Abstract
Introduction:
Persistent organic pollutants (POPs) occur as correlated mixtures, and serum concentrations may vary over time due to changes in exposure and toxicokinetic processes. We investigated associations between a mixture of POPs measured repeatedly before type 2 diabetes mellitus diagnosis (T2DM).
Material and methods:
In a longitudinal nested case-control study within the Tromsø Study, serum concentrations of 19 chlorinated POPs were measured at three pre-diagnostic timepoints: T1 (1986/87; 15–21 years before diagnosis, n = 259), T2 (1994/95; 7–13 years before diagnosis, n = 257), and T3 (2001; 0–6 years before diagnosis, n = 260). Lipid-adjusted POP concentrations were analyzed using positive-direction weighted quantile sum (WQS) logistic regression. Single-pollutant analyses were conducted for comparison.
Results:
The POP mixture was positively associated with subsequent T2DM at T2 [odds ratio (OR) 2.14, 95% confidence interval (CI) 1.34–3.42] and T3 (OR 1.78, 95% CI 1.19–2.66), whereas no association was detected at T1. Cis-heptachlor epoxide and p,p'-DDT received the largest weights at T2. At T3, weights were more distributed, with cis-nonachlor, cis-heptachlor epoxide, β-HCH, PCB-194, and PCB-187 exceeding the equal-weight threshold. Single-pollutant analyses identified similar leading compounds at T2 and T3.
Conclusion:
POP mixtures measured at T2 and T3 were positively associated with subsequent T2DM. These associations may partly reflect preclinical metabolic changes, altered POPs distribution between adipose tissue and blood, residual confounding, or chance. The findings support the complementary use of mixture and single-pollutant approaches and the value of repeated POP measurements. Larger longitudinal studies are needed to clarify temporality, evaluate potential reverse causation, and formally assess time-varying mixture associations.
1 Introduction
Persistent Organic Pollutants (POPs), such as organochlorine pesticides (OCPs) and polychlorinated biphenyls (PCBs) are characterized by their lipophilic, bioaccumulative and toxic properties (). POPs degrade slowly in environmental and biological systems, posing long-term risks to human health. Due to their lipophilic nature, POPs accumulate in adipose tissue, leading to prolonged retention in the body and potential adverse health effects. Although production of most POPs was banned decades ago, their persistence and remobilization driven by climate change (e.g., glacier melt, permafrost thaw) still render them relevant in human health today (). Additionally, global distillation processes transport POPs from warmer to colder regions through wind and ocean currents, resulting in their accumulation in Arctic regions ().
Chronic exposure to POPs has been associated with adverse health outcomes in human, including type 2 diabetes mellitus (T2DM), a disease of increasing global prevalence (). Previous epidemiological studies have often examined the association between individual POPs and T2DM using statistical methods designed to assess one compound at a time, without accounting for the simultaneous exposure to multiple compounds nor time-variant compositions of the exposures. However, evidence suggests that adverse health outcomes may result from the cumulative body burden of chemical mixtures that disrupt the same biological systems. For example, the Agency for Toxic Substances and Disease Registry has noted that the observed association between DDT and T2DM may be influenced by co-exposure to other pollutants (). Similarly, Heindel et al. () emphasized the importance of accounting for co-exposures to chemicals with similar physicochemical properties and mechanisms of action, such as other OCPs and PCBs, when assessing the metabolic effects of DDT/DDE. Additionally, many of the chlorinated POPs share similar exposure pathways, resulting in high correlations that should be taken into account analytically (). The traditional unipollutant approach fails to account for this issue and rather unrealistically assumes that the associations between each compound and T2DM is not influenced by the presence of the other compounds (, ). Moreover, such approaches in epidemiological studies may increase the probability of Type I errors due to multiple testing (). Recent methodological developments have focused on approaches that can investigate the effects of mixtures of pollutants (, ).
Charles et al. previously reported associations between T2DM and OCPs and PCBs in the same study sample as in the present study, using a single-pollutant approach and summing PCBs into a single variable (). However, the potential added value of a multipollutant approach remains unexplored. In this study, we employed Weighted Quantile Sum Regression (WQS), a statistical method designed to evaluate the cumulative effect of chemical mixtures while identifying key contributors in the mixture. WQS generates a single weighted index summarizing correlated mixture components, with weights reflecting the relative allocation of components within the directionally specified index (, ). This framework is ideal for identification of key contributors in a mixture-outcome association, which can guide future mechanistic studies, either experimentally or through mediation analyses in population-based cohorts.
Specifically, in this study we aimed to 1) evaluate the association between a mixture of 19 POPs (11 PCBs and 8 OCPs) and the odds of T2DM using three repeated blood measurements taken 1−6 years, 7−13 years, and 15−21 years prior to T2DM diagnosis, 2) identify the compounds most likely to contribute to this potential association, and 3) assess how the association and the key contributors change over time.
