The impact of 2 years of high‐intensity exercise training on a model of integrated cardiovascular regulation

Abstract

Key points

• Heart rate variability, a common and easily measured index of cardiovascular dynamics, is the output variable of complicated cardiovascular and respiratory control systems. Both neural and non‐neural control mechanisms may contribute to changes in heart rate variability.

• We previously developed an innovative method using transfer function analysis to assess the effect of prolonged exercise training on integrated cardiovascular regulation. In the present study, we modified and applied this to investigate the effect of 2 years of high‐intensity training on circulatory components to tease out the primary effects of training. Our method incorporated the dynamic Starling mechanism, dynamic arterial elastance and arterial–cardiac baroreflex function.

• The dynamic Starling mechanism gain and arterial–cardiac baroreflex gain were significantly increased in the exercise group. These parameters remained unchanged in the controls. Conversely, neither group experienced a change in dynamic arterial elastance. The integrated cardiovascular regulation gain in the exercise group was 1.34‐fold larger than that in the control group after the intervention.

• In these previously sedentary, otherwise healthy, middle‐aged adults, 2 years of high‐intensity exercise training improved integrated cardiovascular regulation by enhancing the dynamic Starling mechanism and arterial–cardiac baroreflex sensitivity.

Abstract

Assessing the effects of exercise training on cardiovascular variability is challenging because of the complexity of multiple mechanisms. In a prospective, parallel‐group, randomized controlled study, we examined the effect of 2 years of high‐intensity exercise training on integrated cardiovascular function, which incorporates the dynamic Starling mechanism, dynamic arterial elastance and arterial–cardiac baroreflex function. Sixty‐one healthy participants (48% male, aged 53 years, range 52–54 years) were randomized to either 2 years of exercise training (exercise group: n = 34) or control/yoga group (controls: n = 27). Before and after 2 years, subjects underwent a 6 min recording of beat‐by‐beat pulmonary artery diastolic pressure (PAD), stroke volume index (SV index), systolic blood pressure (sBP) and RR interval measurements with controlled respiration at 0.2 Hz. The dynamic Starling mechanism, dynamic arterial elastance and arterial–cardiac baroreflex function were calculated by transfer function gain between PAD and SV index; SV index and sBP; and sBP and RR interval, respectively. Fifty‐three participants (controls: n = 25; exercise group: n = 28) completed the intervention. After 2 years, the dynamic Starling mechanism gain (Group × Time interaction: P = 0.008) and the arterial–cardiac baroreflex gain (P = 0.005) were significantly increased in the exercise group but remained unchanged in the controls. There was no change in dynamic arterial elastance in either of the two groups. The integrated cardiovascular function gain in the exercise group increased 1.34‐fold, whereas there was no change in the controls (P = 0.02). In these previously sedentary, otherwise healthy middle‐aged adults, a 2 year programme of high‐intensity exercise training improved integrated cardiovascular regulation by enhancing the dynamic Starling mechanism and arterial–cardiac baroreflex sensitivity, without changing dynamic arterial elastance.

Key points

• Heart rate variability, a common and easily measured index of cardiovascular dynamics, is the output variable of complicated cardiovascular and respiratory control systems. Both neural and non‐neural control mechanisms may contribute to changes in heart rate variability.

• We previously developed an innovative method using transfer function analysis to assess the effect of prolonged exercise training on integrated cardiovascular regulation. In the present study, we modified and applied this to investigate the effect of 2 years of high‐intensity training on circulatory components to tease out the primary effects of training. Our method incorporated the dynamic Starling mechanism, dynamic arterial elastance and arterial–cardiac baroreflex function.

• The dynamic Starling mechanism gain and arterial–cardiac baroreflex gain were significantly increased in the exercise group. These parameters remained unchanged in the controls. Conversely, neither group experienced a change in dynamic arterial elastance. The integrated cardiovascular regulation gain in the exercise group was 1.34‐fold larger than that in the control group after the intervention.

• In these previously sedentary, otherwise healthy, middle‐aged adults, 2 years of high‐intensity exercise training improved integrated cardiovascular regulation by enhancing the dynamic Starling mechanism and arterial–cardiac baroreflex sensitivity.

Introduction

Cardiovascular function adapts with exercise training to meet increased metabolic demands. One tool that is often used to assess control of the circulation after training comprises heart rate (HR) and blood pressure (BP) variability (Okazaki et al. 2005). For example, the quantification of HR variability is often used to examine autonomic function in patients with cardiovascular diseases (del Paso et al. 1996) and its impairment is associated with early mortality (Stein & Kleiger, 1999; Villareal et al. 2002). However, assessing the effects of exercise training on cardiovascular variability is challenging because of the complexity of multiple input and output cardiovascular and respiratory control mechanisms.

