Leukocyte Profiles Differ Between Type 1 and Type 2 Diabetes and Are Associated With Metabolic Phenotypes: Results From the German Diabetes Study (GDS)

Hyperglycemia defines both type 1 and type 2 diabetes irrespective of the different underlying pathogenesis. In type 1 diabetes, T cell–mediated β-cell destruction accompanied by islet-directed autoantibodies persists for a variably long period until hyperglycemia occurs (1). In addition, innate immunity is upregulated and systemic immune markers are altered, which may affect β-cell function (2,3). Altered immune reactivity also accompanies and precedes type 2 diabetes and its associated complications by mechanisms related to subclinical inflammation (4). Although adipose tissue and innate immune cells contribute to increased release of cytokines and chemokines, cell types responsible for subclinical inflammation in type 2 diabetes remain to be precisely defined (5).

White blood cell (WBC) count has long been used for the diagnosis of inflammatory diseases. A recent systematic meta-analysis reported higher WBC counts in persons with than without type 2 diabetes (6). Of note, patients with poorly controlled diabetes have a greater risk for infections and atherosclerosis (7), both being independent causes or consequences of inflammation and abnormal WBC.

Recent analyses addressed the role of pro- and anti-inflammatory immune cell subsets in diabetes with a focus on regulatory T (Treg) cells, which can downregulate and modulate proinflammatory or autoimmune cellular immunity (8,9). In type 1 diabetes, function or life span of Treg cells was found to be increased (10), decreased (8,9,11), or unchanged (12,13). This discrepancy is likely due to differences in disease duration and progression as well as methodological aspects (12,14). Similarly, the few published studies on type 2 diabetes reported either decreased (15,16) or unchanged (9) frequencies of Treg cells.

Here, we tested the hypothesis that the systemic cellular immune status relates not only to type of diabetes but also to the metabolic phenotype. To this end, we analyzed patients of the prospective observational German Diabetes Study (GDS), which consecutively includes patients with recently diagnosed diabetes at adult age, aiming at characterization of their phenotypes and monitoring their disease progression.

Patients aged 18–69 years with either type 1 or type 2 diabetes are recruited for the ongoing GDS. General inclusion criteria of GDS are new-onset type 1 or type 2 diabetes (diabetes duration <12 month) and age between 18 and 69 years. Exclusion criteria of GDS are pregnancy; diabetes other than type 1 or type 2; clinical coronary artery disease; hepatic, renal, psychiatric, or addictive diseases; or immunosuppressive treatment. Diagnosis of type 1 diabetes is based on ketoacidosis with immediate insulin substitution, detection of at least one islet cell–directed autoantibody (islet cell autoantibody, GAD, islet antigen-2 antibody) or of C-peptide (C-Pep) below the detection limit. Those characterized as type 2 diabetic patients are tested to be islet antibody negative and have a typical history of type 2 diabetes presenting with hyperglycemia and are not requiring immediate insulin treatment. GDS patients are reinvestigated 5 and 10 years after their first visit.

For the current study, 194 patients (62 type 1 and 132 type 2 diabetic, diabetes duration ≤5 years) of the GDS were consecutively included between February 2011 and May 2012. Mean duration of diabetes here was 1.29 ± 1.85 years for type 1 diabetic and 1.98 ± 2.25 years for type 2 diabetic patients (P = 0.035). Healthy volunteers (n = 60) were fulfilling the inclusion and exclusion criteria of GDS except for the presence of diabetes. They underwent a standardized 75-g oral glucose tolerance test to exclude dysglycemia. All participants gave written informed consent for the study protocol, which was approved by the ethics board of Heinrich-Heine University.

Automated blood cell count for leukocytes including lymphocytes, monocytes, and granulocytes was performed on the Sysmex KX21 (Sysmex Corporation, Kobe, Japan). Fasting blood glucose (FBG) was measured by the hexokinase method (Epos analyzer 5060; Eppendorf, Hamburg, Germany), C-Pep chemoluminimetrically (Immulite 1000; Siemens, Erlangen, Germany) and hemoglobin A_1c (HbA_1c) using high-pressure liquid chromatography (Varianz II; Bio-Rad Laboratories, Richmont, CA). Plasma total cholesterol (TC), HDL and LDL cholesterol, and triglycerides (TG) were measured at the Institute for Clinical Chemistry and Laboratory Diagnostic at the University Clinics Düsseldorf with standardized methods.

