Persistent Organic Pollutants and Biomarkers of Diabetes Risk in a Cohort of Great Lakes Sport Caught Fish Consumers

Abstract

Background

Exposure to persistent organic pollutants (POPs) is associated with increased diabetes risk, although the mechanism of action is not well delineated.

Methods

We investigated established diabetes biomarkers that could implicate potential mechanistic pathways, including C-reactive protein (CRP), a marker of systemic inflammation; gamma glutamyl transferase (GGT), a liver enzyme associated with oxidative stress; and adiponectin, an adipokine modulating glucose regulation and fatty acid oxidation. These biomarkers as well as hemoglobin A1c (HA1c), and POPs [polychlorinated biphenyls (PCBs), p,p-dichlorodiphenyldichloroethylene (DDE) and polybrominated diphenyl ethers (PBDEs)] were measured in a cohort of Great Lakes sport caught fish (GLSCF) consumers. We examined associations of POPs and fish consumption with HA1c and incident diabetes, and evaluated mediation and moderation by diabetes biomarkers.

Results

Odds of incident diabetes were elevated with exposure to DDE and PCBs. DDE and PCB 118 were positively, and fish meals were inversely, associated with HA1c. CRP was inversely associated with saltwater and total fish meals, particularly in persons with higher adiposity, but did not mediate the associations of fish meals with HA1c. There were few associations of adiponectin, CRP and GGT with POPs, with the exception of positive associations of GGT with PCB 118 and with PBDEs in older persons, and a positive association of adiponectin with PBDEs. Adiponectin, CRP and GGT did not mediate associations of DDE and PCBs with HA1c or incident diabetes. However, the association of DDE with HA1c was stronger in persons with higher CRP, GGT and BMI, and lower adiponectin, while the association of PCB 118 with HA1c was stronger in persons with higher GGT.

Conclusions

These findings suggest that adiponectin, CRP and GGT did not mediate effects of POPs on diabetes or HA1c. However, POPs may have stronger effects on blood glucose in persons at higher risk for diabetes.

1. Introduction

There is accumulating evidence that environmental contaminants are associated with increased diabetes risk. Many studies have investigated the risk of diabetes with exposure to one or more persistent organic pollutants (POPs), including polychlorinated biphenyls (PCBs), dibenzo-p-dioxins and chemicals with related toxicological properties, and persistent pesticides such as DDT and its metabolite p,p-dichlorodiphenyldichloroethylene (DDE) (reviewed by(Kuo et al. 2013;Taylor et al. 2013). While associations have been noted in many studies, chemicals with multiple modes of action have been implicated.

Possible biologic pathways through which POPs could affect diabetes incidence have been hypothesized, including insulin resistance (Kern et al. 2004), pancreatic beta cell destruction (De Tata 2014), mitochondrial dysfunction (Lee 2011), alterations in steroid metabolism (Persky et al. 2011;Persky et al. 2012), antagonism of PPARγ expression (Remillard and Bunce 2002), and induction of low grade chronic inflammation (Fujiyoshi et al. 2006), oxidative stress (Lee et al. 2008), and autoimmunity (Langer et al. 2002). This study explored some of these potential pathways using biomarkers of diabetes risk. Selected biomarkers were C reactive protein (CRP), adiponectin, and gamma glutamyl transferase (GGT). C reactive protein is a general marker of systemic inflammation, which is increased in obese persons and associated with increased risk of type 2 diabetes, independent of obesity (Dehghan et al. 2007;Duncan and Schmidt 2006). Adiponectin, an adipokine, is positively associated with insulin sensitivity, is decreased in obesity and in patients with type 2 diabetes, and has been related to decreased risk of incident diabetes independent of adiposity (Li et al. 2009). Adiponectin has anti-inflammatory properties, including inhibition of tumor necrosis factor α (TNFα) and IL-6 production and induction of anti-inflammatory cytokines (Fantuzzi 2005). GGT is a liver enzyme associated with oxidative stress that is related to increased type 2 diabetes risk, with stronger effects in persons with higher BMI (Nakanishi et al. 2003).

A significant route of exposure to POPs is through ingestion of contaminated food, in general, and fish in particular. Sport caught fish from the Great Lakes have elevated levels of PCBs and other POPs and frequent consumers of these fish have higher exposures than the general population (Turyk et al. 2012). Increased risk of diabetes with elevated POP levels has been noted in Great Lakes fish consuming populations (Codru et al. 2007;Turyk et al. 2009a), but investigations of the effects of fish consumption on diabetes risk have been inconsistent (Wu et al. 2012;Zhang et al. 2013). We previously reported a relationship of DDE with incident diabetes (Turyk et al. 2009a) and of DDE and PCB 118 with prevalent diabetes (Turyk et al. 2009b) in participants in the Great Lakes Fish Consumption Study.

Hemoglobin A1c (HA1c), or glycated hemoglobin, is useful as a measure of glucose control over the lifespan of the red cell (3-4 months). It therefore represents more long-term glucose regulation than acute fluctuations and recently has been adopted as a test for diabetes (≥6.5%) and prediabetes (5.7-6.4%) (American Diabetes Association 2010). Because diabetes risk increases for persons with fasting plasma glucose values at the higher end of the normal range, it has been suggested that diabetes risk prediction may be more accurate if glycemic measures are treated as continuous rather than categorical variables (Tabak et al. 2012). Continuous glycemic measurements (i.e. fasting plasma glucose and HA1c) have been analyzed in relation to POP exposures in several investigations (Calvert et al. 1999;Grandjean et al. 2011;Henriksen et al. 1997;Jorgensen et al. 2008;Langer et al. 2014;Michalek et al. 1999;Suarez-Lopez et al. 2015), but the effects of POP exposures on continuous HA1c levels have not yet been evaluated in the Great Lakes Fish Consumption Study.

The current study measured diabetes biomarkers in participants in the Great Lakes Fish Consumption Study. Our purpose was to determine 1) if POPs or fish consumption are related to levels of adiponectin, CRP, and GGT; 2) if adiponectin, CRP, and GGT are related to incident diabetes or to continuous levels of HA1c; and 3) if the associations of POPs or fish consumption with incident diabetes and continuous HA1c are mediated or modified by adiponectin, CRP, and GGT.

