Inflammatory Cytokines and the Risk to Develop Type 2 Diabetes: Results of the Prospective Population-Based European Prospective Investigation into Cancer and Nutrition (EPIC)-Potsdam Study

The EPIC-Potsdam study, as part of the multicenter population-based cohort study EPIC, aims to explore the relation between dietary and lifestyle factors and the development of complex diseases. Details of the recruitment procedure of EPIC-Potsdam have been published (14). In brief, 27,548 subjects (women aged 35–65 years and men aged 40–65 years) were recruited from the general population. Baseline examinations, including anthropometric measurements, blood sampling, a self-administered food frequency questionnaire, and a personal interview on lifestyle habits and medical history were conducted between 1994 and 1998. Follow-up questionnaires are sent to the study participants every 2–3 years to obtain information on, among other things, current medication and newly developed diseases, including diabetes.

Anthropometric measurements (body height and weight, waist and hip circumference) were performed by trained personnel, with the participants wearing only light underwear and without shoes (15). BMI was calculated as body weight (in kilograms) divided by body height (in meters) squared. Waist-to-hip ratio (WHR) was calculated as waist divided by hip circumference.

Information on lifestyle characteristics were obtained from self-administered questionnaires and a personal, computer-guided interview by trained and quality-monitored personnel (16). Sporting activities (hours per week) were calculated as the average of hours of sports per week during the summer and the winter season.

We studied 27,548 participants of the population-based EPIC-Potsdam study. The follow-up procedure was successful in receiving completely filled-in questionnaires from 96% of all cohort participants attending the baseline examination (17).

Case subjects were those who were free of type 2 diabetes at baseline and developed type 2 diabetes during the first 2- to 3-year follow-up, depending on the time of recruitment. Potential cases of incident diabetes were identified from self-reports on incident disease, current medications, and/or current dietary treatment for diabetes (n = 399). For each potentially incident subject, a special questionnaire was sent to the primary care physician. The study subject was considered as a case subject only if the diagnosis of newly developed diabetes was confirmed by this physician. A total of 201 cases of incident diabetes were identified by this verification process until 1 November 2001. At that time, another 10 potential incident cases were pending confirmation by the primary care physician. These subjects were not further considered. The biochemical analysis regarding diabetes-associated antibodies GAD65 and IA-2 revealed that nine of the case subjects should be considered as case subjects with type 1 diabetes, leaving 192 medically confirmed cases of type 2 diabetes. Each of the 192 individuals with confirmed type 2 diabetes was matched with two control subjects by age (±1 year) and sex (n = 384). For statistical analysis, those individuals with missing values in one of the variables being used in the statistical models were not considered (n = 4 for case subjects, n = 7 for control subjects), thus leaving 188 case subjects and 377 control subjects for the final analysis.

Peripheral venous citrate-blood samples were taken at enrollment into the study. The blood samples were centrifuged at 1,000g for 10 min at 4°C. Plasma was then removed and stored in aliquots in freezers at −80°C until assays of the markers of interest were performed. IL-1β, IL-6, and TNF-α were measured by enzyme-linked immunosorbent assays (ELISAs; R&D Systems, Minneapolis, MN). CRP was determined by high-sensitivity ELISA (Immun Diagnostik, Bensheim, Germany). All assay procedures were performed as described by the manufacturer. Blood samples were analyzed in random order to exclude systemic bias due to interassay variation. Control specimens were analyzed simultaneously on each plate for every marker. The intra-assay coefficient of variation (CV) ranged between 6.4 and 10.2% for IL-1β, 3.8 and 11.1% for IL-6, 8.7 and 14.8% for TNF-α, and 4.9 and 6.2% for CRP. The interassay CV was 10.3% for IL-1β, 9.9% for IL-6, 16.1% for TNF-α, and 13.2% for CRP. The limits of detection were 0.1 pg/ml for IL-1β, 0.094 pg/ml for IL-6, 0.12 pg/ml for TNF-α, and 0.124 ng/ml for CRP. HbA_1c was determined by enzyme immunoassay (DAKO Diagnostika, Hamburg, Germany). The intra-assay CV ranged from 2 to 3%, and the interassay CV was 1.9%. Values of subjects without diabetes have been shown to range from 4.8 to 6.9% in this assay, according to the information of the manufacturer. Diabetes-associated antibodies GAD65 and IA-2 were analyzed by radioimmunoassay (Medipan Diagnostica, Selchow, Germany), which was performed as described by the manufacturer.

