Complement Factor 3 Is Associated With Insulin Resistance and With Incident Type 2 Diabetes Over a 7-Year Follow-up Period: The CODAM Study Open Access

Complement factor 3 (C3) is an emerging risk marker for cardiovascular and metabolic diseases (1,2), and systemic C3 concentrations are closely linked to several measures of body fat and components of the metabolic syndrome (3,4). Complement C3 is produced mainly by the liver (5), but other production sites, such as adipose tissue, may also contribute to systemic C3 levels (6). C3 is the central component of the complement system, and activation via any of the three major complement pathways results in cleavage of C3 into C3a and C3b and subsequent activation of the terminal complement pathway with concurrent formation of C5a and C5b-C9 (also known as the membrane attack complex) (7). Both the anaphylatoxins C3a and C5a, by acting on their respective receptors, and the (sublytic) membrane attack complex have been shown to induce inflammatory responses (8–12).

Impairment of immune and inflammatory homeostasis is thought to cause type 2 diabetes mellitus (T2DM) through affecting insulin resistance (IR) and β-cell function (13,14). Local and/or systemic low-grade inflammation (LGI) is believed to play a major role in the development of IR in several organs (15–17). As examples, intravenous administration of tumor necrosis factor (TNF)-α and interleukin (IL)-6 induced IR in humans (18), and IL-6 and C-reactive protein (CRP) have been associated with development of T2DM in epidemiological studies (19,20). The complement system, as a regulator of both the innate and the adaptive immune system, represents an important part of this inflammatory response and may also contribute to the pathogenesis of T2DM.

Several cross-sectional studies have shown associations of plasma C3 with IR, measured by homeostasis model assessment of IR or hyperinsulinemic–euglycemic clamps, and with T2DM (21–23). However, prospective or longitudinal studies on the relation between C3 and the progression of IR and/or the incidence of T2DM are scarce. To the best of our knowledge, there are no longitudinal data on the relation between (changes in) C3 levels and (changes in) IR. One prospective cohort study showed a positive association of serum C3 with incident T2DM, but this cohort consisted of only adult men (24), and data on women are not yet available. For these reasons, we explored the relation of plasma C3 levels with progression of IR and glucose tolerance, as well as with incidence of T2DM in a prospective cohort of Caucasian subjects with 7-year follow-up and an extensive (metabolic) characterization of participants.

The Cohort on Diabetes and Atherosclerosis Maastricht (CODAM) study is a prospective, observational study on, among others, the natural progression of IR and glucose tolerance. A total of 574 individuals were selected from a large population-based cohort as described in detail elsewhere (4,25) and were extensively characterized at baseline with regard to lifestyle and cardiovascular and metabolic profile. After a median of 7.0 years (interquartile range 6.9–7.1), 495 subjects participated in the follow-up measurements (Fig. 1). The CODAM was approved by the medical ethics committee of the Maastricht University Medical Centre, and all subjects gave written informed consent.

For the current study, we excluded participants on insulin therapy (n = 13 at baseline, n = 40 at follow-up). Baseline data were complete in 545 subjects, except for data derived from the oral glucose tolerance test (OGTT), which were complete in 503 subjects. Follow-up data were complete in 394 subjects, of whom 342 also had complete OGTT data at follow-up. Generally, the 545 included participants had a better metabolic profile than the 29 excluded subjects (data not shown). Similarly, the 394 subjects with complete follow-up data had better metabolic health at baseline than the 180 subjects without or with incomplete follow-up data (data not shown).

Some biomarkers that were obtained in the CODAM study, i.e., C3, insulin, and the inflammatory markers (IL-6, high-sensitivity [hs] CRP, serum amyloid A [SAA], and soluble intercellular adhesion molecule [sICAM]-1), were measured at the time the baseline evaluation of the CODAM study was completed (4,25). More recently, paired (re)measurements of these biomarkers at baseline and follow-up were performed in nearly all subjects (n = ∼550), but with different methods. To make use of all available data and to ensure comparability of baseline and follow-up values, we calculated the mean of both baseline measurements after calibration of previous to recent baseline measurements using Deming regression models, which is recommended for method comparison (26), as described before (27).

