Introduction

Major risk factors for cardiovascular disease (CVD) include both lifestyle choices (e.g., physical inactivity, poor diet) and physiological conditions (e.g. diabetes mellitus, hypertension, obesity) [1]. These physiological conditions have been associated with altered autonomic nervous system (ANS) activity [2–6]. Observed ANS activity reflects a balance between the sympathetic and parasympathetic branches, which oppose one another but are flexible and dynamic in nature. The sympathetic branch provides an overarching accelerating stimulus, promoting metabolic substrate mobilization and utilization in addition to increasing heart rate and cardiac contractile force [7]. In contrast, the parasympathetic branch is the overarching decelerator of the body, promoting substrate storage and decreased heart rate [8]. Bodily functions are regulated by the balance between these opposing systems and any prolonged disturbance or asymmetry in activity can result in the development of several clinical and physiological conditions [9,10].

ANS activity is often evaluated indirectly through indices of heart rate variability (HRV). HRV, or the time variance among intervals of consecutive cardiac cycles, is a widely used and accepted means of assessing ANS activity at the cardiac level [11]. Irregular ANS activity assessed via HRV has been heavily evaluated for its association with CVD and mortality. Specifically, chronically reduced parasympathetic and elevated sympathetic activities are associated with premature mortality, myocardial infarction, and development of type II diabetes mellitus and hypertension [12–14]. Various time and frequency domain HRV metrics can be used to estimate the level of parasympathetic output at the cardiac level [15]. The root mean square of successive differences between normal (non-ectopic) beats (RMSSD) is the primary time domain metric used to reflect vagally mediated changes, with an increase in RMSSD reflecting an increase in parasympathetic activity [16].

Some HRV metrics have been suggested to estimate sympathetic activity (namely the low frequency spectral power component and the ratio of low frequency to high frequency spectral power components), but these values are inherently limited in evaluating sympathetic output given that they are also influenced by parasympathetic activity [17]. An alternative and more independent method of evaluating sympathetic activity at the cardiac level is to monitor systolic time intervals, i.e., the amount of time between electrical stimulation of the heart and its mechanical response. Specifically, the segment of time known as the pre-ejection period (PEP), an estimate of the electromechanical delay of the left ventricle [18], can be used to accurately estimate sympathetic activity [19]. PEP has been shown to have a pronounced inverse relation with sympathetic activity in that sympathetic neural blockade stimulates an increase in PEP length, while parasympathetic blockade results in no change in PEP [20]. When used concurrently, PEP and HRV measurements can provide holistic details of ANS activity at the cardiac level.

While increased adiposity is an independent risk factor for CVD development [21], it is also a well-established risk factor for CVD comorbidities such as diabetes mellitus [22] and hypertension [23]. While significant associations between metrics of increased adiposity and altered ANS activity have been reported [24–26], it is unclear if increased adiposity has both a direct and indirect relation to ANS activity. However, evaluating the influence of adiposity on ANS activity through its risk for diabetes mellitus and hypertension development is a complex undertaking, as there is no single metric that can accurately reflect diabetic or hypertensive risk. Latent variable modeling constructs an unobserved variable that reflects the combined influence of multiple measured variables. Applying this methodology to reflect relations within our conceptual model (Fig 1) provides a more comprehensive assessment of the influences of adiposity on ANS activity. In this model, adiposity is reflected by waist circumference (WC) and has direct effects on RMSSD and PEP (markers of parasympathetic and sympathetic nervous system activities, respectively). Our model also suggests that adiposity has indirect effects on ANS activity through latent constructs representing glycemic impairment (GI) and cardiac stress (CS). We therefore aimed to more accurately model the complex associations between adiposity and ANS function, by fitting a model that included possible indirect effects through CS and GI.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Proposed model.

A priori model including observed variables (squares) depicting the proposed pathways for the direct and indirect effects of waist circumference on markers of ANS activity. Construction of the latent variables (circles) Cardiac Stress (CS) and Glycemic Impairment (GI) are depicted. WC, Waist Circumference; HR, Heart Rate; DBP, Diastolic Blood Pressure; Hb_A1c, Glycosylated Hemoglobin; FBG, Fasting Blood Glucose; RMSSD, Root Mean Square of Successive Differences between normal-to-normal beats; PEP, Pre-Ejection Period.