2 Material and methods
2.1 Study sample
We conducted a longitudinal nested case-control study within The Tromsø Study, a health survey performed among the inhabitants of Tromsø municipality in Northern Norway, which has been repeated approximately every 7th year since 1974 (). In short, we included 145 T2DM cases, 76 women and 69 men, that participated in and had available serum samples from at least three surveys; T1 (1986/87), T2 (1994/95) and T3 (2001). Incident T2DM cases were identified through the local diabetes registry in Tromsø, Norway. Cases were required to have a registry-recorded T2DM after the individual date of T3 blood sampling and before the individual date of blood sampling in the following 2007 survey (T4). Consequently, the T3 sample preceded the diagnosis date for every included case, including diagnosis recorded later in 2001. We randomly selected 145 diabetes-free controls from the overall cohort, 76 women and 69 men that did not develop T2DM during the study period and that also had available serum samples from at least T1, T2 and T3. To avoid possible prevalent undiagnosed T2DM, we excluded 29 cases and one control with HbA1c levels ≥ 6.5% (48 mmol/mol) at T1, T2 or T3. Additionally, one control missed information on triglycerides on T1 (used to calculate lipid-adjusted POPs concentrations), and two controls and one case did not have sufficient serum at T2. Hence, our study included three repeated measurements of the same individuals prior to T2DM-diagnosis in cases, with 259 participants at T1 (116 future cases, 143 controls), 257 at T2 (115 future cases, 142 controls) and 260 at T3 (116 future cases, 144 controls). An overview of inclusions and exclusions is shown in Figure 1, and participant characteristics are presented in Supplementary Table S1. The adjusted WQS analyses were conducted using complete-case data for the outcome, all 19 mixture components, and model covariates. Consequently, the corresponding analytical sample sizes were 253 at T1, 236 at T2, and 228 at T3.
Figure 1
Participation in The Tromsø Study was voluntary, and a signed informed consent form was obtained from all participants. The study was approved by the Regional Committee for Medical and Health Research Ethics (REK, case number 2015/1780). The study conforms to the standards of The Declaration of Helsinki.
2.2 Chemical analyses
Blood samples collected as part of the Tromsø Study were kept at room temperature for 30 min to allow coagulation, followed by centrifugation at 2000g for 10 min. Serum was aliquoted into secondary plastic sample containers within 1 h and stored in a biobank at −70 °C. For the present study, frozen serum samples were retrieved from the biobank, thawed and aliquoted into separate vials appropriate for measurements of lipids and measurements of POPs. Lipids were measured immediately. Before POPs analysis, samples were stored at −30 °C, thawed overnight at 4 °C and allowed to reach room temperature before sample preparation and analysis.
PCBs and OCPs were analyzed using gas chromatography atmospheric pressure ionization coupled to tandem mass spectrometers (Waters, Milford, MA, USA), where a detailed description of the analysis has been described elsewhere (, ). A description of the analytical procedure is provided in Supplementary Material Section 1.2.
In total 13 polychlorinated biphenyl (PCB) congeners (PCB-28, −52, −74, −99, −118, −138, −153, −156, −170, −180, −183, −187, and −194) and 13 OCPs [cis-chlordane, trans-chlordane, 1,1-bis-(4-chlorophenyl)-2,2- dichloroethene [p,p’-DDE]], 1,1,1-trichloro-2,2-bis- (4-chlorophenyl) ethane [p,p’-DDT], alpha-hexachlorocyclohexane [α-HCH], beta-hexachlorocyclohexane [β-HCH], gamma-hexachlorocyclohexane [ɣ-HCH], heptachlor, cis-heptachlor epoxide, hexachlorobenzene [HCB], cis-nonachlor, trans-nonachlor, and oxychlordane) were detected above the method detection limit (MDL) in the samples, but only the compounds with detection percentages over 70% at all timepoints were included in further statistical analyses (Supplementary Table S2).
Total cholesterol and triglycerides, used to calculate total lipids, were measured applying the Cobas® 8000 platform (Roche diagnostics, Switzerland) (). These analyses are accredited according to the ISO 15189 standard ().
2.3 Statistical analyses
All statistical models were performed using STATA 17 (StataCorp, TX, USA) and R version 4.5.3 (R Foundation for Statistical Computing, Vienna, Austria) in RStudio (Posit Software, PBC, Boston, MA, USA). WQS analyses were performed using the using the gWQS package (version 3.0.5).
2.3.1 Lipid-adjustment and transformation of POP concentrations
Total serum lipids were calculated according to Phillips et al. as:
Total lipids (g/L) = 2.27 * total cholesterol (g/L) + triglycerides (g/L) + 0.623 ()
Lipid-adjusted POP concentrations (ng/g lipid) were calculated by dividing wet-weight concentrations (pg/mL) by total serum lipids (g/L). Lipid-adjusted concentrations were natural-log transformed because of a skewed distribution. Further, the log-transformed lipid-adjusted concentrations were standardized to a mean of 0 and standard deviation (SD) 1 within each imputed dataset.
2.3.2 Handling concentrations below the method detection limit
Concentrations below the method detection limit (MDL) were treated as left-censored observations. Censoring -aware multiple imputation was conducted for the three compounds with the greatest proportion of values below the MDL: cis-heptachlor epoxide, oxychlordane and p,p’-DDT. For the remaining compounds, which had low proportions of nondetectable values, concentrations below the MDL were replaced with MDL/√2. See Supplementary Material Section 1.3.1 for details on the multiple imputation method.