Participants were asked to stop their lipid-lowering drugs 14 days before the visit and stop all other medication the day before the visit. After an overnight fast, venous blood samples were collected for assessment of biomarkers. Serum was allowed to clot at room temperature for 45 min; after centrifugation at 3,000 rpm for 15 min, serum aliquots were stored at −20°C, and plasma (EDTA) aliquots were stored at −80°C until use. Samples were thawed only once prior to measurements. At baseline, C3 levels were determined in serum by an automatic analyzer (Hitachi 912) with a Roche kit (Roche Diagnostics Nederland B.V.)—interassay coefficient of variation (CV) of 2.1%. Paired measurements of plasma C3 levels at baseline and follow-up were also performed using the IMMAGE immunochemistry system C3 assay (Beckman Coulter)—interassay CV was 7%. For baseline C3 values, the mean of the two C3 measurements (after calibration) was used for further analyses. At baseline, C3a levels were determined in EDTA plasma by ELISA (MicroVue C3a Plus EIA kit, Quidel) as previously described (28), and soluble C5b-C9 (sC5b-C9) was measured in citrate plasma by ELISA (MicroVue SC5b-9 EIA kit, Quidel).

A standard 75-g OGTT was performed with measurement of glucose levels at 0, 30, 60, and 120 min and at follow-up also at 15 min. At follow-up, an additional fasting glucose measurement was obtained from a second visit, and the mean of these fasting glucose levels was used for further analyses. All plasma glucose levels were measured in NaF/KOx plasma with a hexokinase glucose-6 phosphate dehydrogenase method (ABX Diagnostics)—interassay CV <5%.

Fasting insulin concentrations at baseline were determined in EDTA plasma using a two-sided immunoradiometric test (immunoradiometric assay) using paired monoclonal antibodies (Medgenix Diagnostics)—interassay CV was 6.0% (4,25). Paired measurements of plasma insulin levels were performed at baseline and follow-up at all OGTT time points, using a multiarray detection system based on electrochemiluminescence technology (SECTOR Imager 2400, Meso Scale Discovery, Gaithersburg, MD)—interassay CV was 9.7%. For fasting insulin at baseline, the mean of the two measurements (after calibration) was used for analyses. Fasting nonesterified free fatty acids (NEFAs) at baseline were measured in EDTA plasma using an enzymatic calorimetric NEFA C method (Wako Diagnostics, Richmond, VA)—interassay CV <5%. In addition, paired measurements of NEFAs at baseline and follow-up were performed using the same method. At baseline, the mean of the two measurements was taken for analyses.

Total areas under the curve (AUCs) for glucose and insulin during the OGTT (for the first 30 min and for 120 min) were calculated using the trapezoidal method. Homeostasis model assessment (HOMA2) of IR, which has been shown to correlate well with (muscle) glucose disposal in clamp studies (29), was computed using the HOMA2 calculator (http://www.dtu.ox.ac.uk/homacalculator/index.php). Hepatic IR was estimated using the square root of the product of the AUCs for glucose and insulin during the first 30 min of the OGTT—i.e., SQRT(glucose_0–30 [AUC] × insulin_0–30 [AUC])—converted to conventional units ([mg/dL·h] × [uIU/mL·h]) (30,31). This index has been developed and validated against the product of fasting plasma insulin and endogenous glucose production in clamp studies and has been postulated to represent hepatic IR, as it takes into account not only fasting glucose, but also the early suppression of fasting glucose by insulin during the first phase of the OGTT. Adipocyte IR was calculated as the product of fasting insulin and fasting NEFA concentrations, as described before (25,32,33).

BMI, waist circumference, smoking behavior, family history of T2DM (first-degree relatives), dietary calorie intake, mean daily alcohol consumption, physical activity, and use of antihypertensive, glucose-lowering, and lipid-lowering medication were determined at baseline and follow-up as previously described (4,25). IL-6, hs-CRP, SAA, and sICAM-1 were determined at baseline using methods described before (4). In addition, paired measurements of baseline and follow-up IL-6, hs-CRP, SAA, sICAM-1, IL-8, and TNF-α were determined in EDTA on a multiarray detection system based on electrochemiluminescence technology (SECTOR Imager 2400, Meso Scale Discovery) (27,34). At baseline, when applicable, the mean of two measurements was used for analyses after calibration. Estimated glomerular filtration rate (eGFR) and prior cardiovascular disease (CVD) at baseline, and glucose metabolism status (i.e., normal glucose metabolism, impaired glucose metabolism, or T2DM) both at baseline and follow-up, were determined as previously described (4,35,36).