For assessment of insulin sensitivity, a subgroup of 114 type 2 and 52 type 1 diabetic patients and 29 normoglycemic control subjects underwent a clamp test. No participants consumed alcohol for at least 24 h, ingested food for 10–12 h, or smoked for at least 8 h before the clamp. Patients also stopped their glucose-lowering medication for at least 3 days or their regular insulin after the evening dose before the day of the clamp (17). As part of the GDS study, patients underwent an intravenous glucose tolerance test prior to the clamp test (18). The hyperinsulinemic-euglycemic clamp was started with a priming insulin dose (10 mU ⋅ [kg body wt]⁻¹ ⋅ min⁻¹ for 10 min) followed by continuous insulin infusion of 1.5 mU ⋅ (kg body wt)⁻¹ ⋅ min⁻¹, corresponding to 66 mU/m² ⋅ min (Insuman Rapid; Sanofi, Frankfurt, Germany). Blood glucose concentrations were measured every 5 min and maintained at 5 mmol/L with a variable intravenous 20% dextrose infusion. In healthy control subjects, primed continuous insulin infusion (40 mU/m² ⋅ min) was used. Rates of whole-body insulin sensitivity are given as mean glucose infusion rates (M value: mg glucose ⋅ kg⁻¹ ⋅ min⁻¹) per individual mean plasma insulin concentrations (I: mU insulin ⋅ L⁻¹) during steady state (M/I).

Fresh venous blood was drawn in the morning into sodium-heparin tubes from fasted individuals and examined within 2 h. Leukocytes were analyzed by FACS and flow cytometry using a dual laser FACSCalibur cytometer (Becton Dickinson, Heidelberg, Germany) and Cellquest software (Becton Dickinson). Briefly, blood cells were stained with fluorescence-conjugated antibodies in four different colors. After lysis of erythrocytes (Lysis buffer; Becton Dickinson) and two washes, stained PBMC were resuspended and fixed with CellFIX (BD Biosciences). Twenty thousand lymphocytes were collected in a forward scatter/side scatter (FSC/SSC) lymphocyte gate and saved together with the monocytes and granulocytes. The flow cytometer was calibrated daily with appropriate single-stained samples for setting compensation, and acquired data were analyzed by FlowJo software (version 7.6.1; TreeStar, Inc., Ashland, OR). The following fluorescence-conjugated antibodies were used to identify cell populations: CD3 (T cells), CD4 (T helper cells [Th cells]), CD8 (cytotoxic T cells [Tc cells]), CD14 (monocytes), CD19 (B cells), CD25 (activated/regulative T cells [Treg cells]), CD56 (natural killer cells [NK/NKT cells]) (all from BD Biosciences, Heidelberg, Germany), CD183 (Th1 cells), CD194 (Th2 cells), CD103 (integrin on Treg cells) (all from BD Pharmingen, Heidelberg, Germany), and CD127 (FoxP3 cells) (BioLegend, San Diego, CA).

Data are presented as means ± SD or as individual data and median values. Comparisons between groups were done using the nonparametric Kruskall-Wallis test followed by the Dunn test. ANCOVA was performed to compare groups adjusting for the potentially confounding variables age, sex, and BMI. A closed testing procedure was applied for adjusted pairwise comparisons. All reported significant differences between groups are results of the adjusted analyses unless stated otherwise. Spearman nonparametric correlations and partial correlations adjusted for age, sex, and BMI were estimated. Two-sided P values ≤0.05 were considered to indicate statistical significant differences. All analyses were performed using GraphPad PRISM 4 (GraphPad Software, San Diego, CA) and IBM SPSS software (version 21, 2013; SPSS, Chicago, IL).

Type 2 diabetic patients were older and had higher BMI, fasting C-Pep, TC, LDL, and TG but lower HDL than type 1 diabetic and control subjects (Table 1). Type 2 and type 1 diabetic patients had higher FBG and higher HbA_1c than control subjects. Whole-body insulin sensitivity (M/I, expressed as mg glucose ⋅ kg⁻¹ ⋅ min⁻¹/mU insulin ⋅ L⁻¹) was 66% lower in type 2 diabetic (0.042 ± 0.025) and 45% lower in type 1 diabetic (0.067 ± 0.035) than in control subjects (0.122 ± 0.072; Kruskal-Wallis test: P < 0.001; Dunn multiple comparison test: type 2 diabetic vs. type 1 diabetic P < 0.001, type 2 diabetic vs. control P < 0.001, type 1 diabetic vs. control P < 0.05).

