Evaluation of the Association between Persistent Organic Pollutants (POPs) and Diabetes in Epidemiological Studies: A National Toxicology Program Workshop Review

Background: Diabetes is a major threat to public health in the United States and worldwide. Understanding the role of environmental chemicals in the development or progression of diabetes is an emerging issue in environmental health. Objective: We assessed the epidemiologic literature for evidence of associations between persistent organic pollutants (POPs) and type 2 diabetes. Methods: Using a PubMed search and reference lists from relevant studies or review articles, we identified 72 epidemiological studies that investigated associations of persistent organic pollutants (POPs) with diabetes. We evaluated these studies for consistency, strengths and weaknesses of study design (including power and statistical methods), clinical diagnosis, exposure assessment, study population characteristics, and identification of data gaps and areas for future research. Conclusions: Heterogeneity of the studies precluded conducting a meta-analysis, but the overall evidence is sufficient for a positive association of some organochlorine POPs with type 2 diabetes. Collectively, these data are not sufficient to establish causality. Initial data mining revealed that the strongest positive correlation of diabetes with POPs occurred with organochlorine compounds, such as trans-nonachlor, dichlorodiphenyldichloroethylene (DDE), polychlorinated biphenyls (PCBs), and dioxins and dioxin-like chemicals. There is less indication of an association between other nonorganochlorine POPs, such as perfluoroalkyl acids and brominated compounds, and type 2 diabetes. Experimental data are needed to confirm the causality of these POPs, which will shed new light on the pathogenesis of diabetes. This new information should be considered by governmental bodies involved in the regulation of environmental contaminants.

Diabetes is a major threat to public health in the United States and worldwide [Centers for Disease Control and Prevention (CDC) 2011; Danaei et al. 2011; World Health Organization (WHO) 2011]. Whereas type 1 diabetes (T1D) is largely thought to be of an auto immune origin, type 2 diabetes (T2D) is mainly associated with obesity and metabolic syndrome, although T2D can occur indepen dently of over weight or obesity. Based on data from the 2005-2008 National Health and Nutrition Examination Survey (NHANES), 25.6 million, or 11.3%, of all people in the United States ≥ 20 years of age are estimated to have diagnosed or undiagnosed diabetes, with associated direct medical costs and indi rect costs (disability, work loss, premature death) of $174 billion in 2007 alone (CDC 2011). Another 35% of people ≥ 20 years of age are believed to be pre diabetic, a condi tion in which fasting blood glucose, blood glucose following a 2hr oral glucose toler ance test (OGTT), or plasma HbA1c levels are above normal but not sufficiently elevated to be classified as diabetes (CDC 2011). The pre diabetic condition often portends the sub sequent development of T2D and is a risk factor for micro and macro vascular diseases (Tabák et al. 2012).
Approximately 11% of prediabetic patients who participated in the Diabetes Prevention Program, a large multi center randomized clini cal trial developed by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK), developed T2D each year during the average 3 years of followup (American Diabetes Association 2011; Knowler et al. 2002). Recently, T2D is being diagnosed in individuals earlier in life, including adolescents (NIDDK 2011). Given the number of people impacted by the disease, an estimated 346 mil lion people worldwide (WHO 2011), and the longterm consequences of diabetes in terms of morbidity, mortality, and economic costs, there is considerable interest in under standing the contribution of "non traditional" risk fac tors, such as environmental chemicals, to the diabetes epidemic. Environmental exposures that have been linked to diabetes in at least some study populations include persistent organic pollutants (POPs), arsenic, bisphe nol A, phthlatates, organotins, non persistent pesticides (Thayer et al. 2012), and air pollu tion (Coogan et al. 2012;Hathout et al. 2006;Krämer et al. 2010;O'Neill et al. 2007;Pearson et al. 2010).
