In the summer of 1973, a fire-retardant product (FireMaster FF-1, Michigan Chemical Corporation, St. Louis, MI) containing polybrominated biphenyls (PBBs) was mistakenly added to animal feed and distributed to farms throughout Michigan. As a result, approximately 295 kg of PBBs entered the food web.1 The first effects became apparent in late September 1973, when animals that ingested the contaminated feed began to exhibit symptoms of a mysterious consumptive disease, the cause of which was not identified until April 1974. As a result, Michigan residents, especially farm families, consumed contaminated meat, eggs, and dairy products containing high levels of PBB for approximately 8 months before any preventive measure was taken. After the contaminant was identified, approximately 30,000 head of cattle, 2000 swine, 400 sheep, and more than 2,000,000 chickens had to be destroyed.1
The Michigan Department of Agriculture identified farms that bought the tainted feed and then measured PBB levels in their animals and products, resulting in a list of quarantined farms. In addition, the Michigan Department of Agriculture identified households that bought and consumed products from these farms. To monitor for potential health sequelae in humans, the Michigan Department of Public Health invited people from these farms and households to enroll in a long-term study. In total, 5076 individuals have been enrolled since July 1976. Follow-up of this cohort has continued to the present time.
PBBs are brominated hydrocarbons, with a structure related to other chemical compounds such as polychlorinated biphenyls (PCBs), dioxins, and furans. One of their most important physical properties is that they have a high flammability point and, for this reason, they have been used as flame retardants. PBBs are lipophilic compounds, with a long half-life (approximately 11 years) in animals and humans and with the potential for biomagnification.2 More than 200 PBB congeners are known; in the Michigan PBB incident, the main congener involved was 2,2′,4,4′,5,5′ hexabromobiphenyl (PBB 153). Another 24 congeners were identified in the FireMaster-FF1® mixture, but in much smaller concentrations.1
As for PCBs, their pollution of the Great Lakes has led to higher levels in fish. Because of the frequent consumption of sport-caught fish, higher concentrations were found in human samples in Michigan in areas that also were affected by the PBB incident.3–5 PCBs are also lipophilic, accumulate in fat, and biomagnify in the food chain. From the 1920s until they were banned in the United States in 1977, PCBs had widespread industrial use in capacitors and dielectric fluids. Reports also have indicated that the half-lives of various PCB congeners can range as long as 70 years.6,7 PBBs and PCBs are hypothesized to be endocrine disruptors. Possible health effects of PBBs and PCBs in humans may include adult-onset diabetes mellitus.
It is commonly accepted that adult-onset diabetes (noninsulin-dependent diabetes mellitus) is caused by insulin resistance or a relative insulin deficit. The most important risk factors have been identified as age, obesity, lack of physical activity, and a family history, indicating a genetic predisposition.8,9
No other published study has investigated whether exposure to PBBs increases the incidence of adult-onset diabetes. However, several studies have reported associations of PCBs and dioxins with diabetes mellitus.10–18 Because of similarities in chemical structures, PBBs may have a comparable effect. We hypothesize that exposure to PBBs and PCBs are risk factors for the subsequent development of adult-onset diabetes in this Michigan cohort.
Enrollment in the Michigan PBB cohort began in 1976. At that time, 4126 subjects entered the cohort, undergoing an extensive survey and the measurement of serum PBB and PCB levels (Fig. 1). A second major survey of the cohort took place between 1991 and 1993, with 3581 subjects participating. The first 2 surveys (1976 and 1991) comprised both general and detailed questions concerning health status and lifestyle factors. The third major survey in 2001, with 3449 participants, consisted of a short mailed questionnaire containing questions on selected health conditions, including diabetes.
For our investigation, we analyzed a subset that resulted after applying several exclusion criteria, as illustrated in Figure 1. Of the 4126 cohort members at enrollment in 1976, we excluded those who were younger than 20 years at enrollment (n = 1850) to focus on adult-onset diabetes and to rule out possible juvenile diabetes cases (more common before 20 years of age).19,20 To estimate the incidence of diabetes, we then excluded individuals who reported having diabetes at enrollment (n = 182). We also excluded study subjects who did not participate in at least 1 follow-up (n = 562). We additionally excluded 148 subjects because of missing or inaccurate data. These exclusions resulted in a final analysis population of 1384.