After adjustment for age, sex, and BMI, type 2 diabetic patients had more leukocytes (P ≤ 0.001 and P ≤ 0.001), lymphocytes (P ≤ 0.001 and P = 0.005), and granulocytes (P = 0.004 and P ≤ 0.001) than type 1 diabetic patients and control subjects, whereas there were no differences between type 1 diabetic and control subjects (Fig. 1A–C). Monocyte counts were higher in type 2 than in type 1 diabetic patients (P = 0.006) but comparable in control subjects (Fig. 1D).

NK cells are identified by their CD56 positivity and are part of the innate immune system owing to their ability to rapidly lyse infected cells at the first encounter without prior immunization. Type 2 diabetic patients had lower proportions of CD3⁻CD56⁺ NK cells than type 1 diabetic patients (P = 0.037) and control subjects. In the total study population, NK cells associated positively with age (r = 0.161, P ≤ 0.01) and negatively with sex (r = −0.24, P ≤ 0.001), indicating higher values for men. CD19⁺ B cells belong to the adaptive immune system and are responsible for the humoral immune response by the production of specific antibodies. Numbers of CD19⁺ B cells did not differ between groups. The frequency of CD3⁺ T cells, the main cell subset of cell-mediated specific immunity, was higher in type 2 than in type 1 diabetic patients (P = 0.038), with no differences in other group comparisons. Males had fewer CD3⁺ T cells than females (P ≤ 0.001) in all groups combined (Supplementary Table 1).

Upon contact with the first antigen, immature T cells develop to CD4⁺ Th cells or CD8⁺ Tc cells, both being key players in the adaptive immune system. The CD4⁺/CD8⁺ ratio can be used to describe the immune balance in immune disorders such as HIV or autoimmune diseases. In our study, proportions of CD4⁺ Th cells were higher (P ≤ 0.001), whereas those of CD8⁺ Tc cells were lower (P = 0.002) in type 2 diabetic patients than in control subjects. These differences were lost after adjustment for age, sex, and BMI, mainly due to opposing correlation with age (CD4: r = 0.337, P < 0.001; CD8: r = −0.286, P < 0.001) (Supplementary Fig. 1). NKT cells act at the interface of the innate and adaptive immune system. Upon activation, NKT cells produce large amounts of cytokines (interleukin [IL]-2, IL-4, interferon-γ, tumor necrosis factor [TNF]α), and impaired function of this cell type contributes to the development of autoimmune diseases. In our investigations, the percentage of NKT cells did not differ between groups (data not shown). Th cells were further classified into proinflammatory Th1 and anti-inflammatory Th2 cells according to their surface epitopes, chemokine receptors CXCR3 (CD183) and CCR4 (CD194). After adjustment for age, sex, and BMI, the proportion of CD183⁺ Th1 cells was not different between type 2 diabetic and control subjects but was higher in type 1 diabetic compared with control subjects (P = 0.007) (Fig. 2A). CD194⁺ Th2 cells were not different between the groups, while double-positive CD183⁺CD194⁺ Th cells were more frequent in both type 2 diabetic (P = 0.005) and type 1 diabetic (P ≤ 0.001) than in control subjects.

Subsets of Th cells (CD4⁺) were gated for surface expression of CD25, the IL-2 receptor α-chain, which is expressed in association with CD127, the α-chain of the receptor of IL-7 (14,19). Intensity of CD25⁺ staining distinguishes activated/effector (CD4⁺CD25⁺) and regulatory T cells (CD4⁺CD25⁺⁺), while low or negative staining of CD127 identifies positivity for the intracellular transcription factor, FoxP3 (Forkhead box protein 3: CD4⁺CD25⁺⁺CD127⁻) (14). Regulatory T cells suppress immune responses by a cell contact–dependent mechanism or by secretion of cytokines (IL-10, transforming growth factor-β). The frequency of CD4⁺CD25⁺-activated Th cells was higher in type 2 diabetic (P < 0.001) and type 1 diabetic (P < 0.001) than in control subjects (Fig. 2B). Also, the subset of CD4⁺CD25⁺⁺ Treg cells was slightly higher in type 2 diabetic (P = 0.046) and clearly higher in type 1 diabetic (P = 0.005) compared with control subjects (Fig. 2C). The percentage of FoxP3⁺ Treg cells of activated Th cells tended to be lower in type 2 diabetes (P = 0.052) and was lower in type 1 diabetes (P = 0.017) compared with control subjects (Fig. 2D). Various regulatory immune cell types associated positively with age and/or BMI in the different cohorts (Supplementary Table 1).