https://doi.org/10.1371/journal.pone.0303117.g001

Methods

Study population

This secondary data analysis study was reviewed (IRB #:19–0119) by the Institutional Review Board at the University of North Carolina, Greensboro on September 27th 2018 and deemed ‘NHSR’ (Not Human Subjects Research), since the data analysis did not include identifiable data. The data used in the present study was obtained from the Atherosclerosis Risk in Communities (ARIC) study. Access to the data was granted by the Biologic Specimen and Data Repository Information Coordinating Center on October 17^th, 2018 and the current study did not have access to information that could identify individual participants during or after data collection. The ARIC dataset consists of 15,792 participants recruited from four different field centers throughout the United States of America (Minneapolis, Minnesota; Washington County, Maryland; Forsyth County, North Carolina; and Jackson, Mississippi). These participants were given a comprehensive physical examination at baseline between the years of 1987–1989 and were reevaluated every three years. Detailed methods about data collection and study design for ARIC have been published [27]. Data for the present study was derived from a fifth cycle of visits that was conducted with 5,900 participants between 2011 and 2013. For the analyses outlined here, individuals with known cardiovascular disease, those being pharmaceutically treated for hypertension or diabetes mellitus, anyone with a systolic blood pressure ≥140 mmHg and/or diastolic blood pressure ≥90 mmHg, those who identified as a smoker or that reported smoking in the last 6 months, and those taking beta-blockers at the time of evaluation were excluded. The resulting participant pool (n = 1145) was 61% female and 39% male, aged 74.6 ± 4.8 years. The majority (89%) of subjects self-identified as “white” with 11% self-identifying as “black”. Further sample characteristics are provided in Table 1.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of study population.

https://doi.org/10.1371/journal.pone.0303117.t001

Model construction

Central adiposity, more than overall adiposity, is strongly related to the development of cardiometabolic diseases [32]. As such, the present model explored the effects of waist circumference (WC) on RMSSD (parasympathetic) and PEP (sympathetic), both directly and through the latent variables GI and CS. Given the inherent nature of FBG and Hb_A1c and their high correlation (r = 0.61, p < .001), the model allowed for a covariance between the two variables. Similarly, the latent variables of GI and CS were allowed to covary, given their strong correlation (r = 0.78, p < .001) and their shared relation to CVD through similar metabolic pathways [33]. Since structural equation model calculations, by default,assume that outcome variables in a latent variable analysis have a shared covariance, RMSSD and PEP were allowed to covary in the present analysis, even though their correlation was not statistically significant (Table 2).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Pearson’s correlation coefficients of observed variables.

https://doi.org/10.1371/journal.pone.0303117.t002

Results

Model analyses

The structural equation model displayed an adequate fit for the sample population with significant goodness of fit indices (RMSEA = 0.050 90% CI [0.036–0.066], SRMR = 0.028, CFI = 0.970, TLI = 0.929). Both latent variables (GI and CS) displayed significant factor loadings with their indicator variables (Table 3), suggesting a robust latent construct. Direct variance and covariance parameter estimates for the entire model are listed in Table 4.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Latent variable parameter estimates.

https://doi.org/10.1371/journal.pone.0303117.t003

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Variance and covariance estimates.

https://doi.org/10.1371/journal.pone.0303117.t004

Direct effects.

The direct, indirect, and total effects of WC on RMSSD and PEP are provided in Table 5. WC failed to have a significant direct effect on RMSSD but had a significant direct effect on PEP (β = 0.208, p = .018), with a one standard deviation increase in WC being related to a 0.208 standard deviation increase in PEP.

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Standardized parameter estimates of direct, indirect, and total effects of waist circumference on RMSSD and PEP.

https://doi.org/10.1371/journal.pone.0303117.t005

Indirect and total effects.

In this model, WC failed to have a significant indirect effect on RMSSD through the latent variables of GI or CS, resulting in the lack of a significant total indirect effect (the sum of all indirect effects on the dependent variable). The total effect (direct effect + total indirect effect) of WC on RMSSD (p = 0.070) and the indirect effect of WC on PEP through CS (p = 0.205) were not statistically significant. The indirect effect of WC through GI had a significant effect on PEP in that a one standard deviation increase in WC was associated with a 0.217 standard deviation decrease in PEP. Furthermore, the total indirect effect of WC on PEP was significant with a one standard deviation increase in WC being related to a 0.173 standard deviation decrease in PEP. However, the total effect of WC on PEP was not significant.

Other significant pathways.