2.3.3 Descriptive analyses and correlations
Lipid-adjusted POP concentrations have previously been reported in this cohort by Charles et al. 2021 (); however, previous analyses combined the included PCB congeners into summary measures. Therefore, medians with 25th and 75th percentile for all 19 individual POPs are presented separately for cases and controls in Supplementary Table S3. Differences in POP concentration between cases and controls were assessed using Wilcoxon rank-sum tests.
Correlations between the 19 included compounds were assessed by Spearman’s rank-order correlation.
2.3.4 Weighted Quantile Sum regression
Weighted quantile sum regression (WQS) was used to assess the association between the POP mixture and future T2DM diagnosis, and to estimate the relative contribution of each POP to the mixture index (). Analyses were conducted separately at each of the three timepoints.
The primary WQS analyses included 19 lipid-adjusted, natural-log transformed and standardized POP concentrations. Cis-heptachlor epoxide, oxychlordane and p,p’-DDT were represented by their censoring-aware imputed concentrations. All other POPs were represented by observed concentrations or MDL/√2-substituted values.
WQS regression was fitted within a generalized linear model framework using a binomial distribution and logit link. POP concentrations were categorized into quartiles. The WQS index was constructed as a weighted sum of component-specific quartile scores, with non-negative weights constrained to sum to one. The estimated odds ratios (OR) therefore represent the association between T2DM and a one-unit increase in the weighted quantile mixture index. WQS weights represent the relative contribution of each component to the directionally specified mixture index.
To reduce dependence on a single random training/validation split and to evaluate stability of the WQS weights, repeated-holdout WQS regression was performed. In each repetition, the analytic dataset was randomly divided into training and validation subsets. The training subset was used to estimate WQS weights using bootstrap resampling, and the validation subset was used to estimate the association between the WQS index and T2DM. Unless otherwise specified, a validation proportion of 0.50 was used, weights were estimated using 100 bootstrap samples and 20 repeated holdout splits. The repeated-holdout procedure was conducted separately in all 20 imputed datasets. A fixed random seed (123) was used for all WQS analyses to ensure reproducibility and to hold the resampling scheme constant across imputed datasets. For each timepoint, WQS regression coefficients and standard errors were pooled across imputed datasets using Rubin's rules. Results are presented as ORs with 95% confidence intervals. Weight stability was summarized across the 20 repeated holdouts and 20 imputed datasets, corresponding to 400 weight estimates per compound. For each component, the mean weight, standard deviatoin (SD), median, and empirical 2.5th–97.5th percentile range were calculated.
Potential confounders were identified using a Directed Acyclic Graph (DAG, Supplementary Figure S1). The primary model was adjusted for age, sex, BMI, parity, physical activity and months of breastfeeding (only available at T2 and T3). All confounders were measured contemporaneously with POP measurements at each study timepoint. Unadjusted WQS models were also fitted. Although dietary habits may influence both POP exposure and risk of future T2DM, sufficient dietary information for confounder adjustment was unavailable. However, fish intake, a major dietary source of POP exposure in this population, has previously been reported to be similar among cases and controls ().
BMI may represent a confounder and/or a potential mediator of associations between POP exposure and T2DM. BMI was included as a confounder in the primary model. However, we performed sensitivity analyses excluding BMI as a confounder in the WQS model to assess the total effect of the POPs mixture on future T2DM that may also operate through BMI (Supplementary Table S3). Additional sensitivity analyses used wet-weight, natural-log transformed POP concentrations as mixture components, with natural-log transformed total serum lipids included as an additional confounder (Supplementary Table S3).
To inform the directional specification of the primary WQS model, we examined the directions of unadjusted associations between each individual POP and T2DM using separate logistic regression models (Supplementary Table S3). As the individual POP estimates were predominantly positive or null, with no statistically supported inverse associations, the primary WQS analyses used positive-direction weight estimation. As a directional sensitivity analysis, negative-direction WQS models were fitted at all timepoints using the same mixture components, covariate adjustment sets, imputed datasets and resampling settings as the primary analyses. For these models, component-weight estimation was directed toward a negative association with T2DM (b1_pos = FALSE); however, the final validation-stage regression coefficient was not sign-constrained to assess whether an inverse mixture association was supported by the data.
The influence of WQS implementation choices was assessed using validation proportions of 0, 0.4, 0.6, and 0.8 and using 100 or 300 bootstrap samples. These settings analyses were conducted using one representative imputed dataset because WQS estimates and weights were highly similar across the 20 imputed datasets.
Sex-stratified WQS analyses were conducted separately in women and men, in one representative imputed dataset. These analyses were considered exploratory because of limited sample sizes within sex strata and instability in some bootstrap-resampled models.
