Organochlorine Exposure and Incidence of Diabetes in a Cohort of Great Lakes Sport Fish Consumers

1A worldwide epidemic of diabetes is under way. Although obesity is a dominant risk factor, increased risk of diabetes is also related to sedentary lifestyle, poor diet, older age, ethnicity, family history of diabetes, hypertension, and dyslipidemia. Recently, a number of investigations have found associations of diabetes with persistent organic pollutants (POPs), such as polychlorinated biphenyls (PCBs), polychlorinated dibenzo-p-dioxins, and p,p′-diphenyldichloroethene (DDE), which are detectable in most of the U.S. population at relatively low levels. Diabetes has been associated with dioxin-like chemicals, non-dioxin-like PCBs, DDE, and/or other organochlorine pesticides in several cross-sectional investigations.

2Despite this increasing body of evidence, there is no clear evidence of a causal relationship between these exposures and diabetes because the cross-sectional study design does not allow for assessment of temporality of the exposure–disease association. The possibility that diabetes-related metabolic changes may affect POP metabolism, particularly with low-level exposures, increases the importance of investigations using a cohort study design. measured serum DDE + p,p′-dichlorodiphenyltrichloroethane (DDT) levels in 1971–1973 in workers exposed to pesticides and followed 3,669 workers to 1977–1978 for newly diagnosed diabetes. Workers with incident diabetes were found to have significantly higher DDE + DDT at enrollment compared with workers without diabetes. studied 1,384 adults prospectively for 25 years and found that PCBs were associated with incident diabetes in women but not in men and remained associated when cases occurring during the first 15 years of follow-up were excluded, suggesting that reverse causality was an unlikely explanation for the relationship. Investigations of Operation Ranch Hand veterans, who were exposed to 2,3,7,8-tetra-chlorodibenzo-p-dioxin (TCDD) through the application of herbicides in Vietnam, found associations of TCDD exposure with incident diabetes. Furthermore, no evidence was found to support decreased dioxin metabolic rates in relation to diabetes in Vietnam veterans.

3In the present study we examined data from a large cohort of frequent and infrequent Great Lakes sport fish consumers collected during 1992–2005. DDE, but not dioxin-like mono-ortho PCBs or total PCBs (∑PCBs), was related to prevalent diabetes in this cohort in cross-sectional data from 2004–2005. In the present study, we used a cohort study design to investigate the associations of POP body burdens in 1994–1995 with incident diabetes from 1995 to 2005. We also examined the hypothesis that associations of POPs with diabetes are attributable to reverse causality—that is, that metabolic changes related to diabetes slowed POP metabolism resulting in differential metabolism rates by diabetes status, by calculating annual percent change in POPs between 1994 and 2005 and comparing rates in participants with and without diabetes.

4The Great Lakes Consortium for the Health Assessment of Great Lakes Sport Fish Consumption was organized in 1992. Figure 1 shows the time line for the study. Originally, > 4,200 participants with frequent and infrequent Great Lakes sport fish consumption were recruited and completed a telephone survey assessing demographic characteristics such as age, sex, height, weight, and fish consumption habits. The cohort included Great Lakes fishing charter boat captains, anglers who fished in inland Wisconsin lakes, and infrequent consumers (reporting consumption of fewer than six meals of Great Lakes sport fish in any year in the previous 20 years); spouses were included if the participant had had a child in 1970 or later. Based on the sport fish consumption reported in the survey, a stratified sample of 619 (520 frequent and 99 infrequent) Great Lakes sport fish consumers agreed to donate a blood specimen in 1994–1995, and 209 participants donated blood again in 2001–2003. Health information was collected from some of these cohorts in 1995–1996. Finally, health information was collected in 2003 from 1,788 of the initial 4,200 participants, and blood samples were collected in 2004–2005 from 515 these participants. In total, 293 individuals from the original blood collection in 1994–1995 provided a second blood sample for analysis in 2001–2005. The study protocol was approved by institutional review boards at the University of Wisconsin–Madison and University of Illinois–Chicago, and all subjects gave written informed consent before participation.