2. Methods

2.1 Participants

The Great Lakes Consortium for the Health Assessment of Great Lakes Sport Fish Consumption was organized in 1992 (Anderson et al. 1996), and study design of the Great Lakes Fish Consumption Study through 2004-2005 has been previously described (Turyk et al. 2009b). Briefly, approximately 4,200 participants with frequent and infrequent Great Lakes sport fish consumption were recruited, including Great Lakes fishing charter boat captains, anglers who fished from inland Wisconsin lakes, and infrequent consumers (reporting consumption of fewer than six meals of Great Lakes sport fish in any year of the previous 20 years). Between 1994 and 2005, biological samples were collected at least once from 948 participants and tested for persistent pollutants (Anderson et al. 2008;Hanrahan et al. 1999;Persky et al. 2001). The current analysis incorporates data collected at follow up in 2004-2005 and in 2010.

2.2 Data Collection

Self-reported diagnosis of diabetes, date of diagnosis, demographics, height, weight, smoking, alcohol use, medication use, and fish consumption were assessed by survey. In 2004-2005, 515 participants were surveyed (Turyk et al. 2009b) and data on fish meals consumed in the past year was assessed, and summarized as commercial (fresh water and saltwater) fish meals and sport caught fish meals (Great Lake or other body of water). Health data assessed in 2010 was available from 598 participants, of whom 402 also participated in the 2004-2005 data collection.

In 2004-2005, non-fasting blood was collected in red-top vacutainer tubes, allowed to clot for 20 min at room temperature, centrifuged for 15 min, transferred to solvent-rinsed glass vials, and stored at -20° C until analysis. All laboratory tests were performed by technicians blinded to participant characteristics. Sera samples were analyzed for DDE, and for PCB and PBDE congeners as previously described (Anderson et al. 2008). Briefly, sera were extracted with hexanenethyl ether, with clean-up and fractionation using Florisil, silica-gel and concentrated sulfuric acid. PBDEs were analyzed by gas chromatography–mass spectrometry and PCBs and DDE by gas chromatography.

Total cholesterol and triglycerides were measured by Quest Diagnostics. Total serum lipids were calculated by the formula:

$$\text{Total lipid}=(\text{total cholesterol}\mathrm{mg}∕\mathrm{dL}+2:{27}^{\ast }\text{triglycerides}\mathrm{mg}∕\mathrm{dL})+62:3.$$

HA1c was measured in whole blood by Quest Diagnostics (Auburn Hills, MI and Wood Dale, IL) through affinity chromatography, which measured total glycosylated hemoglobin, from which HA1c is calculated.

Stored sera samples were assayed in 2010 for adiponectin, CRP, and GGT. Because sera samples had been through prior freeze-thaw cycles during testing for hormones in 2004-2005, we evaluated the stability of test results for these biomarkers after a series of freeze-thaw cycles in blood obtained from four donors. These tests did not indicate systematic decline in biomarker measurements with increasing freeze-thaw cycles.

Quality control was monitored using internal positive and negative controls. Inter-assay coefficients of variation (CVs) were calculated from repeated assays on 4-5% of the participant samples. For adiponectin and CRP, each participant sample was assayed in duplicate and CVs were calculated for each duplicate pair. The average CV was then calculated for each assay plate and the intra-assay CV was the average of all the plate averages. For GGT, a single measurement was obtained for each participant sample precluding calculation of an intra-assay CV.

Circulating serum levels of adiponectin and CRP were measured using Quantikine ELISA kits from R&D Systems (Minneapolis, MN). The human adiponectin/Acrp30 immunoassay (DRP300) recognizes recombinant and natural (low, middle, and high molecular weight) human total adiponectin, and had a sensitivity of 0.246 ng/mL. Intra and inter-assay CVs for participant samples were 4.8% and 9.9%, respectively. The human CRP immunoassay kit (DCRP00) had a sensitivity of 0.01 ng/mL. Intra- and inter-assay CVs were are 7.9% and 13.5%, respectively. GGT was measured by the University of Illinois Hospital, Pathology Laboratory Services by an enzymatic rate method using the SYNCHRON LX® System (Beckman Coulter, Inc.). The assay sensitivity was 5 U/L and the inter-assay CV for repeated measurements of participant samples was 4.6%.

2.3 Statistical Analysis

Participants were included in the analysis if data were available from 2004-2005 for POPs exposures, HA1c, serum cholesterol and triglycerides, and at least one of the diabetes biomarkers measured, as well as health data assessed in 2010. Participants were excluded because of missing data and use of hormone medications, including systemic corticosteroids (n=2), steroid hormones (n=26), steroid hormones and corticosteroids (n=1), or thyroid hormones (n=41). We excluded participants taking hormone medications (sex steroids, corticosteroid or thyroid) because of potential effects on diabetes biomarkers and HA1c (Brand et al. 2011;Fallah et al. 2012;Sargeant et al. 2000). The final sample size for analysis was 413.

HA1c% was examined as a continuous outcome in these 413 participants. Included in this sample were 349 persons without diagnosed diabetes and with HA1c<6.5%, and 64 persons with diagnosed diabetes and/or HA1c≥6.5%: 13 with HA1c≥6.5% and without diagnosis, 11 with HA1c<6.5% and with diagnosis, and 40 with HA1c≥6.5% and with diagnosis. Subgroup analyses were performed in the 349 persons without diagnosed diabetes and with HA1c<6.5%.

Incident diabetes was investigated in the subgroup of 287 participants who did not have diagnosed diabetes and did not use thyroid and steroid hormone medications at the 2004-2005 follow up and who also responded to the follow up survey in 2010. Incident diabetes (n=16) was defined as no diagnosed diabetes in 2004-2005 and reported diabetes diagnosis in 2010.

POP exposures (DDE, sum PBDEs, PBDE 47, sum PCBs and individual congeners PCB 118, 180, 132/153, 138/163), fish consumption measures, adiponectin, CRP and GGT were analyzed as continuous variables with natural log transformations (Ln). POPs and fish consumption were also modeled as quartiles or tertiles (see supplemental tables 3 and 4).

