute to the etiology of type 2 diabetes (T2D) independently and/or through complex interactions in a Taiwanese population.
Methods There were 416 patients with a diagnosis of T2D and 188 age- and sex-similar control subjects. To investigate gene-gene interactions, we used both the generalized multifactor dimensionality reduction method and logistic regression models.
Results Allelic and genotypic analyses showed significant main effects of ENPP1 rs1044498 (P = 0.000005 and 0.00007, respectively) on the risk of T2D after Bonferroni correction (P < 0.05/2 = 0.025). Compared to the carrier of the AA genotype of the ENPP1 rs1044498 polymorphism, the likelihood of T2D was 2.442 (95% confidence interval, 1.592–3.747) for the carrier of combined AC+CC genotypes after adjustment of sex and body mass index. In addition, the carriers of AA variant in the PLIN rs894160 polymorphism had a higher risk to T2D than those with the combined AG+GG variants (adjusted odds ratio, 1.856; 95% confidence interval, 1.106–3.115) after adjustment of sex and body mass index. Furthermore, the significant 2-locus (P = 0.001) generalized multifactor dimensionality reduction model was identified between ENPP1 and PLIN. Analyses using logistic regression models confirmed the gene-gene interaction.
Conclusions The results suggest that the ENPP1 and PLIN genes may contribute to the risk of T2D independently and/or in an interactive manner in a Taiwanese population.
Type 2 diabetes (T2D) is a chronic disorder characterized by high blood glucose in the context of insulin insensitivity and pancreas β-cell dysfunction.1The disease affects more than 150 million individuals worldwide.2During the study years 1999–2004 among adults in Taiwan, the prevalence of T2D was 6.5% for men and 6.6% for women, respectively.3Owing to the high prevalence of T2D, extensive research is still being performed to identify disease-susceptibility loci using candidate gene approaches, genome-wide association studies, gene expression profiling, and family linkage studies.4,5Type 2 diabetes is genetically heterogeneous as more and more genetic components associated with T2D are being discovered.4,5The genetic studies of T2D will offer unique insights into the underlying pathogenesis, provide better individual prediction of disease risk, and may eventually lead to new therapies for treatment and prevention.2
Association studies based on individual single nucleotide polymorphisms (SNPs) may overlook the associations that can only be found when the combinations of multiple genomic regions are investigated.6,7Therefore, it is essential to address gene-gene interactions for defining a trait that implicates complex disease-related mechanisms, particularly when each involved variant only manifests a minor effect.8–10Evidence is accumulating to suggest that gene-gene interactions may play important roles in determining susceptibility to T2D by identifying more and more interactions between several gene pairs.11–16
Several statistical methods, such as logistic regression and the generalized multifactor dimensionality reduction (GMDR) method,17have been applied to SNP association studies for detecting gene-gene interactions associated with a number of complex diseases.18–21Generalized multifactor dimensionality reduction is a nonparametric data mining approach and is applicable to both continuous and dichotomous phenotypes. Moreover, GMDR permits adjustment for discrete and quantitative covariates in various population-based studies with unbalanced case-control samples. The GMDR approach has been used to evaluate potential gene-gene interactions among 5 genes for the risk of diabetic nephropathy among Taiwanese T2D individuals including sex, hypertensive status, systolic and diastolic blood pressure levels, duration of diabetes, and body mass index as covariates.19It has also been shown that the SNPs from the obesity candidate genes may contribute to the risk of T2D in an interactive manner according to the presence or absence of obesity using GMDR.20
In this work, we assessed the main effects of both single-locus and multilocus interactions to test the hypothesis that the T2D-related genes such as the ectonucleotide pyrophosphatase/phosphodiesterase 1 (ENPP1) and perilipin (PLIN) genes may contribute to the etiology of T2D independently and/or through complex interactions among Taiwanese individuals. Both the ENPP1 and PLIN genes have previously been associated with T2D risk,22–29but little is known whether these genes interact with each other. We determined whether significant gene-gene interactions exist between these 2 genes in affecting T2D using the GMDR method and logistic regression models.
MATERIALS AND METHODS
Subjects
The patients were partially original to the previous study by Lin et al.20The type 2 diabetic group consisted of 416 Taiwanese patients who were recruited from the Tri-Service General Hospital in Taipei, Taiwan, in 2002. All the recruited patients fulfilled the following criteria: (1) the age was between 30 and 75 years; (2) had a diagnosis of diabetes for more than 5 years; (3) the fasting plasma glucose was greater than 6.93 mmol/L (126 mg/d L); (4) the hemoglobin A1C was greater than 6%. In addition, we recruited 188 volunteers as the control group from the same hospital. The control subjects were given clinical examinations to rule out T2D. Approval was obtained from the Internal Review Board of the Tri-Service General Hospital before conducting the study, and the approved informed consent form was signed by each subject.