5Because the study was originally designed as an exposure investigation, data on health conditions were not collected at baseline. Follow-up surveys in 1995–1996, 2001–2003, 2003, and 2004–2005 assessed self-reported diagnosis of diabetes, date of diagnosis, demographics, height, weight, smoking, alcohol use, medication use, and years of sport fish consumption.

6Blood was collected in red-top Vacutainer tubes, allowed to clot for 20 min, centrifuged for 15 min, transferred to solvent-rinsed glass vials, and stored at –20°C until analysis. We analyzed samples for DDE and PCB congeners as previously described. We extracted sera with hexane/ethyl ether, with cleanup and fractionation using Florisil, silica gel, and concentrated sulfuric acid. PCB congeners and DDE were analyzed by gas chromatography. We monitored quality control with method blanks, spikes of bovine serum, duplicates of bovine serum spikes or sample duplicates, surrogate spikes, and confirmation of the analytes by second column or gas chromatography–mass spectrometry, as appropriate.

7Total cholesterol and triglycerides were measured by Quest Diagnostics (Auburn Hills, MI, and Wood Dale, IL) in samples collected in 2004–2005 and by Meriter Laboratories (Madison, WI) in samples collected in 2001–2003. We calculated total serum lipids by the formula total lipid = [total cholesterol (mg/dL) × 2.27] + triglycerides (mg/dL) + 62.3.

8In this report, we present analyses of data from two subgroups of participants (Figure 1). First, we studied incident diabetes from 1995–2005 in participants who were without diagnosed diabetes at exposure assessment in 1994–1995. Second, we examined the effect of diabetes on rates of metabolism of DDE and PCB-132/153 in individuals who had repeat exposure measurements in 1994–1995 and 2001–2005.

9We summed congener-specific values for selected PCBs to yield ∑PCBs (sum of PCB congeners 74, 99, 118, 146, 180, 194, 201, 206, 132/153, 138/163, 170/190, 182/187, and 196/20). We included PCB-118, a dioxin-like mono-ortho PCB congener, in the sum and also examined it separately. Values for individual PCB congeners that were below the limit of detection (LOD) were imputed as zero. Analyses repeated with values below the LOD imputed as the LOD/2 for each specific congener yielded similar results (data not shown).

10In the group of participants with PCB and DDE (n = 619) measurements in 1994–1995, we identified incident diabetes through questionnaires administered in 1995–1996, 2001–2003, 2003, and 2004–2005 that assessed diabetes diagnosis and year of diagnosis. In total, 471 participants remained we excluded 127 without follow-up data, 13 with prevalent diabetes in 1994–1995, 1 with juvenile diabetes, and 7 with missing date of diabetes incidence (Figure 1).

11We computed person-years of follow-up for each participant as the amount of time since the contaminants were measured in 1994–1995 to the date of the first diabetes diagnosis or the date of the last completed survey. We calculated incidence rates for diabetes (incidence/1,000 person-years) by dividing the number of cases of disease by the number of person-years of follow-up and multiplying by 1,000. We used the Cox proportional hazards regression model to evaluate the effect of exposure to POPs and sport fish consumption on diabetes incidence, using ordinal exposure variables or natural log transformations of continuous exposure variables. We obtained estimates of parameters and standard errors using PROC PHREG in SAS (version 9.1; SAS Institute Inc., Cary, NC). We retained relevant cofounders, including age, age squared, body mass index (BMI), sex, serum lipids, smoking, alcohol use, all fish meals in the last year, and Great Lakes sport-caught fish meals in the last year, in the models based on biologic and statistical criteria. Although serum lipids, smoking, and alcohol use were not assessed at baseline, we obtained estimates of these variables during the follow-up. Age, sex, and BMI were included in all adjusted models, and additional covariates were added individually to this adjusted model. Confounding was defined as a change of > 10% in the incidence rate ratio (IRR) for the exposure after control for the covariate. We investigated effect modification using variables indicating the product of the potential effect modifier with the exposure, and by stratification of regression models by the potential effect modifier. We performed a sensitivity analysis after exclusion of participants who reported incident diabetes < 6 years after the exposure measurements in order to reduce the potential that associations were influenced by reverse causality.