Associations of adiponectin, GGT and CRP with POPs and fish consumption were examined in linear regression models, adjusting for age, BMI, gender, and serum lipids. Education, smoking, alcohol use, and specific types of medication use were considered as covariates, but were not included in the final models because they did not confound the POP/biomarker associations. Age, BMI, and gender were evaluated for effect modification in stratified models.

Analyses of the relationships of GGT, CRP, adiponectin, POPs, and fish consumption with incident diabetes and with HA1c were conducted using logistic and ordinary least squares regression models, respectively, adjusting for gender, age, BMI, and serum lipids for POP models only. We also adjusted HA1c models for use of any diabetes medication. Education, smoking, alcohol use, serum lipids, and selected medications (beta blocker, diuretic, calcium channel blocker, antilipemic, angiotensin-converting enzyme inhibitor, any antidepressive) were included in the model if confounding was present (>10% change in effect estimate). Additional potential confounders available only for incident diabetes models were family history of diabetes and long-term prednisone use. For models with POPs as predictor variables, different types of fish meals were considered potential confounders, and the strongest confounder was selected. For models with fish meals as predictor variables, DDE and sum PCBs were considered potential confounders, and the strongest confounder was selected.

Effect modification by adiponectin, CRP, GGT and BMI was investigated using multiplicative interaction terms. For interaction terms with p-values<0.15, modification was illustrated using estimates of the exposure/outcome association at selected percentiles of the modifier. Gender and age did not modify associations of POPs and fish consumption with HA1c.

Mediation analyses were used to explore the hypothesis that these diabetes biomarkers are partial mediators in associations between POPs and diabetes. Mediation would be supported if the following factors were satisfied 1) diabetes was associated with the POP, 2) the diabetes biomarker was associated with the POP, and 3) the association of diabetes with the POP was attenuated by including the biomarker as a covariate in the regression model.

Information about fasting time prior to blood sample collection was missing for 50 persons. Adiponectin differed by fasting status prior to blood draw, with mean=10.5 mg/L for fasting ≤12 h and mean=9.0 mg/L for fasting >12 h (p=0.03). However, adjusting final adiponectin models for fasting time did not change results substantially (not shown). Fasting was not associated with the GGT, CRP, POPs and fish consumption variables, nor did adjustment for fasting substantially affect results of cross sectional or prospective models (not shown). All statistical analyses were performed using SAS Version 9.3 (SAS Institute, Inc., Cary, NC).

3. Results

3.1 Associations among Diabetes Biomarkers, Exposures and Demographics

Associations of BMI, age, gender, education, smoking, alcohol use, and serum lipids with prevalent and incident diabetes are shown in Supplemental Table 1 and associations of these characteristics with adiponectin, GGT, CRP, HA1c, POPs and fish consumption in Tables 1 and 2. Correlations among diabetes biomarkers, POPs and fish consumption variables are found in Supplemental Table 2.

Table 1.

Associations of diabetes biomarkers with participant characteristics, n=413

Characteristic		LnAdiponectin, mg/L	LnCRP, mg/L	LnGGT, U/L	HA1c, %
All Participants	Meana	10.0	2.2	21.9	5.77
	95% CI	9.4,10.6	2.0, 2.5	20.7, 23.2	5.70,5.84
Men, n=313	Meana	8.9	2.4	24.0	5.8
Women, n=100	Meana	14.0	1.8	16.7	5.6
	p-valuec	<0.0001	0.07	<0.0001	0.008
Age <50 years, n=93	Meana	8.9	2.0	21.7	5.5
Age 50-64.9 years, n=219	Meana	10.2	2.2	22.4	5.8
Age ≥65 years, n=101	Meana	10.5	2.6	21.1	6.0
	p-valuec	0.15	0.19	0.69	<0.0001
BMI <25 kg/m2, n=74	Meana	14.2	1.1	17.3	5.5
BMI 25-29.9 kg/m2, n=174	Meana	9.8	2.2	22.2	5.7
BMI >30 kg/m2, n=167	Meana	8.7	3.0	24.0	5.9
	p-valuec	<0.0001	<0.0001	0.0004	<0.0001
≤<High School, n=142	Meana	9.5	2.7	23.7	5.9
Some College, n=132	Meana	10.9	2.1	20.5	5.7
≥4-year College, n=139	Meana	9.6	1.9	21.6	5.7
	p-valuec	0.15	0.01	0.11	0.08
Smoker, n=43	Meana	9.5	2.7	22.2	5.8
Non-smoker, n=370	Meana	10.0	2.2	21.9	5.8
	p-valuec	0.58	0.22	0.89	0.93
Alcohol Use, n=301	Meana	10.0	2.2	23.0	5.7
No Alcohol Use, n=107	Meana	9.9	2.1	19.2	6.0
	p-valuec	0.83	0.77	0.002	0.003
Prevalent Diabetesd, n=64	Meana	8.7	2.2	23.9	6.9
No Prevalent Diabetesd, n=349	Meana	10.2	2.4	21.6	5.6
	p-valuec	0.13	0.44	0.14	<0.0001
Incident Diabetese, n=16	Meana	6.8	4.9	26.4	6.3
No Incident Diabetese, n=271	Meana	10.5	2.1	21.2	5.5
	p-valuec	0.003	<0.0001	0.18	0.003
Hemoglobin A1c, %	r	−0.15	0.10	0.10	-
	p-valueb	0.003	0.04	0.05
Cholesterol, mg/dL	r	0.10	0.04	0.11	−0.13
	p-valueb	0.04	0.40	0.03	0.009
Triglycerides, mg/dL	r	−0.36	0.18	0.32	0.17
	p-valueb	<0.0001	0.0002	<0.0001	0.0005

^aGeometric mean for adiponectin, CRP, GGT, arithmetic mean for HA1c.

^bP-values are for Pearson's correlation coefficients.

^cP-values are from Student's t-test or ANOVA.