Over the past several years, research addressing the role of environmental chemicals in T2D has rapidly expanded. The February 2011 Diabetes Strategic Plan (NIDDK 2011) acknowledged the growing science base in this area and cited the need to understand more about the role of environmental exposures as part of future research and prevention strate gies. To help develop such a research strategy, the National Toxicology Program (NTP) at the National Institute of Environmental Health Sciences (NIEHS) organized a state ofthescience workshop in January 2011 titled "Role of Environmental Chemicals in the Development of Diabetes and Obesity" (NTP 2011). The objective of this workshop was to examine the literature for evidence of associations between certain chemicals and obesity or diabetes. Epidemiological studies of associations between diabetes and POPs, particularly the halogenated POPs, were considered at the workshop, along with studies of diabetes in association with arse nic, maternal smoking during pregnancy, bisphenol A, phthalates, organotins, and non persistent pesticides (Thayer et al. 2012). A wide variety of chemicals were included in the POPs category, including organo chlorines [2,3,7,8tetrachlorodibenzopdioxin (TCDD or dioxin), Agent Orange, other nonTCDD poly chlorinated dibenzopdioxins (PCDDs), poly chlorinated dibenzo furans (PCDFs), polychlorinated biphenyls (PCBs), dichloro diphenyl trichloro ethane (DDT), dichloro diphenyl dichloro ethylene (DDE), and dichloro diphenyl dichloroethane (DDD)]; brominated compounds [polybrominated diphenyl ethers (PBDEs) and polybrominated biphenyls (PBBs)]; and perfluorinated com pounds [perfluoro octane sulfonate (PFOS), perfluoro octanoic acid (PFOA), perfluoro hexane sulfonate, and perfluoro nonanoic acid].
For the present review we evaluated the literature in terms of consistency, strengths and weaknesses (including power and statistical methods) of the clinical diagnosis, exposure assessment, and study population charac teris tics in order to identify data gaps and areas for future evaluation and research in the area of POPs exposure and diabetes outcomes.

Literature search.
We developed a PubMed (http://www.ncbi.nlm.nih.gov/pubmed) Medical Subject Headings (MeSH)based and keyword search-based strategy to iden tify epidemiological studies of POPs expo sure (organochlorine, organofluorine, and organobromine compounds) and health outcomes related to T1D, T2D, and child hood obesity [for detailed information on the literature search strategy, see Supplemental Material, pp. 2-3 (http://dx.doi.org/10.1289/ ehp.1205502)]. We conducted an initial search on 24 August 2009 and subsequently updated the search through 15 December 2010. Studies of POPs and T2D or diabetes related outcomes (e.g., metabolic syndrome) in both adults and children were eligible for review. We excluded studies from considera tion if they were occupational studies, used death certificates to identify T2D, or did not present original data. Because of time con straints, we formally assessed only studies with T2D as the outcome, excluding stud ies with metabolic syndrome as the outcome. Our search identified 2,752 publications (after removal of duplicates), 72 of which pre sented original data on diabetesrelated stud ies (see Supplemental Material, Figure S1). We excluded 28 studies from consideration because the health outcome was not T2D or because the method used to measure expo sure or classify T2D was not adequate (see Supplemental Table S1). We considered blood or target tissue levels the most informa tive exposure measures; however, this informa tion was not always available (e.g., studies of Vietnam veterans). Studies on Vietnam vet erans were excluded if they were not specific enough to imply exposure to Agent Orange or TCDD; for example, studies comparing veterans who were in Vietnam with those who were not in Vietnam were excluded because they did not specify exposed versus unexposed veterans. We did not consider occupational studies because exposure may be more tar geted depending on the occupation, nor did we consider a study by AndersonMahoney et al. (2008) because the population studied comprised plaintiffs involved in a lawsuit filed due to unusally high PFOA levels in drinking water. In addition, we chose to limit the intro duction of potential biases that are unique to these studies, such as the healthy worker effect. We also excluded studies that used death cer tificates to identify diabetes cases because the prevalence of diabetes is under estimated from mortality data. For example, in a U.Sbased study that characterized the sensitivity and specificity of death certificates for diabetes (Cheng et al. 2008), diabetes was listed as a direct or contributing cause of death on only 6.2% of the death certificates for adults who were known to have diabetes.