CD8⁺ Tc cells were further classified into CD8⁺ Tc1 cells with Th1-like and CD8⁺ Tc2 cells known for a Th2-like cytokine secretion pattern. Although CD8⁺ Tc cell count was similar in all groups after adjustment of the data, proinflammatory CD8⁺CD183⁺ Tc1 cells were more frequent in type 2 diabetic (P ≤ 0.001) and type 1 diabetic (P = 0.006) than in control subjects (Fig. 2E). Percentage of CD8⁺CD25⁺ regulatory Tc cells was also higher in type 2 diabetic (P = 0.023) and type 1 diabetic (P = 0.007) subjects after adjustment for age, sex, and BMI (Fig. 2F). On Treg cells, we further examined the expression of integrin αEβ7 (CD103), a type I transmembrane glycoprotein binding to E-cadherin and mediating homing of lymphocytes to sites of inflammation. After adjustment, type 2 diabetic patients exhibited higher frequency of CD103⁺ cells of CD4⁺CD25⁺⁺ Treg cells (3.4% vs. 1.7%, P = 0.045) and of CD8⁺CD25⁺ Treg cells (10.8% vs. 9.9%, P = 0.02) than control subjects. There were no differences between type 1 diabetic patients and control subjects (data not shown).

Out of the described 21 immune cell subsets, 16 correlated with age, 7 with sex, and 14 with BMI in all groups combined (Supplementary Table 1). Because of the higher age and BMI at the onset of disease in type 2 diabetes and overall more male than female participants in the groups (Table 1), we performed partial correlation analyses after adjustment for age, sex, and BMI.

In type 2 diabetic patients, the proportion of Treg cells and their subgroup of FoxP3 Treg cells associated positively with parameters of glycemia, FBG, and HbA_1c. Activated T cells correlated negatively with HDL and positively with TG. Leukocyte, lymphocyte, and monocyte counts exhibited weak positive correlations with C-Pep. Furthermore, M/I was positively associated with T cells and CD103⁺ Treg cells (r = 0.190, P = 0.046, and r = 0.366, P ≤ 0.001). All correlations between immune cells and metabolic parameters are shown in Table 2.

In type 1 diabetic patients, monocyte counts related positively to M/I (r = 0.469, P = 0.003) (Supplementary Fig. 2) and negatively with HDL (r = −0.364, P ≤ 0.05), while total leukocyte and lymphocyte counts were not correlated with metabolic parameters. Granulocyte count associated positively with C-pep. The frequency of CD3⁺ T cells associated negatively with HDL and positively with TG. CD4⁺ Th cells correlated positively and CD8⁺ Tc cells negatively with TC. Anti-inflammatory CD194⁺ Th2 cells and double-positive CD183⁺CD194⁺ Th cells, but not Treg cells, were positively associated with HbA_1c. All correlations between immune cells and metabolic parameters in type 1 diabetic patients are shown in Table 2.

Finally, M/I negatively correlated with FBG (r = −0.352, P ≤ 0.001), HbA_1c (r = −0.235, P = 0.013), C-Pep (r = −0.36, P ≤ 0.001), and TG (r = −0.298, P ≤ 0.001) but positively with HDL (r = 0.367, P ≤ 0.001) in type 2 but not in type 1 diabetic patients.

This study demonstrates upregulation of cellular immune activity in both metabolically well-controlled type 2 and type 1 diabetes patients despite short disease duration. Long-standing type 2 diabetes is frequently accompanied by subclinical inflammation (20,21). While our finding of increased total WBC as well as lymphocytes, monocytes, and granulocytes counts in new-onset type 2 diabetic patients is in agreement with a recent meta-analysis (6), our study extends these observation by demonstrating distinct associations of WBC with metabolic parameters.