2.3.5 Single-pollutant logistic regression
The association between a one-standard-deviation higher concentration in each POP separately (with the PCBs summarized) and T2DM have been reported by Charles et al. (). To facilitate comparison with WQS results, separate logistic regression models were fitted for each of the 19 POPs, where POP concentrations were categorized into quartiles, and ORs with 95% confidence intervals were estimated per one-quartile increase in the pollutant concentration. Models were fitted separately at each timepoint. The models were adjusted for the same set of confounders as the primary WQS models and were restricted to the same analytic sample as the corresponding WQS analysis. Estimates were pooled across the 20 imputed datasets using Rubin's rules. To account for multiple comparisons in the unipollutant analyses, p-values were adjusted using the Benjamini-Hochberg false discovery rate procedure at a 0.05 level.
3 Results
3.1 Participant characteristics
Characteristics of the study participants have been reported in Supplementary Table S1, and previously in Charles et al. 2021 (). In short, cases were older than controls (47.5 and 45.1 years at T1, respectively) and had higher BMI (27.3 versus 24.2 kg/m2 at T1, 29.0 versus 25.4 kg/m2 at T2 and 30.6 versus 26.5 kg/m2 at T3, respectively) and total lipids (7.9 versus 7.0 g/L at T1, 8.1 versus 7.5 g/L at T2, and 7.5 versus 7.0 g/L at T3, respectively) compared to controls at the different timepoints.
3.2 Concentrations of POPs
Of the 26 detected POPs in the serum samples, 19 compounds were detected in >70% of study participants at all timepoints and were therefore included in further analyses [Supplementary Table S2 (),]. They included PCB-74, PCB-99, PCB-118, PCB-138, PCB-153, PCB-156, PCB-170, PCB-180, PCB-183, PCB-187, PCB-194, p,p’-DDE, p,p’-DDT, β-HCH, cis-heptachlor epoxide, HCB, cis-nonachlor, trans-nonachlor, and oxychlordane.
Medians with 25th and 75th percentiles for cases and controls at each timepoint, along with differences in median concentrations, are presented in Supplementary Table S4. In summary, cases had significantly higher concentrations of PCB-99, PCB-118, cis-heptachlor epoxide, β-HCH and p,p’-DDT at T1. At T2 and T3, significant differences were observed for PCB-74, PCB-99, PCB-118, PCB-153, PCB-138, PCB-183, PCB-187 (only at T3), trans-nonachlor, cis-nonachlor, oxychlordane, cis-heptachlor epoxide, β-HCH, p,p’-DDE and p,p’-DDT.
3.3 Results from multipollutant analyses
All PCB congeners were highly correlated at all timepoints, where correlation coefficients ranged from 0.71-0.99 (Supplementary Figure S2). At T1 and T2, chlordanes (oxychlordane, trans-nonachlor and cis-nonachlor) were also highly correlated with the PCBs. HCB, β-HCH, cis-heptachlor epoxide, p,p’-DDE and p,p’.DDT were less correlated with each other and the other compounds at all timepoints, where correlation coefficients ranged from 0.15−0.67.
In the main model, the POPs mixture was not associated with subsequent T2DM at T1. At T2, a one-unit increase in the WQS index was associated with higher odds of subsequent T2DM (OR 2.14, 95% CI 1.34, 3.42; Figure 2; Supplementary Table S5). Cis-heptachlor epoxide, p,p’-DDT and hexachlorobenzene (HCB) collectively received 85% of the total estimated index weight, with weights of 51%, 27%, and 7%, respectively (Figure 3; Supplementary Table S5). At T3, a one-unit increase in the WQS index was also associated with higher odds of subsequent T2DM (OR 1.78, 95% CI 1.19, 2.66; Figure 2; Supplementary Table S5). Five compounds had weights exceeding the prespecified equal-weight threshold of 5.3%: cis-nonachlor (26%), cis-heptachlor epoxide (21%), β-HCH (14%), PCB-194 (9%), and PCB-187 (7%) (Figure 3; Supplementary Table S5).
Figure 2
Figure 3
Sensitivity analyses excluding BMI yielded stronger positive mixture associations at T2 and T3, with ORs of 2.87 (95% CI 1.74, 4.74) and 2.08 (95% CI 1.53, 2.82), respectively. Analyses using wet-weight POP concentrations with adjustment for total lipids produced similar results to the primary lipid-normalized analyses, with ORs of 2.03 (95% CI 1.27, 3.23) at T2 and 1.79 (95% CI 1.28, 2.52) at T3 (Supplementary Table S5).
Negative-direction WQS analyses did not provide evidence of an inverse association between the POP mixture and subsequent T2DM at any timepoint. The estimated ORs were 1.20 (95% CI 0.85–1.66) at T1 and 1.45 (95% CI 0.84–2.51) at T2. At T3, the negative-direction WQS index was positively associated with subsequent T2DM (OR 1.56, 95% CI 1.06–2.30; Supplementary Table S5).
In sex-stratified analyses, the WQS mixture was not significantly associated with subsequent T2DM at any timepoint for neither men nor women (Supplementary Table S6).