^dDiagnosed diabetes reported in 2004-2005 and/or HA1c≥6.5%.

^eNo diagnosed diabetes reported in 2004-2005 and reported diabetes diagnosis in 2010

Table 2.

Associations of POPs and fish consumption with participant characteristics, n=413

Characteristic		LnDDE ng/g	LnPCB 118 ng/g	LnΣPCB ng/g	LnΣPBDE ng/g	LnGLSCF meals/yr	LnSaltwater fish meals/yr	LnTotal fish meals/yr
All Participants	Meana	2.0	0.12	2.4	0.30	6.6	11.4	40.2
	95% CI	1.8, 2.1	0.11,0.13	2.2, 2.5	0.28,0.32	5.5, 7.9	9.6,13.5	35.7,45.3
Men, n=313	Meana	2.19	0.13	2.71	0.30	8.1	10.8	41.2
Women, n=100	Meana	1.39	0.10	1.54	0.30	3.4	13.3	37.1
	p-valuec	<0.0001	0.004	<0.0001	0.92	<0.0001	0.31	0.46
Age <50 years, n=93	Meana	1.19	0.09	1.46	0.25	5.2	11.0	36.5
Age 50-64.9 years, n=219	Meana	2.05	0.12	2.50	0.32	6.3	12.9	41.1
Age ≥65 years, n=101	Meana	2.86	0.15	3.28	0.31	9.0	8.9	41.8
	p-valuec	<0.0001	<0.0001	<0.0001	0.02	0.10	0.23	0.69
BMI <25 kg/m2, n=74	Meana	1.27	0.10	1.95	0.33	4.1	9.9	35.9
BMI 25-29.9 kg/m2, n=174	Meana	1.99	0.11	2.51	0.27	6.8	10.4	37.8
BMI ≥30 kg/m2, n=167	Meana	2.34	0.13	2.42	0.32	7.9	13.2	45.0
	p-valuec	<0.0001	0.005	0.04	0.08	0.04	0.37	0.30
≤High School, n=142	Meana	2.16	0.13	2.61	0.31	8.5	9.0	44.7
Some College, n=132	Meana	1.97	0.11	2.22	0.31	7.1	13.2	42.5
≥4-year College, n=139	Meana	1.79	0.11	2.28	0.29	4.8	12.6	34.2
	p-valuec	0.22	0.17	0.13	0.76	0.04	0.23	0.57
Smoker, n=43	Meana	1.66	0.10	2.57	0.34	15.4	16.0	57.4
Non-smoker, n=370	Meana	2.01	0.12	2.35	0.30	6.0	10.9	38.6
	p-valuec	0.20	0.14	0.44	0.22	0.001	0.19	0.02
Alcohol Use, n=301	Meana	2.09	0.12	2.46	0.30	7.6	12.2	44.2
No Alcohol Use, n=107	Meana	1.76	0.11	2.15	0.29	4.3	9.6	30.9
	p-valuec	0.16	0.10	0.07	0.79	0.005	0.25	0.02
Prevalent Diabetesd, n=64	Meana	3.28	0.16	3.03	0.31	6.6	11.1	40.4
No Prevalent Diabetesd, n=349	Meana	1.79	0.11	2.27	0.30	6.6	11.4	40.2
	p-valuec	<0.0001	0.0009	0.003	0.78	0.99	0.91	0.98
Incident Diabetese, n=16	Meana	3.8	0.22	3.4	0.26	6.7	6.8	32.5
No Incident Diabetese, n=271	Meana	1.8	0.11	2.3	0.30	6.3	12.3	39.7
	p-valuec	<0.0001	0.0002	0.03	0.23	0.89	0.19	0.53
Hemoglobin A1c, %	r	0.30	0.19	0.17	0.01	−0.04	−0.07	−0.08
	p-valueb	<0.0001	0.0001	0.0006	0.77	0.40	0.18	0.10
Cholesterol, mg/dL	r	0.06	0.11	0.12	0.09	−0.02	0.01	0.002
	p-valueb	0.25	0.03	0.01	0.07	0.73	0.78	0.97
Triglycerides, mg/dL	r	0.21	0.20	0.21	0.08	0.02	−0.02	−0.03
	p-valueb	<0.0001	<0.0001	<0.0001	0.09	0.76	0.67	0.54

^aGeometric mean.

^bP-values are for Pearson's correlation coefficients.

^cP-values are from Student's t-test or ANOVA.

^dDiagnosed diabetes reported in 2004-2005 and/or HA1c≥6.5%.

^eNo diagnosed diabetes reported in 2004-2005 and reported diabetes diagnosis in 2010

3.2 Associations of Adiponectin, CRP, and GGT with POPs and Fish Consumption

Associations of adiponectin, CRP, and GGT with POPs and fish consumption, after adjustment for age, gender and BMI, were examined in the full cohort and in models stratified by gender, median age and median BMI (Table 3 and Supplemental Table 3). LnDDE, Lnsum PCBs and individual LnPCB congeners were not significantly associated with Lnadiponectin, LnCRP, or LnGGT (Table 3) with the exception of LnGGT, which was positively associated with tertiles of PCB 118 (Supplemental Table 3).

Table 3.

Association of POPs and Fish Consumption with Adiponectin, CRP and GGT, n=413

Biomarker	Exposure	Beta-coefficient, p-valuea
LnAdiponectin	LnDDE	0.02, 0.68
	LnPCB 118	−0.01, 0.89
	LnΣPCBs	0.02, 0.68
	LnPBDE 47	0.04, 0.26
	LnΣPBDEs	0.08, 0.07b
	LnGLSCF Meals	−0.01. 0.38
	LnSaltwater Fish Meals	−0.01, 0.71
	LnTotal Fish Meals	0.03, 0.24
LnCRP	LnDDE	−0.01, 0.90
	LnPCB 118	0.02, 0.82
	LnΣPCBs	−0.04, 0.64
	LnPBDE 47	0.02, 0.74
	LnΣPBDEs	0.04, 0.56
	LnGLSCF Meals	−0.01, 0.79
	LnSaltwater Fish Meals	−0.06, 0.045
	LnTotal Fish Meals	−0.03, 0.54
LnGGT	LnDDE	0.05, 0.12
	LnPCB 118	0.07, 0.10
	LnΣPCBs	0.04, 0.33
	LnPBDE 47	0.04, 0.13
	Ln ΣPBDEs	0.06, 0.13
	LnGLSCF Meals	0.01, 0.68
	LnSaltwater Fish Meals	0.002, 0.92
	LnTotal Fish Meals	−0.02, 0.95

^aLinear regression mode s adjusted for age, gender, and BMI. Models with POPs also adjusted for serum lipids.

