Body Mass Index and Obesity- and Diabetes-Associated Genotypes and Risk for Pancreatic Cancer

In 2009, pancreatic cancer accounted for around 35,270 deaths and ranked as the fourth leading cause of cancer-related death in the United States (1). The etiology of pancreatic cancer is poorly understood. Cigarette smoking, obesity, long-term type II diabetes, and a family history of pancreatic cancer are known risk factors for pancreatic cancer (2). Since 1990, the prevalence of obesity has doubled in the United States (3). Statistics from 2007 to 2008 indicate that 33.8% (32.2% of men and 35.5% of women) of American adults are obese (4). A recent meta-analysis of 21 independent prospective studies found that the relative risk of pancreatic cancer per 5 kg/m² increase in body mass index (BMI) to be 1.16 (95% CI, 1.06–1.17) in men and 1.10 (95% CI, 1.02–1.19) in women (5). BMI was more consistently associated with risk of pancreatic cancer than physical activity in both cohort and case–control studies (6–10). It has been estimated that obesity contributes to as much as 26.9% of all cases of pancreatic cancer in the United States (11). Our previous study showed that increased risk of pancreatic cancer was more strongly associated with obesity at younger adulthood (30–40 years) than was weight gain at older age (older than 40 years; ref. 12). However, genetic factors that might predispose obese individuals to develop pancreatic cancer have not yet been identified.

Previous studies have implicated long-term diabetes as a modifiable risk factor for pancreatic cancer (13, 14). Our recent study showed that use of the antidiabetic drug metformin was associated with lower risk of pancreatic cancer among diabetic patients (15). Similarly, genetic factors that predispose diabetics to develop pancreatic cancer are unknown.

The fat mass and obesity-associated (FTO) gene, coding a 2-oxoglutarate–dependent nucleic acid demethylase, has been linked with obesity (16, 17) or type II diabetes (18, 19) in genome-wide association (GWA) studies. Whether this gene has a role in obesity-associated pancreatic cancer has not been examined to date. A recent GWA study identified the nuclear receptor family 5 member 2 (NR5A2, also known as liver receptor homolog 1) gene as one of the top hits for pancreatic cancer (20). The NR5A2 gene plays an important role in lipid metabolism (21) and has been associated with BMI in a previous GWA study of heart disease (22). Peroxisome proliferator-activated receptor gamma (PPARγ) has been linked as one of the top hits in type II diabetes in a previous GWA study (23). Fesinmeyer and colleagues reported a possible association between the PPARγ P12A variant and the risk of pancreatic cancer in a nested case–control study of 83 patients and 166 controls from a high-risk cohort of smokers (24).

AMP-activated protein kinase (AMPK) is an energy-sensing protein that regulates energy metabolism and cell division (25). It is a heterotrimer consisting of a catalytic alpha-subunit and noncatalytic beta- and gamma-subunits. We selected the PRKAA2, PRKAB1, and PRKAB2 genes that code for the catalytic unit and the regulatory subunits of AMPK in this study. AMPK serves as the molecular target of metformin. Adiponectin, coded by the ADIPOQ gene, is an adipocyte-secreted hormone that has been hypothesized to play an important role in diabetes, cardiovascular disease, and cancer (26). A higher plasma level of adiponectin also has been associated with decreased risk of pancreatic cancer in male smokers (27).

The goal of this study was to investigate whether the GWA study finding on NR5A2 would be found in our study population and to further explore the interaction of gene variants with BMI and diabetes in modifying the risk of pancreatic cancer. To investigate our hypothesis that gene variants can predispose obese or diabetic individuals to develop pancreatic cancer, we selected genes that have been associated previously with obesity or diabetes and genes that are known to play an important role in regulating the balance of energy, specifically NR5A2, FTO, PPARγ, ADIPOQ, PRKAA2, PRKAB1, and PRKAB2 gene variants. We investigated the associations of these gene variants with pancreatic cancer in a case–control study.

