The present study was conducted among employees (n = 1,103; 88% male; mean age 40 years [range 18–64 years]) of a large European airplane manufacturer in the south of Germany who took part in a voluntary company health check in 2011. Participant characteristics are presented in Table 1. Participants received a personalised comprehensive health report. All data were anonymised before analysis. This study was approved by the ethics committee of the Medical Faculty Mannheim, Heidelberg University. All participants gave written informed consent.
Participants arrived at a location away from their usual workplace between 06:45 and 08:45 hours in the morning for their health check. After drawing fasting venous blood and a medical examination, participants were seated in a quiet room to fill out questionnaires on demographic, medical and health behaviour data. Anthropometric (e.g. height, weight, waist and hip circumference) and BP measurements were carried out by trained study personnel. Demographic data, including age, sex and marital status, along with socioeconomic status (SES) indicators (measured as hierarchical job position, manual occupation and shift work), self- and doctor-diagnosed medical conditions, and lifestyle factors (e.g. smoking, alcohol intake, exercise) were obtained by questionnaires used and validated in the MONICA (‘Monitoring trends and determinants in cardiovascular disease’) study .
T cell phenotypes were assessed by flow cytometry. Whole blood samples were collected in EDTA-coated tubes (Sarstedt, Nümbrecht, Germany), stored at room temperature and prepared within 1 h of collection. Briefly, 30 μl whole blood was stained with a combination of the following conjugated monoclonal antibodies: anti-CD3 allophycocyanin (APC)–cyanine dye 7 (Cy7; clone SK7), anti-CD4–peridinin chlorophyll protein (clone SK3), anti-γδ T cell receptor (γδTCR)–phycoerythrin (PE; clone B1), anti-CD8–APC (clone SK1; BD Biosciences, San José, CA, USA); and anti-CD45RA–FITC (clone HI100) and anti-CD27–PE-Cy7 (clone M-T271; BD Pharmingen, San Diego, CA). All antibodies were purchased from and validated by BD Biosciences and BD Pharmingen at pre-diluted concentrations for use at the recommended volume per test. Following a 20 min incubation at room temperature in the dark, 1.5 ml BD FACS lysing solution (BD Biosciences) was added to the mixture and incubated for another 15 min. After centrifugation for 7 min at 700 g, the supernatant was removed and both lysed erythrocytes and unbound antibody were washed away. The pellet was subsequently re-suspended in 250 μl 2% paraformaldehyde solution until analysis. Data were collected using a FACSCanto II flow cytometer and dedicated FACSDiva software (BD Biosciences). Spectral overlap was electronically compensated for using single labelled antibody tubes. Following data acquisition, files were transferred to a third party software program (FlowJo v7.6.5, Tree Star, Ashland, OR, USA) for analysis. Representative plots of the gating strategy are shown in electronic supplementary material Fig. 1. Lymphocyte numbers were obtained by multiplying the total leucocyte count by the percentage of gated lymphocytes. The lymphocyte number was further multiplied by the percentages of gated CD3+ cells and their subsequent subsets to calculate the numbers of cells per microlitre used in the analyses.
CMV status determination
Fasting plasma samples were stored in small aliquots at −80°C until analysis. Evidence of a previous CMV infection (serostatus) was determined using a commercially available ELISA (BioCheck, Foster City, CA, USA) according to the manufacturer’s instructions. Optical density values obtained from participants’ samples were fitted to a standard curve. These concentrations were then compared with a cut-off value to compute CMV index scores. Participants with a borderline seropositive result, i.e. a calculated index score of >0.85 and <1.15, were re-tested (n = 9). If they remained borderline, participants with index scores above and below 1.00 were considered CMV+ and CMV−, respectively, as per the manufacturer’s instructions. The sensitivity, specificity and accuracy of the test are reported as 95.0%, 96.7% and 96.0%, respectively.
HbA1c, fasting glucose, triacylglycerol, LDL-C, HDL-C and high-sensitivity C-reactive protein levels were measured by an accredited clinical laboratory (Synlab Laboratories, Augsburg, Germany) according to standard laboratory procedures that comply with International Organization for Standardization norms (DIN EN ISO 15189). HbA1c levels were measured using a second-generation HbA1c immunoassay (Roche Diagnostics, Mannheim, Germany), and fasting glucose levels were measured using the glucose hexokinase enzymatic assay (Glucose OSR6121, Beckman Coulter, Brea, CA, USA), in accordance with the latest standardised guidelines and recommendations for laboratory analysis in the diagnosis of diabetes . Cholesterol and triacylglycerol were automatically measured enzymatically (Cobas 8000 analyser; Roche Diagnostics). HDL-C was measured using a competitive homogeneous assay (Roche Diagnostics), and LDL-C was calculated using the Friedewald equation . These values were also used to calculate the ratio of LDL-C to HDL-C.