^bsignificant with control for fasting.

PCB132153, PCB138163, PCB180 were not associated with biomarkers and are not shown in table.

In stratified models (not shown), LnSum PBDEs and LnPBDE 47 were associated with LnGGT in persons of above median age (β=0.13, p=0.02 and β=0.11, p=0.008, respectively); Lnsum PBDEs was associated with Lnadiponectin in persons with above median BMI (β=0.11, p=0.05) and above median age (β=0.18, p=0.003); and LnPBDE 47 was associated with Lnadiponectin in persons with above median age (β=0.13, p=0.009).

Lnsaltwater fish meals were inversely associated with LnCRP in the full cohort (Table 3), and in males and persons with above median BMI (β=-0.08, p=0.005 and β=-0.08, p=0.02, respectively, not shown). Quartiles of total fish meals were inversely associated with LnCRP (Supplemental Table 3). Lntotal fish meals and LnGLSCF meals were inversely associated with LnCRP only in persons with above median BMI (β=-0.08, p=0.03, and β=-0.13, p=0.02, respectively, not shown).

3.3 Associations of Incident Diabetes and Hemoglobin A1c with Diabetes Biomarkers

After adjusting for confounding, Lnadiponectin was inversely associated with incident diabetes and HA1c; LnCRP was positively associated with incident diabetes, but not with HA1c; and GGT quartiles were positively but not significantly associated with incident diabetes, while LnGGT was positively but not significantly associated with HA1c (Tables 4 and 5).

Table 4.

Associations of Incident Diabetes with POPs, Fish Consumption, Adiponectin, GGT and CRP.

Exposure or Biomarker	Odds of Incident Diabetes: Model 1a			Odds of Incident Diabetes: Model 2b
	OR	95% CI	P Value	OR	95% CI	P Value	Additional confounders
LnAdiponectin	0.20	0.07, 0.56	0.002	0.20	0.07, 0.56	0.002	none
LnCRP	2.87	1.19, 6.89	0.02	3.22	1.25, 8.29	0.02	Lnsaltwater fish meals
LnGGT	1.75	0.77, 3.96	0.18	1.50	0.57, 3.90	0.41	Family history diabetes
Quartile GGT	1.86	1.04, 3.30	0.04	1.70	0.93, 3.11	0.08	Family history diabetes
LnDDE	2.08	1.01, 4.29	0.05	2.63	1.17, 5.89	0.02	Calcium channel blocker use
LnPCB118	2.83	1.35, 5.90	0.006	3.28	1.50, 7.18	0.003	Lntotal fish meals
LnPCB132153	2.59	1.05, 6.40	0.04	3.24	1.17, 9.03	0.02	Calcium channel blocker use, Lntotal fish meals, education
LnPCB138163	2.64	1.12, 6.24	0.03	3.62	1.38, 9.52	0.009	Calcium channel blocker use, Lntotal fish meals
LnPCB180	1.89	0.74, 4.87	0.19	2.33	0.86, 6.35	0.10	Lntotal fish meals
LnΣPCB	2.43	0.96, 6.12	0.06	3.38	1.19, 9.59	0.02	Calcium channel blocker use, Lntotal fish meals
LnPBDE47	0.71	0.38, 1.34	0.29	0.71	0.38, 1.34	0.29	none
LnΣPBDE	0.44	0.16, 1.21	0.11	0.44	0.16, 1.21	0.11	none
LnTotal Fish Meals	0.79	0.51, 1,23	0.30	0.66	0.40, 1,09	0.10	LnΣ PCBs
LnSaltwater Fish Meals	0.80	0.59, 1.08	0.14	0.80	0.59, 1.11	0.18	none
LnGLSCF Meals	1.02	0.75 1.39	0.89	0.89	0.63, 1.26	0.50	LnΣ PCBs

N=287, including 16 incident diabetes cases

^aModel 1 adjusted for age, BMI, gender and serum lipids for POP models

^bModel 2 adjusted for all covariates in Model 1 and for additional covariates determined to be confounders.

Table 5.

Cross sectional Associations of Hemoglobin A1C with POPs, Fish Consumption, Adiponectin, GGT, and CRP in 2004-2005

Group	Exposure or biomarker	Association of Exposure with Hemog obin A1c
		Model 1a			Model 2b
		Beta	95% CI	P-value	Beta	95% CI	P-value	Additional Confounders
With and without diabetes N=413	LnAdiponectin	−0.159	−0.246,−0.072	0.0004	−0.159	−0.246,−0.072	0.0004	none
	LnCRP	0.027	−0.026, 0.081	0.31	0.011	−0.044, 0.065	0.70	education, serum lipids, ace inhibitor use
	LnGGT	0.100	0.005, 0.195	0.04	0.076	−0.026, 0.178	0.15	Alcohol use, serum lipids, antilipid use, LnDDE
	LnDDE	0.099	0.034, 0.164	0.003	0.118	0.051, 0.186	0.0006	LnGLSCF meals
	LnPCB118	0.055	−0.024, 0.133	0.17	0.091	0.007, 0.175	0.03	Alcohol use, LnGLSCF meals
	LnTotal Fish Meals	−0.059	−0.102, −0.016	0.007	−0.072	−0.115, −0.029	0.001	LnDDE
	LnSaltwater Fish Meals	−0.030	−0.060, −0.001	0.045	−0.030	−0.060, −0.001	0.045	None
	LnGLSCF Meals	−0.019	−0.049, 0.010	0.20	−0.039	−0.069, −0.008	0.01	smoking, LnDDE
No diabetes and HA1c<6.5% N=349	LnAdiponectin	−0.084	−0.148, −0.019	0.01	−0.074	−0.138, −0.009	0.03	Alcohol use
	LnCRP	0.007	−0.030, 0.044	0.70	−0.011	−0.048, 0.026	0.55	Alcohol use, smoking, lnsaltwater fish, serum lipids, beta-blocker use, ace inhibitor use, diuretic use
	LnGGT	0.037	−0.026, 0.100	0.25	0.035	−0.034, 0.103	0.32	Alcohol use, antilipid use, serum lipids
	LnDDE	0.011	−0.035, 0.057	0.64	0.025	−0.022, 0.071	0.30	smoking, Lntotal fish meals
	LnPCB118	0.010	−0.047, 0.067	0.73	0.033	−0.027, 0.093	0.27	Alcohol use, LnGLSCF meals
	LnTotal Fish Meals	−0.040	−0.070, −0.011	0.008	−0.040	−0.070, −0.011	0.008	none
	LnSaltwater Fish Meals	−0.029	−0.050, −0.009	0.005	−0.029	−0.050, −0.009	0.005	none
	LnGLSCF Meals	−0.008	−0.029, 0.012	0.41	−0.014	−0.036, 0.007	0.18	Smoking, LnDDE

