Environmental contaminants as etiologic factors for diabetes.

For both type 1 and type 2 diabetes mellitus, the rates have been increasing in the United States and elsewhere; rates vary widely by country, and genetic factors account for less than half of new cases. These observations suggest environmental factors cause both type 1 and type 2 diabetes. Occupational exposures have been associated with increased risk of diabetes. In addition, recent data suggest that toxic substances in the environment, other than infectious agents or exposures that stimulate an immune response, are associated with the occurrence of these diseases. We reviewed the epidemiologic data that addressed whether environmental contaminants might cause type 1 or type 2 diabetes. For type 1 diabetes, higher intake of nitrates, nitrites, and N-nitroso compounds, as well as higher serum levels of polychlorinated biphenyls have been associated with increased risk. Overall, however, the data were limited or inconsistent. With respect to type 2 diabetes, data on arsenic and 2,3,7,8-tetrachlorodibenzo-p-dioxin relative to risk were suggestive of a direct association but were inconclusive. The occupational data suggested that more data on exposure to N-nitroso compounds, arsenic, dioxins, talc, and straight oil machining fluids in relation to diabetes would be useful. Although environmental factors other than contaminants may account for the majority of type 1 and type 2 diabetes, the etiologic role of several contaminants and occupational exposures deserves further study.