Questionnaires and Definition of Variables
Three standardized questionnaires were used, one for each survey. The first 2 questionnaires were administered as in-person interviews, whereas the third one was a mail-in questionnaire. Information about diabetes was ascertained by answers to the questions below.
- In the enrollment questionnaire (1): Have you had sugar in your urine, high blood sugar, or diabetes?
- In the 1991 questionnaire (1): Has a physician ever told you that you had diabetes? (2) Has diabetes been a problem during the past year? (3) Have you taken a prescription medication for diabetes during the past year?
- In the 2001 questionnaire (1): Have you ever had diabetes? (2) Did you have diabetes in the past 5 years? (3) Have you seen a doctor for diabetes in the past 5 years? (4) Were you hospitalized for diabetes in the past 5 years? (5) Did you take any prescription medication for diabetes in the past 5 years?
For our analyses, we also used questionnaire information at enrollment on the following characteristics: sex, age, height and weight at enrollment (to calculate body mass index [BMI]), smoking status, and alcohol consumption (liquor, wine, beer, or total abstainer).
PBB and PCB Serum Determinations
Serum samples were analyzed at the Michigan Department of Public Health laboratories using gas chromatography with electron capture detection.21–23 Before being measured for PBB serum levels, the denatured serum sample went through an ether-ethyl or hexane-ether extraction and then through either a Florisil or Florisil and silica gel column. The size of the gas chromatography peak was then compared with that of a control sample containing a known quantity of FireMaster-FF1®. The limit of detection for PBB in serum was 1 part per billion (ppb = μg/L). The method of PBB detection was based on the main PBB congener, 2,2′,4,4′,5,5′ hexabromobiphenyl.
A modified Webb–McCall packed column gas chromatography technique was used to measure serum PCB levels. The methodology for laboratory analysis has been previously reported.24–26 Three different standards were used (Aroclor 1016, 1254, and 1260). An internal comparison conducted by the Michigan Department of Community Health found the results of the 3 standards to be similar. The technical detection limit for PCBs was initially 5 ppb, after 1982, it was 3 ppb. For purposes of statistical analyses, the lowest group of serum levels of PCBs was set at 5 ppb or lower.
To estimate associations between PBB and PCB serum levels and diabetes incidence, we calculated the incidence density (number of diabetes cases/person-years of follow-up) and applied Poisson regression models to estimate incidence density ratios (IDRs) and their associated 95% confidence intervals (CIs).
Our main outcome variable, diabetes, was treated as a dichotomous outcome (yes/no). Diabetes was coded as “yes” if the subject reported having the condition at least once during the second or third survey (as described above), and as “no” if they never reported the condition. In 1991, 84 subjects reported diabetes; 71 of these responded positively to at least 2 of the 3 questions. No one reported prescription medication or diabetes as a “problem” without having a doctors’ diagnosis. Thirty of the 84 cases did not receive prescription medication. In the 2001 mailed questionnaire, individuals were considered to have diabetes if they answered “yes” to the question “ever had diabetes” or, for participants who lacked data on that question, if they answered questions (2) to (5) positively.
Exposure variables were defined using PBB and PCB serum measurements at enrollment. The serum PBB and PCB levels were grouped into 4 levels, based on their distribution within the study group (1 group below detection limits, and tertiles for the remainder: PBBs: ≤1.0 ppb [detection limit], 1.1–3.0, 3.1–7, >7 ppb; PCBs: ≤5, 5.1–7.0, 7.1–10, >10 ppb).
Age at enrollment was categorized into 3 groups: 20–44 years, 45–59 years, and ≥60 years. This approach was chosen in view of the fact that the age of 45 years is considered a critical point in the epidemiology of diabetes mellitus and screening most often is recommended after this age.27,28 BMI at enrollment (kg/m2) was calculated from self-reported weight and height. We grouped BMI into 3 categories (<25, 25–29, and ≥30 kg/m2), reflecting underweight and normal, overweight, and obese.29 We chose to include both the underweight and normal categories in a single group (<25 kg/m2) because of small sample size in the underweight group. Smoking status and alcohol consumption at enrollment were defined using 4 groups each: “total abstainer,” “former user,” “current user,” and “status not available.”