For all analyses, we used SAS software release 8.0 (SAS Institute, Cary, NC). In a first step, measurement values of the laboratory parameters (IL-1β, TNF-α, and CRP) below the limit of quantification (LOQ) were set at 0.7 × the respective LOQ (18). Means, standard deviations, and proportions of baseline characteristics of case and control subjects were calculated. The means ± SD are reported. Significance was considered at two-tailed α < 0.05.

The nonparametric Wilcoxon’s rank-sum test was used to test for differences in continuous variables between case and control subjects, and a χ² test with 1 degree of freedom (Mantel-Haenszel test) was used to describe differences in proportions between case and control subjects. Spearman correlation coefficients were used to test the association between anthropometric, lifestyle variables, and cytokines.

Associations were initially investigated separately for IL-1β, IL-6, and TNF-α. IL-6 and TNF-α were divided into quartiles, and CRP was dichotomized. Because of the high numbers of undetectable values, IL-1β was dichotomized. Because of case and control exclusions due to missing values, primarily unconditional logistic regression analysis was used to estimate odds ratios (ORs) and corresponding 95% CIs. As previously demonstrated, the estimation of the OR approximates the relative risk given an infrequent disease occurrence (19). We therefore calculated the OR to estimate the relative risk to develop type 2 diabetes that is associated with increasing categories of the investigated cytokine. Because the design of the study also allows conditional logistic analysis, study results were recalculated using conditional logistic regression analysis. In the article, the results of the unconditional regression and major results of the conditional regression analysis are shown. Estimates of relative risk were first obtained from age-adjusted (continuous) and sex-adjusted (categorical: female and male) analyses; this analysis was followed by further adjustment for BMI (continuous). The subsequent model also considers sex-normalized WHR (continuous), sporting activities in hours/week (continuous), smoking status (categorical: current smoker and nonsmoker), alcohol consumption in grams/day (continuous), and educational attainment (categorical: basic training, technical school, and university). In the last model, we additionally added HbA_1c (continuous) to the fully adjusted model to reduce potential bias caused by undetected cases of prevalent diabetes.

To investigate the role of cytokine patterns, formal interaction terms including the cytokines in question were analyzed. Product terms were built from dichotomized subgroups. IL-1β was therefore dichotomized as previously described (setting individuals with undetectable levels as 0 and those with detectable levels as 1), whereas IL-6 and TNF-α were dichotomized by using the 75th percentile as the cutoff point (setting individuals who were <75th percentile for both IL-6 and TNF-α as 0 and those who were >75th quartile as 1). Quartile cut points were estimated from the combined group of control and case subjects. In addition, ORs and 95% CIs were estimated for each combination, including two of these cytokines with the low-level category (<75th percentile for IL-6 and TNF-α and nondetectable values of IL-1β, respectively) as the reference category.

To reduce bias by individuals with undetected prevalent type 2 diabetes at baseline, we repeated the analysis, including only those case and control subjects with HbA_1c <5.8%.

Baseline characteristics of the participants are shown in Table 1. As expected, case subjects had higher BMI and WHR, and they exercised less (Table 1). Elevated glucose levels, such as impaired fasting glucose or impaired glucose tolerance, are well known to be a risk factor for future type 2 diabetes (20), which is reflected in our study group by significantly elevated HbA_1c levels in case subjects (6.39 ± 2.16%) compared with control subjects (4.73 ± 0.74%). Of the participants, 2.8% had undetectable levels of CRP (5% for TNF-α and 62% for IL-1β, respectively). IL-6 levels were detectable in all participants. As demonstrated in previous studies (8,9), elevated levels of CRP were found to be associated with an increased risk of type 2 diabetes in the fully adjusted model (OR 1.9, 95% CI 1.2–3.2). With respect to clinical parameters and CRP, subgroup analysis (HbA_1c <5.8%) yielded results comparable to the analysis that included all participants. For example, the risk of individuals with elevated CRP levels to develop type 2 diabetes was 2.1 (95% CI 1.2–3.7) in the fully adjusted model in this subgroup. The mean HbA_1c was 4.5 ± 0.5% for control and 4.9 ± 0.5% for case subjects (P < 0.001) within the restricted subcohort. Correlations of cytokines, CRP, BMI, WHR, and HbA_1c are demonstrated in Table 2.

