This study is to our knowledge the largest study to evaluate the potential effects of pesticides on diabetes incidence in adults. The prospective design of the study ensures that exposures were reported prior to the diagnosis of diabetes and reduces the potential for recall bias. Of the fifty pesticides evaluated, seven displayed suggestive evidence of an association with diabetes incidence in both ever use and cumulative days of use models: aldrin, chlordane, heptachlor, dichlorvos, trichlorfon, alachlor, and cyanazine. It is noteworthy that all of these pesticides are chlorinated compounds while only half of the pesticides investigated were chlorinated.
Few studies, if any, have considered the potential diabetogenic effects of alachlor and cyanazine, which both showed a dose-response association with diabetes in the present study. However, the biologic plausibility of a diabetogenic effect of exposure to persistent organic pollutants (POPs, e.g., dioxins, polychlorinated biphenyls, and organochlorine insecticides), and organophosphate insecticides is supported by numerous studies.
Persistent organic pollutants (organochlorine insecticides)
Because POPs are lipid soluble and bioaccumulate in animal tissues, studies of the relationship of chronic exposure to POPs to diabetes can be conducted using human biologic samples (3
). Of the seven pesticides for which the odds of diabetes was increased in both ever-never and dose-response analyses in the present study, three (aldrin, chlordane, and heptachlor) are POPs. Although the organochlorine insecticides in this study are no longer available on the market, measurable levels of these and other POPs are still detectable in the general population and in food products, making these findings potentially relevant to the general population (2
Studies using NHANES data have found associations of POPs with both diabetes and insulin resistance and have noted in particular the association of diabetes with organochlorine insecticides (2
). A metabolite and an impurity of chlordane were most strongly associated with insulin resistance in non-diabetics (12
). Animal studies of exposure to chlordane have demonstrated increased lipids and triglycerides in liver (13
) and altered glucose metabolism (14
). Our finding that chlordane exposure followed a dose-response association with diabetes incidence strengthens the chlordane-diabetes hypothesis. Heptachlor is a frequent component of chlordane mixtures and is structurally very similar (16
), but few studies have considered the diabetogenic actions of heptachlor itself. There is some evidence that heptachlor affects lipid metabolism (17
). Similarly, few studies have examined aldrin in relation to diabetes although it has been shown that aldrin disrupts carbohydrate metabolism in fish (18
Dioxin, a frequenct contaminant of herbicides used for military purposes, is a POP that has been studied repeatedly for its potential diabetogenic effect in humans. Studies have suggested that exposure to dioxin as a contaminant of the herbicide Agent Orange increased the risk of diabetes and disrupted glucose and insulin homeostasis among exposed veterans (20
). Although the number of exposed diabetics was small, our findings that participants who reported mixing herbicides in the military had an increased odds of diabetes incidence compared to participants who did not mix herbicides in the military or who were not in the military are consistent with these studies.
Unlike POPs, organophosphate insecticides (OPs) are readily degraded, and consequently, studies have more frequently been conducted in animal models where the outcome is typically short-term disruption of glucose homeostasis. An advantage of the Agricultural Health Study is the ability to consider the risk of diabetes in humans in relation to long-term exposure to lower levels of OPs. Of the ten OP insecticides investigated, we found seven (chlorpyrifos, coumaphos, diazinon, dichlorvos, phorate, terbufos, and trichlorfon) that had increased odds of diabetes, three of which (chlorpyrifos, diazinon, and trichlorfon) were associated in a dose-dependent manner.
Type 2 diabetes is characterized by insulin resistance, which initially is compensated by an increase in insulin production. Over time, the pancreas fails to produce sufficient insulin to stimulate adequate glucose uptake in adipose and muscle tissues, leading to hyperglycemia and type 2 diabetes. Pancreatic β-cells contain muscarinic acetylcholine receptors, which are involved in the glucose-dependent production of insulin (22
). OPs are known inhibitors of acetylcholinesterase, the enzyme responsible for the degradation of acetylcholine. Thus, exposure to sufficiently high levels of OPs would be expected to result in increased accumulation of acetylcholine, potentially leading to over-stimulation and eventual down-regulation of its receptors (23
) and reducing insulin production.