Both neural and non‐neural control mechanisms may contribute to changes in HR variability (Saul et al. 1991; del Paso et al. 1996; Wang et al. 2004). For example, changes in left ventricular (LV) volume during respiration produce changes in LV end‐diastolic pressure (LVEDP) via the end‐diastolic pressure–volume relationship. Sequentially, changes in LVEDP produce changes in stroke volume (SV) via the dynamic Starling mechanism, which reflects dynamic ventricular–arterial coupling (Shibata et al. 2008). Changes in SV produce changes in systolic BP (sBP) via arterial compliance and distortion of arterial baroreceptors (Tanaka et al. 2000). The arterial–cardiac baroreflex finally transfers sBP variability into changes in HR via the modulation of autonomic neural activity (Tsuji et al. 1996). HR variability at the respiratory frequency thus is produced, at least partly, by the cascade mechanism of LV filling pressure → SV → arterial distortion → arterial–cardiac baroreflex (Fig. 1).

Figure 1. Novel concept of the three‐component cascade model of integrated cardiovascular regulation.

Simple model of the three‐compornent cascade of integrated cardiovascular regulation, including the dynamic Starling mechanism (DSM) (ventricular–arterial coupling), dynamic arterial stiffness and arterial–cardiac baroreflex function. Gain_DSM represents dynamic ventricular–arterial coupling: the relationship between PAD and SV on the Starling curve. The input signal of Gain_DSM is PAD, and the output signal is SV. Gain_{dynamic arterial elastance} represents dynamic arterial stiffness: between SV and sBP. The input signal of Gain_{dynamic arterial elastance} is SV, and the output signal is sBP. Gain_baroreflex represents arterial–cardiac baroreflex function: between sBP and RR interval. The input signal of Gain_baroreflex is sBP, and the output signal is RR interval. Gain_{integrated cardiovascular regulation} represents whole total cardiovascular regulation from LVEDP to SV, from SV to sBP, from sBP to RR interval via dynamic ventricular–arterial coupling, dynamic arterial elastance and arterial–cardiac baroreflex function. Ea, arterial elastance. [Color figure can be viewed at wileyonlinelibrary.com]

Previously, our team developed a two‐component cascade model to determine both the individual and the combined effects of dynamic ventricular–arterial coupling and arterial–cardiac baroreflex and their contribution to sBP and HR variability (Shibata et al. 2006). In the present study, we updated this method to include a three‐component cascade model: (i) dynamic Starling mechanism (changes in SV induced by changes in LVEDP); (ii) dynamic arterial elastance (changes in BP induced by changes in SV); and (iii) arterial–cardiac baroreflex (changes in HR induced by changes in BP) (Fig. 1). Using transfer function analysis, this model allows the quantification of beat‐by‐beat haemodynamic relationships between the input and output of each component of the cascade in the frequency domain.

We hypothesized that, as a result of an improvement in cardiac compliance (via 2 years of high‐intensity exercise training), cardiovascular regulation of HR variability would be altered, demonstrating the critical effect of cardiac filling on cardiovascular control.

Methods

Participant population

All participants in the present study were drawn from a previously published study for which a detailed description of the subject population and sample size calculation has been provided (Howden et al. 2018). Our previous results suggest that middle aged hearts retain some degree of plasticity and may respond to an adequate dose of training to restore youthful LV compliance (Howden et al. 2018). In brief, the present study comprised a prospective, parallel group, randomized controlled study, analysing 2 years of controlled exercise training. Sixty‐one healthy, sedentary, middle‐aged (aged 45–64 years) participants were recruited from the Dallas Heart Study (Victor et al. 2004) and from employees at Texas Health Resources and the University of Texas Southwestern Medical Centre, as well as via local media, between September 2012 and February 2014. After obtaining informed consent, all participants were rigorously screened for comorbidities and were excluded if any of the following conditions were present: hypertension, body mass index ≥30 kg m⁻², untreated hypo‐ or hyperthyroidism, obstructive sleep apnoea syndrome, chronic obstructive pulmonary disease, tobacco use during the past 10 years, coronary artery disease or structural heart disease. Participants were also excluded if they reported a consistent exercise history that involved exercising for >30 min, three times a week. The experimental procedures were explained to all participants and their informed consent was obtained as approved by the institutional review boards of the University of Texas Southwestern Medical Centre at Dallas and Texas Health Presbyterian Hospital Dallas. All procedures conformed with the standards set by the Declaration of Helsinki. The trial has been registered at ClinicalTrials.gov (NCT02039154) and was overseen by an independent data safety and monitoring board.