Laboratory Methods
Table 1 provides detailed information on the selected SNPs, which includes their gene characteristics, allelic variants, and minor allele frequency.
DNA was isolated from blood samples using QIAamp DNA blood kit following the manufacturer’s instructions (Qiagen, Valencia, CA). The qualities of isolated genomic DNAs were checked using the agarose gel electrophoresis and the quantities determined using spectrophotometry.
All SNP genotypings were performed using the Taqman SNP genotyping assay (Applied Biosystems Inc (ABI), Foster City, CA). The primers and probes of SNPs were from ABI assay on demand kit. Reactions were carried out according to the manufacturer’s protocol. The probe fluorescence signal detection was performed using the ABI Prism 7900 real-time polymerase chain reaction system. Four standard DNA samples with known genotypes were used for quality control.
Statistical Analysis
The categorical data were analyzed using the χ2 test. Furthermore, continuous variables were analyzed using the Student t test. Hardy-Weinberg equilibrium of the SNPs was calculated using the Fisher exact test. Logistic regression was conducted to adjust for sex and body mass index (BMI). The covariates included in the analysis were sex and BMI because these 2 covariates were statistically different between the cases and controls as compared to age. Odds ratios (ORs) and their 95% confidence intervals were evaluated. Multiple testing was adjusted by Bonferroni correction. The criterion for significance was set at P < 0.05 for all tests. The data were presented as mean ± standard deviation.
The power to detect significant associations was calculated by the Power for Association With Errors software (PAWE; http://linkage.rockefeller.edu/pawe/).30,31
Generalized Multifactor Dimensionality Reduction
To investigate gene-gene interactions, we used the GMDR method, which is described in detail elsewhere.17Briefly, the n-dimensional space formed by a given set of SNPs is reduced to a single dimension to analyze n-way interactions. Score-based statistics using maximum-likelihood estimates are then calculated to classify multifactor cells into 2 different groups (either high risk or low risk). This is performed for all possible combinations of SNPs, and the combination with the lowest misclassification error is selected. Moreover, we tested 2-loci interactions using 10-fold cross-validation in an exhaustive search, which considers all possible SNP combinations.
The GMDR software provides a number of output parameters, including cross-validation consistency, the testing balanced accuracy, and empirical P values, to assess each selected interaction.17The cross-validation consistency score is a measure of the degree of consistency with which the selected interaction is identified as the best model among all possibilities considered. Furthermore, the testing balanced accuracy is a measure of the degree to which the interaction accurately predicts case-control status with scores between 0.50 (indicating that the model predicts no better than chance) and 1.00 (indicating perfect prediction). Finally, permutation testing obtains empirical P values of prediction accuracy as a benchmark based on 1000 shuffles.
In addition, GMDR allows covariate adjustment by providing both discrete and continuous covariates.17In this study, we analyzed the data while adjusting for sex and BMI. The covariates included in the analysis were sex and BMI because these 2 covariates were statistically different between the cases and controls as compared to age. Logistic regression models were also performed to confirm the results from GMDR analyses.
RESULTS
Subjects
Table 2 describes the demographic and clinical characteristics of the study population. Unrelated T2D cases and controls had similar distribution of sex. In addition, the distribution of age in the patients with T2D and in the control population was well matched. Body mass index was not similar between the patients with T2D and the control population (P = 0.0001). All SNPs were evaluated for their contribution to T2D in the complete sample population, including 416 T2D cases and 188 controls.
Single-Locus Analyses
Table 3 shows the genotype and allele distributions of the 2 SNPs in the T2D cases and controls for the complete sample population. An association with T2D revealed by the allelic or genotypic test was detected in the ENPP1 and PLIN genes. ENPP1 rs1044498 also displayed statistically significant difference in the genotypic test after Bonferroni correction (P < 0.05/2 = 0.025). Given 416 cases, 188 controls, adjusted P values after Bonferroni correction, and data without error, the allelic test had 86% power for ENPP1 rs1044498. In the allelic test, the error parameter means that one allele was incorrectly coded as the other allele. Similarly, the genotypic test had 77% power for ENPP1 rs1044498. The genotype distribution of ENPP1 rs1044498 deviated from the Hardy-Weinberg equilibrium in the case subjects but not in the control subjects (data not shown).