Mean baseline levels of IL-6 were higher among case subjects compared with control subjects (P < 0.0001). Elevated levels of IL-6 were associated with an increased risk of type 2 diabetes (risk estimates of individuals according to IL-6 quartiles are demonstrated in Table 3). After adjustment for all covariables (BMI, WHR, sports, age, sex, smoking status, educational attainment, alcohol consumption, and HbA_1c), IL-6 was found to be an independent predictor of type 2 diabetes (OR 2.57, 95% CI 1.24–5.47) This result was confirmed in conditional regression analysis (2.6, 1.2–5.9). Comparable and significant results were observed in analyses restricted to case subjects with HbA_1c <5.8% only. Within this subgroup analysis, the risk of individuals within the 4th quartile of IL-6 was also substantially increased (3.1, 1.3–7.4 [unconditional]; 3.7, 1.16–11.9 [conditional]) in the fully adjusted model.

Mean concentrations of TNF-α were higher in case subjects (2.04 ± 1.51 pg/ml) compared with control subjects (1.79 ± 1.28 pg/ml, P < 0.01). The risk of type 2 diabetes increased with increasing quartiles of TNF-α (Table 3). However, after adjustment for BMI or WHR, this association was no longer significant. Correlation analysis showed a mild correlation between TNF-α and BMI (r = 0.17, P < 0.001) or WHR (r = 0.15, P < 0.001). Again, analysis of participants with HbA_1c <5.8% yielded a similar picture as analysis of all participants. For example, the ORs in the fully adjusted model including HbA_1c were 1.0, 1.8 (95% CI 0.8–3.8), 0.9 (0.4–1.9), and 1.3 (0.6–3.1) for the increasing quartiles. Using conditional regression analysis also yielded no significant results regarding TNF-α–dependent relative risks after adjustment for BMI and/or WHR.

We found that 41.0% of case subjects and 36.6% of control subjects had detectable levels of IL-1β, which was not a statistically significant difference. According to these data, the relative risk of developing type 2 diabetes was not associated with circulating levels of IL-1β, independent of the model or type of analysis applied (Table 3).

Participants with a combined elevation of IL-6 levels and detectable levels of IL-1β were found to have an increased risk of future type 2 diabetes (OR 3.3, 95% CI 1.7–6.8 [unconditional]; 3.479, 1.078–11.222 [conditional]) compared with the low-level reference group in the fully adjusted model. In contrast, individuals with elevated levels of IL-6 but nondetectable levels of IL-1β had no significantly increased risk to develop type 2 diabetes (1.2, 0.6–2.5 [unconditional]; 1.5, 0.5–4.4 [conditional]) compared with the low-level reference group in the fully adjusted model. These results were confirmed by the inclusion of a formal interaction term between detectable IL-1β and elevated IL-6; this term was 3.3 (95% CI 1.1–9.7) and was significant (Table 4). Similar results were obtained including only individuals with HbA_1c <5.8%; here, the interaction term was 2.3 (1.1–5.1).

In crude analysis, individuals with a combined elevation of IL-6 and TNF-α or with a combined elevation of TNF-α and IL-1β had a substantially increased risk compared with individuals with elevated levels of IL-6 alone or compared with the low-level reference group. These effects of a combined elevation of IL-6 and TNF-α (OR 3.2, 95% CI 1.6–6.4) or IL-1β and TNF-α (2.3, 1.1–4.9) was still significant in the analysis restricted to participants with HbA_1c <5.8%. However, as described for TNF-α alone, this effect did not remain significant in the fully adjusted model. Results were again confirmed using conditional regression analysis and by calculation of formal interaction terms (Table 4).

We evaluated the effects of various inflammatory cytokines on the risk of type 2 diabetes. Participants with a combined elevation of IL-6 and IL-1β had a roughly threefold increased risk of developing type 2 diabetes compared with the low-level reference group. In contrast, participants with elevated levels of IL-6 alone (and undetectable levels of IL-1β) had no substantial increase of their diabetes risk. In this regard, IL-1β appears to have a permissive role in the IL-6–mediated acute-phase response preceding the onset of type 2 diabetes. Elevated levels of TNF-α were associated with an increased diabetes risk in the crude analysis. However, this TNF-α-dependent effect was no longer significant after adjustment for BMI or WHR. Furthermore, we found no significant effects of TNF-α on IL-1β- or IL-6-dependent risk estimates.