Diabetes and metabolic syndrome classification
Diabetes was classified according to the ADA guidelines in individuals with fasting glucose levels of >6.94 mmol/l and/or HbA1c levels of ≥6.5% (48 mmol/mol) in the absence of known diabetes. Those with self-reported, doctor-diagnosed diabetes were also classified as diabetic. Prediabetes was classified as a fasting glucose level between 5.55 and 6.94 mmol/l and/or an HbA1c level between 5.7% (39 mmol/mol) and 6.4% (46 mmol/mol) . The remaining normal-glycaemic individuals, therefore, had fasting glucose and HbA1c levels of <5.55 mmol/l and <5.7%, respectively. The metabolic syndrome components were assessed as the following: (1) waist circumference >102 cm (men) or >88 cm (women); (2) plasma triacylglycerol >1.70 mmol/l; (3) plasma HDL-C <1.03 mmol/l (men) or <1.29 mmol/l (women); (4) BP ≥130 mmHg (systolic) and/or ≥85 (diastolic) mmHg; and (5) plasma fasted glucose ≥5.55 mmol/l. Each of these components were dichotomised (yes or no) and added together to create a metabolic syndrome component score (range 0–5). Those with a score of ≥3 were classified as having the metabolic syndrome .
To approximate a normal distribution of the variables used in the current analyses, we applied transformations based on information criteria obtained from the Ladder-of-Powers in Stata 12 (StataCorp, College Station, TX, USA). The transformation with the least statistical deviation from a normal distribution, indicated by the smallest χ
2 (or the most non-significant p value) was used, as previously recommended . Missing data (<7% for all variables) was handled by multiple imputation in IBM SPSS (version 20, Chicago, IL, USA). Briefly, a fully conditional specification method was automatically chosen to replace missing data. In this method, each variable was fitted in a univariate (single dependent variable) model using all other available variables in the model as predictors, and missing values were imputed for each variable being fitted. Linear and logistic regressions were used for continuous and categorical variables, respectively. Relevant variables with already complete data were entered only as predictors to improve estimates. After ten iterations for each of the five imputation datasets, pooled estimates were used for all subsequent analyses below.
First, participant characteristics (i.e. demographics and lifestyle behaviours) were compared between CMV+ and CMV− individuals. Student’s t tests and χ
2 analyses were used for continuous and categorical variables, respectively.
Second, differences in CMV status by HbA1c and fasting glucose levels and diabetic status were explored using binary logistic regressions. CMV status was entered as the dependent variable, and each factor was entered, in turn, as an independent variable. Potential confounders known to impact CMV infection and reactivation, including age, sex, marital status, SES (job status, manual occupation), and lifestyle factors (smoking, alcohol intake, BMI, and physical activity) [30–32], were statistically controlled in hierarchical models (Models 1–3): Model 1 was adjusted for age and sex; Model 2 was Model 1 additionally adjusted for marital status and SES (job status and manual occupation); and Model 3 was Model 2 further adjusted for smoking, alcohol, BMI and physical activity. These models were entered stepwise as covariates throughout the remaining analyses.
Third, numbers of CD8+ EM and EMRA T cells were compared between levels of glycaemic control (indicated by diabetic classification) using ANOVA and ANCOVA. These analyses were stratified by CMV status and the abovementioned potential confounders were entered as covariates (Models 1–3).
Finally, separate linear regressions were used to explore the individual associations of HbA1c and fasting glucose levels with EM and EMRA T cell subset numbers. Potential confounders were entered as covariates using the same models as above.
The above analyses were repeated with each of the dyslipidaemia and CVD risk factors (i.e. total cholesterol, LDL-C, HDL-C, the LDL-C to HDL-C ratio and triacylglycerol) entered separately as independent predictors of EM and EMRA T cell subset numbers. For significant associations, HbA1c was added as a potential mediator to examine the role of glucose levels on lipid metabolism. All analyses were performed with IBM SPSS version 20.