We identified an additional 17 articles by reviewing the reference lists in the primary litera ture and review articles, for a total of 43 studies.
Data extraction. NTP Office of Health Assessment and Translation staff extracted the main findings from the included studies [see Supplemental Material, Table S2 (http:// dx.doi.org/10.1289/ehp.1205502)]. The identification of the main findings was based on the following strategy: • When a study did not report a statisti cally significant associa tion (i.e., p > 0.05) between POPs exposure and T2D at any exposure level, we extracted the main finding from the highest exposure group compared with the referent group (e.g., fourth quartile vs. first quartile). • When a study reported a statistically sig nificant association (i.e., p ≤ 0.05) between POPs exposure and T2D and that associa tion displayed a monotonic dose response, we extracted the main finding based on the lowest exposure group with a statistically significant association (e.g., third quartile vs. first quartile). • When associations were non monotonic in nature, we identified the main findings on a casebycase basis and considered any sta tistical trend analyses that might have been conducted, consistency of the overall pattern across exposure groups, and/or the biological significance of the non monotonic finding. POPs represent a toxicologically diverse range of chemicals, all of which are persis tent in the body (i.e., have a long halflife) and the environment. Chemicals are broadly divided into categories based on the halogen group (e.g., chlorinated, fluorinated, bromi nated). Chemicals in the chlorinated group were further divided into common chemical class designa tions (i.e., dioxins, PCBs, DDT/ DDE/DDD). In assessing the PCB studies, we evaluated both total PCBs and PCB153 together because PCB153 is a major contribu tor to total PCB exposure and is used as an indicator PCB. PCB153 is often used as a surro gate measure for total PCBs because it is less expensive to measure (Cote et al. 2006;Meeker and Hauser 2010). Assessing patterns of association for individual PCBs across stud ies is particularly challenging because the class contains 209 structures that are not easy to categorize on the basis of structural similarity and/or biological activity. Even the categoriza tion of "dioxinlike" or "non dioxinlike" is not sufficient because both categories of PCBs are linked to diabetes (Giesy and Kannan 1998;Lee et al. 2006Lee et al. , 2010Lee et al. , 2011a. In general, the findings for individual PCB congeners other than PCB153 are less suggestive for an over all association [see Supplemental Material, Figure S2 (http://dx.doi.org/10.1289/ ehp.1205502)] (Codru et al. 2007;Everett et al. 2007;Lee et al. 2010;Patel et al. 2010;Turyk et al. 2009a).
Study quality. We categorized studies into groups on the basis of study design and nature of the exposure: a) cohort studies with a prospective or nested case-control design, b) crosssectional studies, c) case-control studies, d) occupational studies, e) ecological studies, f ) studies of maternal exposure, and g) studies of Vietnam veterans.
We included a study for consideration if it identified T2D as the outcome and the expo sure measure was deemed adequate. Study quality was evaluated by panel members dur ing workshop deliberations. Aspects of study volume 121 | number 7 | July 2013 • Environmental Health Perspectives quality included potential selection bias, pos sibility of association resulting from reverse causation, or loss to followup. These aspects were not summarized for each study but were considered during the discussion.
Use of Meta Data Viewer to assess patterns of findings. The POPs literature on diabetes is quite complex, consisting of 72 epidemiological studies that often reported findings for multiple compounds in the same study. To visually assess patterns of pri mary study findings from this literature, we used a newly developed software program, the Meta Data Viewer (Boyles et al. 2011). In brief, the Meta Data Viewer is a graph ing program that can display up to 15 text columns and graph 1-10 numerical values. The input data file is an Excel document, and users can sort, group, and filter data to look at patterns of findings across studies. We used this software program to visually display data during the workshop and to gen erate the figures presented below. The odds ratios (ORs) and 95% confidence intervals (CIs) are presented as they were reported by the study's authors; in some cases, rounding may affect the appearance of symmetry for the 95% CIs. The graphing program, accom panying data file, and instructions for use are publicly accessible from the NTP (http://ntp. niehs.nih.gov/go/tools_metadataviewer). The data file currently contains 870 main findings from > 200 human studies on diabetes and childhood obesity-related outcomes for POPs, as well as other exposures such as metals (e.g., arsenic, cadmium, lead, mercury), bisphe nol A, non persistent pesticides, phthalates, and maternal smoking during pregnancy.