Variables with a skewed distribution (i.e., HOMA2-IR, hepatic IR, adipocyte IR, fasting glucose, postprandial [2 h] glucose, AUC_glucose, all inflammatory markers, C3a) were log_e transformed prior to further analyses. A LGI score was calculated by averaging the Z-scores (i.e., [individual’s observed values – population mean]/SD) of the six (log_e transformed) inflammatory markers (IL-6, IL-8, TNF-α, hs-CRP, SAA, and sICAM-1) (4,25,34), with the population mean and SD for each individual circulating biomarker based on its average of the baseline and follow-up measurements. Characteristics at baseline and at follow-up were compared using the paired t test or Wilcoxon signed rank test for continuous variables and the McNemar test for dichotomous variables.

We performed three different statistical analyses to examine the association of plasma C3 levels with IR and glucose tolerance, each examining different aspects of this relationship and providing additional information (37,38). First, we investigated the overall longitudinal associations of C3 levels with IR and glucose tolerance over the 7-year period in all subjects (n = 545) with generalized estimating equations (GEE) with an exchangeable correlation structure to account for the correlation of repeated measurements within subjects (38). This model indicates whether there is any association at all between C3 and the outcome, either between subjects (differences) and/or within subjects (changes through 7 years). Because outcome variables were log_e transformed, regression coefficients were expressed in percentage change (e.g., percentage increase in HOMA2-IR per 0.1 g/L increase in C3). Second, we investigated whether changes in plasma C3 levels were associated with (absolute) changes in IR and glucose tolerance during the 7-year follow-up in subjects with follow-up data (n = 394), using linear regression, to address specifically the within-subject component of these associations. As changes in IR and glucose tolerance were normally distributed, the regression coefficients were expressed as changes in outcome variables (e.g., unit increase in the change in HOMA2-IR per 0.1 g/L increase in the change in C3). Finally, to investigate clinical outcomes, we investigated whether baseline plasma C3 levels were associated with 7-year cumulative incidence of T2DM in the subjects without T2DM at baseline (n = 333), using logistic regression analysis.

According to the outcome (IR versus glucose tolerance) and the model that was used (GEE, change, logistic), analyses were adjusted for relevant potential confounders. Analyses of IR were adjusted for glucose metabolism status, but analyses of glucose tolerance were not, because of potential collinearity. The GEE analyses were adjusted for both measurements (i.e., baseline and follow-up) of the potential confounders (when available) and adjusted for follow-up time. The change analyses were adjusted for changes in (time-dependent) covariates. In the prospective logistic regression analyses, baseline C3 and baseline covariates were used. As such, analyses were adjusted for age, sex, follow-up time, and (when appropriate) glucose metabolism status (model 1) and prior CVD, eGFR, smoking status, alcohol consumption, dietary energy intake, physical activity, family history of T2DM, and use of medication (model 2). Subsequently, (changes in) waist circumference (model 3) and the LGI score (model 4) were added. We also investigated whether the associations differed between sexes by adding interaction terms between C3 and sex to the (fully adjusted) models. A two-sided P value of <0.05 was considered statistically significant. For interaction, P < 0.1 was considered statistically significant. However, since interaction analyses are often underpowered, we included the results of sex-stratified analyses in supplemental files regardless of the P value for interaction. All analyses were performed using STATA Data Analysis and Statistical Software for Windows, version 9.0/SE (StataCorp LP, College Station, TX).

Table 1 shows characteristics of the study population at baseline (n = 545) and follow-up (n = 394). Generally, most variables remained unchanged over the 7-year follow-up period. Indices of IR showed a small decrease over time, while 2-h glucose and AUC glucose and markers of LGI increased over time. The mean plasma C3 level increased over time from 1.01 to 1.14 g/L, while adiposity measures remained relatively stable. Pearson correlation coefficients of C3, LGI, and HOMA2-IR with the individual inflammatory markers at baseline are shown in Supplementary Table 1.