To the best of our knowledge, this is the first study to describe the combined effects of the three inflammatory cytokines IL-1β, IL-6, and TNF-α on the risk to develop type 2 diabetes. However, some limitations of this study need to be considered. It is well known that there is a relatively high proportion of individuals with undiagnosed type 2 diabetes among the general population (21,22). We therefore adjusted for blood glucose control by including HbA_1c into the fully adjusted model. To reduce the remaining potential bias of prevalent diabetes in case or control subjects at baseline, we also performed separate analyses among participants (case and control subjects) with HbA_1c <5.8%. A total of 94 individuals in our cohort provided fasting blood specimens. Within this subgroup, all case individuals with HbA_1c <5.8% had a baseline fasting glucose <7.0 mmol/l, which is sufficient to exclude diabetes in epidemiological studies, according to American Diabetes Association criteria. Thus, the number of individuals with prevalent diabetes at baseline (in case and control subjects) appears to be small in the analyses restricted to individuals with HbA_1c <5.8%. However, physician-diagnosed cases (as in this study) are likely to represent more progressed stages of diabetes compared with those diagnosed, for example, by an oral glucose tolerance test. Given that cytokine changes at baseline may be associated with the stage of diabetes, one might expect weaker associations if earlier stages of diabetes are investigated. In addition, some of the control subjects may have developed diabetes by the end of follow-up but were not diagnosed by a physician.

The robustness of results was additionally confirmed by inclusion of further covariates (preexisting hypertension and hyperlipidemia) into the model, although there is no clear functional evidence that these factors influence the development of type 2 diabetes. It is important to note that all of these additional analyses confirmed the findings described. Although BMI and WHR are more common clinical measures of obesity and central obesity, respectively, they may not fully account for the metabolic consequences of obesity and residual confounding even though adjustment for these parameters may exist. We analyzed various components of the metabolic syndrome, and cytokine effects were found to be independent of hypertension or hyperlipidemia. Although previous studies demonstrated that inflammatory markers are associated with future type 2 diabetes, even after adjustment for fasting insulin (6,8), it remains to be elucidated whether this holds true for the cytokine changes described here. Another potential bias may result from different precision of cytokine measurement. When two or more cytokines are entered into the statistical models, the relative strength of association between cytokines and disease can be expected to be highest for the cytokine with the least measurement error. In our study, inter- and intra-assay coefficients of variation were comparable in measurements of the three cytokines. However, additional sources of measurement variability (related to venipuncture, blood processing, or effects of long-term storage) may have differed for the three cytokines. Thus, we cannot entirely exclude that some of the above-mentioned limitations may have influenced our results.

Our data suggest that the pattern of inflammatory cytokines is important in the pathogenesis of type 2 diabetes. These findings are in line with the fact that inflammatory reactions depend on a cluster of cytokines rather than on single cytokines only. Patterns of cytokine production differ with different inflammatory conditions, and cytokines are components of a large complex signaling network (11–13). Several mechanisms, such as considerable influence of cytokines on lipid metabolism, may be important for the effects of combined elevations of different cytokines. For example, both IL-6 and IL-1β act on the liver to produce the characteristic dyslipidemia of the metabolic syndrome, with increased VLDL and decreased HDL (3). Combined elevation of IL-6 and IL-1β dramatically increased the expression of the acute-phase proteins, compared with the effect of each cytokine alone (11). Another potential molecular mechanism how inflammation may be involved in the pathogenesis of type 2 diabetes has been elucidated in recent elegant studies showing that sensitizing of insulin signaling by salicylates is induced via inhibition of the activity of IκB kinase β (23–25). IL-1β is well known to activate the IκB kinase β and might thereby induce insulin resistance.

In conclusion, our data support the concept that subclinical activation of the immune system is involved in the pathogenesis of type 2 diabetes. We demonstrated that a specific pattern of cytokines was associated with an increased risk of type 2 diabetes, rather than isolated elevation of the respective cytokines.

This project was supported by grants from Gottfried-Wilhelm-Leibnitz-Gesellschaft. Further grants to the authors were from the German Diabetes Association (104/03/2001 [to J.S. and M.M.] and 103/03/2001 [to M.R.]), Fritz-Thyssen-Stiftung (10.01.2.102), Deutsche Forschungsgemeinschaft (RI 1076/1-1), the Eli-Lilly International Foundation (to J.S. and A.F.H.P.), the European Union (SOC 95 201408 OSF02), and the Deutsche Krebshilfe (70-2488-HAI).

We thank K. Sprengel and S. Richter for laboratory assistance and U. Fiddicke and W. Bernigau for assistance with the study data. The HbA_1c analyses were conducted at the Department of Clinical Biochemistry, University of Greifswald, under the responsibility of Dr. Hans-Joachim Rose. We thank C.A. Barth for critical discussion of the project.