Adiponectin, biomarkers of inflammation and changes in cardiac autonomic function: Whitehall II study

Participants are from the Whitehall II Study, an occupational cohort of 10,308 British civil servants (6896 men and 3412 women aged 35–55 years) of mainly white ethnicity recruited between 1985 and 1988 (phase 1) [13].

The cohort has been followed at eight subsequent phases. Every second phase (~ 5 years apart) included a clinical health examination (phases 1, 3, 5, 7, and 9). Data from phases 3, 5, 7, and 9 form the basis for this study as measurements of inflammation, HR, HRV and a standard 2-h 75 g OGTT were performed at those phases. Diabetes was diagnosed at screening by OGTT (WHO criteria) or by self-reported diagnosis. Participants fasting ≥ 8 h were classified as having diabetes if fasting- or 2-h plasma glucose value were ≥ 7.0 or 11.1 mmol/l respectively. Participants fasting < 8 h were classified with diabetes if any of their two glucose measurements were ≥ 11.1 mmol/l. Blood pressure and lipids were measured as described [13]. Patients with heart disease including ischemic heart disease and arrhythmias at baseline were excluded from analyses.

During follow-up, participants were censored if they died, were lost to follow-up or developed heart disease e.g. ischemic heart disease and arrhythmias either diagnosed at study examination or by register based follow-up, which comprised 1809 (6.0%) person-examinations.

Serum levels of hsCRP, IL-6, IL-1Ra and total adiponectin were measured as described [14, 15]. We excluded 543 (1.9%) person-examinations with hsCRP > 10 mg/l as indicator of acute infections.

Five-minute resting 12-lead ECG recordings were obtained subsequent to 5 min of supine rest. Recordings were filtered through an automated algorithm, allowing the analyses of suitable R–R intervals without the presence of arrhythmias, ectopic beats and branch blocks. Indices of HRV were analysed in the time domain [the standard deviation of all intervals between R waves with normal-to-normal conduction (SDNN) and the root mean square of the sum of the squares of differences between consecutive R–R intervals (RMSSD)] and by utilizing a Blackman-Tukey algorithm also in the frequency domain low-frequency (LF) power (in the 0.04–0.15 Hz frequency band), high-frequency (HF) power (in the 0.15–0.4 Hz frequency band) and total power (in the ≤ 0.4 Hz frequency band). The ratio between LF and HF power (LF/HF ratio) was calculated.

Information on smoking habits, physical activity and medication use were gathered by self-administered questionnaires [13].

This analysis used HR and six HRV indices (SDNN, RMSSD, HF power, LF power, LF/HF ratio and total power) as outcomes. All except HR were log-transformed prior to analysis to fulfil the assumption of normally distributed model residuals.

The following biomarkers of subclinical inflammation were included as exposures: hsCRP, IL-6, IL-1Ra and adiponectin (all log₂ transformed). Adiponectin and IL-1Ra were measured only in a case-cohort subsample nested within the Whitehall II study. [14, 15].

Up to 25,050 person-examinations for 8469 participants were analysed (9459 person-examinations for the 3421 participants in the subcohort). In each 5-year observation window of two consecutive phases, we studied the associations of baseline levels of biomarkers of subclinical inflammation and follow-up levels of the different outcomes, adjusting for baseline level of the outcome.

All analyses were adjusted for age, sex, ethnicity, study phase and diabetes (model 1). Additional adjustments included BMI and physical activity (model 2) and smoking, systolic blood pressure, total cholesterol, triglycerides, tricyclic antidepressants, diuretics and beta blockers (model 3). For all outcomes except HR, analyses were also adjusted for HR obtained as part of the HRV analyses. To compare the estimated associations across models 1–3 for a given outcome and exposure, we used a complete-case approach, limiting the analyses to data with complete information on all covariates in model 3. The same individual may contribute with more than one observation to the analyses. To account for the likely correlation of repeated measurements within the same participant, we used mixed-effects models with a random intercept and a random slope for time. Individuals changing diabetes state during the study contributed with person-examinations first to the non-diabetes group and later to the diabetes group.

Since prospective studies on adiponectin and cardiovascular outcomes have indicated that the direction of the association may depend on diabetes status (positive associations between adiponectin and cardiovascular events or mortality in individuals with diabetes rather than inverse associations in non-diabetic individuals) [16], we tested for a modifying effect of diabetes on the association between HR and HRV indices with adiponectin. As sensitivity analysis, we also calculated models including interaction terms with diabetes for the three biomarkers of inflammation (i.e. hsCRP, IL-6 and IL-1Ra). To assess if body composition estimated by waist circumference instead of BMI would affect model estimates, we repeated the analyses replacing BMI with waist circumference.

Heart rate and HRV indices are highly interrelated, and the same is true for the inflammation-related biomarkers and adiponectin. Therefore, correction for multiple testing would likely inflate the risk of type II errors and was therefore not applied.

Statistical analyses were performed in R version 3.3.1 (The R Foundation for Statistical Computing) and SAS version 9.2 (SAS Institute, Cary, NC).

Our data on biomarkers of subclinical inflammation, HR and HRV are novel, as only two small prospective studies that investigated the temporal relationship between hsCRP and vagal tone assessed by HRV measurement of HF power [17, 18].