We examined the overall association of plasma C3 levels with IR and glucose tolerance using GEE analyses (n = 545 at baseline, of which n = 394 also at follow-up). Over the 7-year follow-up period, C3 levels (per 0.1 g/L) were longitudinally associated with HOMA2-IR, hepatic IR, and adipocyte IR after adjustments for age, sex, follow-up time, and glucose metabolism status (Table 2) (model 1) and remained significant after subsequent adjustment for potential confounders (model 2): β = 15.2% (95% CI 12.9–17.6) for HOMA2-IR, β = 6.1% (95% CI 4.7–7.4) for hepatic IR, and β = 16.0% (95% CI 13.0–19.1) for adipocyte IR. C3 levels (per 0.1 g/L) were also longitudinally associated with fasting glucose (β = 1.8% [95% CI 1.2–2.4]), 2-h glucose (β = 5.2 [95% CI 3.7–6.7]), and AUC_glucose (β = 3.6% [95% CI 2.7–4.6]) (model 2). When these associations were further adjusted for waist circumference (model 3) and the LGI score (model 4), these positive associations were attenuated but remained statistically significant: β = 8.7% (95% CI 6.5–11.0) for HOMA2-IR, β = 3.9% (95% CI 2.4–5.3) for hepatic IR, β = 8.0% (95% CI 5.3–10.9) for adipocyte IR, β = 0.8% (95% CI 0.2–1.5) for fasting glucose, β = 3.5% (95% CI 1.9–5.2) for 2-h glucose, and β = 2.2% (95% CI 1.2–3.2) for AUC_glucose. We observed similar effect sizes in sex-stratified analyses (Supplementary Table 2) (all P values for interaction >0.10).

To address specifically the within-subject component of these associations, we next investigated whether changes in C3 levels were associated with changes in IR and glucose tolerance in the 394 subjects with baseline and available follow-up data. During the 7-year follow-up, greater changes in C3 levels (per 0.1 g/L) were positively associated with changes in HOMA2-IR (β = 0.18 [95% CI 0.12–0.24]), changes in hepatic IR (β = 1.31 [95% CI 0.64–1.98]), and changes in adipocyte IR (β = 3.23 [95% CI 1.09–5.37]) after adjustment for changes in age, sex, and glucose metabolism status (Table 3) (model 1). These associations, again, remained unchanged after further adjustment for changes in potential confounders (model 2). Further adjustment for changes in waist circumference (model 3) and changes in the LGI score (model 4) resulted in attenuation, but the associations of changes in C3 with changes in HOMA2-IR (β = 0.08 [95% CI 0.02–0.15]) and changes in hepatic IR (β = 0.87 [95% CI 0.12–1.61]) remained statistically significant. Finally, no statistically significant associations were observed between changes in C3 levels and changes in fasting glucose, 2-h glucose, or AUC_glucose during the 7-year follow-up. We observed similar effect sizes in sex-stratified analyses (Supplementary Table 3) (all P values for interaction > 0.10).

After exclusion of subjects with T2DM at baseline (n = 61), 333 subjects were available for logistic regression analysis on incident T2DM. During the 7-year follow-up, 57 out of 333 subjects developed T2DM (17%). Baseline C3 levels (per 0.1 g/L) were associated with a higher risk of developing T2DM (odds ratio [OR] = 1.6 [95% CI 1.3–2.0]) after adjustment for baseline age and sex. Also, after adjustment for prior CVD, eGFR (baseline only), smoking status, alcohol consumption, dietary energy intake, physical activity, family history of T2DM, use of medication (model 2), waist circumference (model 3), and the LGI score (model 4), baseline C3 remained independently associated with incident T2DM (OR model 4 = 1.5 [95% CI 1.1–2.0]). Similar associations were obtained in sex-stratified analyses (Supplementary Table 4, all P values for interaction >0.10). Additional adjustment for baseline HOMA2-IR did also not change this effect size (OR = 1.5 [95% CI 1.1–2.0]).

To account for the possibility that C3 levels were affected by the presence of infectious or inflammatory diseases at the time of blood sampling, we also repeated all analyses excluding subjects with hs-CRP levels >10 mg/L (n = 37 at baseline; n = 41 at follow-up). This did not materially change the associations of C3 with metabolic outcomes (Supplementary Tables 5–7). In addition, similar associations were observed after stringent exclusion of subjects with any self-reported (history of) pulmonary, renal, gastrointestinal, thyroid, or rheumatic disease or cancer at baseline (n = 200, Supplementary Tables 5–7). Finally, C3 levels may be influenced by female sex hormones (39,40), but exclusion of premenopausal women (21%) and women using hormone replacement therapy (9%) at baseline did not influence our findings (total excluded n = 63) (Supplementary Tables 5–7).