Sensitivity analyses varying the training/validation proportion and number of bootstrap samples are shown in Supplementary Table S7. Point estimates were broadly similar across settings, although precision varied and confidence intervals included the null for the T2 model with no validation subset and the T3 model with a validation proportion of 0.6. The leading-weighted compounds were broadly consistent across most validation proportions and bootstrap settings. At a validation proportion of 0.8, the training subset used for weight estimation was relatively small, and the relative ordering of component weights was less consistent.
The distribution of WQS component weights across repeated-holdout analyses and censoring-aware imputed datasets is presented in Supplementary Table S8. At T2, cis-heptachlor epoxide and p,p'-DDT consistently received the largest weights, exceeding the equal-weight threshold in 100% and 98.5% of analyses, respectively. At T3, cis-nonachlor consistently exceeded the equal-weight threshold, while cis-heptachlor epoxide and β-HCH exceeded this threshold in 90% of analyses. The percentile ranges were wide for several components, particularly at T3, indicating uncertainty in the precise magnitude and relative ranking of individual component weights across resampling.
3.4 Results from unipollutant analyses
When tested individually, none of the compounds were significantly associated with T2DM at T1 after adjusting for confounding variables (Supplementary Table S9). At T2, a one-quartile increase in cis-heptachlor epoxide (OR 1.94 95% CI 1.39, 2.70) and p,p’-DDT (OR 1.59, 95% CI 1.17, 2.15) displayed a significant association with future T2DM after adjustment for multiple comparisons (Supplementary Table S9). At T3, cis-nonachlor (OR 1.72 95% CI 1.26, 2.34) and cis-heptachlor epoxide (OR 1.59 95% CI 1.18, 2.15) were significantly associated with T2DM after adjustment for multiple comparisons (Supplementary Table S9).
4 Discussion
In this study, a mixture of 19 POPs measured at T2 (7−13 years before diagnosis in cases) and T3 (0−6 years before diagnosis) was significantly and positively associated with future T2DM, whereas no association was detected at T1 (15−21 years before diagnosis) At T2, cis-heptachlor epoxide and p,p’-DDT received the largest positive-direction WQS weights, accounting for 51% and 27% of the total index weight, respectively. HCB received a smaller wight of 7%. At T3, five compounds had weights exceeding the equal-weight threshold; cis-nonachlor (26%), cis-heptachlor epoxide (21%), β-HCH (14%), PCB-194 (9%), and PCB-187 (7%). After correction for multiple comparisons, the single-pollutant analyses identified the same leading compounds as the WQS analyses at T2 and T3. These findings highlight the value of repeated POP measurements and complementary mixture and single-pollutant approaches for evaluating associations between highly correlated POP exposures and subsequent T2DM.
The positive associations detected at T2 and T3 should not be interpreted as evidence that associations differed statistically across sampling occasions or that a specific biological susceptibility window was identified. The WQS models were fitted separately at each timepoint, with timepoint-specific quantile cut-points and mixture weights. Consequently, a one-unit increase in the WQS index does not necessarily represent the same absolute exposure contrast at T1, T2, and T3. Additionally, the timepoint-specific analyses were not restricted to participants with complete exposure and covariate data at all three study occasions. Consequently, differences in participant availability and analytical sample composition across timepoints may have contributed to variation in the observed associations and should be considered when comparing findings across timepoints. Moreover, participant age, calendar time, exposure distributions, time before diagnosis, and possible preclinical disease progression changed concurrently across the three sampling occasions. The present analyses could therefore not distinguish biological time-dependent effects from age-related factors, calendar-period effects, changes in exposure or measurement characteristics, preclinical disease processes, or random variation. Larger studies using longitudinal mixture methods and formal tests of time-varying associations are needed to address these questions. The available sample size and number of T2DM cases limited the precision and feasibility of such complex interaction models.
At T2, the prominence of (p,p’)-DDT in the positive-direction WQS index is consistent with previous experimental and epidemiologic literature, suggesting that DDT and related compounds may be associated with metabolic processes relevant to T2DM, including altered steroid- and thyroid hormone homeostasis, insulin sensitivity, glucose production, adipose tissue distribution, inflammation, β-cell function, weight regulation and obesity (–). In contrast, mechanistic evidence for cis-heptachlor epoxide, a metabolite of heptachlor, is limited. Heptachlor is metabolized to heptachlor epoxide by the CYP-450 enzyme system, and the liver has been identified as a potential target organ for heptachlor toxicity, but the exact mechanisms potentially underlying the association between cis-heptachlor epoxide and T2DM remain poorly understood (). However, the observational findings in the present study cannot establish that DDT, cis-heptachlor epoxide, or other individual compounds caused the observed associations. In particular, WQS weights summarize the relative contribution of components to a directionally specified mixture index and should not be interpreted as independent or causal compound-specific effects.