WC had significant effects on GI and CS with a one standard deviation increase in WC being associated with 0.682 and 0.178 standard deviation increases in GI and CS, respectively. CS had a significant effect on RMSSD in that a one standard deviation increase in CS would elicit a 0.524 standard deviation decrease in RMSSD. However, CS failed to have a significant effect on PEP. GI had a significant effect on PEP with a one standard deviation increase in GI being associated with a 0.318 standard deviation decrease in PEP. GI did not have a significant effect on RMSSD. A comprehensive depiction of the structural equation model with standardized factor loadings is provided in Fig 2.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Structual equation model.

Structual equation model including the standardized solutions for the direct and indirect effects of waist circumference on markers of RMSSD and PEP. Cardiac Stress, CS; Glycemic Impairment, GI; WC, Waist Circumference; HR, Heart Rate; DBP, Diastolic Blood Pressure; Hb_A1c, Glycosylated Hemoglobin; FBG, Fasting Blood Glucose; RMSSD, Root Mean Square of Successive Differences between normal-to-normal beats; PEP, Pre-Ejection Period. Note: SE, standard error; *p < .05; **p < .001.

https://doi.org/10.1371/journal.pone.0303117.g002

Discussion

The results of this study support the concept that central adiposity, as measured by WC, has significant direct and indirect effects on markers of ANS activity. Specifically, WC displayed robust associations with markers of sympathetic nervous system activity, while having little influence on parasympathetic nervous system activity when accounting for the impact of glycemic impairment markers and indices of cardiac stress. This highlights the need to directly evaluate both branches of the autonomic nervous system when discussing the relation of ANS activity and adiposity. Additionally, there is evidence that the impact of central adiposity goes beyond direct effects and that nuanced mechanisms may influence ANS function.

In this model, WC failed to have a significant direct, indirect, or total effect on RMSSD. These findings differ from earlier studies that have demonstrated direct relations between adiposity and metrics of parasympathetic activity. Windham et al. (2012) found that an increase in WC was significantly associated with a decrease in RMSSD and SDNN in middle-aged (∼45 years old), but not older (∼85 years old), participants when controlling for age, sex, race, hypertension, glucose tolerance, and physical activity. The age difference between the Windham et al. younger cohort and the present study could be a contributing factor for this difference in statistical magnitude as it has been shown that the age-related decrease in time domain metrics of HRV subsides after 60 years of age in apparently healthy individuals [35].

Our results displayed a significant direct effect of WC on PEP, suggesting a decrease in sympathetic activity with a higher WC. However, the total effect of WC on PEP was neutralized through the negative influence of GI. This model suggests that increases in central adiposity (WC) might have a blunting effect on sympathetic activity (direct effect) while simultaneously increasing sympathetic drive by increasing (worsening) indices of glycemic impairment (indirect effect). These opposing actions mitigate observable physiological changes in PEP, as reflected by the lack of a significant total effect of WC on PEP.

The direct effect of WC on PEP, suggesting a decrease in sympathetic activity with an increase in WC, seems counterintuitive, but is biologically plausible. Aging is often accompanied by an increase in sympathetic nervous system activity [36], particularly due to an impaired ability of baroreflexes to buffer changes in blood pressure [37]. Esler et al suggest that sympathetic activity at the level of the heart is normal or lower in obese individuals compared to normal weight individuals in elderly populations with hypertension [38]. The suggested concept of the obesity paradox lends credence to the notion that central adiposity could decrease sympathetic activity in elderly poulations such as the one used in the currect study. While those diagnosed with overt hypertension under older criteria (blood pressure ≥140/90 mmHg and/or under pharmaceutical treatment) were excluded from the current study, this observation of lower and/or normal sympathetic activity in obese individuals with normal to slightly eleveated blood pressure coincides with the observed significant direct effect and the general lack of a total effect of WC on PEP.