Our charge was to assess whether monitoring of diabetes rates could be used as a method for detecting community exposure to critical pollutants. This question arose from the observation that the frequency of hospital admissions for diabetes varied substantially among several areas (Areas of Concern) on the border of the United States and Canada (1). The possibility that a corresponding distribution of toxic substances might account for the variation was suggested by scientists at the International Joint Commission, who advise the two governments on the Great Lakes Water Quality Agreement (2). We responded to their charge by reviewing data on whether environmental factors might be responsible for variation in rates of diabetes.
The notion that environmental contaminants could increase risk of diabetes is fairly new (3)(4)(5). That occupational exposure could increase risk, however, has been recognized since the 1970s, when an association with carbon disulfide was reported (6). Here we review the relevant epidemiologic data, which can be categorized as follows: type 1 diabetes in relation to nitrates, nitrites, and nitrosamines, and in relation to polychlorinated biphenyls (PCBs); and type 2 diabetes in relation to arsenic, 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD), and occupational exposures. We included occupational exposures in our review because they could affect local disease rates if the area around the industry were polluted or if a large proportion of the work force in a given area were employed by one industry. We begin with a brief description of the main types of diabetes and their epidemiology.
Type 1 diabetes results from decreased insulin production by the pancreatic β cells. β-cell deficiency is due to autoimmune processes, or, in some cases, to idiopathic destruction (7). The autoimmune process against β cells is thought to be triggered by a combination of genetic predisposition and environmental factors. The concordance of type 1 diabetes among monozygotic twins is 20-35% (8,9), suggesting that environmental factors play a large role in the etiology. The environmental factors usually considered etiologically relevant are infectious agents or dietary factors that stimulate an immune response (10). Poisonings with the rodenticide Vacor (Rohm and Haas Co., Philadelphia, PA), however, have caused type 1 diabetes (11). Furthermore, certain drugs, such as pentamidine, can be toxic to β cells (12). Agents that cause type 1 diabetes in animal models act through a variety of mechanisms (13), though all rely on toxins fairly specific for pancreatic β cells, such as alloxan and streptozotocin. Vacor, alloxan, and streptozotocin all include a urea structure; streptozotocin is also an N-nitroso compound.
The onset of type 1 diabetes is typically before adulthood. The incidence in whites is greater than in blacks or Asians (14). The incidence of type 1 diabetes has been increasing worldwide for approximately 40 years, with an average yearly increase in incidence of 3% (15). A typical incidence rate is 10/10 5 person-years (world population standard), though rates vary markedly and are much higher in selected developed countries. Clustering of cases of type 1 diabetes, diagnosed among children who were together during a defined period, further supports an environmental component to the etiology (16)(17)(18). These cluster studies, however, offer little that allows one to distinguish the effect of contaminants from those of infectious agents or exposures that stimulate an immune response.
Type 2 diabetes is due to resistance to insulin action and a relative deficiency of insulin. Age, obesity, central adiposity, lack of physical activity, and dietary glycemic load are the main factors identified as responsible for the disease (19). The concordance rate among monozygous twins is about 30% (9). Whether chemical agents can cause type 2 diabetes in humans is not as clearly established as for type 1 diabetes, though suggestive data exist (6). Many drugs, however, exacerbate type 2 diabetes (20). As with type 1 diabetes, animal models of type 2 diabetes rely on a variety of mechanisms, and many include an element of impaired insulin action (21).
The onset of type 2 diabetes is typically during adulthood. Disease is more frequent among blacks, Mexican-Americans, and Native Americans (14). The prevalence of diabetes of all types was 6.5% in the United States in 1998 (22), with approximately 90-95% of cases due to type 2 diabetes (23). The prevalence of type 2 diabetes in the United States has increased by 33% during the past decade (22,23). This increase has been attributed to the rise in the prevalence of obesity (22). Worldwide the prevalence of type 2 diabetes varies roughly 10-fold, and the number of people with diabetes has increased 11% in the past 5 years (24).
For both type 1 and type 2 diabetes mellitus, the rates have been increasing in the United States and elsewhere; rates vary widely by country, and genetic factors account for less than half of new cases. These observations suggest environmental factors cause both type 1 and type 2 diabetes. Occupational exposures have been associated with increased risk of diabetes. In addition, recent data suggest that toxic substances in the environment, other than infectious agents or exposures that stimulate an immune response, are associated with the occurrence of these diseases. We reviewed the epidemiologic data that addressed whether environmental contaminants might cause type 1 or type 2 diabetes. For type 1 diabetes, higher intake of nitrates, nitrites, and N-nitroso compounds, as well as higher serum levels of polychlorinated biphenyls have been associated with increased risk. Overall, however, the data were limited or inconsistent. With respect to type 2 diabetes, data on arsenic and 2,3,7,8-tetrachlorodibenzo-pdioxin relative to risk were suggestive of a direct association but were inconclusive. The occupational data suggested that more data on exposure to N-nitroso compounds, arsenic, dioxins, talc, and straight oil machining fluids in relation to diabetes would be useful. Although environmental factors other than contaminants may account for the majority of type 1 and type 2 diabetes, the etiologic role of several contaminants and occupational exposures deserves further study. Regardless of type, patients with diabetes mellitus are at increased risk of small and large blood vessel disease and hyperlipidemia, resulting in retinopathy, neuropathy, vascular diseases such as myocardial infarction, stroke, aneurysm, and kidney failure, and they are also at increased risk of depression (25). The cost of diabetes in the United States in 1992 was estimated at $90 billion (26).
For both type 1 and type 2 diabetes, genetic factors by themselves appear to account for less than half of the disease; incidence rates have increased over relatively short periods, and incidence rates vary widely across geographic areas. These observations suggest that environmental factors, broadly defined, account for much of the disease (27). Temporal and spatial variations in disease frequency, however, are nonspecific with respect to the etiologic agents involved (28). Because environmental contaminants that are diabetogenic in humans are plausible, we will consider the evidence that any might be associated with risk of diabetes. In our discussion we briefly address whether the variation in rates of diabetes in the Areas of Concern might be related to the contaminants identified in this review.