Intervention

Exercise training

The exercise training programme followed a periodized approach, where increases in training frequency, duration and intensity progressed over time. Each participant was provided with an individualized training plan which increased over the first 10 months, and then remained stable for the final 14 months. Details of the training programme have been reported previously (Howden et al. 2018). At peak training load (6–10 months), participants were training 5–6 hours per week, including two interval sessions and one long base pace session (at least 1 h) each week. For the final 14 months, one interval session per week was dropped and the training load continued at 4–5 h a week with one interval and one long session at lower intensity per week, plus two or three moderate intensity training sessions. This ‘dose’ of exercise (Bhella et al. 2014; Shibata et al. 2018) imparted in middle age (Howden et al. 2018) has been demonstrated to induce substantial changes in cardiac compliance.

Controls

In the control group, a combination of yoga, balance and strength training was prescribed three times a week for 2 years. This prescription allowed for a similar level of interaction with the research staff in both groups.

Measurements

Exercise testing

Measurements of maximal oxygen uptake were performed at baseline and at 2 years using the Douglas bag technique; gas fractions were analysed by mass spectrometry and ventilatory volumes were analysed using a Tissot spirometer, as reported previously (Howden et al. 2018). Maximal oxygen uptake (${\dot{V}}_{O_2\max }$) was defined as the highest oxygen uptake measured from at least a 30 s Douglas bag.

Steady‐state haemodynamics

Right heart catheterization was performed before and after the 2 year intervention. A 6 Fr Swan‐Ganz catheter under fluoroscopic guidance was placed into the pulmonary artery via an antecubital vein. Mean pulmonary capillary wedge pressure (PCWP) and right atrial pressure were determined visually at end expiration. Cardiac output was measured with the C₂H₂ rebreathing method (Jarvis et al. 2007). Blood pressure was measured at the brachial artery by ECG gated electrosphygmomanometry (Tango; SunTech Medical, Witney, UK) during cardiac output measurements. Total arterial compliance was determined by the ratio of stroke volume and pulse pressure to evaluate central aortic function. Effective arterial elastance was defined as the ratio of end‐systolic pressure over stroke volume (Kelly et al. 1992), with end‐systolic pressure estimated by the use of the single‐beat method, as previously described and validated (Chen et al. 2001).

Dynamic cardiovascular variability

Pulmonary artery diastolic (PAD) pressure was used as a surrogate of LVEDP to avoid the risks associated with prolonged balloon inflation. Finger photoplethysmography (Portapres; FMS, Amsterdam, The Netherlands) was used to measure arterial blood pressure continuously, with beat‐to‐beat SV calculated from the Model‐flow method (Beat‐Scope; FMS). After confirmation of haemodynamic stability, subjects were asked to breathe at a controlled frequency (0.2 Hz: 12 breaths min^–1) by visually tracking a moving cursor displayed on a computer for 8 min at the same time as beat‐by‐beat PAD, SV index, sBP and RR interval were continuously collected. The final 6 min data segment was used for the transfer function analysis.

Data analysis

Transfer function analysis

PAD, sBP and ECG waveforms were sampled at 1 kHz and digitized at 12 bits with an analogue‐to‐digital converter (Das‐20; Metrabyte, Keithley Instruments Cleveland, OH, USA). Digitized signals were stored in a laboratory computer and processed with a custom designed program for PAD, sBP and R‐wave detection. Beat‐to‐beat PAD (used as a surrogate for LVEDP), sBP and RR interval were linearly interpolated and then resampled at 2 Hz for spectral analysis. Time series of PAD, sBP and RR interval were first detrended with third‐order polynomial fitting and then subdivided into 256 point segments with 50% overlap for spectral estimation. This process resulted in five segments of data over the 6 min period recordings. Fast Fourier transforms were implemented with each Hanning‐windowed data segment and then averaged to calculate autospectra [S _X–X(f)] and cross‐spectra [S _X–Y(f)] for PAD, sBP and RR interval variability, respectively. The spectral resolution for these estimates is 0.0078125 Hz. Transfer function analysis between these variables was used as described previously (Shibata et al. 2006; Hieda et al. 2017). Therefore, transfer functions [H(f)] of the dynamic Starling mechanism [H _{PAD–SV index}(f)], dynamic arterial elastance [H _{SV index–sBP}(f)], arterial–cardiac baroreflex [H _{sBP–RR interval}(f)] and the integrated cardiovascular regulation [H _{PAD–RR interval}(f)] were obtained with the equations:

$$H_{\mathrm{PAD}-\mathrm{SV}\mathrm{index}}(f)=S_{{\mathrm{PAD}}^{\ast }\mathrm{SV}\mathrm{index}}(f)/S_{{\mathrm{PAD}}^{\ast }\mathrm{PAD}}(f)$$

$$H_{\mathrm{SV}\mathrm{index}-\mathrm{sBP}}(f)=S_{{\mathrm{SV}}^{\ast }\mathrm{sBP}}(f)/S_{\mathrm{SV}{\mathrm{index}}^{\ast }\mathrm{SV}\mathrm{index}}(f)$$