The more distributed weight profile at T3, compared to T2, may reflect a less clearly dominated mixture profile at this timepoint. As T3 was closer to T2DM diagnosis among cases, circulating POP concentrations may have reflected not only external exposure but also compound-specific pharmacokinetic processes, including persistence, metabolism, elimination, and redistribution between adipose tissue and blood. Changes in body composition, lipid metabolism, or adipose-tissue mobilization during the pre-diagnostic period could potentially alter serum concentrations differently across compounds according to their physicochemical properties and toxicokinetic behavior. Thus, the more distributed T3 weight profile may partly reflect differences in the circulating POP profile rather than distinct causal effects of individual compounds. However, differences in WQS weights across timepoints should be interpreted cautiously because the POPs were strongly correlated and timepoint-specific weights are not directly comparable.
Reverse causation is an alternative explanation for the observed associations. Preclinical metabolic changes preceding T2DM may alter lipid metabolism and the distribution of POPs between adipose tissue and blood. For example, adipose-tissue dysfunction and increased lipolysis may mobilize lipophilic compounds from adipose tissue into the circulation (). Previous analyses in this cohort reported higher HbA1c concentrations among future cases than controls at T2, suggesting that metabolic changes may have been present years before clinical diagnosis (). In addition, previous work in this cohort found a smaller decline in concentrations of several POPs among cases than controls from T1 to T2 and (). These observations are compatible with altered POPs toxicokinetics during the pre-diagnostic period, but do not establish that such processes explain the associations observed here. Residual confounding and chance findings cannot be excluded.
The role of adipose tissue in the POPs-T2DM association is particularly complex; when healthy, adipose tissue may serve as a storage unit, protecting other organs from harmful effects of POPs, but dysfunctional adipose tissue with uncontrolled lipolysis may release POPs into circulation, potentially amplifying their harmful effects (). While BMI was included as a confounder in the main analysis, it may also lie on a potential pathway between POP exposure and T2DM, given the proposed obesogenic effects of some POPs (). Sensitivity analyses excluding BMI yielded stronger positive associations between the POP mixture and subsequent T2DM at T2 and T3. This pattern is compatible with, but does not establish, mediation through BMI. The difference between BMI-adjusted and non-adjusted estimates may also reflect residual-confounding by BMI, changes in precision, or model specification. Furthermore, BMI may not fully capture variations in adipose-tissue function, lipid turnover, or the capacity to store and mobilize POPs.
The agreement between the leading compounds in the single-pollutant and WQS analyses is noteworthy, but neither approach can isolate independent effects of individual compounds in the presence of substantial co-exposure and correlation. Single-pollutant models may reflect associations with correlated co-exposures, whereas WQS provides an estimate for the overall mixture under a directional assumption and summarizes the relative contribution of components to that mixture index. The WQS approach also reduces the influence of extreme exposure values by using quantile scores. Nevertheless, WQS results depend on the chosen direction, quantile specification, training/validation split, and sample size, and component weights can be unstable when exposures are strongly correlated, and the sample size is relatively modest. The repeated-holdout analyses were included to reduce dependence on a single training/validation split. Sex-stratified analyses were exploratory and should be interpreted cautiously because of limited sample sizes and instability in some bootstrap-resampled models. Negative-direction WQS analyses did not support an inverse association between the POP mixture and subsequent T2DM. At T3, the negative-direction-weighted index was positively associated with T2DM. Because the final validation-stage coefficient was not sign-constrained, this result indicates that the negative-direction weight-estimation procedure did not identify a stable inverse mixture pattern, rather than evidence of a protective or additional positive mixture effect. This may reflect limited evidence for inverse component-specific associations, high correlations among POPs, and instability in estimated weights across training and validation subsets. However, conventional directional WQS estimates positive and negative mixture patterns separately and does not simultaneously estimate opposing component-specific effects.
We interpret the positive association observed in the negative-direction WQS as a lack of support for an inverse mixture association. This may reflect instability in the estimated weights or heterogeneity across training-validation splits, particularly because the individual compounds showed null or positive associations and there was little substantive reason to expect an inverse mixture effect. Although negative-direction WQS analyses did not support an inverse mixture association, conventional WQS estimates one direction at a time and cannot simultaneously estimate potentially opposing component effects.
This study has several strengths, including repeated measurements covering a prolonged period relevant to both temporal changes in POP concentrations and T2DM progression and detailed assessment of a broad POP mixture. However, serum POP concentrations at a given timepoint reflect complex processes involving historical exposure, diet, age, birth cohort, body composition, lipid metabolism, and individual toxicokinetics (). The study was conducted in one Arctic population with a specific exposure profile and concentration range; therefore, the findings may not generalize to populations with different POP mixtures, exposure levels, or metabolic characteristics. Future studies should combine repeated serum measurements with formal longitudinal mixture methods and, where feasible, measurements of POP concentrations in adipose tissue to improve understanding of total body burden, blood–adipose distribution, and their relation to metabolic health.
5 Conclusion
A mixture of 19 POPs measured at T2 and T3 was positively associated with subsequent T2DM, whereas no association was detected at T1. At T2, p,p'-DDT and cis-heptachlor epoxide received the largest positive-direction WQS weights and were also the only two compounds associated with future T2DM in the single-pollutant analyses. At T3, the mixture association was accompanied by a more distributed WQS weight profile. These findings support the use of complementary mixture and single-pollutant approaches when evaluating highly correlated POPs exposures. Repeated exposure measurements may help characterize associations between circulating POP profiles and subsequent T2DM across the pre-diagnostic period; however, larger longitudinal studies are needed to formally evaluate time-varying mixture associations, reverse causation, and potential biological mechanisms.