In order to substantiate the possibility that C3 may play an active role in the development and progression of HOMA2-IR and/or glucose intolerance, we performed cross-sectional analyses of baseline levels of C3a and sC5b-C9 as readouts of complement activation. Indeed, plasma levels of C3a and sC5b-C9 were significantly associated with HOMA2-IR and adipocyte IR as well as with 2-h glucose and AUC_glucose in models adjusted only for age, sex, and glucose metabolism (Supplementary Table 8). However, the associations were attenuated and lost significance after adjustments for other covariates, with only the association of sC5b-C9 with AUC_glucose remaining statistically significant (β = 0.06 [95% CI 0.01–0.12]). Finally, plasma C3a at baseline (per 10% increase) was significantly associated with incident T2DM after adjustment for age and sex (OR = 1.09 [95% CI 1.01–1.18]), but this association was attenuated after adjustment for the other covariates (OR = 1.03 [95% CI 0.94–1.14]) (Supplementary Table 9).

In this prospective cohort study we showed that during a 7-year follow-up, baseline levels of C3, the central complement component, were independently associated with incident T2DM in both men and women and that plasma C3 levels were independently associated with estimated IR in muscle, liver, and adipocytes, as well as with glucose tolerance. Moreover, changes in C3 levels were associated with changes in muscle IR, hepatic IR, and adipocyte IR (although the latter not independently of changes in obesity), but not with changes in glucose tolerance. The magnitude of the effects observed by GEE analyses was comparable to those of the change models (e.g., using the median HOMA2-IR of 1.51 at baseline, a 8.7% increase for HOMA2-IR [model 4] corresponds to an absolute change of 0.13 units in HOMA2-IR).

To the best of our knowledge, this is the first longitudinal study on the association of circulating C3 with detailed measures of IR as well as glucose tolerance. Our observations extend the previously reported association of plasma C3 with incident T2DM in a prospective cohort study in men (24) by showing that this association was present not only in men, but also in women. The effect size observed in our study (OR 1.51 per 0.1 g/L) was somewhat larger than the effect reported by Engström et al. (24) (OR 1.37 per 0.2 g/L), potentially related to a selection of subjects with increased risk of cardiometabolic diseases in the current study (4,25). The 7-year incidence of T2DM in CODAM was 17%, which was substantially higher than the 6-year incidence of 4.5% reported by Engström et al. (24) for the cohort of men participating in a screening program. Both studies adjusted for a comparable set of potential confounders, with our present study also adjusting for six common markers of LGI. Importantly, our change analyses showed that within-subject changes in C3 levels are also associated with changes in each of the three measures of IR used, but not with changes in glucose tolerance. Altogether, these findings are consistent with the hypothesis, but do not prove, that plasma C3 levels may induce progression of IR and eventually lead to T2DM.

In some analyses, we observed an attenuation of effect sizes after adjustment for waist circumference, particularly for adipocyte IR (Table 3) (model 3). This may to some extent be interpreted as simple confounding since abdominal obesity is a known determinant of both IR and plasma C3 levels (3,41–43). However, the relationship of waist circumference with C3 is not straightforward, because abdominal obesity may also be considered an ascending proxy in the causal pathway between C3 and IR (Supplementary Fig. 1B). Adjustments for an ascending proxy are not always prudent, as they can attenuate estimates due to differences in bias and variance between waist circumference and C3 (44). Moreover, it has been suggested that C3 may actually cause obesity (45), in which case changes in obesity should be considered a mediator for which adjustment is not wanted (Supplementary Fig. 1C). Likewise, LGI may be considered an intermediate factor and/or an ascending proxy in the pathway between C3 and IR, but we cannot not exclude the possibility that LGI (partly) represents a simple confounder, as proinflammatory cytokines that cause IR (17,18) may also increase C3 levels (46,47). Because of the observational nature of our data, we are unable to disentangle whether waist circumference and LGI are simple confounders, ascending proxies, or mediators. Therefore, the true effect sizes may actually lie between models 2 and 3/4, as these latter models may have been underestimations due to unnecessary adjustment for waist circumference and LGI. Nevertheless, the associations of changes in C3 with changes in HOMA2-IR and hepatic IR remained statistically significant in all models.