$$\begin{array}{c}{H}_{\mathrm{sBP}-\mathrm{RR}\mathrm{interval}}(f)=S_{{\mathrm{sBP}}^{\ast }\mathrm{RR}\mathrm{interval}}(f)/S_{{\mathrm{sBP}}^{\ast }\mathrm{sBP}}(f)\hfill \end{array}$$

$$\begin{array}{ccc}\hfill H_{\mathrm{PAD}-\mathrm{RR}\mathrm{interval}}(f)& =& S_{{\mathrm{PAD}}^{\ast }\mathrm{RR}\mathrm{interval}}(f)/S_{{\mathrm{PAD}}^{\ast }\mathrm{PAD}}(f)\hfill \\ & =& H_{\mathrm{PAD}-\mathrm{SV}\mathrm{index}}(f)\times H_{\mathrm{SV}\mathrm{index}-\mathrm{sBP}}(f)\hfill \\ & & \times H_{\mathrm{sBP}-\mathrm{RR}\mathrm{interval}}(f)\hfill \end{array}$$

where $S_{{\mathrm{PAD}}^{*}\mathrm{SV}\mathrm{index}}(f)$, $S_{\mathrm{SV}{\mathrm{index}}^{*}\mathrm{sBP}}(f)$, $S_{{\mathrm{sBP}}^{*}\mathrm{RR}\mathrm{interval}}(f)$ and $S_{{\mathrm{LVEDP}}^{*}\mathrm{RR}\mathrm{interval}}(f)$ are cross‐spectra between PAD and SV index, SV index and sBP, sBP and RR interval, and PAD and RR interval, respectively; and $S_{{\mathrm{PAD}}^{*}\mathrm{PAD}}(f)$, $S_{\mathrm{SV}{\mathrm{index}}^{*}\mathrm{SV}\mathrm{index}}(f)$ and $S_{{\mathrm{sBP}}^{*}\mathrm{sBP}}(f)$ are the autospectra of PAD, SV index and sBP, respectively.

Transfer function gain was derived from the real part [H _R(f)] and the imaginary part [H _I(f)] of the complex transfer function as:

$$\left|\mathrm{Gain}(f)\right|={\left\{{\left[H_{\mathrm{R}}(f)\right]}^2+{\left[H_{\mathrm{I}}(f)\right]}^2\right\}}^{1/2}$$

Gain reflects the magnitude of the relationship between the input and output signals of the system modelled by H(f) over a specified frequency range of controlled breathing (0.19–0.21 Hz). In other words, the transfer function gain is similar to the slope of a linear regression but in the frequency domain. The integrated cardiovascular regulatory gain was obtained by the product Gain_{PAD‐SV index,} Gain_{SV index–sBP,} and Gain_{sBP–RR interval}.

Coherence function was derived from S _X–X(f), S _Y–Y(f) and S _X–Y(f) as:

$$\mathrm{Coherence}={\left|S_{\mathrm{X}-\mathrm{Y}}(f)\right|}^2/\left[S_{\mathrm{X}-\mathrm{X}}{(f)}^{\ast }{S}_{\mathrm{Y}-\mathrm{Y}}(f)\right]$$

The reliability of the linear transfer function was evaluated by estimates of coherence, which ranges between 0 and 1. The coherence is similar conceptually to the r ² value in a linear regression.

The group mean values of gain and coherence were calculated in the frequency range (0.19–0.21 Hz) and averaged for all subjects. The spectral power of LVEDP, SV index, sBP and RR interval was also calculated in the same frequency range (0.19–0.21 Hz) by integrating the corresponding autospectra.

Statistical analysis

Continuous variables are expressed as mean (95% confidence intervals) and categorical variables are expressed as n (%). The primary analysis included all participants who completed the 2 year follow‐up. Continuous end‐points were compared between groups using mixed‐effects model repeated measures analysis with the restricted maximum likelihood method. The repeated measures models included the intervention group factor (controls vs. exercise group), a repeated factor for study visits time and a Group × Time interaction; in addition to the overall average effect, the slope and intercept were allowed to vary from participant to participant (random effect). The difference in response between the controls and the high‐intensity exercise group was assessed via the interaction effect. Post hoc multiple comparisons were made by the Mann–Whitney U test with the use of the Bonferroni correction. P < 0.05 was considered statistically significant. Statistical analysis was performed by computer software (JMP, version 11.0; SAS Institute Inc., Cary, NC, USA).

Results

Baseline characteristics

Sixty‐one participants were randomized in the present study. In total, 53 participants completed the 2 year study: 25 in the control group and 28 in the high‐intensity exercise group. Although some of these data have been reported previously (Howden et al. 2018), key variables such as subject characteristics and overall response to training are also reported here for ease of access. The baseline characteristics are summarized in Table 1. Participants in the high‐intensity exercise group maintained excellent compliance with the 2 year exercise intervention (mean of 88 ± 11%).