Statements
Data availability statement
The datasets presented in this article are not readily available because the data that support the findings of this study are available from The Tromsø Study. Restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. The corresponding author will on request detail the restrictions and any conditions under which access to some data may be provided. Requests to access the datasets should be directed to The Tromsø Study, https://uit.no/research/tromsostudy.
Ethics statement
The studies involving humans were approved by the Regional Committee for Medical and Health Research Ethics (REK, case number 2015/1780). The study conforms to the standards of The Declaration of Helsinki. The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.
Author contributions
MN: Data curation, Investigation, Conceptualization, Visualization, Writing – original draft, Formal analysis. AB: Formal analysis, Supervision, Writing – review & editing. CR: Writing – review & editing, Project administration, Supervision, Methodology, Conceptualization. DC: Writing – review & editing. SH: Writing – review & editing, Investigation, Resources. TN: Writing – review & editing. TW: Writing – review & editing. TS: Writing – review & editing. TJ-T: Writing – review & editing. VB: Project administration, Writing – review & editing, Conceptualization, Supervision, Funding acquisition.
Funding
The author(s) declared that financial support was received for this work and/or its publication. The project is financially supported by the Northern Norway Regional Health Authorities (project number HNF-1518-20). The funders had no influence on the design of the study, collection and interpretation of data or writing of the manuscript.
Acknowledgments
The authors are grateful for the participants of The Tromsø Study for their contribution and donation of blood samples.
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that generative AI was used in the creation of this manuscript. An AI tool provided by UiT The Arctic University of Norway was used for language editing. The authors reviewed the edited text and are responsible for the final manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fenvh.2026.1946520/full#supplementary-material
References
1.
(AMAP) AMaAP. AMAP Assessment 2002: persistent organic pollutants in the Arctic. In: de WitCAFiskATHobbsKEMuirDCG, (eds.)Oslo: Arctic Monitoring and Assessment Programme (AMAP) (2004), p. 21.
2.
HungHHalsallCBallHBidlemanTDachsJDe SilvaAet al. Climate change influence on the levels and trends of persistent organic pollutants (POPs) and chemicals of emerging Arctic concern (CEACs) in the Arctic physical environment - a review. Environ Sci Process Impacts. (2022) 24(10):1577–615. 10.1039/D1EM00485A
3.
LeeDHPortaMJacobsDRJr.VandenbergLN. Chlorinated persistent organic pollutants, obesity, and type 2 diabetes. Endocr Rev. (2014) 35(4):557–601. 10.1210/er.2013-1084
4.
Agency for Toxic Substances and Disease Registry (ATSDR) Toxicological Profiles. Toxicological Profile for DDT, DDE, and DDD. Atlanta (GA): Agency for Toxic Substances and Disease Registry (US) (2022).
5.
HeindelJJHowardSAgay-ShayKArrebolaJPAudouzeKBabinPJet al. Obesity II: establishing causal links between chemical exposures and obesity. Biochem Pharmacol. (2022) 199:115015. 10.1016/j.bcp.2022.115015
6.
AxmonARignell-HydbomA. Association between biomarkers of exposure to persistent organochlorine compounds (POCs). Chemosphere. (2006) 64(4):692–4. 10.1016/j.chemosphere.2005.11.002
7.
VermeulenRSchymanskiELBarabasiALMillerGW. The exposome and health: where chemistry meets biology. Science. (2020) 367(6476):392–6. 10.1126/science.aay3164
8.
BraunJMGenningsCHauserRWebsterTF. What can epidemiological studies tell US about the impact of chemical mixtures on human health?Environ Health Perspect. (2016) 124(1):A6–9. 10.1289/ehp.1510569
9.
PanSLiZRubboBQuon-ChowVChenJCBaumertBOet al. Applications of mixture methods in epidemiological studies investigating the health impact of persistent organic pollutants exposures: a scoping review. J Expo Sci Environ Epidemiol. (2024) 35:522–34. 10.1038/s41370-024-00717-3
10.
CharlesDBergVNøstTHBergdahlIAHuberSAyottePet al. Longitudinal changes in concentrations of persistent organic pollutants (1986−2016) and their associations with type 2 diabetes mellitus. Environ Res. (2022) 204(Pt B):112129. 10.1016/j.envres.2021.112129
11.
BellaviaA. Statistical Methods for Environmental Mixtures. 1 edn. Cham: Springer (2025). p. 99.
12.
CarricoCGenningsCWheelerDCFactor-LitvakP. Characterization of weighted quantile sum regression for highly correlated data in a risk analysis setting. J Agric Biol Environ Stat. (2015) 20(1):100–20. 10.1007/s13253-014-0180-3
13.
JacobsenBKEggenAEMathiesenEBWilsgaardTNjølstadI. Cohort profile: the tromso study. Int J Epidemiol. (2012) 41(4):961–7. 10.1093/ije/dyr049
14.