The study participants were identified from a previously conducted hospital-based case–control study of pancreatic cancer at The University of Texas MD Anderson Cancer Center from January 2000 to January 2009 (12, 28). Cases were patients with a pathologically diagnosed pancreatic ductal adenocarcinoma, and 80% of patients had the diagnosis of their disease at MD Anderson and 20% had their diagnosis prior to their MD Anderson visit. We prospectively identified patients with pancreatic cancer during their initial visit to the Gastrointestinal Clinic at MD Anderson by daily reviewing clinical schedule of all known medical oncologists and surgeons who see pancreatic cancer patients. With the physicians' approval, we approached all patients with either confirmed or suspected pancreatic adenocarcinoma for consent and the response rate (consented/approached) was 83.9%. Failure to recruit was mostly related to emotional stress and severity of illness (45.4%), patients or physician refusal (25.4%), and limited time to conduct the interview (19.3%; ref. 28). There is no significant difference in the distribution of gender, age, and tumor stage between the recruited and the nonrecruited patients (28). Among recruited patients, those who were initially thought to have pancreatic adenocarcinoma but later diagnosed with cholangiocarcinoma, neuroendocrine tumor, ampullary tumor, or pancreatitis were excluded from the study. Control subjects were cancer-free, healthy, non–blood relatives (mainly spouses or friends) of patients with cancers other than gastrointestinal cancer or lung cancer and head and neck cancer. Controls were recruited from the diagnostic radiology clinic of MD Anderson. A short, structured questionnaire was used to screen for potential controls on the basis of the eligibility criteria. The response rate for controls was 83.6%. Both cases and controls must be U.S. residents and could communicate in English. Controls were frequency-matched with cases by age (±5 years), gender, and race/ethnicity. For each study participant, we obtained written informed consent for an interview and a blood sample. The Institutional Review Board at MD Anderson Cancer Center approved our study.

We collected information on patient demographics, BMI, smoking, alcohol use, history of diabetes, and family history of cancer by conducting personal interviews by trained interviewers and a structured and validated questionnaire. We asked each individual about his or her usual and current height in inches and weight in pounds at ages 14 to 19, mid-20s, mid-30s, mid-40s, mid-50s, mid-60s, mid-70s, and in the year prior to recruitment. We then calculated the BMI (kg/m²) by using the usual height and the weight at each age period. Because of our previous finding that BMI at age 30s had the strongest association with risk of pancreatic cancer (12), we used BMI at age 30s as the BMI variable in the current data analysis. Diabetes was defined by self-reported diagnosis and use of antidiabetic medications. Information on the year or the age of diabetes diagnosis was also collected. We cross-checked the diabetes history for cases from their available medical records, and no inconsistency was found.

The inclusion criteria for this study included availability of a DNA sample and complete information on risk factors. However, BMI was missing from 166 cases and 370 controls because information on BMI was not collected until 2004 by using a revised questionnaire. Individuals with missing BMI were excluded from the analysis on BMI–gene interaction or from the multivariate models with justification for BMI.

DNA was extracted from peripheral lymphocytes by using a FlexiGene DNA kit (Qiagen) and a Maxwell 16 automated system (Promega). We conducted genotyping using the TaqMan method and an ABI 7900HT Fast Real-time PCR system. About 5% of the samples were analyzed in duplicate and more than 98% consistency was achieved. We did not include samples with inconsistent results in our final analysis.

We conducted all statistical analyses using Stata 10.1. All tests were based on α = 0.05 level (2-sided). We analyzed the demographics, risk factors, and genotype frequency distributions between the case and control groups, as well as the Hardy–Weinberg equilibrium (HWE) of genotype distribution in the control group using the χ² test. The main effects of genotype and interaction with risk factors were measured on the basis of adjusted ORs (AOR) and corresponding 95% CIs in unconditional logistic regression models, including all demographic and risk factors [age (continuous), sex, race, or ethnicity (non-Hispanic white, Hispanic, black, and others), education (less than high school through bachelor's degree, advanced degree), smoking status (nonsmoker, ≤20 pack-years, >20 pack-years), alcohol use (nondrinker, ≤420 g/wk, >420 g/wk), history of diabetes (yes or no), BMI (≤24.9, 25–29.9, ≥30 kg/m²), and family history of cancer among first-degree relatives (yes or no)]. In the interaction analysis, the homozygous and heterozygous genotypes containing the variant allele were combined if (i) the frequency of the homozygote variant was very low (≤10 patients) and (ii) both genotypes had the same trend of effect (e.g., increased or decreased risk of pancreatic cancer compared with the referent group). The homozygous and heterozygous genotypes containing the common allele were combined if the genotype was in recessive inheritance mode.