Table 1.

Baseline characteristics

	Controls	Exercise group
	(n = 28)	(n = 33)
Age (years)	51.4 (49.4–53.4)	53.2 (51.5–54.9)
Male, n (%)	13 (46)	15 (45)
Weight (kg)	75.4 (70.0–80.9)	75.1 (70.2–80.0)
Height (cm)	169 (165–173)	170 (167–173)
BMI (kg/m–2)	26.2 (25.0–27.5)	25.8 (24.7–26.8)
BSA (m2)	1.88 (1.79–1.96)	1.88 (1.80–1.96)
Race, n (%)
White	23 (82)	26 (79)
Black	–	1 (3)
Hispanic	2 (7)	2 (6)
Asian	3 (11)	4 (12)
24 h ABPM sBP (mmHg)	123 (120–126)	120 (118–123)
24 h ABPM dBP (mmHg)	74 (72–76)	72 (70–74)
${\dot{V}}_{O_2\max }$ (mL kg–1 min–1)	29.5 (27.6–31.4)	29.0 (27.3–30.7)

Values are the mean (95% confidence intervals) or n (%). BMI, body mass index; BSA, body surface area; ABPM, ambulatory blood pressure monitoring, dBP, diastolic blood pressure.

The effect of the high‐intensity training on haemodynamics and exercise capacity

The effect of the exercise intervention on resting, steady‐state haemodynamic parameters and exercise capacity is summarized in Table 2, as reported previously (Howden et al. 2018). Exercise training significantly improved ${\dot{V}}_{O_2\max }$, decreased resting HR, increased SV and improved LV compliance.

Table 2.

Haemodynamics and exercise capacity

	Controls (n = 25)		High‐intensity exercise group (n = 28)
	Pre	Post	Pre	Post	Group × Time
HR (bpm)	64 (60–67)	64 (61–67)	63 (60‐67)	58 (55–61)a	0.0003
sBP (mmHg)	109 (106–113)	107 (103–110)	107 (104–110)	104 (102–107)	0.95
dBP (mmHg)	69 (67–72)	70 (67–73)	67 (65–69)	68 (65–71)	0.95
CI (L min–1 m–2)	2.5 (2.4–2.7)	2.5 (2.4–2.7)	2.5 (2.4–2.7)	2.5 (2.3–2.6)	0.42
SV index (mL m–2)	41 (38–43)	42 (39–44)	42 (39–45)	45 (42–49)	0.32
TAC index (mL mmHg–1 m–2)	1.05 (0.96–1.13)	1.20 (1.08–1.32)	1.07 (0.97–1.17)	1.29 (1.16–1.42)	0.50
Ea index (mmHg mL–1 m–2)	2.47 (2.30–2.64)	2.35 (2.18–2.52)	2.37 (2.18–2.57)	2.16 (2.00–2.32)	0.48
${\dot{V}}_{O_2\max }$ (mL kg–1 min–1)	29.5 (27.6–31.4)	28.7 (26.5–30.9)	29.0 (27.3–30.7)	34.4 (32.0–36.7)a	<0.0001

^a P < 0.05 compared to pre within group (post hoc multiple comparisons).

Values are the mean (95% confidence intervals). dBP, diastolic blood pressure; CI, cardiac output index; TAC, total arterial compliance; Ea index, effective arterial compliance.

Transfer function analysis

Dynamic Starling mechanism (dynamic ventricular–arterial coupling)

The gain of the dynamic Starling mechanism (Gain_{PAD–SV index}: the transfer function gain between PAD and SV index) increased from pre to post in the exercise group but was unchanged in the controls (Fig. 2A). Power spectral density of PAD (quantifying the magnitude of the variability in PAD at the respiratory frequency), the input variable of the dynamic Starling mechanism, varied at the controlled respiratory frequency (0.2 Hz) and was significantly lower in the exercise group at post compared to pre but did not change in the controls (Table 2). Power spectral density of SV index, the output variable of the dynamic Starling mechanism, was comparable between the two groups at pre and post (Table 3).

Figure 2. Transfer function gain in each component of the integrated cardiovascular regulation.

Transfer function gain of the dynamic Starling mechanism (A), dynamic arterial elastance (B) and arterial–cardiac baroreflex (C).

Table 3.