Nitrates, Nitrites, and Nitrosoamines
In the gastrointestinal tract, nitrates can be converted to nitrites, and nitrites can react with amines to form N-nitroso compounds. Because of this interrelatedness, we have considered the data for these nitrogencontaining compounds together. Drinking water can be contaminated with nitrates resulting from fertilizer application. Foods contain nitrates, nitrites, and N-nitroso compounds.
In 1981 Helgason and Jonasson (29) first drew attention to the possibility that N-nitroso compounds may cause type 1 diabetes in humans. Consumption early in pregnancy of cured mutton, a source of N-nitroso compounds, was followed by a high incidence of type 1 diabetes in male offspring. They proposed that dietary nitrosamine activity was inhibited by estrogen and promoted by testosterone, as with streptozotocin, thus accounting for the male specificity. Helgason et al. (30) subsequently induced diabetes in the progeny of mice fed N-nitroso-laden mutton, and males were preferentially affected. Subsequent in vitro work revealed that selected N-nitroso compounds were especially toxic to pancreatic β cells (31). Associations between regional water nitrate levels and the incidence of type 1 diabetes have been reported in three ecologic studies (Table  1). Kostraba et al. (3) found a positive correlation even though the range of average nitrate levels was relatively limited. The association, however, appeared to be influenced greatly by data from one or two counties. Parslow et al. (32) reported that the rate of diabetes was 30% higher among those in water supply zones with average nitrate levels of 14.9-40 mg/L compared with those in areas with means less than 3.2 mg/L. Van Maanen et al. (33) reported a 50% increase in rates when nitrate levels exceeded 25 mg/L, compared with levels less than 10 mg/L, though the confidence interval (CI) for this increase was wide and included one. The average nitrate concentrations in all areas of all three studies were below the current World Health Organization standard (50 mg/L) (34). Although the data from these studies are consistent with the nitrate hypothesis, they provide only modest support because of the possibility that correlates of water nitrate level could account for the associations observed. This possibility is better addressed by data from case-control studies. In addition, drinking water exposure data could possibly be improved if seasonal variation in nitrate levels were assessed.
Data from case-control studies on childrens' dietary intakes of nitrogencontaining compounds and type 1 diabetes (35)(36)(37) show a mixture of results (Table 2). Virtanen et al. (36) estimated intake of nitrate and nitrite from drinking water and found no relation with risk (not shown in table). The intake of nitrates and nitrites from food was much greater than from water (36). Because the absolute intake of Nnitroso compounds, nitrates, and nitrites was not reported in most studies (35,37), it is possible that low exposure levels in some studies was responsible for lack of association (37). A further uncertainty is whether nitrate is equally toxic in food and in water.
In summary, some data suggest an association between intake of nitrogen-containing compounds and risk of type 1 diabetes. Overall, however, the data are limited and inconsistent.

Polychlorinated Biphenyls
In a recent small study of pregnant women, serum levels of PCBs were 30% higher among those with diabetes (primarily type 1) than among those without (38). Because the study was cross-sectional, whether the association was causal could not be determined. If the association is confirmed by others, studies designed to assess whether the association is causal would be in order.

Potential Risk Factors for Type 2 Diabetes
Arsenic Epidemiologic data for populations with high exposure to arsenic, including selected industrial groups, are generally consistent with an increased risk of type 2 diabetes (Table 3). In several studies of other populations with high arsenic exposure, investigators have not specifically reported results for diabetes (44)(45)(46)(47), raising the possibility that no notable associations were present. Compared with the arsenic exposure levels among the populations represented in Table 3 (4,39,40), exposure levels in the general U.S. population are much lower, with a mean drinking water level of about 0.001 mg/L (48). Within the United States, some areas have higher levels of exposure, for example, in parts of Utah where the average water arsenic levels are roughly 0.1 mg/L (49). A study among the Utah population that had increased exposure (49) showed that the overall rate of death from diabetes was not increased compared with the rate from the rest of Utah.
Arsenic is metabolized in vivo to trivalent arsenic. A trivalent arsenical, phenylarsine oxide, has adverse effects on the insulin receptor and glucose transport in in vitro experiments (50).
The available epidemiologic data on arsenic and diabetes are suggestive but inconclusive because of the limited number of studies, their small size, and the possibility of publication bias. The available data do not address the dose-response issue in detail. If arsenic exposure via drinking water does increase risk of type 2 diabetes, this may occur only among those consuming water with an arsenic concentration of more than 0.1 mg/L. This is potentially an extremely serious problem in Bangladesh, where up to 30 million people may be drinking arseniccontaminated water (51).