12In this analysis we included the 293 participants who had repeat organochlorine measurements at 1994–1995 and 2001–2005; four were excluded because diabetes status was missing in all follow-up surveys (Figure 1). estimated dioxin elimination rates based on a first-order kinetic assumption. However, in the present study a half-life for DDE and PCB congeners could not be calculated for all participants because not all participants had decreased POP levels at the later measurements in 2001–2005, perhaps due to continued sport fish consumption. Therefore, we chose to examine the hypothesis that metabolic changes related to diabetes slowed POP metabolism using annual percent change in DDE or PCB-132/153 concentration.

13We calculated annual percent change in DDE and PCB-132/153 as [(follow-up – baseline concentration)/baseline concentration]/time interval in years × 100. Values for DDE and PCB-132/153 that were below the LOD were imputed as the LOD/2. Annual percent changes were log-transformed, after the addition of a constant to move the minimum value of the distribution > 1, to approximate a normal distribution. We compared annual percent change for DDE and PCB-132/153 in participants who reported diabetes at any time during the study with those who reported no diabetes using least squares geometric means and 95% confidence intervals (CIs) estimated by analysis of covariance. We adjusted estimates for age, sex, BMI at baseline, percent change in BMI, and log baseline exposure level. We repeated this analysis using DDE and PCB-132/153 half-lives calculated for participants who had a decrease in body burden at the 2001–2005 measurement (half-lives were calculated as described by ), which yielded similar results to the analysis of annual percent change in DDE and PCB-132/153 (data not shown).

14Participants who were without diagnosed diabetes when they had exposure measurements in 1994–1995 were followed for incident diabetes for an average of 8.4 years. Follow-up data was not available for 127 participants. These participants were more likely to be female, were significantly younger, and had lower ∑PCB and DDE levels compared with those with follow-up data (data not shown). Table 1 shows the characteristics of the participants with follow-up.

15A total of 36 cases of diabetes were reported after the initial exposure assessment, for an incidence rate of 9.1/1,000 person-years in the entire cohort, 6.5/1,000 person-years in women, and 11/1,000 person-years in men (data not shown). We calculated incidence rates for subgroups of the cohort: infrequent Great Lakes sport fish consumers, Wisconsin inland lake anglers, and Great Lakes captains who fished from Lake Michigan, Lake Huron, or Lake Erie. Incidence/1,000 person-years was 6.9 in infrequent Great Lakes consumers and was higher for sport fish consumers for all lakes (Michigan, 9.5; Huron, 10.6; Wisconsin anglers, 13.3) except Lake Huron (4.4).

16Geometric mean DDE, PCB-118, and ∑PCBs, but not years of sport fish consumption, were higher in participants who subsequently developed diabetes (Table 1). However, in proportional hazard models adjusting for age, BMI, and sex, only DDE remained associated with incident diabetes (Table 2). Age confounded all diabetes–exposure associations, and BMI confounded the associations of diabetes with DDE and PCB-118 (data not shown). A one-tertile increase in DDE corresponded to an IRR of 2.01 (95% CI, 1.20–3.66; p = 0.008) (data not shown), and a 1-natural-log increase in DDE corresponded to an IRR of 1.52 (95% CI, 1.03–2.24; p = 0.04; data not shown). The association DDE with incident diabetes remained significant with further adjustment for smoking, alcohol use, serum triglycerides, serum cholesterol, fish meals in last year, Great Lakes sport fish meals in last year, years of sport fish consumption, PCB-118, and ∑PCBs (Table 2).

17When the 15 participants who developed diabetes during the first 6 years of follow-up were excluded from the analysis, the association of DDE with incident diabetes remained significant (IRR for increase in 1 tertile = 2.37; 95% CI, 1.20–4.66; p = 0.01) (data not shown).

18We found no synergistic and/or antagonistic effects among exposures or between BMI or age and exposures in models using multiplicative interaction terms for exposure, BMI, and age tertiles (data not shown). In analyses stratified by BMI, the IRR for a one-tertile increase in DDE was ≥ 1.7 in all BMI strata, although the number of diabetes cases in the two lower-BMI groups was small and the IRRs in did not reach significance (Table 3). We found similar results in age-stratified models (Table 3). In sex-stratified analyses, the associations of DDE with incident diabetes were of borderline significance (p-value > 0.05) (Table 4), but the association of DDE with diabetes reached significance in men after further adjustment for ∑PCB tertiles.