Power spectral density and coherence

	Controls (n = 25)		Exercise group (n = 28)
	Pre	Post	Pre	Post	Group × Time
PSD of PAD (mmHg2)	1.66 (1.14–2.18)	2.09 (1.39–2.79)	3.15 (1.94–3.99)	2.14 (1.25–3.01)*	0.02
PSD of SV index [(mL m–2)2]	1.30 (0.75–1.84)	1.39 (0.84–1.93)	1.47 (0.95–1.98)	1.93 (1.42–2.45)	0.35
PSD of sBP (mmHg2)	3.08 (1.88–4.27)	3.04 (1.84–4.23)	4.23 (3.50–5.76)	3.84 (2.72–4.97)	0.28
PSD of RR interval (ms2)	254 (143–366)	248 (136–360)	263 (157–368)	397 (291–502)	0.053
Coherence of PAD–SV index	0.69 (0.61–0.78)	0.68 (0.60–0.76)	0.69 (0.62–0.77)	0.74 (0.66–0.81)	0.46
Coherence of SV index–sBP	0.74 (0.65–0.83)	0.69 (0.60–0.77)	0.73 (0.65–0.81)	0.73 (0.65–0.82)	0.44
Coherence of sBP–RR interval	0.77 (0.71–0.84)	0.76 (0.69–0.82)	0.76 (0.69–0.82)	0.72 (0.66–0.78)	0.74

^* P < 0.05 compared to pre within group.

Values are the mean (95% confidence intervals). PSD, power spectral density.

Dynamic arterial elastance

The gain of dynamic arterial elastance (Gain_{SV index–sBP}: the transfer function gain between SV index and sBP) at the respiratory frequency was similar between the two groups at pre and post (Fig. 2B). Power spectral density of sBP, the output variable of dynamic arterial elastance, was also similar between the two groups at pre and post (Table 3).

Arterial–cardiac baroreflex

The arterial–cardiac baroreflex gain (Gain_{sBP–RR interval}: the transfer function gain between sBP and RR interval) at the respiratory frequency increased in the high‐intensity exercise group compared to the controls (Fig. 2C). At a lower frequency range (0.05–0.15 Hz), the arterial–cardiac baroreflex gain similarly increased in the high‐intensity exercise group compared to the controls [Group × Time, P = 0.043, Controls: from 6.01 (4.88–7.14) to 6.57 (5.26–7.88) ms mmHg^–1, P = 0.98, vs. Exercise group: from 7.09 (5.85–8.34) to 11.9 (8.23–15.7) ms mmHg^–1, P = 0.0066]. Power spectral density of RR interval, the output variable of the arterial–cardiac baroreflex, increased by a greater amount from pre to post in the high‐intensity exercise group (Table 3).

Integrated cardiovascular regulation

The integrated cardiovascular regulation gain (Gain_{PAD–RR interval}) was significantly increased by a factor of 1.34 in the high‐intensity exercise group compared to the controls (Group × Time interaction, P = 0.02) (Fig. 3). The integrated cardiovascular regulation gain in both males and females in the exercise group was significantly improved compared to the controls [females, Controls: from 16.4 (10.0–22.8) to 14.7 (8.3–21.1) vs. Exercise group: from 14.3 (8.1–20.4) to 19.2 (13.1–25.3), Group × Time interaction, P = 0.008; and males: Controls: from 8.6 (5.4–11.8) to 8.4 (5.2–11.5) vs. Exercise group: from 8.5 (5.6–11.4) to 11.3 (8.4–14.2), Group × Time interaction, P = 0.05]. PAD‐SV index, the SV index–sBP, and sBP‐RR interval in both the control and high‐intensity exercise group at pre and at post was similar and sufficiently high (range: 0.68–0.77) to support the validity of the transfer function analysis (Table 3).

Figure 3. Transfer function gain.

Transfer function gain of integrated cardiovascular regulation (the cascade of the dynamic Starling mechanism, dynamic aortic elastance and arterial–cardiac baroreflex function).

Discussion

The primary findings of the present study are: (i) after 2 years of high‐intensity exercise training, the dynamic Starling mechanism gain and the arterial–cardiac baroreflex gain were significantly increased in the exercise group, although there was no significant change in dynamic arterial elastance, and (ii) the integrated cardiovascular regulation gain, considered as an integrated multiplicative linear function, increased significantly by 1.34‐fold in the exercise group compared to the controls. These findings demonstrate that vigorous exercise training can improve total cardiovascular function by enhancement of LV compliance and arterial baroreflex sensitivity.