^aModel 1 adjusted for age, BMI, and gender, for all models; serum lipids for POP models; and diabetes medication use for models for participant group: “with and without diabetes”.

^bModel 2 adjusted for all covariates in model 1 and for additional covariates determined to be confounders.

Sum PCB, PCB132153, PCB138163, PCB180, Sum PBDE, PBDE47 were not associated with HA1c and are not shown in table.

3.4 Associations of Incident Diabetes with POPs and Fish Consumption: Evaluation of Mediation and Moderation by Diabetes Biomarkers

Odds of incident diabetes were significantly elevated by LnDDE, LnPCB 118, LnPCB 132153, and LnPCB 138163 exposure, controlling for age, gender, BMI, and serum lipids (Table 4). LnsumPBDEs, LnPBDE 47, Lntotal fish meals and Lnsaltwater fish meals decreased odds of incident diabetes, but these associations did not reach significance (Table 4).

We found no evidence for mediation, as no substantial changes in associations of diabetes with DDE and PCBs were seen with adjustment for CRP or GGT, and associations of diabetes with DDE, sum PCB and PCB congeners were strengthened with further adjustment for adiponectin (not shown).

Because of the small number of diabetes cases, we did not explore modification of associations of incident diabetes with POPs and fish consumption by BMI, adiponectin, CRP and GGT.

3.5 Associations of HA1c with POPs and Fish Consumption: Evaluation of Mediation and Moderation by Diabetes Biomarkers

HA1c was positively associated with LnDDE and LnPCB118, and inversely associated with Lnsaltwater fish meals, Lntotal fish meals, and LnGLSCF meals after adjustment for confounding factors (Table 5). In models that included the subgroup of participants without diabetes and with HA1c<6.5%, Lntotal fish meals and Lnsaltwater fish meals, but not LnDDE, LnPCB 118 and LnGLSCF meals, remained significantly and inversely associated with HA1c (Table 5).

We found evidence for negative confounding, or strengthening of effect, by GLSCF meals for associations of HA1c with DDE and PCB 118. Furthermore, associations of HA1c with GLSCF and total fish meals were negatively confounded by DDE and PCBs, but not PBDEs. However, associations of saltwater fish meals with HA1c were not confounded by POPs. The joint effect of DDE and GLSCF meals on HA1c is illustrated in Figure 1. Mean percent HA1c was 0.43 higher in persons with the highest tertile of DDE and the lowest tertile of GLSCF meals compared to those with the lowest tertile of DDE and the highest tertile of GLSCF meals.

Figure 1.

Joint effect of DDE exposure and GLSCF meals in past year on HA1c levels, adjusted for age centered, gender, BMI centered, diabetes medication use, and serum lipids centered. Total N=413, number of participants in each group is shown in on bar.

Mediation was assessed by further adjusting multivariable models for Lnadiponectin, LnCRP, or LnGGT, but little change was observed in the magnitude of associations of HA1c with LnDDE, LnPCB118, and Lntotal fish, Lnsaltwater fish, and LnGLSCF meals (not shown).

GGT, CRP, adiponectin and BMI modified the positive association of LnDDE with HA1c, with stronger effects in persons with higher GGT, CRP, and BMI and with lower adiponectin (Table 6). Similarly, the effect of LnPCB118 on HA1c was stronger in persons with higher GGT. Inverse associations of HA1c with Lntotal fish meals and LnGLSCF meals were stronger in persons with higher CRP and GGT, respectively. The decline in HA1c with saltwater fish ingestion was larger in those with lower BMI and adiponectin.

Table 6.

Modification of associations of Hemoglobin A1c with POPs and fish consumption by BMI, adiponectin, CRP and GGT

Participant Group	Exposure	Modifier	Interaction p-value	Conditional associations of Hemoglobin A1c with POPs and fish consumption at selected percentiles of modifying variables.a
				25th Percentile Betab, p-value	50th Percentile Betab, p-value	75th Percentile Betab, p-value
With and Without Diabetes, n=413	LnDDE	BMI	0.03	0.08, 0.04	0.11, 0.002	0.15, <0.0001
		Adiponectin	0.04	0.16, 0.0002	0.13, 0.0002	0.10, 0.01
		CRP	0.05	0.07, 0.07	0.12, 0.0005	0.15, 0.0001
		GGT	0.09	0.08, 0.05	0.11, 0.002	0.14, 0.0002
	LnPCB118	GGT	0.05	0.03, 0.56	0.07, 0.09	0.12, 0.01
	LnTotal Fish Meals	CRP	0.13	−0.06, 0.01	−0.08, 0.0004	−0.10, 0.0007
	LnSaltwater Fish Meals	BMI	0.11	−0.05, 0.01	−0.03, 0.03	−0.02, 0.38
		Adiponectin	0.02	−0.06. 0.003	−0.03, 0.03	−0.01, 0.54
	LnGLSCF Meals	GGT	0.14	−0.03, 0.12	−0.04, 0.01	−0.05, 0.004
No Diabetes and HA1c<6.5%, n=349	LnDDE	CRP	0.09	−0.001, 0.98	0.32, 0.22	0.05, 0.08
	LnTotal Fish Meals	GGT	0.12	−0.03, 0.13	−0.04, 0.01	−0.05, 0.001
	LnSaltwater Fish Meals	BMI	0.03	−0.04, 0.0004	−0.03, 0.002	−0.02, 0.13

^aA total of 40 mode s were examined but results shown in table only if p-value for interaction of modifier with exposure <0.15.