We show that higher IL-1Ra levels were associated with more pronounced increases in resting HR after adjustment for a range of potentially confounding factors. Given the established link between higher resting HR and increased risk of cardiovascular outcomes (e.g. coronary heart disease, sudden cardiac death, stroke, atrial fibrillation) and mortality [19], this finding implicates cytokines of the IL-1 family in the regulation of HR and in cardiovascular risk. As antagonist of the proinflammatory cytokine IL-1β, IL-1Ra has no agonist activity of its own and is thus unlikely to impair cardiac function [20]. Higher IL-1Ra levels are related to a higher risk of cardiovascular events in population-based studies [21] and have been proposed to reflect the presence of metabolic and/or inflammation-related cardiovascular risk factors that may involve the activation of the NLR family pyrin domain containing 3 (NLRP3) inflammasome and increased IL-1β activity [21, 22]. Higher IL-1Ra could thus be interpreted as result of an endogenous but eventually insufficient counter-regulation.

For the three biomarkers of subclinical inflammation selected in this study, we found five significant associations with changes in various HRV indices (LF power, LF/HF ratio and total power) out of 18 associations tested. Therefore, our study provides some evidence that subclinical inflammation may precede both impaired sympathetic and vagal activity, but at the same time indicates that BMI represents a major confounder in this context. Overall, effect sizes were small and no clear pattern of associations emerged, so that we did not find a pronounced impact of baseline levels of these biomarkers on impairments in HRV in our population-based study population. Therefore, our results differ from associations between inflammation-related biomarkers and cardiac autonomic dysfunction in cross-sectional studies [7]. Moreover, we could not confirm the association between CRP levels and increases in HF power from a small prospective study [18]. Although direct effects of immune activation e.g. induced by lipopolysaccharide on HR and HRV have been demonstrated, it is important to emphasise that these studies involve acute elevations of proinflammatory cytokines that exceed circulating levels in the general population by one to two orders of magnitude [23]. Collectively, these findings raise the question whether associations between biomarkers of inflammation and HRV in cross-sectional studies are mainly caused by effects of the autonomic nervous system on inflammatory pathways [8]. Another small prospective study indeed showed an inverse association between HF–HRV and CRP 4 years later [17]. It is also possible that subclinical inflammation is so closely linked to the autonomic regulation that changes in inflammation are associated with more immediate changes in HRV. This would explain that no temporal association to future HRV was seen when adjusting for baseline HRV in the present study.

The present study demonstrates that higher serum levels of total adiponectin were associated with beneficial changes in HR and several measures of HRV 5 years after blood sampling in patients with diabetes. The associations were present for three of the six HRV indices, SDNN, RMSSD and LF power. RMSSD is a marker of parasympathetic autonomic tone whereas SDNN and LF power are mainly sympathetic measures. This suggests that in diabetes patients adiponectin could have neuroprotective effects on both branches of the autonomic nervous system. There were no significant associations between adiponectin and changes in HR or HRV in non-diabetic individuals indicating that adiponectin may not be associated with autonomic regulation in healthy people. Reduced HRV is not only strongly associated with diabetes in the general population, but already found in youth with type 2 diabetes and in adults with prediabetes or insulin resistance [24‐27]. However, further research is necessary to elucidate to what extent determinants of reduced HRV and CAN such as adiponectin and other biomarkers may depend on age and on the degree of glucose intolerance and hyperglycaemia.

We are not aware of previous prospective studies that assessed the potential link between adiponectin levels and future changes in HR or HRV. Three cross-sectional studies reported that higher adiponectin levels were associated with a reduced LF/HF ratio [28], reduced very-low-frequency (VLF) power [11] and higher odds of CAN [10] in individuals with type 2 diabetes, whereas a study based on an occupational sample (prevalence of diabetes 2.3%) observed no associations between adiponectin and HRV indices [29]. Thus, the direction of associations between adiponectin and HRV or CAN differs between these aforementioned cross-sectional studies and our prospective study. Associations between higher adiponectin levels and higher cardiovascular risk have been reported before in individuals with type 2 diabetes [16] and may be explained as counterregulatory phenomenon as discussed above for IL-1Ra [30] The second explanation is of interest because adiponectin is upregulated by natriuretic peptides which are direct markers of coronary heart disease and could mediate such associations [31].

Our findings of favourable associations between baseline adiponectin levels and changes in both HRV and indices of HRV is in line with the inverse association between higher adiponectin levels and lower incidence of distal sensorimotor polyneuropathy in an older population-based cohort with more than half of the participants having prediabetes or type 2 diabetes [32]. However, the latter association was mediated by obesity and other confounders, whereas the association with measures of early CAN in the present study was robust to extensive adjustment.

Strengths of this study are the large sample size, the comprehensive measurements of outcomes and exposures assessed simultaneously, the extensive adjustment for confounders, the stratification by diabetes status and the exclusion of individuals with elevated hsCRP to reduce the impact of acute inflammatory states. Importantly, the prospective design allowed us to assess the temporal association between inflammation and HRV changes, whereas associations in cross-sectional studies may be attributable to reverse causality, i.e. impact of HR and HRV on biomarkers of inflammation. Limitations include the observational design that precludes causal inferences and the selection of four pro- and anti-inflammatory biomarkers whereas others (e.g. IL-18) also merit further research. Furthermore, the study population was predominantly of European descent, so that data cannot be generalised to other ethnic groups. Also, the cohort is based on people employed as civil servants at the study start excluding e.g. unemployed people and people employed in the private sector. The study may therefore not be fully generalizable to the general population.