Dynamic Starling mechanism

The beat‐by‐beat and breath‐to‐breath dynamic relationship between PAD and SV, which we have termed ‘dynamic Starling mechanism’ gain, is determined by both the end‐diastolic pressure–volume relationship and ventricular–arterial coupling (Hieda et al. 2017). In a previous study of 1‐year training in sedentary seniors (>65 years), there was a modest but highly variable increase in the dynamic Starling mechanism, without a change in cardiac compliance (Fujimoto et al. 2013). In the present study, the dynamic Starling mechanism gain in the exercise group was clearly improved by 2 years of high‐intensity training as a result of the reduction of the power spectral density of PAD (the input variability), without changing the power spectral density of the SV index (the output variability). This outcome is probably explained by a more compliant LV in the exercise group resulting in smaller fluctuations in PAD despite similar respiratory induced changes in cardiac filling, facilitating improvements in the efficiency of the Starling mechanism and perhaps cardiac energetics. By contrast, if exercise training is started too late in life (e.g. after age 65 years), there may be little effect on LV stiffness as a result of a reduction of cardiac plasticity in sedentary, senior hearts (Fujimoto et al. 2010; Fujimoto et al. 2013). These findings indicate that 2 years of exercise training can improve the dynamic Starling mechanism in sedentary, middle‐aged individuals which incorporates: (i) a vigorous exercise prescription (≥4 days week^–1 including high‐intensity interval training for 2 years) and (ii) intervening during the ‘sweet spot period’, in which potential cardiac plasticity is sufficient to reverse age‐related LV stiffening, if still present (Bhella et al. 2014; Shibata et al. 2018).

Dynamic arterial elastance

Several randomized controlled trials report that various exercise modalities have a beneficial effect on endothelial function (Munk et al. 2009; Vona et al. 2009; Figueroa et al. 2011; Ho et al. 2012; Beck et al. 2013). Vigorous exercise training strategies improved arterial stiffness significantly, and such an effect was enhanced with high‐intensity exercise training and in participants with greater arterial stiffness at baseline (Ashor et al. 2014). This diminishing augmentation in central pressure leads to decreased ventricular afterload and increased coronary perfusion pressure, which may reduce subsequent cardiovascular disease including myocardial hypertrophy, ischaemia and infarction (Sakuragi & Abhayaratna, 2010). As opposed to these specific and direct measures of arterial structure and function, we assessed dynamic arterial elastance fluctuations caused by breathing fluctuations as part of our assessment of integrated cardiovascular regulation. By contrast to our initial hypothesis, there was no convincing change in the gain of dynamic arterial elastance (Gain_{SV index–sBP}) despite 2 years of intensive training that was sufficient to alter cardiac compliance. There are several possible explanations for this finding. Possibly, more muscular and/or stiffer peripheral arteries allow larger adaptations of arterial wall properties in response to exercise training compared to central elastic arteries (Green et al. 2017). In agreement with this hypothesis, one study reported that exercise training improves peripheral artery stiffness (i.e. popliteal artery) in the absence of changes in a central (carotid) artery (Rakobowchuk et al. 2008). It may be that structural adaptations to exercise in large blood vessels (e.g. increased elastin content and inhibition of collagen activity within the arterial wall) require the passage of longer periods of time than structural changes in the heart that may be more plastic (Ferreira et al. 2006). We have previously demonstrated that functional changes in the arteries are much more responsive to training than structural ones (Shibata et al. 2011). It is also possible that, although critical to the analysis strategy of the present study, this particular method of evaluating the dynamic change in BP with changes in SV is not the most sensitive tool to assess changes in arterial structure and function after exercise training. Other measures, such as the augmentation index, beta stiffness index and steady‐state measures of arterial compliance, are probably more appropriate for examining the specific question about changes in arterial function with prolonged exercise training.

Arterial–cardiac baroreflex

The gain of the arterial–cardiac baroreflex was significantly improved in the high‐intensity exercise group compared to the controls. The present study indicates that 2 years of high‐intensity exercise training enhances the arterial–cardiac baroreflex and increases HR variability at the respiratory frequency. In the present analysis, it must be acknowledged that the sBP–RR relationship at the respiratory frequency may be influenced by direct neural effects of respiration rather than exclusively feedback from baroreceptors. However, analysing the gain of the arterial–cardiac baroreflex at a lower frequency range (0.05–0.15 Hz), which avoids the effect of respiration on HR, showed similar results.

Cardiac autonomic function is primarily controlled by the arterial baroreflex, located in the aortic arch and carotid sinuses, which responds to a drop in blood pressure by eliciting inverse changes in HR. The arterial–cardiac baroreflex control has prognostic significance in patients with increased cardiovascular risk (La Rovere et al. 2002). Furthermore, impaired arterial–baroreflex sensitivity is associated with all‐cause and cardiovascular mortality (Gerritsen et al. 2001). Previous studies demonstrated that exercise training can normalize the arterial–cardiac baroreflex in disease conditions such as heart failure, perhaps mediated by a reduction in renin–angiotensin system activity (Mousa et al. 2008), or a decreased oxidative stress response (Gao et al. 2007). Indeed, lowered arterial–baroreflex sensitivity was found to double the risk of mortality in patients who had other cardiovascular risk factors, such as diabetes or hypertension (Gerritsen et al. 2001). Several studies have demonstrated resetting of the baroreflex stimulus–response curve during exercise, indicating a vertical upward shift on the response arm and a lateral rightward shift to higher operating pressures (Fadel et al. 2001; Ogoh et al. 2003; Raven et al. 2006). It may be that the repetitive and sustained shifts in baroreflex function during exercise training could be one mechanism by which 2 years of vigorous exercise training in the present study could improve arterial baroreflex sensitivity.