^bAdjusted for age, BMI, and gender, for all models; serum lipids for POP models; and diabetes medication use for models for participant group: “with and without diabetes”. Additional covariates were included if determined to be confounders (See Table 5).

4. Discussion

We previously reported positive associations of DDE and PCB118, but not PBDEs, with prevalent diabetes in the 2004-2005 cross-sectional data form this cohort (Turyk et al. 2009a;Turyk et al. 2009b). The current study extends our investigations of the same dataset, finding positive associations of DDE and PCB118 with a continuous measure of glycemic control, HA1c. However, associations of DDE and PCB 118 with HA1c did not reach significance in the subgroup analysis that focused on participants without diagnosed diabetes and with HA1c levels below 6.5% [the threshold used to identify potential diabetes for population screening (American Diabetes Association 2010)], suggesting that the effect of POPs on glucose dysregulation may be more important for development of diabetes than prediabetes. Analyses of cross sectional National Health and Nutrition Examination Survey (NHANES) data that defined prediabetes by HA1c% found that PCB 118 was associated with HA1c% in the higher range (5.9%-6.4%), but not in the lower range of 5.7%-5.8% (Everett and Thompson 2012), and that DDE was associated with diabetes but not prediabetes (Everett and Matheson 2010). However, a 23-year longitudinal investigation found that PCB exposure was associated with increased HA1c% during follow up in persons who became diabetic by year 20 as well as in non-diabetic controls (Suarez-Lopez et al. 2015).

Our findings on elevated incident diabetes risk with DDE exposure are consistent with published results in this cohort based on follow up from 1993 to 2005 (Turyk et al. 2009a). While we did not previously detect an association of PCBs with incident diabetes, the current analysis revealed increased diabetes risk with PCB exposure. The association of PCBs with incident diabetes is consistent with some (Lee et al. 2010;Lee et al. 2011;Vasiliu et al. 2006;Wu et al. 2013), but not all prospective investigations on this topic (Rignell-Hydbom et al. 2009;Zani et al. 2013).

Our past work did not identify any associations of years consuming sport caught fish with incident and prevalent diabetes (Turyk et al. 2009a;Turyk et al. 2009b). The current analysis focused on different fish consumption metrics, namely meals of fish consumed in the past year, including total fish meals, saltwater fish meals and GLSCF meals. Odds of incident diabetes were reduced, but not significantly, with increasing total fish meals and saltwater fish meals. HA1c was inversely and significantly associated with total fish meals and saltwater fish meals in the full cohort and remained significantly and inversely associated in the subgroup analysis that included only participants without diagnosed diabetes and with HA1c levels below 6.5%, suggesting that fish consumption may affect blood glucose levels in the normal and prediabetic range as well as the diabetic range. In persons without diabetes, HA1c was inversely associated with eating oily fish in women but not men, but the association was attenuated with adjustment for family history of diabetes, smoking, physical activity, alcohol, total energy and fruit and vegetable intake (Harding et al. 2004). Among diabetics, intake of fish has been associated with decreased HA1c (Lee et al. 2012). In the current study we were not able to determine which component of fish could account for its effect on blood glucose. Protective effects of fish consumption on diabetes have been attributed to omega-3 fatty acids, mediated in part through reduction of triglycerides and inflammation (Carpentier et al. 2006), but interventions (Akinkuolie et al. 2011;Fedor and Kelley 2009) and prospective observational studies (Wu et al. 2012;Zhang et al. 2013) have not consistently reported a beneficial effect of marine omega-3 fatty acids on insulin resistance and diabetes.

GLSCF meals were inversely associated with HA1c only after adjustment for DDE exposure, while positive associations of DDE and PCB 118 with HA1c were strengthened after control for GLSCF meals. These findings emphasize the importance of adjusting for both POP exposures and fish intake in investigations of populations ingesting contaminated fish, and are consistent with a study of sport fish consumers that noted elevated odds of diabetes with DDE and PCB exposure, but decreased odds of diabetes with increasing total fish and trout meals, with adjustment for PCBs (Philibert et al. 2009). On the other hand, a cross sectional investigation of Finnish fishermen did not find that co-adjustment for environmental contaminants, including POPs, and omega-3 fatty acids affected associations of either factor modeled separately with glucose, insulin resistance, or inflammatory markers (Turunen et al. 2013). Inconsistent effects of fish and marine omega-3 fatty acid intake on diabetes risk in prospective observational studies could potentially be confounded by unmeasured contaminant exposures (Wu et al. 2012;Zhang et al. 2013).

Our findings of elevated odds of incident diabetes with elevated CRP and GGT and decreased odds with elevated adiponectin levels were consistent with the literature (Dehghan et al. 2007;Duncan and Schmidt 2006;Nakanishi et al. 2004). However, we detected few associations of these diabetes biomarkers with POPs and fish ingestion, with the exception of positive associations of GGT with PCB 118 and with PBDEs in older persons, a positive association of adiponectin with PBDEs, and inverse associations of CRP with fish ingestion, particularly in persons with higher adiposity. Furthermore, adiponectin, CRP and GGT did not attenuate associations of POPs with either HA1c or incident diabetes, and this study does not lend support for any of these biomarkers as mediators of associations of POPs and diabetes.