We report that the number of total leukocytes in type 2 diabetes associates positively with BMI, C-Pep, and LDL and the number of monocytes correlates negatively with FBG. Even more interesting is the positive correlation of ambient glycemia (as assessed by FBG and HbA_1c), with anti-inflammatory CD4⁺CD25⁺⁺ Treg cells and CD4⁺CD25⁺⁺CD127⁻ FoxP3 cells. The regulation of nutrient uptake and use is critically important for the control of immune cell number and function (19). While quiescent T cells mainly use oxidative phosphorylation to generate ATP, they switch from phosphorylation of glucose, amino acids, and fatty acids to the more glycolytic metabolism during T-cell activation, cytokine production, and memory development (22). Of note, CD4⁺ T-cell subsets show distinct metabolic differences in that Treg cells are the least glycolytic cell type of CD4⁺ T cells but exhibit greater lipid oxidation and mitochondrial membrane potential (22). In our study, percentages of regulatory T cells correlate positively with parameters of glycemia but not of lipidemia. As reported, glucose lowering by medication or gastric banding surgery can decrease WBC and improve immune cell-mediated inflammation (23,24). Taken together, the observed positive relationship of Treg cells with FBG and HbA_1c is in line with the contention that hyperglycemia specifically affects the immune system in type 2 diabetes.

In our type 2 diabetic patients, insulin sensitivity correlated positively with the adhesion molecule CD103 on regulatory Th cells. Insulin resistance, defined by reduced glucose disposal (M/I) to skeletal muscle and liver, not only is typical for type 2 diabetes but may also occur in long-standing type 1 diabetes (25,26). Not only chronic blood glucose (glucose toxicity) (27) but also lipid elevation (lipotoxicity) drives the insulin resistance in both diabetes types (28). Here, we confirm the inverse relationship of glycemia with M/I for recent-onset type 2 diabetes, which has previously been reported mainly for longer-standing diabetes (29,30). The absence of such relationships in our type 1 diabetes cohort is likely due to the short duration of hyperglycemia until disease onset, while subclinical hyperglycemia generally exists long before the clinical manifestation of type 2 diabetes. In addition, subclinical inflammation can contribute to insulin resistance in type 2 diabetes (30,31), but less is known about its role and the impact of autoimmunity in type 1 diabetes (32).

The current study did not address serum concentrations of cytokines. Interestingly, a recent analysis showed that an inflammatory score derived from the proinflammatory plasma cytokines, IL-6, TNFα, osteopontin, fractalkine, MCP-1, and anti-inflammatory adiponectin, inversely relates to insulin sensitivity. The inflammatory score independently predicted fasting plasma glucose and insulin resistance in type 2 diabetic patients with high sensibility and specificity (33). These results are supported by our findings of greater fractions of T cells and activated Th and Tc cells in type 2 diabetic patients. Also, type 1 diabetic patients showed higher insulin resistance and had elevated parts of activated Th and Tc cells than control subjects. Of note, studies in mice suggest an interaction between insulin action, TNFα-converting enzyme, and TNFα (34).

In our type 2 diabetic cohort, low HDL and high TG associated with activated CD4⁺CD25⁺ T cells in line with a proinflammatory condition. Lipids, particularly fatty acids, may directly suppress lineage-specific cytokine production in different T-cell subsets (22). In addition, the low HDL and high TG reflect a dyslipidemic profile typical for insulin-resistant states. Restribution and accumulation of lipids in nonadipose tissues such as skeletal muscle can also occur in poorly controlled type 1 diabetes (35) and may induce dysregulation of cellular metabolism and function (28). This may coexist with impaired adaptation of mitochondrial function to prevalent metabolic states in type 2 diabetes (36) but also in insulin-resistant patients with poorly controlled type 1 diabetes (26). Thus, lower capacity for lipid oxidation and accumulation of toxic lipid intermediates such as diacylglycerols, ceramides, or acylcarnitines may impair not only insulin signaling but also functionality of immune cells commonly in type 1 and type 2 diabetes.