The observed indirect effect of increased WC being related to increased sympathetic activity (decrease in PEP) via an increase in GI is commonly observed in recent studies evaluating sympathetic activity and various markers of GI [39–41]. The latent variable GI reflects the combined influence of FBG, Hb_A1c, and insulin, with any increase in these metrics resulting in increases in the GI variable. Insulin had the highest factor loading for the GI construct and is likely the primary driver for the observed effects. Elevated serum insulin concentrations are postulated as one of the major pathways linking increased adiposity (specifically central adiposity) to sympathetic nervous system overactivity [32]. In addition, direct exposure of cardiac tissue to insulin can induce positive inotropic effects through calcium dependent and independent mechanisms [42]. The present study also showed FBG and Hb_A1c to have significant factor loadings for the GI construct. Previous research has demonstrated increases in FBG [43–45] and Hb_A1c [46–48] with increases in markers of sympathetic activity. Furthermore, ANS activity reflective of cardiac autonomic neuropathy complications have been associated with diabetes mellitus [39]. Linkage between adiposity, as determined by WC, and autonomic function has been observed in groups stratify by glucose tolerance, with a higher prevalence of cardiovascular autonomic neuropathy in those with worsening glucose tolerance [49]. Similar findings were observed using dicotomus groupings based on metabolic syndrome status, with an increased prevalence of cardiovascular autonomic neuropathy in those with metabolic syndrome compared to those without [49]. These findings were independent of worsening cardiac parameters, as makers of endothelial function were similar across groups regardless of stratification method. Given this, the authors postulated that glucose tolerance was the most impactful driving force for autonomic function deterioration [49]. This concept is in agreement with the current study results, since WC only displayed a significant indirect relation with autonomic function via the latent varible GI.

Body fat distribution has been suggested to be a more potent risk factor for the development of cardiovascular disease and hypertension compared to markers of general adiposity [50]. One of the more recently explored pathways is the potential role of visceral adiposity on hypertension development via increased sympathetic nervous system activity [50]. As such, markers for visceral fat (WC) should display a more significant relation with markers of sympathic activity (PEP). However, such relations were not exhibited in the current study as shown by the insignificant indirect effects of WC on PEP through the CS construct.

Given that HRV metrics have been shown to be altered with diabetic status and impaired glycemic control, those previously diagnosed with diabetes mellitus were excluded from the current study. However, the sample population used in this secondary data analysis had a mean fasting blood glucose of 104.79±15.85 mg^.dL^-1, suggesting that a large portion of the group (n = 561, approximately 49% of the sample population) would be classified as pre-diabetic. In addition, there is increasing evidence that people with pre-diabetes may demonstrate signs of cardiac autonomic neuropathy [51]. Similarly, blood pressure just below the hypertensive cutoff is associated with autonomic dysfunction [4]. While the mean diastolic blood pressure of the sample population did not classify as hypertensive (66.4±9.3 mmHg), there were 102 participants (approximately 9% of the sample population) that had a DBP of ≥80 mmHg. As such, those at risk for, but not formally diagnosed with, hypertension were included in the sample. Therefore, despite the exclusion of those diagnosed with diabetes and/or hypertension, the ranges of glycemic and cardiovascular parameters in the present data set are broad enough for the investigation of their influence on autonomic function. Given that the only significant in-direct effect of WC on autonomic function was through the latent variable of GI, it is possible that the increased prevalence of elevated glycemic control markers in this cohort (compared to the prevalence of elevated blood pressure) influenced our results.

The choice of WC as our marker of adiposity over other population-based metrics (e.g., body mass index) could have influenced the observed results. Compared to body mass index, WC better reflects the amount of visceral fat [52]. Since visceral fat is more metabolically deleterious, it is a greater predictor of mortality than subcutaneous fat [53]. However, the interrelations between WC, visceral fat, and markers of parasympathetic neural activity are less straightforward. Increases in visceral fat deposits as measured by computed axial tomography (an even more accurate method of visceral fat determination) have been shown to have significant inverse relations with RMSSD in the absence of a significant correlation between WC and RMSSD, despite WC and visceral fat area being positively related [54]. Still, studies with larger cohorts (n = 8,538) suggest that WC has a stronger inverse relation to HRV metrics, specifically RMSSD, than body mass index or waist-to-hip ratio in apparently healthy populations [55]. These previously observed decreases in markers of parasympathetic activity, including RMSSD, were attributed to an increase in sympathetic activity due to an increase in adiposity, which the current study observed in the significant indirect effects of WC on PEP via GI. As a whole, the results of our study suggest that a greater association of sympathetic tone with centralized adiposity occurs without direct alterations to parasympathetic activity.