We detected a positive association of adiponectin with sum PBDEs in the entire cohort, and in persons with higher body mass or older age, suggesting improved insulin sensitivity with increased PBDEs. Supporting this observation is our finding of a non-significant trend towards reduction of incident diabetes risk with PBDE exposure. To our knowledge, no other epidemiologic studies have investigated effects of PBDEs on adiponectin; however, decreased insulin-stimulated glucose oxidation was observed in adipocytes isolated from penta-BDE treated rats (Hoppe and Carey 2007). We did not observe associations of adiponectin with DDE or PCBs, overall or by age, gender, or BMI strata. In contrast, three other epidemiological investigations found inverse associations (Lim and Jee 2014;Mullerova et al. 2008) or trends (Kern et al. 2004) of adiponectin with PCBs or related chemicals, and stronger effects in subgroups with elevated BMI (Lim and Jee 2014;Mullerova et al. 2008). Experimental evidence that PCB 153 (Taxvig et al. 2012) and DDE (Howell, III and Mangum 2011) stimulate release of adiponectin from mature adipocytes supports these results. Reasons for difference in the effect of PCBs on adiponectin in these investigations and in the current study are not clear, but could be related to variability in population characteristics and exposures. Finally, fish ingestion was not related to adiponectin levels, although increased adiponectin has been detected in fish intake intervention studies (Kondo et al. 2010;Lara et al. 2007;Neale et al. 2013;Zhang et al. 2012).

POPs have been associated with increased inflammation, assessed by a variety of biomarkers (Fujiyoshi et al. 2006;Imbeault et al. 2012;Kumar et al. 2014b;Kuwatsuka et al. 2014). In the current study, we did not find any associations of CRP with POPs, and other investigators have inconsistently detected effects of POPs on CRP. Elevated CRP was associated with occupational PCB exposure in women (Persky et al. 2011), but not men (Persky et al. 2012); CRP was positively associated with organochlorine pesticides and inversely associated with non-dioxin like PCBs in NHANES participants (Ha et al. 2007;Kim et al. 2012); and dioxin-like chemicals, PCBs, and organochlorine pesticides did not affect CRP in two other studies (Kumar et al. 2014b;Turunen et al. 2013). Our findings of inverse associations of saltwater fish meals and total fish meals with CRP are consistent with observational and intervention studies (He et al. 2009;Ouellet et al. 2008;Ramel et al. 2010;Smith et al. 2009).

GGT, an independent risk factor for the development of cardiovascular disease and diabetes, is induced by oxidative stress and catalyzes the first step in the degradation of glutathione, which functions as a conjugating ligand for phase II metabolism of xenobiotics, such as POPs (Lee et al. 2008). Exposures to POPs have been associated with increased GGT, even within the normal reference range in some (Chen et al. 2006;Lee and Jacobs, Jr. 2006;Sweeney and Mocarelli 2000), but not all reports (Kumar et al. 2014a;Persky et al. 2011;Persky et al. 2012). In the current study, PCB 118 was positively associated with GGT, as were sumPBDEs and PBDE 47 in persons of above median age.

BMI, adiponectin, GGT and CRP modified the association of DDE with HA1c, with stronger effects of DDE on HA1c in persons with higher levels of risk for diabetes (higher BMI, GGT and CRP and lower adiponectin). Similarly, GGT modified the association of PCB118 with HA1c. A cross sectional analysis of NHANES data examined modification of the effect of POPs on insulin resistance by CRP, with results similar to ours, namely stronger effects of POPs on insulin resistance in persons with higher concentrations of CRP (Kim et al. 2012). There is also evidence of effect modification of associations of diabetes and POPs by body weight, with stronger effects in persons with higher adiposity (Lee et al. 2014). With respect to fish intake, our findings suggest that the protective effect of fish consumption on HA1c is stronger in persons with higher levels of the diabetes biomarkers. However, this was not the case for adiposity, as the protective effect of saltwater fish meals on HA1c was stronger in those with lower body mass, which is consistent with a meta-analysis concluding that that fish consumption was related to lower risk of diabetes in populations with lower BMI (Wu et al. 2012).

There are several limitations to this study. First, with respect to measurement of biomarkers, there may have been some loss of activity due to long term storage and prior freeze/thaw cycles. However, all of the serum samples were handled similarly, so we do not expect differential bias in statistical estimates. Second, diabetes diagnosis and fish consumption were collected by self report and may be affected by measurement error, although misclassification is unlikely to be differential. Third, there may be residual confounding for variables with measurement errors or for unmeasured variables, such as other dietary factors (Boeing et al. 2000;Harding et al. 2004;Prynne et al. 2009), physical activity, and other contaminants. Fourth, adjustment of POPs models for serum lipids is appropriate for non-fasting samples, but could potentially result in attenuated effect estimates if serum lipids are intermediates in the causal pathway between POPs and diabetes (Taylor et al. 2013). Nevertheless, this study takes advantage of an established, well characterized cohort and is one of the first studies to explore the role of diabetes biomarkers in the associations of POPs.

5. Conclusions

This investigation found that odds of incident diabetes were elevated with exposure to DDE and PCBs and that DDE and PCB 118 were positively associated with HA1c. Fish intake was inversely associated with HA1c, suggesting the presence of nutrients in fish that modify glucose levels. Co-adjustment for POPs and fish consumption strengthened associations of either factor modeled separately, highlighting the importance of assessing both POP exposures and fish intake in populations ingesting contaminated fish.

Adiponectin, CRP, and GGT were not associated with DDE and PCB congeners, with the exception of a positive association of GGT with PCB 118. CRP was inversely related to fish consumption. However, diabetes biomarkers did not mediate associations of POPs or fish consumption with HA1c or with incident diabetes. Our findings suggest that while adiponectin, CRP and GGT do not mediate the effect of POPs on HA1c, positive associations of POPs with HA1c are stronger in persons with higher levels of these diabetes risk factors.

Highlights.

Diabetes biomarkers, adiponectin, γ-glutamyl transferase and CRP, were measured

POPs were associated with incident diabetes and higher Hemoglobin A1c levels

Associations of diabetes with POPs were not mediated by diabetes biomarkers

POPs had stronger effects on blood glucose in persons with higher diabetes risk