HuberSAverinaMBroxJ. Automated sample preparation and GC-API-MS/MS as powerful tool for analysis of legacy POPs in human serum and plasma. Anal Methods. (2020) 12:912–29. 10.1039/C9AY02400J
15.
AllaouiGRylanderCAverinaMWilsgaardTFuskevågOMBergV. Longitudinal changes in blood biomarkers and their ability to predict type 2 diabetes mellitus-the tromsø study. Endocrinol Diabetes Metab. (2022) 5(2):e00325. 10.1002/edm2.325
16.
Accreditation N. Accreditation Scope for TEST 209: Universitetssykehuset Nord Norge HF, Laboratoriemedisin UNN. Lillestrøm: Norsk akkreditering (2022).
17.
PhillipsDLPirkleJLBurseVWBernertJTJr.HendersonLONeedhamLL. Chlorinated hydrocarbon levels in human serum: effects of fasting and feeding. Arch Environ Contam Toxicol. (1989) 18(4):495–500. 10.1007/BF01055015
18.
BergVCharlesDDHuberSNøstTHSandangerTMAverinaMet al. Temporal changes in per and polyfluoroalkyl substances and their associations with type 2 diabetes. Sci Rep. (2025) 15(1):22026. 10.1038/s41598-025-05422-1
19.
Burgos-AcevesMAMigliaccioVDi GregorioIPaolellaGLeprettiMFaggioCet al. 1,1,1-trichloro-2,2-bis (p-chlorophenyl)-ethane (DDT) and 1,1-dichloro-2,2-bis (p, p'-chlorophenyl) ethylene (DDE) as endocrine disruptors in human and wildlife: a possible implication of mitochondria. Environ Toxicol Pharmacol. (2021) 87:103684. 10.1016/j.etap.2021.103684
20.
LindPMLindL. Endocrine-disrupting chemicals and risk of diabetes: an evidence-based review. Diabetologia. (2018) 61(7):1495–502. 10.1007/s00125-018-4621-3
21.
TimokhinaEPYaglovVVNazimovaSV. Dichlorodiphenyltrichloroethane and the adrenal gland: from toxicity to endocrine disruption. Toxics. (2021) 9(10):243. 10.3390/toxics9100243
22.
ZhuangSZhangJWenYZhangCLiuW. Distinct mechanisms of endocrine disruption of DDT-related pesticides toward estrogen receptor alpha and estrogen-related receptor gamma. Environ Toxicol Chem. (2012) 31(11):2597–605. 10.1002/etc.1986
23.
RossiMTaddeiARFascianiIMaggioRGiorgiF. The cell biology of the thyroid-disrupting mechanism of dichlorodiphenyltrichloroethane (DDT). J Endocrinol Invest. (2018) 41(1):67–73. 10.1007/s40618-017-0716-9
24.
HeindelJJBlumbergBCaveMMachtingerRMantovaniAMendezMAet al. Metabolism disrupting chemicals and metabolic disorders. Reprod Toxicol. (2017) 68:3–33. 10.1016/j.reprotox.2016.10.001
25.
ZhaoXZhangHShenXZhengQWangS. Theoretical study on the metabolic mechanism of heptachlor in human cytochrome P450 enzymes. Int J Mol Sci. (2025) 26(5):2021. 10.3390/ijms26052021
26.
LeeYMJacobsDRJr.LeeDH. Persistent organic pollutants and type 2 diabetes: a critical review of review articles. Front Endocrinol (Lausanne). (2018) 9:712. 10.3389/fendo.2018.00712
27.
AgudoAGoñiFEtxeandiaAVivesAMillánELópezRet al. Polychlorinated biphenyls in Spanish adults: determinants of serum concentrations. Environ Res. (2009) 109(5):620–8.
Summary
Keywords
longitudinal studies, obesity, organochlorines, persistent organic pollutants, type 2 diabetes mellitus, weighted quantile sum regression
Citation
Nilsen MM, Bellavia A, Rylander C, Charles D, Huber S, Nøst TH, Wilsgaard T, Sandanger TM, James-Todd T and Berg V (2026) Associations between serum persistent organic pollutants and type 2 diabetes in a multipollutant analysis using repeated measurements in the Tromsø Study. Front. Environ. Health 5:1946520. doi: 10.3389/fenvh.2026.1946520
Received
23 July 2026
Revised
26 August 2026
Accepted
31 August 2026
Published
15 September 2026
Volume
5 - 2026
Edited by
Rakesh Kumar, Auburn University, United States
Reviewed by
Xuehai Wang, Karolinska Institutet (KI), Sweden
Md Saif Mahabub, Hamdard University Bangladesh, Bangladesh
Zuhai Chen, Sun Yat-sen University, China
Updates
Copyright
© 2026 Nilsen, Bellavia, Rylander, Charles, Huber, Nøst, Wilsgaard, Sandanger, James-Todd and Berg.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Mia Mathisen Nilsen mia.m.nilsen@uit.no
ORCID Mia Mathisen Nilsen orcid.org/0009-0007-7734-4046
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.