No information was available on the date of diagnosis. The number of person-years of follow-up was calculated from the date the subject enrolled until the date of the first survey when the subject reported having been diagnosed with diabetes; otherwise, the end of the follow-up period is represented by the completion date of the last questionnaire.
We constructed a dataset with groups that represented all combinations of explanatory variables. For each group we calculated the total number of person-years of follow-up (denominator) and the count of diabetes cases (numerator). We used person-years of follow-up as the offset variable when applying the GENMOD procedure. All analyses were carried out using SAS software, release 8.2 (SAS Institute Inc. Cary, NC). Because we found substantially different serum levels in men and women, and because the contaminants may be endocrine disruptors, we stratified all analyses by sex and estimated sex-specific incidence density ratios.
Almost all (99.8%) of the study participants were white. Both sexes were equally represented (49.7% men and 50.3% women). Approximately two-thirds were in the 20- to 45-year-old range at enrollment (Table 1). Regarding BMI, 50% were normal or underweight, 37% overweight, and 13% obese. At enrollment, 60% were nonsmokers, 14% were past smokers, and 24% were actively smoking; we could not establish the smoking status for 24 subjects (2%). We found that 33% of the subjects were total alcohol abstainers, whereas 60% were consuming alcohol in various quantities, and 3% were categorized as former alcohol consumers; we could not determine alcohol consumption for 50 subjects (4%).
Regarding PBB exposure, 355 (26%) of the study subjects had a serum PBB level below or equal to the limit of detection (Table 1). More women than men were in this lowest serum levels group (273 vs 82), whereas more men were in the highest serum levels group. PCBs showed the same pattern of distribution. PBB-serum level categories were distributed equally over the age categories, whereas for PCBs, younger participants more frequently had lower serum levels (38% of the participants 24–44 years of age).
The crude cumulative incidence of diabetes was 13% (180 cases): 12.9% (89 cases) in men and 13.1% (91 cases) in women. The proportion of diabetes cases identified in the second survey who also reported diabetes in the third survey was 81%. The crude cumulative incidence increased with age in women, but not in men (Table 2). This cumulative proportion was highest in study subjects who were obese (≥30 kg/m2): 34% in men and 31% in women. The number of cases did not increase with PBB serum levels, but did increase with higher PCB levels.
Total follow-up time was 30,676 person-years, 15,223 in men and 15,453 in women. Our final model, stratified by sex, included the exposure variables PBB and PCB, along with the covariates age, BMI, and smoking and drinking status at enrollment (Table 3). The highest BMI group had the highest incidence density ratios: 3.52 times higher among men (95% CI = 1.79–6.92), and 3.81 times higher among women (2.11–6.89). Men who were current or former smokers had an increased IDR for developing diabetes (for current smokers, IDR = 1.72; 95% CI = 1.01–2.94; for former smokers 1.99; 1.15–3.44). Women who were current or former smokers also appeared to follow the same pattern of an increased risk.
We found modestly higher IDRs with higher serum PBB levels in women but not in men (Table 3). For PCBs, we detected an increased incidence density ratio of having adult-onset diabetes in women in all 3 higher PCB groups when compared with the reference level.
In the Michigan PBB cohort with a 25-year follow-up, an increased serum level of PBB was not a risk factor for the incidence of self-reported adult-onset diabetes mellitus. However, in women, but not in men, higher PCB serum levels were related to increases in the incidence of diabetes.
An important strength of this study is that PBB and PCB serum measurements were available at enrollment, before to the development of diabetes. Hence, serum concentrations of PBBs and PCBs were not altered as a result of diabetes,18 a circumstance that could have led to a misclassification of exposure. However, reverse causation may have taken place if a participant with a prediabetic state had elevated lipid levels at enrollment, which in turn could have resulted in increased serum concentrations of PBBs and PCBs and the subsequent detection of diabetes. One way to minimize the potential of reverse causation is to increase the latency period. When we excluded the 84 cases that occurred between enrollment in 1976 and the second survey (1991–1993), the IDR for women with highest PCB levels did not change substantially (IDR = 2.42; 95% CI = 1.04–5.65, compared with 2.33 in Table 3). Thus, in our cohort, reverse causation seems unlikely.