To investigate possible interactions of genotypes with known risk factors, we collapsed our data into groups in logistic regression; for example, for diabetes: nondiabetic patients with the common allele, nondiabetic patients with the mutant allele, diabetic patients with the common allele, and diabetic patients with the mutant allele. We also generated interaction terms using the cross product of the genotype and the exposure variable, and we assessed the interaction by using the likelihood ratio test by comparing the full model including the interaction term with the reduced model excluding the interaction term. Both the full model and the reduced model contained the genotype and the exposure variable of interest, as well as the other factors described previously. We calculated a pairwise correlation matrix for the variables of interest before constructing a multivariate model. Because diabetes can be a manifestation of pancreatic cancer, we excluded subjects with a diabetes onset within 2 years prior to their cancer diagnosis in the interaction analysis to limit the influence of reverse causation. Multiple comparison was adjusted using the Single Nucleotide Polymorphism (SNP) Spectral Decomposition software by Nyholt (29). Through adjustments, the effective number of markers (SNPs) was 13.5 and the statistical significance threshold for the main effect was set at 0.004.

Table 1 summarizes the demographics and risk factor distributions between cases and controls, which were well matched by sex but controls were overrepresented with individuals younger than 60 years and underrepresented by minorities because of the difficulties in recruiting older healthy individuals or minorities in this study setting. History of diabetes, heavy smoking, BMI of 25 kg/m² or more, and family history of cancer among first-degree relatives, respectively, were significantly associated with 2.9-, 1.5-, 2.5-, and 1.5-fold increased risk of pancreatic cancer. We analyzed the distribution of age, race and sex between study participants with or without the BMI information. There was no significant difference in any of these factors between patients with or without the BMI information (data not shown). Controls who had missing BMI were younger and females than those with BMI data. However, including BMI in the multivariable models did not significantly change the risk estimates for smoking, alcohol, and family history of cancer but reduced the OR from 2.9 to 2.2 for diabetes.

We selected 15 SNPs of FTO, PPARγ, NR5A2, ADIPOQ, PRKAA2, PRKAB1, and PRKAB2 genes (Table 2) from the top hits of previous GWA studies of obesity (30), type II diabetes (31), pancreatic cancer (20), and insulin sensitivity (32). All genotype frequency distributions in the controls were in HWE (0.14 ≤ P ≤ 1.00). There was statistically significant 79%, 43%, 23%, or 39%, 29%, and 33% lower risk of pancreatic cancer associated with PPARγ Ex4-49C>G GG (P = 0.004), NR5A2 −91147C>T TT (P < 0.001), IVS2+1901C>T CT (P = 0.02), or TT (P = 0.01), IVS1-1354A>G AG (P = 0.002), and ADIPOQ Ex3+117T>C CT (P = 0.047) genotypes, respectively (Table 3). The associations between risk of pancreatic cancer and PPARγ Ex4-49C>G GG, NR5A2 −91147C>T TT, and IVS1-1354A>G AG remained statistically significant after adjusting for multiple comparisons (P ≤ 0.004). Compared with risk estimates from the full model including BMI, excluding BMI from the model (including study participant with missing BMI information) did not make any significant changes in the risk estimates (Table 3).

Genotype distributions for 12 of 15 SNPs were significantly different between racial or ethnic groups (P < 0.01; data not shown). Thus, we conducted additional analysis among non-Hispanic whites only. Most of the risk estimates showed minimal changes (<10%) from the observations made in the entire study population (Table 3). The PRKAA2 Ex9+2011A>C AC genotype was significantly associated with increased risk of pancreatic cancer in the entire study population (P = 0.038) but not in non-Hispanic whites only (P = 0.11). The PRKAB2 Ex8+1306G>A AA was significantly associated with increased risk of pancreatic cancer among non-Hispanic whites (P = 0.035). The PPARγ Ex4-49C>G GG, NR5A2−91147C>T TT, IVS2+1901C>T CT, or TT, IVS1-1354A>G AG, and ADIPOQ Ex3+117T>C CT genotypes were similarly associated with lower risk of pancreatic cancer, as was observed in the entire study population (P < 0.05; Table 3). After multiple test adjustments, only the NR5A2 −91147C>T TT genotype (P = 0.001) remained statistically significant and IVS1-1354A>G AG (P = 0.006) became borderline statistically significantly associated with reduced risk of pancreatic cancer in non-Hispanic whites (Table 3).