2,3,7,8-Tetrachlorodibenzo-p-dioxin
We identified 11 reports that addressed the relation of TCDD with type 2 diabetes, hyperglycemia, or hyperinsulinemia ( Table 4). Most of the studies were of workers exposed to TCDD (52)(53)(54)(55)57,59,60). Cranmer et al. (61) studied community members living near a toxic waste disposal site, and Pesatori et al. (58) reported on the experience of those living near a TCDD-laden plume resulting from an out-of-control reaction in a chemical plant. Vietnam veterans exposed to TCDD in Agent Orange were the subjects in the report by Henriksen et al. (56). A group of veterans not exposed to Agent Orange were the subjects in the report by Longnecker and Michalek (5). We excluded from this review two studies that examined diabetes in veterans who had served in the Vietnam War but for whom no serum TCDD levels were available (62,63). Most Vietnam veterans were not exposed to TCDD more than other groups with background-level exposure (64,65); exceptions were veterans in the Chemical Core and in Operation Ranch Hand who came in contact with Agent Orange.
The results of the 11 studies were categorized according to type of outcome and exposure level (Table 5). Of the studies showing an unequivocally positive association, none are very convincing on close examination.
The results from the Seveso study (58) were unequivocally positive among women but not among men. The relative risk of 1.9 (95% CI, 1.1-3.2) among the highly exposed women in Zone B was adjusted for age but not other risk factors.
The group of highly exposed Czech workers (mean age 46 years) (52) had a higher prevalence of diabetes than those 20-79 years of age in the Czech Republic (24). But in the absence of a statistical comparison of prevalences and the possibility that confounding factors accounted for the increase, whether the prevalence of diabetes was notably greater than expected remains in doubt.
Although the study by Henriksen et al. (56) was a cohort study, the TCDD serum levels used to assess exposure were measured within a few years of when diabetes and related outcomes were ascertained. Thus, if diabetics or subjects with subclinical glucose intolerance had a slower rate of excretion of TCDD, this could account for the association observed (66). Furthermore, the prevalence of diabetes in veterans exposed to Agent Orange was not greater than in the unexposed comparison group. The results from the study by Longnecker and Michalek (5) suffer from the same weakness outlined for the study by Henriksen et al. (56).
The positive finding in the study by Cranmer et al. (61) was an association of TCDD level with hyperinsulinemia on a glucose tolerance test (GTT). Although this is consistent with a TCDD-type 2 diabetes relation, no association with glucose was found.
Environmental Health Perspectives • VOLUME 109 | SUPPLEMENT 6 | December 2001 Table 3. Risk of type 2 diabetes in groups highly exposed to arsenic relative to less-exposed groups, by type of exposure.  Abbreviations: -, inverse relation; ∅, no association; ±, equivocally positive association; +, unequivocally positive association. a Study design was case series; n = 1 based on one-fifth of cases with abnormal GTT. b Results for females were unequivocally positive. c Increased insulin level was the positive finding. Mortality studies Medium-high exposed | | | Morbidity studies Low-moderate exposed || Medium-high exposed | | || || Abbreviations: -, inverse relation; ∅, no association; ±, equivocally positive association; +, unequivocally positive association. a The studies from which results were tallied are listed in Table 4.

Number of
Several of the studies with equivocally positive results also have notable weaknesses. The populations studied by Ott et al. (54) and Zober et al. (55) were basically the same, but the results of these two studies appear to be inconsistent with each other (Table 4). Calvert et al. (57) found that the workers with the highest serum TCDD levels also had the highest serum glucose levels compared with those of an unexposed group, but within the group of exposed workers there was no dose-response relation between TCDD levels and serum glucose or prevalence of diabetes.
Overall, the data on TCDD exposure in relation to diabetes and hyperglycemia are mixed. Compelling studies supporting a causal effect of TCDD on diabetes are absent. We note, however, that Enan et al. (67) have shown that TCDD decreases cellular glucose update, thus a diabetogenic effect of TCDD is biologically plausible.