Lack of serum lipid levels at the time of enrollment may be considered as a limitation. However, lipid standardization (eg, the division of serum PCB or PBB concentrations by serum lipids) has been reported to be prone to bias.30 One strength of this study is that we had access to a large cohort, which allowed us to precisely define our study population without sacrificing power. Another major strength is that the study population was followed for 25 years, an extended period of time, which provides the opportunity for the development of diabetes.
We do not believe that the choice of our study population (no diabetes at enrollment and >20 years of age, Fig. 1) has introduced a selection bias. However, we were concerned that 562 of 2094 subjects were lost to follow-up, and therefore we investigated possible differences between our study group and those lost to follow-up. The latter group was older and had a higher BMI. Because both age and BMI are known risk factors for adult-onset diabetes, it is possible that we underestimated the diabetes incidence in this population. However, there were no differences in the distribution of PBB and PCB levels between the loss-to-follow-up group and the sample we analyzed.
It is also possible that PBB or PCB exposure may be related to higher mortality (“harvesting effect”), which could lead to an underestimation of the diabetes incidence and, thus, bias our result toward the null. To investigate the likelihood of this bias, we compared PCB and PBB levels for cohort member who were deceased and the rest of the study population, and found no differences in their distributions.
Age and BMI categories, although based on established categories, are wide, which may have led to residual confounding. To investigate this possibility, we repeated our models using a larger number of categories (9 age and 7 BMI classes), and the findings did not change substantially. Again, there was no effect in men but an association was found in women. Compared with the results shown in Table 3, in women the dose–response dependency became more pronounced (IDRs of 1.0, 1.39, 1.57, and 2.32 for the same 4 PCB categories).
Detection bias is also a concern, namely, the fact that self-reporting may not be the ideal way of measuring disease occurrence. However, we believe that in the case of the PBB accident, which received a great deal of publicity, it is likely that study participants were prompted to see their physician and therefore would have received a diagnosis of diabetes if the disease were truly present. In fact, it has been shown that Michigan farmers after the PBB incident had a higher rate of reporting several health conditions and unspecific symptoms when compared with a similar group of Wisconsin farmers.31 However, the prevalence of these disorders was not associated with serum PBB levels. For the enrollment survey, we found evidence that over-reporting of diabetes occurred in all exposure groups. The probability that reporting diabetes in the first survey (enrollment) would be followed by a second diabetes report in the follow-up was only 35% (positive predictive value). However, in this analysis, we excluded all previously existing diabetes cases from the study population to consider only new cases of diabetes. The agreement on reporting diabetes between the second and third surveys was high (80%), indicating a greater degree of reliability.
Another potential source of information bias was the difficulty in differentiating between juvenile and adult-onset diabetes from self-reports. We minimized this bias by only selecting participants who were at least 20 years of age at enrollment because the onset of most of cases of juvenile diabetes is expected to occur before the age of 20.32,33
A further area of potential bias was the fact that the study population was formed by enrollment of farm households as well as farm-produce consumer households. The enrollment by households implies that the individual observations may not be independent. However, additional analysis revealed that the within-household effect was negligible in our study.
An increase in diabetes incidence with higher BMI is evident in both men and women (Tables 2 and 3) and is in accordance with a number of prior studies.34–38 Both current and former smoking showed associations with diabetes in both sexes. Other studies have also reported that smoking is an independent risk factor for diabetes.39,40
Regarding the 2 halogenated compounds, PBBs and PCBs, we found that, adjusting for other risk factors, women in groups with higher serum PCB levels had a 2- to 2.3-fold increased IDR of diabetes (Table 3). This association was not observed in men. Women also showed a modest linear association of increasing IDR for diabetes with PBB levels. In contrast, the incidence density for diabetes in men seemed to decline with increasing PBB levels. We have no explanation for the sex difference found in our study. Longnecker et al18 and Glynn et al17 also reported that PCBs were a risk for diabetes mellitus in women. Other studies with exposures to a mixture of several organochlorines have reported, a higher mortality from diabetes mellitus in workers,41 and a higher prevalence of diabetes in Swedish fisherman and their wives.42
Several potential mechanisms have been suggested to explain the higher relative risk of diabetes with exposure to dioxins/furans or PCBs. Langer et al43 found that antibodies against glutamic acid decarboxylase are increased in factory workers exposed to PCBs, suggesting an immunomodulatory effect. Such antibodies are typical markers among patients clinically classified as either juvenile or adult-onset patients with diabetes.44 Other reports either suggested an antagonism of dioxin with the peroxisome proliferator-activated receptor gamma45 or found that dioxins up-regulated insulin-like growth factor binding protein-1, which is down-regulated by insulin.46 The Michigan PBB cohort offers a unique chance to further evaluate these different mechanisms.