The observed associations of C3 with IR over time and with incident T2DM can be explained by at least two pathophysiological mechanisms that may occur simultaneously. First, higher plasma levels of C3 may be associated with more complement activation and higher levels of its activated product C3a as well as the terminal pathways activation products C5a and sC5b-C9. The anaphylatoxins C3a and C5a, by acting on their respective receptors (48–50), attract and activate leukocytes and increase the synthesis of proinflammatory cytokines by several cell types, including leukocytes and Kupffer cells (8–12,50,51). Indeed, blocking of anaphylatoxic pathways in C3aR- or C5aR-knockout mice fed a high-fat diet led to decreased adipocyte size, less liver steatosis, less adipose tissue infiltration by macrophages, a reduction in tissue and plasma proinflammatory cytokines, and less systemic IR (48,51), and inhibition of C3a and C5a signaling by blocking their receptors appeared to induce similar effects (50). As such, complement activation and the subsequent local and/or systemic inflammation may actively contribute to the development of IR and T2DM. In agreement, we observed significant associations of C3a and sC5b-C9 with metabolic outcomes at baseline and for C3a also with incident T2DM, although not in the fully adjusted analyses. In addition, we observed a modest attenuation of the observed associations of C3 with HOMA2-IR and adipocyte IR after adjustment for LGI (model 4), which could be interpreted as mediation. This mediating effect might even have been underestimated since our LGI score did not include IL-β, which is an important proinflammatory cytokine that is upregulated upon inflammasome activation by C3a and (sublytic) C5b-C9 (8–10).

Second, C3 may not by itself contribute actively to the development of IR, but rather serve as a marker of adipocyte dysfunction. Adipocytes have been shown to produce C3 and activate it to C3a and finally C3a-desarg, also known as acylation-stimulating protein (ASP), which promoted storage of glucose and lipids in adipocytes in a way similar to insulin (52–54). Because C3 levels also correlate with postprandial triglyceride levels in vivo (23,55), it was hypothesized that high levels of C3 are a marker of high postprandial triglycerides, for example, due to ASP resistance. ASP resistance, which may be initiated by fatty acids or the proinflammatory cytokine TNF-α (56,57), would then result in a redistribution of glucose and lipids to other organs, leading to IR in liver and muscle (23,47,54,55). Indeed, in mouse models, interruption of the ASP system led to decreased triglyceride storage in adipose tissue, delayed lipid clearance, impaired glucose tolerance, and IR (58,59). Data in humans are, however, not yet available.

A major strength of the present longitudinal cohort study is the comprehensive analysis of the association of C3 with detailed measures of both IR and glucose tolerance at baseline and 7-year follow-up, adjusted for major (potential) confounders. As for any longitudinal study, this set of analyses is limited by the fact that not all subjects attended the follow-up evaluation, and our effect sizes may have been underestimated because of selective attrition (unhealthy subjects not attending follow-up) (60). Also, despite the availability of longitudinal data, the observational nature of our study does not allow us to draw definite conclusions on causality, and therefore we cannot fully exclude a reverse causation scenario. Insulin may have several effects on C3 transcription and production (61), although hyperinsulinemic clamp studies did not observe changes in adipose tissue C3 mRNA expression or serum C3 levels (22,62). Finally, our study consisted of middle-aged and older Caucasian subjects who were selected on the basis of an increased risk for metabolic disease and CVD, and extrapolation to other study populations or other ethnicities should be done with caution.

In conclusion, we have shown that baseline C3 levels were independently associated with incident T2DM in men and women at 7-year follow-up. In addition, (changes in) C3 levels were strongly associated with (changes in) IR in muscle, liver, and adipocytes. Changes in plasma C3 levels may therefore reflect progression of metabolic dysregulation, including progression of LGI and IR, that eventually leads to abnormalities in glucose tolerance and frank T2DM. Moreover, higher C3 levels may be associated with complement activation in liver and adipose tissue. Whether C3 itself and/or downstream complement activation products can causally contribute to the development of IR and T2DM remains to be investigated in humans.

Funding. Research activities of I.F. are supported by a postdoc research grant from the Netherlands Heart Foundation (grant 2006T050). Part of this work was supported by grants of the Netherlands Organisation for Scientific Research (940-35-034) and the Dutch Diabetes Research Foundation (98.901) and by Project Prediction and Early Diagnosis of Diabetes and Diabetes-Related Cardiovascular Complications (grant 01C-104) performed within the framework of the Center for Translational Molecular Medicine (www.ctmm.nl).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. N.W., M.M.J.v.G., and I.F. researched data, contributed to discussion, and wrote and edited the manuscript. E.J.M.F. researched data, contributed to discussion, and reviewed the manuscript. C.J.H.v.d.K., C.G.S., B.B., and C.D.A.S. contributed to discussion and reviewed and edited the manuscript. N.W. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.