We detected statistically significant interactions of BMI with the FTO IVS1-27777C>A (P_interaction = 0.0001) and IVS1-23525T>A (P_interaction = 0.0015) genotypes in the entire study population (Table 4). In general, the variant allele of these 2 SNPs was associated with 22% to 28% reduced risk of pancreatic cancer among individuals with BMI of less than 25 kg/m² but was associated with 54% to 60% increased risk of pancreatic cancer among those with BMI of 25 kg/m² or more. We observed similar results when we restricted the analyses to non-Hispanic whites only (data not shown). There was a weak but statistically significant interaction of ADIPOQ Ex3+117 CT/CC with BMI (P_interaction = 0.03). Furthermore, stratification analysis showed that both heterozygous and homozygous variants of the FTO IVS1-27777C>A and IVS1-23525T>A SNPs were associated with reduced risk of pancreatic cancer among participants with BMI of less than 25 kg/m² but were associated with an increased risk among participants with BMI of more than 25 kg/m² (Table 5). On the other hand, NR5A2 variant carrier was at lower risk of pancreatic cancer than the common allele carrier regardless of BMI status.

We did not observe any significant interaction of genotype with type II diabetes before and after excluding for recent diabetes (Table 6). A possible differential effect of ADIPOQ Ex3+117T>C variant on risk of pancreatic cancer was observed, that is, the CT/CC variant carriers showing a decreased risk among nondiabetic patients but an increased risk among diabetic patients compared with the TT carriers. However, the effect in diabetic patients disappeared when BMI was included in the model or when the recent diabetes was removed from the model (Table 6). We also did not observe an interaction of genotype with smoking status or alcohol use (data not shown).

Our study replicated the significant association between NR5A2 gene variants and decreased risk of pancreatic cancer, as reported previously by a GWA study (20). In addition, we observed a significant association between PPARγ P12A variant and lower risk of pancreatic cancer and significant interactions of 2 FTO gene variants and 1 ADIPOQ variant with BMI in modifying the risk of pancreatic cancer. These results are consistent with our hypothesis that polymorphic variations of obesity- or diabetes-related genes may modify the risk of pancreatic cancer directly or jointly with BMI.

NR5A2 was identified as one of the pancreatic cancer susceptibility genes in a GWA study (14), but the mechanisms underlying the association between NR5A2 variants and lower risk of pancreatic cancer remain unknown. NR5A2 is expressed in the exocrine pancreas and functions mainly in lipid homeostasis through control of bile acid biosynthesis, cholesterol transport, and steroidogenesis (21). Previous studies found that NR5A2 plays a role in breast and colon cancer via regulating the Wnt/β-catenin signaling pathway (33). NR5A2 is regulated by the pancreatic-duodenal homeobox 1 (PDX-1) gene and may play an important role in the development of the pancreas (34). Because the ADIPOQ gene is known to have an NR5A2 response element (35), we hypothesized that the NR5A2 gene interacts with BMI or diabetes in modifying the risk of pancreatic cancer. However, we found that NR5A2 variant carriers had lower risk of pancreatic cancer than common allele carrier regardless of BMI or diabetes status in our study. Additional study is needed to illustrate the biological mechanism underlying the association of NR5A2 and pancreatic cancer.

Another gene that remained significantly associated with reduced risk of pancreatic cancer after adjusting for multiple comparisons is PPARγ, which has the highest expression in adipose tissue in humans (36). PPARγ communicates with pathways related not only to adipocyte differentiation, lipid accumulation, and glucose homeostasis but also to inflammation suppression via inhibition of or interference with proinflammatory components such as STAT, NF-κB, and AP-1 (36). PPARγ overexpression has been observed previously in various types of cancer, including pancreatic cancer (37, 38). The role of PPARγ in pancreatic cancer is controversial (39). On one hand, PPARγ exerts an anticancer effect through promoting cancer cell differentiation, cell-cycle arrest, and apoptosis; on the other hand, PPARγ promotes tumor development by stimulating angiogenesis (39). A polymorphism in PPARγ Ex4-49C>G (P12A) has been extensively studied (40). The G allele is associated with a reduced protein activity (40). Most studies have linked the G allele or the GG genotype with lower BMI (41), reduced insulin resistance (42), increased insulin sensitivity (43), and reduced risk of type II diabetes (31) and cancer (44, 45). Our current study detected an inverse association of the GG genotype with pancreatic cancer. Because of the low minor allele frequency, the interaction of the GG genotype with BMI or diabetes could not be examined in this study. In contrast to our results, a previous study found the G allele to be associated with an increased risk of pancreatic cancer, but it included only 83 cases and 166 controls (24). Further study in additional study populations is needed.