Meta Data Viewer is a public resource; the program and any associated NTP data files are available for research and publication.
Main findings. We took into account patterns of findings for chemicals or chemi cal classes if at least three different studies reported diabetesrelated outcomes for that chemical or chemical class. We did not con sider epidemiological evidence sufficient to determine whether any of the positive associa tions were causal in nature.
The strongest positive associations were with transnonachlor ( Figure 1) T, tertile; ww, wet weight. Self-report indicates a self-reported diagnosis of T2D; medication refers to medications used to treat T2D; and FBG and HbA1c indicate levels that were sufficiently elevated to be classified as T2D.

Discussion
The purpose of this evaluation was not only to assess the epidemiological literature for evidence of associations between POPS and T2D but also to collaboratively iden tify data gaps and areas for future research in the area of POPs exposure and outcomes related to diabetes. The resulting list of data gaps includes topics that are related to but not specifically discussed here. For example, we found only one epidemiological study on POPs and T1D, a very important health out come (RignellHydbom et al. 2010). The full list of data gaps and research needs recom mended by workshop participants based on the literature review are summarized in Appendix 1.
Vietnam veteran studies. The conclusion from our evaluation, that there is an associa tion between POPs and diabetes in Vietnam veterans, differs somewhat from assessments conducted by the Institute of Medicine (IOM) Committee to Review the Health Effects in Vietnam Veterans of Exposure to Herbicides (IOM 1994(IOM , 2001(IOM , 2011. The evidence for an association between expo sure to herbicides used during the Vietnam War and longterm health effects in vet erans, including diabetes, is assessed every other year by this committee as part of the Agent Orange Act of 1991. The strength ofevidence conclusion from the epidemio logical studies included in the first report (IOM 1994) was for " inadequate/insufficient evidence to determine whether an associa tion exists" between exposure to herbicides [2,4dichlorophenoxy acetic acid (2,4D), 2,4,5trichloro phenoxyacetic acid (2,4,5T) and its contaminant TCDD, cacodylic acid, and Picloram] and diabetes mellitus.
However, a committee convened by the IOM in 1999 to conduct a specific review of the scien tific evidence regarding T2D and Agent Orange in Vietnam veterans concluded that there was limited or suggestive evidence of an association between T2D and exposure to Agent Orange used in Vietnam (IOM 2001). This conclusion was maintained in The Veterans and Agent Orange updates in 2001, 2002, 2004(IOM 2011. In contrast, our conclusion from the present evaluation is that there is evidence for a positive association when the data were considered collectively (Figure 4).
It is less clear whether studies should use lipidstandardized blood measurement for lipo philic chemicals; several different approaches are currently used in models, including a) wet concentrations without considera tion of lipid profiles, b) lipidstandardized concentrations, or c) wet concentra tions with lipid adjustment. Because POPs circulate with serum lipids, high blood lipids increase measured levels of POPs. Therefore, the failure to account for this relation ship may result in the over estimation of relative risks. However, the exposure to certain chlorinated POPs can lead to increased levels of serum lipids, and dyslipidemia is involved in the pathogenesis of T2D, suggesting that dys lipidemia may be an inter mediate factor in the relation ship between POPs and T2D. In this situation, adjusting for this relation ship may under estimate true associations. Even though true associations may be somewhere between unadjusted and adjusted results, there is uncer tainty about the most appropriate way to deal with lipids.
Adjusting for obesity is controversial in studying the association between POPs and diabetes. There is growing evidence that obe sity is on the causal pathway between POPs and diabetes (Lee et al. 2011b;Ruzzin et al. 2010). In addition, this relation ship is poten tially confounded by the consumption of fatty food, which is associated with obesity and with increased POPs levels. However, adipose tissue serves as a reservoir of POPs, thereby reducing the circulating POPs level (Lim et al. 2011). This effect might have a positive role in limiting the exposure to target tissues for diabetes, such as pancreatic βcells.