Occupational Exposures
In this section we review data on diabetes in relation to occupational exposures. Data for workers exposed to arsenic and TCDD, however, were considered above with the relevant nonoccupational data. The statistical power of occupational mortality studies, such as those shown in Table 6, to detect increases due to diabetogenic exposures is limited because a) reporting of diabetes on death certificates is highly variable and death certificates reflect less than half of the diabetes among the deceased (80), b) the assessment of exposure may lack sufficient detail, and c) the number of exposed subjects who develop diabetes is relatively small in typical occupational cohort studies. In addition, exposures in the occupational setting are generally mixed and not specific with respect to toxic substances implicated.
Among rubber workers, a moderately increased mortality from diabetes was reported in two cohorts (68,69,81). Two subgroups of rubber workers were identified by McMichael and colleagues (69,81) as having the greatest risk of diabetes: a) those in inspection, finishing, and repair, and b) those in janitoring, trucking, power plant, and test driving. In the report by Andjelkovic et al. (69), the overall standardized mortality ratio (SMR) for diabetes was only slightly elevated at 117, but when the deaths occurring only during retirement were considered, the SMR was 135 (p < 0.05). Similarly, Weiland and colleagues (70) found that mortality from diabetes was greater among retired rubber workers (SMR = 181; 95% CI, 131-244) than among active workers (SMR = 152; 95% CI, 112-201). The greater risk of occupationassociated diabetes seen after retirement (69,70) fits with the hypothesis that an occupation-induced susceptibility to type 2 diabetes could be unmasked by the increased sedentarism and obesity that accompany retirement. Exposure to N-nitroso compounds in the rubber industry has been high (82), although exposure to other agents such as βnaphthylamine, benzene, polycyclic aromatic hydrocarbons, solvents, fumes from vulcanization, and talc has also been frequent.
Data on mortality from diabetes among pulp and paper mill workers have been presented in three reports (71)(72)(73), with a moderate excess of diabetes evident in two (71,72). Although exposure to numerous chemicals and other substances occurs in the pulp and paper industry, potentially notable agents are dioxins and talc.
Among a group of chemical industry workers, deaths from diabetes were double the expected number (74). Among the many exposures in that group, none were specifically linked to diabetes. In a smaller study among chemical and refinery workers, there was no overall excess of diabetes (75). Among the subset who did at least some work in the chemical plant, however, an SMR of 173 was found, although this was not statistically significant (three observed cases).
Dry-cleaning workers have been exposed to several solvents, with tetrachloroethylene the main agent in use since the 1950s (77). In the two studies of laundry and dry-cleaning workers reporting results specifically for diabetes, an excess of death from diabetes was found in one study (76) but not in a second smaller one (77).
Excess mortality from diabetes was also found among workers involved in engine manufacturing (78). Exposure specifically to machining fluid was associated with increased risk. Some machining fluids contain N-nitroso compounds, but the straight oils implicated in this study did not.
In a study of a group of pesticide users and an unexposed group, Morgan et al. (79) found that subjects with diabetes, compared with those without diabetes, had higher blood levels of 1,1,1-trichloro-2,2-bis(pchlorophenyl)ethane (DDT) and its metabolite 1,1-trichloro-2,2-bis(p-chlorophenyl) ethylene (DDE). As with the studies of TCDD, the possibility exists that subjects with diabetes or prediabetes excreted organochlorines (in this case DDT and DDE) more slowly.
An excess of diabetes has also been reported among workers exposed to heat stress (83) and among those with sedentary occupations (81,84). Because these associations were likely due to confounding by body mass index or low physical activity, we considered them outside the scope of this review.
Trivalent chromium is an essential nutrient, and supplementation improves glucose tolerance in most clinical trials (85). A study of tannery workers occupationally exposed to chromium found a lower prevalence of impaired glucose tolerance and of diabetes mellitus than in the control group, even though the exposed workers were more obese; the results, however, were statistically significant only among workers more than 48 years of age (86). In other studies of chromium-exposed tannery workers, investigators have generally not presented results for diabetes (87,88), although in one study exposed workers had an SMR of 130 (95% CI, 67-227) (89).
In summary, the occupational data show for both rubber and pulp and paper mill workers an excess of diabetes deaths in more than one study (68,(70)(71)(72), but results were mixed (69,73). Complete investigations of occupational risks would need to account for the level of physical activity and body mass index associated with a given exposure.