Moreover, the OR analysis showed the risk genotypes of variants in ENPP1 and PLIN, indicating an increased T2D risk (Table 4). The effects of the aforementioned risk genotypes on T2D were adjusted for the influence of sex and BMI by logistic regression. As demonstrated in Table 4, the carriers of the combined AC+CC variants of the ENPP1 rs1044498 polymorphism had a higher risk to T2D than those with the AA variants (AC+CC vs AA; adjusted OR, 2.442; 95% confidence interval, 1.592–3.747) after adjustment of sex and BMI. Compared to the carrier of the combined AG+GG genotypes in the PLIN rs894160 polymorphism, the likelihood of T2D was 1.856 (95% confidence interval, 1.106–3.115) for the carrier of AA genotype after adjustment of sex and BMI.
Multilocus Analyses
Then, we used the GMDR analysis to assess the impacts of combinations between the 2 SNPs for the complete sample population. There were a significant 2-locus model (P = 0.001) involving ENPP1 and PLIN with covariate adjustment by sex and BMI. Overall, the 2-locus model had the testing accuracy of 60.24% and showed good cross-validation consistency (10/10), indicating a potential gene-gene interaction between ENPP1 and PLIN.
Furthermore, a significant interaction (P = 0.0001) between ENPP1 and PLIN was confirmed by logistic regression models, adjusting for sex and BMI.
DISCUSSION
Our study is the first to date to have examined not only the main effect but also epistatic effects of 2 candidate genes, including ENPP1 and PLIN, which are significantly associated with the risk of T2D among Taiwanese individuals. In this study, single-locus analyses showed significant main effects of the ENPP1 (in allelic and genotypic tests and OR analysis) and PLIN (in both genotypic test and OR analysis) genes on the risk of T2D in a Taiwanese population. In addition, a potential interaction between ENPP1 and PLIN was implicated by the significant 2-locus model involving ENPP1 and PLIN. For the first time, showing the important role in which the ENPP1 and PLIN play in modulating the etiology of T2D independently and in an interactive manner is a promising finding.
The ENPP1 gene encodes a type II transmembrane glycoprotein and may confer susceptibility to insulin resistance.32However, genetic evidence of its effect on T2D has been inconsistent. The ENPP1 rs1044498 (K121Q) polymorphism has been reported to predispose to T2D in a Dominican population,22in South Asians living in the United States and in India, in US white subjects,23in a French and Austrian study,24and in a Tunisian population.25On the contrary, this association with T2D has not been replicated in a study in Denmark,33in UK white subjects,34and in a study of Scandinavian, Polish, and North American white samples.35The PLIN gene encodes perilipins, which are phosphoproteins for coating intracellular lipid droplets.36It has been shown that the PLIN rs894160 SNP is located in the neighborhood of susceptibility loci for obesity, diabetes, and hypertriglyceridemia.26–29
Besides the statistical significance, the potential biological mechanism under this interaction model was our concern. Little is known about the relationship between the ENPP1-PLIN interaction and T2D risk. It has been found that effects of increased ENPP1 expression on adipocyte maturation could have a negative influence on the overall ability of adipose tissue to maintain normal glucose and lipid metabolism.37Furthermore, ENPP1 overexpression in adipose tissue induces key features of the metabolic syndrome such as fatty liver, hyperlipidemia, and dysglycemia.38The PLIN gene was also shown to be associated with increased lipolysis, which may lead to augmentation in fatty acid release from adipose tissue and adversely affect insulin sensitivity.29It is well established that adipose tissue is a key endocrine organ synthesizing and secreting adipokines that are involved in the pathogenesis of T2D.39It is possible that the metabolic changes related to the ENPP1-PLIN interaction may affect the endocrine function of adipose tissue that in turn leads to the development of T2D. Nevertheless, data linking the ENPP1-PLIN interaction and the endocrine function of adipose tissue are sparse. Further studies are needed to test these hypotheses.