The ARIC study utilized data collected from pulse wave velocity and ankle-brachial index measurements to calculate PEP. Studies have used both methodologies independently to provide various markers of CVD risk [56,57] and monitor disease progression [58]. Additionally, these methods have been shown to provide valid approximations of systolic time intervals in the absence of more precise methods such as echocardiogram [59], giving credence to the accuracy and validity of the PEP measurements used for the current study. HRV results were derived from a 2-minute ECG rhythm strip which, depending on the metric used, could have a sizable impact on results. Considered the gold-standard of clinical HRV assessment, a recording length of 24 hours provides the most detailed evaluation of ANS activity [17]. However, shorter recordings of approximately five minutes have been successfully used in numerous studies [30]. Frequency domain measures are preferred since they are influenced to a lesser extent by shorter recording lengths than most time domain measures [11], but only time domain metrics were available in the ARIC dataset. Future studies utilizing more robust HRV metrics and longer ECG recording lengths have the potential to shed more light on the proposed relations of the study variables. Our results are strengthened by the inclusion of validated indices of both branches of the ANS. This provides an accurate, holistic representation of ANS activity without relying on the inferences from a single branch, as is a common practice when solely using HRV results. Additionally, our cohort included both male and female participants. Although sex is a potential confounder, HRV differences between the sexes has been shown to be reduced after approximately 60 years of age [60]. The current study did not evaluated sex differences in the modeled relations due to the smaller number of male subjects in the investigated cohort (n = 446). Such a sample size was insufficient to run such a complex model with this number of parameters. Thus, an accurate sex comparison of the full model could not be made in the current study. Evaluation of sex differences should be conducted in future studies utilizing an adequat sample size of both male and female participants.

The current study is not without its limitations. The cross-sectional design of our study diminishes inferences of causality in the observed relations and the directions of the effects beyond those suggested in the model development. Secondary analysis of the ARIC dataset provided a sample size that was large enough to adequately test our proposed structural equation model. However, we view this as an initial step and future longitudinal investigation is necessary to confirm any causal effects underlying the direct and indirect relations identified in the present study. In addition, while numerous adiposity markers were considered for these analyses (WC, BMI, % fat from BIA), the best model fits were found with WC. However, we acknowledge that none of the adiposity markers available in the dataset are considered gold standard assessments. The use of dual-energy x-ray absorptiometry to assess body composition and/or the use of computed axial tomography to more accurately quantify visceral fat could strengthen the observed effects in this model and explain more of the variance in ANS activity.

Interestingly, similar studies have found that adding markers of cardiovascular fitness could improve model fit parameters and be a stronger determiner of HRV indices of ANS activity [61]. The addition of cardiovascular fitness could have impactful effects on the proposed model. Specifically, an increase in cardiovascular fitness will have considerable effects on adiposity, glycemic management, and cardiac stress parameters, all of which can influence ANS activity. The current model did not account for cardiovascular fitness or physical activity status even though both have direct effects on all the study’s variables, since these variables were available for only small subsets of the ARIC cohort. While the ARIC study recorded self-reports of physical activity as well as conducted clinical measures of physical fitness (grip strength, walking tests, etc.), restricting the current study cohort to include only those that completed these measurments resulted in significant reductions in the sample size. The sample size reductions, coupled with the increased number of model parameters (i.e., adding any of the available physical fitness metrics), resulted in an inability to render a testable structural equation model. Future investigations into how cardiovascular fitness may mitigate or abolish the observed effects of adiposity on ANS activity are warranted.

In summary, our results provide support for a theoretical and statistically sound model for describing the relation of adiposity and ANS activity while accounting for the interactions of contributing physiological parameters. In a population of older, apparently healthy adults, increased central adiposity displayed opposing direct and indirect effects on markers of sympathetic activity, resulting in a lack of significant overall differences in PEP, regardless of adiposity (WC). Moreover, WC presented no significant direct or indirect effects on an index of parasympathetic activity. These results suggest that the true nature of adiposity’s relation to ANS activity is multifaceted. Use of latent structural equation modeling to uncover the complex interrelations of multiple risk factors may better illustrate the biological control systems behind cardiometabolic risk factors and CVD. This approach may more closely mirror multifactored physiological conditions such as metabolic syndrome, which is strongly influenced by the interplay of glycemic regualtion, cardiovascular health, central adiposity, and autonomic function. If it is true that adiposity directly and indirectly effects sympathetic activity, as reflected in the current model by PEP, then reducing adiposity or limiting adiposity gain during aging may promote improved autonomic function by means of both direct influence and indirect influence via alterations to glycemic regulation and cardiovascular health. This utilization of a multipronged approach is in agreement with the current clinical recommendations on how toreduce the risk of developing metabolic syndrome and CVD. Prospective interventional studies employing multiple time points should consider the current proposed and evidenced model to confirm these outcomes.

Acknowledgments

This manuscript was prepared using ARIC research materials obtained from the NHLBI Biologic Specimen and Data Repository Information Coordinating Center and does not necessarily reflect the opinions or views of the ARIC or the NHLBI.