Nonmonotonic exposure-response relationships. Several of the reviewed studies reported evidence of nonmonotonic exposure-response relationships. For example, in the CARDIA (Coronary Artery Risk Development in Young Adults) cohort, estimated associations with dia betes were strongest for the second quartile of exposure to transnonachlor, oxychlordane, mirex, highly chlorinated PCBs, and PBB153 . Other studies (Lee et al. 2011a;RignellHydbom et al. 2009;Turyk et al. 2009a) reported monotonic relation ships. A closer evalua tion of the dose-response curves from each of these studies (Lee et al. 2011a;RignellHydbom et al. 2009;Turyk et al. 2009a) revealed that the risk of diabe tes was substantially increased with only small increases within the lower ranges of POPs con centrations, but only slightly increased with higher increases in concentrations of POPs. For example, in the PIVUS (the Prospective Investigation of the Vasculature in Uppsala Seniors) study, the adjusted ORs across quin tiles of summary measures of PCBs were 1.0, 4.5, 5.1, 8.8, and 7.5 (Lee et al. 2011b).
In this sense, the dose-response curves presented in these studies share the lowdose portion of a wide inverted Ushaped associa tion. Varying background exposure distribu tions may contribute to different forms of the concentration-response curves seen between studies, depending on the relative importance of different POPs in the background mix ture. The inverted Ushaped association has been suspected to be biologically linked to the endocrinedisrupting properties of POPs because an increase from no to low occupancy of hormone receptors has been observed to have linear effects on hormonemediated phe nomena, but that effect sometimes deceler ated or even stopped when the dose increased (Vandenberg et al. 2012). Thus, improving understanding of the biological basis for potential non linear relation ships was consid ered by the workshop participants to be an important research need (Appendix 1).
Meta-analysis or pooled analysis of existing studies. The workshop participants discussed the possibility of conducting a metaanalysis of existing studies, or a pooled analysis of individuallevel data from prospective studies, in particular the five prospective studies of PCB153 and DDE (Lee et al. , 2011aRignellHydbom et al. 2009;Turyk et al. 2009a;Vasiliu et al. 2006). However, the participants concluded that there was too much variation across studies to permit a meta analysis or pooled analysis. For example, the five studies of PCB153 and DDE mentioned above used different diagnostic strategies and approaches to address confounding, particu larly by serum lipid levels . The cohorts also varied with regard to age, from 18 to 30 years  to 70 years (Lee et al. 2011a), and sex, which was exclusively female in one study (RignellHydbom et al. 2009), exclusively male in another , and mixed in the remaining cohorts (Lee et al. 2011a;Turyk et al. 2009a;Vasiliu et al. 2006). In addition, temporal and geographic variation among the cohorts resulted in substantial differences in the chemical mixtures to which the populations were exposed as well as the duration and relative concentrations of exposures.
Causality. Although several organochlo rine compounds showed positive associations with T2D, we cannot determine whether these associations are causal in nature based on observational epidemiologic studies alone; additional animal and in vitro mechanistic studies are needed to clarify the role of POPs in metabolic disease development. Factors to be considered in such studies should address the influence of time windows of exposure; exposure measurements (e.g., the chemi cal analysis of individual POPs); chemical mixtures identifying relevant tissue targets; biologi cal mechanisms that lead to obesity, insulin resistance, lipidemia, and diabetes; and the influence of genetic variation among ani mal models. Combining results from relevant mechanistic and animal studies with findings from epidemiologic studies would enhance our ability to establish a possible causal linkage between POPs and diabetes.
Identification of individual chemicals or chemical mixtures that are associated with T2D in epidemiology studies will help direct further toxicity testing. The combined use of toxicity testing and screening of chemical classes using assays relevant to diabetes will also help epidemiologists determine which chemicals to measure in future studies. The structures of chemicals that are associated with diabetes are highly variable, and it is difficult to link them to a common etiologic mecha nism. Further research to identify all relevant pathways to diabetes will aid in deciphering structure-activity relationships.