The type 1 diabetic patients exhibited a higher proportion of proinflammatory Th1 cells than control subjects along with higher activated CD4⁺CD25⁺ Th cell subsets. This is in line with the function of T cells to destroy the insulin-producing pancreatic β-cells in type 1 diabetes (37). In contrast with the type 2 diabetic patients, the number of WBC counts did not differ between type 1 diabetic and healthy persons. We found that CD3⁺ T cells associate positively with TG and negatively with HDL in type 1 diabetic patients. Additionally, TC associated positively with CD4⁺ Th cells and negatively with CD8⁺ Tc cells, indicating an influence of cholesterol metabolism on the adaptive immune system. These findings also point to an upregulated proinflammatory state with higher TC and lower HDL in type 1 diabetic patients, similar to our findings in type 2 diabetes, but involving different immune cells. Changes in lipid homeostasis are critical for T-cell growth, activation, and function, as metabolic requirements of T cells increase dramatically upon activation by shifting from lipid oxidation to lipid synthesis for creating membranes to enable cell growth (22).

In addition, we found a positive correlation between insulin sensitivity and the number of circulating monocytes in type 1 diabetic patients. These data confirm the recently published suggestion that decreased blood monocytes and elevated neutrophils may be additional biomarkers of insulin resistance in type 1 diabetes (38). Of note, while that study assessed insulin resistance from the surrogate markers waist-to-hip ratio, hypertension, and HbA_1c (38), the present association is based on direct measurement of insulin sensitivity using the standard hyperinsulinemic-euglycemic clamp test. Monocytes are progenitor cells of macrophages, which, together with CD8⁺ T cells, are the first cells found in the inflamed pancreas tissue during the development of the disease (39). Upon activation, they secrete increased levels of proinflammatory cytokines (40). The correlation of innate immune cell alterations with insulin resistance may illustrate that not only metabolic but also cellular factors play a role in the pathogenesis of insulin resistance in type 1 diabetes or vice versa.

Our study has strengths and limitations. We applied a flow cytometry protocol using fresh whole blood and avoiding Ficoll separation or other in vitro procedures, thereby minimizing altered surface expression of phenotypic markers. The prospective set up using fresh blood allowed measurements of ambient metabolic parameters and immune cells in extensively metabolically characterized patients. On the other hand, the cross-sectional design of our study does not allow us to draw conclusions regarding the long-term course of disease. Furthermore, immune cells were not phenotyped upon stimulation with islet antigens or other functional stimuli of innate or adaptive immune cells, and we did not investigate infiltrating cells from endocrine pancreas or adipose tissue.

In conclusion, this study demonstrates that both type 1 and type 2 diabetes show increased immune activation compared with normoglycemic persons. However, immune cell subtypes substantially differ between type 1 and type 2 diabetic patients. In type 2 diabetes, higher WBC counts are present in the face of lower proportions of defensive NK cells of the innate immune system and a generally activated pattern of the adaptive immune system. Hyperglycemia was the most striking parameter modulating the pattern of immune cells. In type 1 diabetes, WBC counts were not increased but featured a general activation of adaptive immunity and a correlation of monocyte counts with insulin resistance.

The GDS Group consists of H. Al-Hasani, J. Eckel, G. Giani, C. Herder, A. Icks, J. Kotzka, K. Müssig, B. Nowotny, P.J. Nowotny, W. Rathmann, J. Rosenbauer, P. Schadewaldt, N.C. Schloot, J. Szendroedi, D. Ziegler, and M. Roden (speaker).

Acknowledgments. The authors thank all individuals for participating in this study and Professor Boege from the Institute for Clinical Chemistry and Laboratory Diagnostic for providing some of the routine laboratory analyses at the University Hospital Düsseldorf.

Funding. This study was supported in part by the German Center for Diabetes Research.

Duality of Interest. N.C.S. is currently on leave of absence and employed by Lilly Deutschland GmbH. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. B.M.-H. designed the study, researched data, calculated statistics, created tables and graphs, wrote the manuscript, and critically reviewed the manuscript. R.R. designed the study, researched data, calculated statistics, created tables and graphs, and critically reviewed the manuscript. B.N., C.K., S.K., M.-C.S., and J.S. performed the clinical characterization of volunteers and critically reviewed the manuscript. J.R. provided advice for statistical analysis and critically reviewed the manuscript. N.C.S. and M.R. designed the study, contributed to data analysis and discussion, and wrote, edited, and critically reviewed the manuscript. M.R. 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.

Prior Presentation. Parts of this study were presented in abstract form at the 50th German Diabetes Society Congress, Berlin, Germany, 28–31 May 2014.