Indeed, organophosphate exposure has been shown repeatedly to be associated with hyperglycemia in animal models (24
). Dichlorvos specifically has been shown to disrupt glucose homeostasis in male Wistar rats (25
). We found that applicators exposed to dichlorvos had an increased odds of diabetes and that the odds increased with increasing cumulative days of use, although the test for trend was of borderline significance. Furthermore, the pesticide most strongly associated with diabetes among applicators was the organophosphate insecticide trichlorfon, which is converted to dichlorvos in mammals (26
In addition to the effects of short-term exposure to OPs, studies that have considered the effects of long-term, low-level exposure may be of greater relevance to our study population. Studies of long-term exposure to OPs with respect to diabetes in humans have not previously been conducted. However, animal studies have demonstrated that tolerance to OP exposure develops over time, likely as a result of decreased expression of muscarinic receptors (27
). Because these 10 receptors mediate the production of insulin in β-cells, a decrease in muscarinic receptors could potentially lead to decreased insulin production. Additionally, prolonged stimulation by acetylcholine may reduce β-cell sensitivity to glucose (28
As opposed to organophosphate insecticides, the inhibition of acetylcholinesterase by carbamate insecticides is reversible and short-lived, and therefore the effects of exposure would be expected to be less severe. We found that the carbamate insecticides showed very weak, if any, evidence of an association with diabetes in fully-adjusted models. Pesticides from the fungicide and fumigant groups also showed no convincing association with diabetes, indicating that our findings were exclusive to organochlorine and organophosphate insecticides and a limited number of herbicides.
The role of body mass index
Because POPs are lipophilic, people with higher BMI may be more likely to store higher levels of POPs than people with lower BMI with equivalent exposure. A study in the NHANES population found that obesity and diabetes were associated only among participants with detectable levels of POPs (2
). The diabetogenic effect of dioxin exposure has also been shown to be stronger among obese compared to lean individuals (29
). The effects of the seven pesticides that showed an association in both ever-never and dose-response analyses were strongest in obese participants. Although BMI may be related to pesticide exposure in the case of lipophilic compounds, it is not clearly in the causal pathway (i.e., pesticide exposure has not been shown to cause weight gain in adults), allaying the concern that adjusting for BMI would result in an over-adjustment of the effect of pesticide exposure.
One limitation of this study was the use of self-reported diagnosis of diabetes. Among the 1,055 participants who indicated a diagnosis of diabetes at baseline and completed the follow-up 15 interview, 92 percent (972) confirmed this diagnosis in the follow-up interview. This suggests a high level of reliability in self-reporting diabetes in this cohort. Furthermore, it is reassuring that age and BMI were associated with the outcome as these are well established risk factors for diabetes. A second limitation was our inability to control for exercise and diet.
A drawback in studies of occupationally exposed cohorts that has been raised in previous studies is the ability of the group identified as unexposed to represent a truly low-risk group (2
). Inclusion of participants exposed to other potentially diabetogenic pesticides in the unexposed group may have resulted in an underestimation of the true effects.
In regard to age, which is associated with cumulative pesticide exposure and causally associated with diabetes, there may be some concern about residual confounding. However, estimates from models with age treated as a continuous or as a quadratic term were nearly identical to those from the model using age as a categorical variable. In age-stratified analyses, the observation that pesticide effects were more prominent among younger applicators may be due to an increased number of competing risk factors for diabetes at older ages.
There was a strong relationship between diabetes incidence and state of residence; applicators from North Carolina had a two-fold increased odds of diabetes compared to applicators in Iowa, even after adjusting for age, BMI, and smoking. This may reflect differences in health and lifestyle status between the states that were not completely controlled for by age, BMI, and smoking alone. State, therefore, was included in all fully-adjusted models and state-stratified analyses were conducted when necessary.
Although we had relatively good follow-up of the cohort after 5 years, participants who did not complete the follow-up interview were more likely to have had diabetes at enrollment. While the cumulative days of pesticide use did not differ significantly, ever use of pesticide groups was lower among participants who were lost to follow-up. The loss of prevalent diabetics does not necessarily imply the loss of incident diabetics, and it is impossible to know whether the loss of diabetics would be related to level of exposure. However, we cannot exclude the possibility that selection would have biased our results.
Pesticide applicators who reported exposure to certain organochlorine and organophosphate insecticides and two herbicides showed an increased risk of diabetes independent of age, state of residence, and body mass index. These results extend previous findings of persistent organic pollutants and organophosphate insecticides to a much larger cohort where diabetes onset was assessed prospectively and exposure was measured in a semi-quantitative manner. Although based in an occupationally-exposed cohort, the findings may have relevance to the general population in the case of environmentally-persistent chemicals. Apart from organochlorine insecticides, most pesticides in this study are considered general use pesticides and are available to the general public. The increasing burden of diabetes in populations world-wide warrants our improved understanding of the possible relationship of diabetes risk to long-term, low-levels of pesticide exposure.