Environmental Contaminants and Increased Rates of Diabetes
Available data on drinking water nitrates do not exclude the possibility that nitrates affect risk of type 1 diabetes, but neither were they strongly supportive of an association. More  Wingren, 1991 (72) Pulp, paper mill 46 130* ,b Wong, 1996 (73) Pulp, paper mill 17 110 Wong, 1984 (74) Chemical 19 220* Marsh, 1991 (75) Chemical, refinery 9 77 Katz, 1981 (76) Laundry, dry cleaning 25 177* ,a Blair, 1979 (77) Laundry, dry cleaning 9 103 a Park, 1996 (78) Engine manufacturing 25 150* ,a Morgan, 1980 (79) Pesticide users 58 >100* ,c a Proportional mortality ratio. b Originally presented as an odds ratio; here shown ×100. c DDE/DDT levels in cases significantly higher than in controls. No SMR or RR was presented, but presumably if one had been presented, it would be greater than 100 if expressed on a scale comparable with the others shown in the table. *p < 0.05. data from case-control studies done in areas where exposure to nitrates in water is unusually high would be particularly useful. The association between PCBs and type 1 diabetes was reported in only one small crosssectional study. The importance of this observation will be clearer if replicated by others, especially using a prospective design.
Data on health effects of arsenic were suggestive of an association with type 2 diabetes. Additional reports on diabetes in groups consuming water contaminated with arsenic, where epidemiologic studies of other outcomes have already been done (44)(45)(46)(47), could be useful. If an arsenic effect exists, however, it is likely to affect few populations because severe contamination is unusual.
TCDD appeared to be the only environmental contaminant identified that could be having widespread affects in the general population. But data among populations with background-level exposure came from just two studies, each with limitations. Prospective data on the relation between TCDD and type 2 diabetes are needed.
Although selected occupational exposures may increase risk of type 2 diabetes, few specific agents were implicated in our review. Additional data on diabetes among those with exposure to N-nitroso compounds, arsenic, TCDD, talc, and straight oil machining fluids would be of interest.

Environmental Contaminants and Local Variation in Rates of Diabetes
The observation that prompted this review was that hospitalization rates for diabetes varied across the Areas of Concern (1). The great majority of diabetes is type 2; therefore, the variation in hospitalization rates is accounted for by corresponding variation either in risk factors for type 2 diabetes or in the medical management of patients. Given that relative body weight is such a strong risk factor for type 2 diabetes, it is highest among the potential suspects accounting for geographic variation in rates of diabetes. Without the ability to take this and other such accepted risk factors for diabetes into account, it would be highly speculative to assume a high rate of diabetes in a given area was due to contamination.
Nonetheless, areas with higher rates of hospitalization could be those containing a substantial portion of the population occupationally exposed to a diabetogenic agent. Among the studies where an occupation was associated with an increased rate of death from diabetes, however, the rate was at most doubled and was usually much lower. An investigation of diabetes among workers involved in engine manufacturing, as in the Park and Mirer study (78), taking occupational physical activity into account, may be worthwhile. (Windsor, Ontario, Canada, one of the Areas of Concern, had an increase in diabetes hospitalizations, is economically like Detroit, and could have this industry there.) Arsenic exposure from drinking water in the Areas of Concern would seem an unlikely culprit, but the possibility of occupational arsenic exposure might be worth investigating. Similarly, substantial variation in contamination of the food supply with TCDD seems unlikely, but again, the possibility of occupational exposure may be worth considering.

β-Cell Toxins and Type 2 Diabetes
Type 1 and type 2 diabetes have no established risk factors in common (see above). In searching for new risk factors, however, it may be worthwhile to consider that there could be overlap in the risk factors for these two types of diabetes. Both type 1 and type 2 diabetes have β-cell insufficiency, to a greater or lesser extent, as part of their pathogenesis. The β-cell toxin streptozotocin, typically used to induce type 1 diabetes in animals, under certain conditions can cause type 2 diabetes (90). Thus, for example, examining nitrosamine intake as a risk factor for type 2 diabetes could be worthwhile.

Summary
With respect to variation in diabetes rates in the Areas of Concern, our review points to no obvious environmental contaminants that might explain the variation in diabetes rates. Several occupations and occupational exposures were identified, however, that may have contributed. In general terms regarding environmental contaminants as etiologic agents for diabetes, data on arsenic and TCDD were suggestive but inconclusive with respect to type 2 diabetes, and for type 1 diabetes data on intake of nitrates, nitrites, and N-nitroso compounds were less suggestive but not completely null. Apart from the exposures considered in this review, other environmental contaminants could be related to risk of diabetes; however, no specific clues were uncovered in the epidemiologic literature.