The C allele of ADIPOQ Ex3+117T>C was known to confer a lower expression of adiponectin (46), and it was positively associated with the risk of type II diabetes (47) but not with risk of breast cancer (48). In our study, we found an inverse association of the CT genotype with risk of pancreatic cancer in the subgroup with a BMI of less than 25 kg/m² (P = 0.005) compared with that of the TT genotype. Furthermore, we did not observe any significant interaction of this genotype with type II diabetes. Because of the low frequency of homozygous variants, additional studies with larger sample size are required to clarify the association of the ADIPOQ genotype with pancreatic cancer.

FTO has a high expression in hypothalamic nuclei and is regulated by fasting and feeding status (49). It negatively regulates lipid metabolism via facilitating lipid cell lipolysis (50). FTO variations have been strongly associated with risk of obesity and type II diabetes (30, 31), implying that common mechanisms are shared by these conditions. Genotype AA (vs. TT) in FTO IVS1-23525T>A has been connected with an average increase of about 3 kg in weight, a unit in BMI, as well as a 1.67-fold increased risk for obesity (BMI ≥ 30; ref. 51). Previous studies found the A allele to be associated with an increased risk of obesity, type II diabetes (30, 31), and kidney cancer (52) but with a decreased risk of lung cancer (52). FTO IVS1-23525T>A and FTO IVS1-27777C>A were both located in the intron region and in strong linkage disequilibrium (r² = 0.92). The minor allele of FTO IVS1-27777C>A or FTO IVS1-23525T>A was significantly associated with a decreased risk of pancreatic cancer in the subgroup with a BMI of less than 25 kg/m² but was associated with an increased risk of pancreatic cancer in the subgroup with a BMI of 25 kg/m² or more. The mechanism underlying the differential association of SNPs on risk of pancreatic cancer is not understood at present. The FTO IVS1-23525T>A variant has been associated with early-onset or adult obesity (30, 51, 53), but a null association with BMI was recently reported in older men (54). We explored whether an age difference existed in the associations between the FTO genotype and risk of pancreatic cancer in our study population. We observed that the AA genotype was positively associated with risk of pancreatic cancer among individuals who became obese before or after 40 or 50 years of age (data not shown). We detected no interaction of genotype with age in this study population.

Our study has inherent limitations that accompany any case–control study. Both the patients' referral pattern and selective analysis of study population (inclusion criteria) increased the possibility of selection bias. Missing the seriously ill patients may not only exacerbate selection bias but also induce information bias because of failure to examine the risk factors correlated with patient survival. Misreporting may have occurred when BMI information was collected, though previous studies have found a high level of consistency among self-reported weight and measured weight (7, 55). Despite these limitations, we detected a significantly positive association between BMI and increased risk of pancreatic cancer in our study population.

Our study has several strengths. Perfect matching of sex between patient and control groups will eliminate information bias related to gender because women tend to underestimate and men tend to overestimate their weight (55). Our large sample size and complete exposure information allow us to explore the interaction of gene with risk factors. All SNPs following HWE in the control group reflected good quality control process as well as representativeness of the control population. Finally, the use of BMI at age 30s eliminates the temporal ambiguity between BMI and risk of pancreatic cancer. Overall, our data support our hypothesis that individual variability in BMI-, obesity-, and diabetes-related genes may modify the risk of pancreatic cancer.

Future studies to understand the mechanisms underlying the associations between these gene variants and risk of pancreatic cancer may provide opportunities for the development of novel strategies in the prevention and treatment of this deadly disease.

No potential conflicts of interest were disclosed.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.