19We examined the hypothesis that associations of POPs with diabetes are attributable to reverse causality—that is, that metabolic changes related to diabetes slowed POP metabolism, resulting in differential metabolism rates by diabetes status. We estimated metabolism using annual percent change in DDE and PCB-132/153 in 289 participants who had repeat measurements of serum PCBs and DDE in 1994–1995 and 2001–2005 and known diabetes status. Least-squares geometric means of the annual percent change in DDE and PCB-132/153 were not significantly different in participants with and without diabetes, with or without adjustment for age, sex, BMI, percent change in BMI, and natural log baseline exposure (Table 5).

20In this investigation we found an association of DDE exposure, but not exposure to dioxin-like PCB-118, ∑PCBs, or years of eating Great Lakes sport fish with incident diabetes in a cohort of sport fish consumers.

21The results of the present study are consistent with the general finding that exposure to lipid soluble POPs is related to diabetes [reviewed by ]. Associations with a single chemical are difficult to interpret because of concomitant exposures to other POPs, although adjustment for multiple exposures may be imprecise because exposures can be strongly correlated due to similar exposure routes. Unmeasured exposures as well as the potentially beneficial effects of omega-3 fatty acids on diabetes might confound associations of diabetes with POPs in populations with high fish consumption, although our results did not suggest that there is confounding by fish-related exposures.

22We noted consistent, dose-related associations of DDE with incident diabetes. This is similar to our findings from cross-sectional analyses of the same cohort. Other investigations have also found an association of DDE or DDT with diabetes. Several investigations concluded that pesticide exposure was related to diabetes, but not all have noted associations with DDT application.

23Our failure to find associations with non-dioxin-like PCBs is also similar to our cross-sectional analyses but not to the findings of most other cross-sectional investigations. Our results also did not confirm the association of PCB exposure with incident diabetes in women found by, a cohort assembled because of accidental dietary exposure to polybrominated biphenyls rather than high Great Lakes sport fish consumption. Potential explanations for these divergent results include lower exposure levels, shorter follow-up time, and smaller number of participants in our cohort.

24The lack of an association of diabetes with dioxin-like PCB-118 is also consistent with our cross-sectional data. However, our estimate of dioxin-like exposure is limited by our measurement of a single dioxin-like mono-ortho PCB congener, possibly biasing the findings toward the null hypothesis. Several investigations have examined the possibility that dioxin-like exposure is related to increased diabetes. The evidence is more consistent for lower-level exposures than for higher exposures, particularly for high-level TCDD exposures in workers from facilities involved in the manufacture of TCDD-contaminated herbicides [reviewed by ].

25POP metabolism is affected by body fat and changes in body fat. We adjusted models for BMI, although if BMI is in the causal pathway this adjustment may be inappropriate. In the present study, BMI did not modify the effect of DDE on incident diabetes, but effect modification by body weight was found in several cross-sectional studies.

26We did not find sex differences in the effects of POPs on diabetes, as noted in several other studies. In this cohort, PCBs, DDE, years of sport fish consumption, and age were higher in men than in women. Although POP body burdens can be decreased by lactation, the differences in PCB and DDE body burdens were explained primarily by years of sport fish consumption and age, respectively, and not by lactation.

27The higher incidence of diabetes in the men (11/1,000 person years) than in women (6.5/1,000 person years) may be related to higher DDE body burdens in the men. Similarly, the higher diabetes incidence rates in three of the four sport-caught fish consumer groups compared with the infrequent Great Lakes sport fish consumer group could reflect the higher DDE body burdens in the anglers. However, the group with the highest DDE body burdens [Lake Michigan] did not have the highest diabetes incidence. Factors that might be related to the low diabetes incidence rate in the Lake Huron anglers are not obvious. Between 1995 and 2005, crude and age-adjusted diabetes incidence rates for adults 18–79 years of age were 5.9 and 6.1 per 1,000, respectively. Extrapolating the CDC rates to the present study, we would have expected approximately 23–24 incident diabetes cases in our study population, if the age distributions were similar.