However, the laboratory animal data on organochlorineinduced changes in glucose and insulin levels are not necessarily consis tent with associations between POPs and an increased incidence of T2D reported by epidemiologic studies (Everett et al. 2007;Uemura et al. 2008). It is unclear whether the lack of consistency results from physio logical differences between rodents and humans in the development of diabetes, or from experimental variables related to dif ferences in exposure levels, the window of exposure, and/or the duration of exposure and length of followup. Much of the work in this area is based on TCDD exposure. In humans, diabetes is charac terized by increased blood glucose levels. In contrast, in differ ent animal models, TCDD has been shown to cause hypo glycemia (Fried et al. 2010;Gorski and Rozman 1987;Viluksela et al. 1998Viluksela et al. , 1999, to have no effect on glucose levels (Unkila et al. 1995), or to cause both hyper glycemia and hypo glycemia at different time points during or after dosing (Ebner et al. 1988;Potter et al. 1983). Although epidemiology studies tend to show a posi tive relationship between TCDD body bur dens and insulin levels (Cranmer et al. 2000;Michalek et al. 1999), TCDD typically causes hypo insulinemia and increased insulin sen sitivity in animals (Ebner et al. 1988;Fried et al. 2010;Gorski et al. 1988;Gorski and Rozman 1987;Potter et al. 1983;Stahl et al. 1992;Weber et al. 1987). Thus, in animal models, exposure to TCDD mimics the fea ture of reduced insulin secretion observed in the clinical progression of pre diabetes to overt diabetes. Inhibition of glucose uptake may at least partially explain why hypo insulinemia is frequently observed in ani mal studies. In most tissues studied, TCDD inhibits glucose uptake by decreasing the activity or protein level of glucose transporter (GLUT) proteins responsible for transporting blood glucose to adipose, muscular, pancre atic, hepatic, and intestinal epithelial tissue (ElSabeawy et al. 2001;Enan et al. 1992b;Liu and Matsumura 1995;Matsumura 1995;Olsen et al. 1994). Decreased glucose uptake into the pancreas could mean that pancreatic βcells do not sense higher blood glucose levels and therefore do not elicit an insulin response to those levels (Matsumura 1995). The level of glucoseuptake inhibi tion appears to correlate with the activation of the aryl hydrocarbon receptor, which is required for TCDDinduced toxicological effects (Matsumura 1995;Olsen et al. 1994). However, the dioxin exposures in these in vivo and in vitro studies are approximately 1,000-100,000 times background body bur dens observed in the U.S. population. The in vivo studies are associated with body weight loss, histopathological findings, and significant decreases in thyroid hormones. Extrapolating these effects and mechanisms to background human exposures is challenging.

Conclusions
Diabetes is a major threat to public health worldwide (WHO 2011); although there are wellestablished risk factors for diabetes (e.g., excess weight), environmental chemi cals might also contribute to the etiology of this disease. On the basis of our review of human epidemiological studies, we conclude that there is support for positive associations between diabetes and certain chlorinated POPs. We identified a number of research needs (Appendix 1), noting in particular the need to a) better understand the relation ships between both develop mental and adult exposure to POPs and obesity, diabetes, and related metabolic disturbances; b) identify mechanisms for the observed associations, which will require basic research to develop better animal models and identify relevant biological pathways that could be assessed using in vitro screening systems; c) under stand the modifying effects of factors such as inflammation, visceral fat, other chemi cal exposures, genotype, age at exposure, and the duration of exposure; and d) develop improved methods to measure POPs in small blood volumes using high throughput tech nologies at a reasonable cost.
T2D is a debilitating disease that affects adults as well as children and adolescents. The economic impact of the disease is enormous, not only in terms of direct medical costs but also on lost productivity. Therefore, under standing the impact of environmental factors such as chemical exposures is a highpriority research goal (NIDDK 2011). Exposure to environmental chemicals may be an additional risk factor that, if prevented, could facilitate a reduction in disease incidence and in the over all associated health and economic burden.