In conclusion, we found no relationship between exposure to PBB and incidence of diabetes. In women, but not in men, there was a two-fold increased incidence density of diabetes associated with higher PCB serum levels at enrollment. This study adds to the body of evidence that PCBs may be a risk factor for adult-onset diabetes in women.
We thank Lawrence Fischer, former director of the former Institute of Environmental Toxicology at Michigan State University, for suggesting this project. Susan Davis provided comments on a previous version of the manuscript.
1. Fries GF. The PBB episode in Michigan: an overall appraisal. Crit Rev Toxicol
2. Rosen DH, Flanders WD, Friede A, et al. Half-life of polybrominated biphenyl in human sera. Environ Health Perspect
3. Swain WR. Effects of organochlorine chemicals on the reproductive outcome of humans who consumed contaminated Great Lakes fish: an epidemiologic consideration. J Toxicol Environ Health
4. Humphrey HEB. Population studies of PCBs in Michigan residents. In: D’Itri FM, Kamrin MA, eds. PCBs: Human and Environmental Hazards
. Boston: Butterworth Publishers; 1983:299–310.
5. Humphrey HE, Budd ML. Michigan’s fisheater cohorts: a prospective history of exposure. Toxicol Ind Health
6. Matthews HB, Dedrick RL. Pharmacokinetics of PCBs. Annu Rev Pharmacol Toxicol
7. Kamrin MA. Pesticide Profiles: Toxicity, Environmental Impact, and Fate
. Boca Raton, FL: Lewis Publishers, Inc.; 1997.
8. O’Rahilly S. Science, medicine, and the future. Non-insulin dependent diabetes mellitus: the gathering storm. BMJ
9. Meigs JB, Cupples LA, Wilson PW. Parental transmission of type 2 diabetes: the Framingham Offspring Study. Diabetes
10. Pazderova-Vejlupkova J, Lukas E, Nemcova M, et al. The development and prognosis of chronic intoxication by tetrachlordibenzo-p-dioxin in men. Arch Environ Health
11. Enan E, Lasley B, Stewart D, et al. 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) modulates function of human luteinizing granulosa cells via cAMP signaling and early reduction of glucose transporting activity. Reprod Toxicol
12. Henriksen GL, Ketchum NS, Michalek JE, et al. Serum dioxin and diabetes mellitus in veterans of Operation Ranch Hand. Epidemiology
13. Pesatori AC, Zocchetti C, Guercilena S, et al. Dioxin exposure and non-malignant health effects: a mortality study. Occup Environ Med
14. Cranmer M, Louie S, Kennedy RH, et al. Exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) is associated with hyperinsulinemia and insulin resistance. Toxicol Sci
15. Longnecker MP, Michalek JE. Serum dioxin level in relation to diabetes mellitus among Air Force veterans with background levels of exposure. Epidemiology
16. Longnecker MP, Daniels JL. Environmental contaminants as etiologic factors for diabetes. Environ Health Perspect
17. Glynn AW, Granath F, Aune M, et al. Organochlorines in Swedish women: determinants of serum concentrations. Environ Health Perspect
18. Longnecker MP, Klebanoff MA, Brock JW, et al. Polychlorinated biphenyl serum levels in pregnant subjects with diabetes. Diabetes Care
19. Leibson CL, O’Brien PC, Atkinson E, et al. Relative contributions of incidence and survival to increasing prevalence of adult-onset diabetes mellitus: a population-based study. Am J Epidemiol
20. World Health Organization. Definition, Diagnosis and Classifications of Diabetes Mellitus and its Complications: Report of a WHO Consultation
. Geneva: World Health Organization, Department of Noncommunicable Disease Surveillance; 1999.