28The present investigation has a number of limitations. As in any observational study, the association between DDE and diabetes does not establish causality. This investigation was not originally designed as a health study, so some important risk factors for diabetes were not determined at baseline, including smoking, alcohol use, serum lipid measurements, physical activity, and family history of diabetes. However, smoking, alcohol use, and serum lipids were estimated from follow-up data, and did not confound the effect estimates for DDE. did not find substantial attenuation of the association of TCDD with incident diabetes after adjustment for triglycerides, nor did they see effect modification by triglycerides. We assessed diabetes by self-report, which could have resulted in underdiagnosis and/or misclassification of disease. In a cross-sectional sample from our cohort taken in 2004–2005, we tested sera for hemoglobin A1c, which has been used as an indication of undiagnosed diabetes. In the cross-sectional sample, we found elevated hemoglobin A1c (> 6.3%) in 4% of the participants without diagnosed diabetes, and 92% of those reporting diabetes also reported diabetes medication use or had elevated hemoglobin A1c levels. These findings suggest that some underdiagnosis, but not misclassification, of diabetes is likely to also be present in the current study.

29The size of the study sample with only 10 years of follow-up limited our ability to examine the concurrent effects of multiple POP exposures. For DDE tertiles, the large 95% CIs in part reflect the small number of cases in the lowest tertile, which was the reference category. However, we also found a significant association of diabetes with the natural log of DDE. Loss to follow-up for the incidence analysis was greater for females, younger participants and those with lower organochlorine body burdens, but because diabetes disease status of these participants is not known, it is difficult to speculate whether these losses bias the analysis.

30The strengths of this study include the longitudinal collection of POP body burdens and survey data in a cohort with exposures near the general population of the United States and the measurement of multiple exposures. Metabolic changes related to diabetes could be present years before diagnosis, and results of prospective studies could be affected by reverse causation if the time between exposure measurement and diagnosis is short. However, in the present study, when we excluded participants who developed diabetes during the first 6 years of follow-up, the association of DDE with incident diabetes remained significant (p = 0.01), suggesting that reverse causality was an unlikely explanation for the relationship. Similarly, in the, the association of PCBs with diabetes remained when cases occurring during the first 15 years of follow-up were excluded. Additionally, the analysis of annual percent change in DDE and PCB-132/153 suggests that diabetes did not affect metabolism rates of these chemicals. A similar investigation also found no evidence to support decreased TCDD metabolic rates in relation to diabetes in Operation Ranch Hand veterans.

31Several mechanisms through which POPs, predominantly dioxin-like chemicals, could affect diabetes incidence have been investigated, including hyperinsulinemia, antagonism of peroxisome proliferator-activated receptor-γ (PPAR-γ) expression, induction of tumor necrosis factor-α (TNF-α), auto-immunity, and alterations in steroid metabolism. Dioxin exposure has been related to hyperinsulinemia in persons without diabetes. Dioxin suppressed PPAR-γ protein expression, induced TNF-α secretion, and decreased glucose transport and lipoprotein lipase activity in vitro and decreased glucose transporter activity in mice. Elevated levels of autoantibodies associated with diabetes have been found in PCB-exposed workers. Potential mechanisms for the effects of DDE on diabetes have not been well studied, but DDE is known to have antiandrogenic effects, whereas DDT and other metabolites of DDE are estrogenic. Prospective studies suggest that testosterone may differentially modulate risk of diabetes in men and women, whereas sex hormone–binding globulin decreases risk in both men and women. Further mechanistic studies are needed in animals and in humans, which should include biological measurements of risk factors for diabetes.

32The present study showing an increased incidence of diabetes among those exposed to higher levels of DDE at baseline, along with the observation that diabetes status did not affect annual percent change in DDE and PCB-132/153, confirms previous cross-sectional studies and suggests the possibility of a causal relationship. Future studies should address the biological mechanisms by which DDE may be affecting glucose homeostasis.