Our results were consistent with the previous findings11–16that gene-gene interactions should be taken into account in the candidate gene studies of T2D. By using logistic regression models and the multifactor dimensionality reduction (MDR) method, Qi et al.11investigated the hepatocyte nuclear factor 4 alpha (HNF4A) and potassium inwardly rectifying channel subfamily J member 11 (KCNJ11) genes and observed significant interactions between HNF4A rs2144908, rs4810424, and rs1884613; and KCNJ11 E23K, resulting in an increased T2D risk in women. Moreover, Morcillo et al.13indicated that the −75G/A SNP of the apolipoprotein A-I (APOA1) gene may interact with the Trp64Arg SNP of the adrenergic beta-3 receptor (ADRB3) gene in predicting T2D in a population from southern Spain. In addition, Keshavarz et al.14searched for a possible interaction among 3 genes including serine/threonine kinase 11 (STK11), cyclic adenosine monophosphate response element-binding-regulated transcription coactivator 2 (CRTC2), and protein kinase adenosine monophosphate–activated alpha 2 catalytic subunit (PRKAA2) and found an interaction between 2 polymorphisms (STK11 rs741765 and CRTC2 6909C>T) and the risk for T2D in Japanese. Qu et al.16conducted another gene-gene interaction study for 3 genes including tumor protein p53 (TP53), Rap guanine nucleotide exchange factor 1 (RAPGEF1) and nuclear respiratory factor 1 (NRF1), and provided evidence for an interaction between TP53 rs1042522 and RAPGEF1 rs11243444 as well as between TP53 rs1042522 and NRF1 rs1882095 influencing susceptibility to T2D in a Chinese population. In French subjects, Cauchi et al.15also demonstrated that the potential interactions of insulinlike growth factor–binding protein 2 (IGFBP2) and solute carrier family 30 member 8 (SLC30A8), IGFBP2 and hematopoietically expressed homeobox (HHEX), chemokine receptor 4 (CXCR4) and neurogenin 3 (NGN3), CXCR4 and cyclin-dependent kinase inhibitor 2A (CDKN2A), as well as CXCR4 and cyclin-dependent kinase inhibitor 2B (CDKN2B) increased the risk of T2D, respectively. Finally, Neuman et al.12suggested that HNF4A rs1884613 synergistically interacted with the rs10010131 SNP of the Wolfram syndrome 1 (WFS1) gene, HNF4A rs1884613 with the rs12255372 SNP of the transcription factor 7-like 2 (TCF7L2) gene, HNF4A rs1884613 with KCNJ11 rs5219, and WFS1 rs10010131 with KCNJ11 rs5219, respectively, led to higher risk for T2D in an Ashkenazi sample.
The specific purpose of this work was to investigate the gene-gene interactions in the hypothesis that small single gene effects could not be detected by single-locus studies. Although the single-locus association between the PLIN rs894160 SNP and T2D was relatively modest (P = 0.0459) in the genotypic test, we further established a significant interaction model between PLIN rs894160 and ENPP1 rs1044498 in T2D using both the GMDR and logistic regression analyses. Our results demonstrated that an independently nonsignificant gene may interact with another gene resulting in an increased disease risk. On the other hand, ENPP1 could be driving the interaction analysis given the significant single-locus result. Similarly, the previous study by Neuman et al. found that KCNJ11 rs5219 was not associated with T2D in single-locus analyses; however, KCNJ11 rs5219 reached significance when paired with either the HNF4A rs1884613 or WFS1 rs10010131.12
One limitation of this study is that the contributions of other markers in the 2 selected genes should be further examined in future work. As mentioned previously, several previous studies demonstrated these selected SNPs as positive associations with T2D. In the current pilot study, we assumed that an SNP might be a good candidate to explore the genetic role of the selected genes if the SNP sustained its association in several studies.19Second, these findings may not be generalizable to other populations.19,20Further, gene-gene interaction analysis may lead to some loss of power owing to the large number of hypothesis tests performed in a single experiment.40Large ethnically matched studies would be necessary to know if such interaction is found in non-Taiwanese subjects. In future work, we will recruit more patients and controls as a replication group for facilitating further analyses.
In conclusion, our study has tested the association between 2 candidate genes and T2D in Taiwanese subjects based on single-locus and multilocus analyses. Our findings support the hypothesis that the SNPs from the ENPP1 and PLIN genes contribute to the risk of T2D independently or in an interactive manner. Further, we acknowledge the limits of evaluation of SNPs that have not been identified as the causal variants. Independent replications in large sample sizes are needed to confirm the role of the polymorphisms found in this study in T2D.
ACKNOWLEDGMENTS
The authors thank Dr Dee Pei of the Cardinal Tien Hospital, Dr Yi-Jen Hung of the Tri-Service General Hospital, and Dr Shi-Wen Kuo of the Buddhist Xindian Tzu Chi General Hospital for research collaboration.