21. Burse VW, Needham LL, Liddle JA, et al. Interlaboratory comparison for results of analyses for polybrominated biphenyls in human serum. J Anal Toxicol
22. Kuwahara SS, Calera F, Perry ES. Distribution of polybrominated biphenyls (PBB) among fractions derived from contaminated human plasma. Transfusion
23. Needham LL, Burse VW, Price HA. Temperature-programmed gas chromatographic determination of polychlorinated and polybrominated biphenyls in serum. J Assoc Anal Chem
24. Humphrey HEB. Evaluation of Humans Exposed to Water-Borne Chemicals in the Great Lakes
. Lansing, MI: Michigan Dept. of Public Health; 1983.
25. Hovinga M, Sowers M, Humphrey H. Historical changes in serum PCB and DDT levels in an environmentally-exposed cohort. Arch Environ Contam Toxicol
26. Price HA, Welch RL, Scheel RH, et al. Modified multiresidue method for chlordane, toxaphene, and polychlorinated biphenyls in fish. Bull Environ Contam Toxicol
27. Mayfield J. Diagnosis and classification of diabetes mellitus: new criteria. Am Fam Physician
. 1998;58:1355–62, 1369–1370.
28. Meltzer S, Leiter L, Daneman D, et al. 1998 clinical practice guidelines for the management of diabetes in Canada. Canadian Diabetes Association. Can Med Assoc J
29. Clinical Guidelines on the Identification, Evaluation, and Treatment of Overweight and Obesity in Adults—The Evidence Report. National Institutes of Health. Obes Res.
30. Schisterman EF, Whitcomb BW, Louis GM, et al. Lipid adjustment in the analysis of environmental contaminants and human health risks. Environ Health Perspect
31. Anderson HA, Lilis R, Selikoff IJ, et al. Unanticipated prevalence of symptoms among dairy farmers in Michigan and Wisconsin. Environ Health Perspect
32. Bingley PJ, Douek IF, Rogers CA, et al. Influence of maternal age at delivery and birth order on risk of type 1 diabetes in childhood: prospective population based family study. Bart’s-Oxford Family Study Group. BMJ
33. Kibirige M, Metcalf B, Renuka R, et al. Testing the accelerator hypothesis: the relationship between body mass and age at diagnosis of type 1 diabetes. Diabetes Care
34. Colditz GA, Willett WC, Stampfer MJ, et al. Weight as a risk factor for clinical diabetes in women. Am J Epidemiol
35. Colditz GA, Willett WC, Rotnitzky A, et al. Weight gain as a risk factor for clinical diabetes mellitus in women. Ann Intern Med
36. Modan M, Karasik A, Halkin H, et al. Effect of past and concurrent body mass index on prevalence of glucose intolerance and type 2 (non-insulin-dependent) diabetes and on insulin response. The Israel study of glucose intolerance, obesity and hypertension. Diabetologia
37. Choi BC, Shi F. Risk factors for diabetes mellitus by age and sex: results of the National Population Health Survey. Diabetologia
38. McPhillips JB, Barrett-Connor E, Wingard DL. Cardiovascular disease risk factors prior to the diagnosis of impaired glucose tolerance and non-insulin-dependent diabetes mellitus in a community of older adults. Am J Epidemiol
39. Ko GT, Chan JC, Tsang LW, et al. Smoking and diabetes in Chinese men. Postgrad Med J
40. Wannamethee SG, Shaper AG, Perry IJ. Smoking as a modifiable risk factor for type 2 diabetes in middle-aged men. Diabetes Care
41. Wong O, Brocker W, Davis HV, et al. Mortality of workers potentially exposed to organic and inorganic brominated chemicals, DBCP, TRIS, PBB, and DDT. Br J Ind Med
42. Rylander L, Rignell-Hydbom A, Hagmar L. A cross-sectional study of the association between persistent organochlorine pollutants and diabetes. Environmental Health: A Global Access Science Source
43. Langer P, Tajtakova M, Guretzki HJ, et al. High prevalence of anti-glutamic acid decarboxylase (anti-GAD) antibodies in employees at a polychlorinated biphenyl production factory. Arch Environ Health
44. Torn C. C-peptide and autoimmune markers in diabetes. Clin Lab
45. Remillard RB, Bunce NJ. Linking dioxins to diabetes: epidemiology and biologic plausibility. Environ Health Perspect
46. Marchand A, Tomkiewicz C, Marchandeau JP, et al. 2,3,7,8-tetrachlorodibenzo-p-dioxin induces insulin-like growth factor binding protein-1 gene expression and counteracts the negative effect of insulin. Mol Pharmacol