Integrated cardiovascular regulation

Our group previously found that sedentary middle‐aged adults have evidence of LV stiffening with a concomitant reduction in distensibility, and a leftward shift in the LV end‐diastolic pressure–volume relationship (Fujimoto et al. 2012). In previously sedentary, healthy, middle‐aged adults, 2 years of exercise training improved peak oxygen uptake and decreased cardiac stiffness (Howden et al. 2018). Therefore, we theorize that the hearts of healthy, sedentary middle‐aged adults retain some degree of plasticity in the integrated cardiovascular regulation as well. Consequently, this period of life may represent an optimal time to ‘intervene’ via an exercise training regimen, with the ultimate goal of preventing/reversing age‐related reductions in distensibility that otherwise occur with sedentary ageing. Interestingly, the integrated cardiovascular regulation gain (Gain PAD–RR interval) was improved by 1.34‐fold, enhancing the dynamic Starling mechanism and especially arterial–cardiac baroreflex sensitivity in the high‐intensity exercise group compared to the controls.

The frequency domain analysis used in the present study allows us to assess the efficiency of a cascade model in the cardiovascular system, including each component of overall cardiovascular function, as well as the integrated cardiovascular regulation that consists of those components. Indeed, the frequency domain analysis enables the quantification of a series of beat‐by‐beat and breath‐to‐breath haemodynamic parameters simultaneously, which is impossible for a time domain analysis. Because the integrated cardiovascular regulation model has the ability to handle data transformed from continuous haemodynamic parameters at a specific frequency range, this allows for a more precise and complete description of haemodynamic power transfer and may be especially useful for evaluating the changes in arterial–cardiac baroreflex sensitivity that might be mediated by multiple underlying haemodynamic influences.

Study limitations

The present study has several limitations. First, SV was measured indirectly using the model flow method, although this technique has been validated (Abdellatif et al. 2016). Although the absolute values of SV as measured by the model flow technique are not particularly accurate as a result of the assumptions of the technique with respect to pulse contour analysis, the relative changes on a beat‐to‐basis are much more accurate, and the continuous nature of this measurement allows the bedside monitoring of beat‐by‐beat SV changes. Second, the absolute value of PAD is not equivalent to PCWP or LVEDP; however, PAD is considered to be a suitable surrogate, as shown by a strong positive linear relationship between PCWP and PAD, and it tracks changes in LV filling pressure quite well (Shibata et al. 2011). Third, we acknowledge that the integrated model parameter of cardiovascular regulation (Starling mechanism × arterial elastance × baroreflex) is susceptible to a potential summation error. Future studies will also need to address whether this characterization of integrated cardiovascular regulation may be associated with clinical outcomes.

Conclusions

In these previously sedentary, otherwise healthy middle‐aged adults, 2 years of high‐intensity exercise training improves integrated cardiovascular regulation by enhancing the dynamic Starling mechanism and arterial–cardiac baroreflex sensitivity without changing the dynamic arterial elastance.

Clinical perspectives

What is new?

• In these previously sedentary, otherwise healthy middle‐aged adults, 2 years of high‐intensity exercise training improved integrated cardiovascular regulation by enhancing the dynamic Starling mechanism and arterial–cardiac baroreflex sensitivity.

• The integrated cardiovascular regulation gain in the exercise group was 1.34‐fold larger than the control group after the intervention.

What are the clinical implications?

• High‐intensity exercise training is an effective strategy to improve the integrated cardiovascular regulation including the dynamic Starling mechanism and arterial–cardiac baroreflex sensitivity. These adaptations may explain at least part of the beneficial effects of exercise training on blood pressure control and cardiovascular health.

Additional information

Competing interests

The authors declare that they have no competing interests.

Author contributions

BDL conceived and designed the study and experiments. MH developed the analysis strategy and analysed the data as presented. EJH, SS, JSL, WC and BDL performed the experiments. MH was responsible for drafting the manuscript, with all authors contributing to writing the paper. All authors approved the final version of the manuscript submitted for publication. All persons designated as authors quality for authorship and all those who qualify for authorship are listed.

Funding

This study was supported by National Institute of Health Grant AG017479. Drs M. Hieda, S. Sarma and B. D. Levine were also supported in part by the American Heart Association Strategically Focused Research Network (14SFRN20600009‐03). Dr. Hieda was also supported by American Heart Association post‐doctral fellowship grant (18POST33960092) and the Harry S. Moss Heart Trust. Dr W. Cornwell is supported by the National Institute of Health/National Heart, Lung and Blood Institute mentored patient‐oriented research career development award (1K23HL132048‐01).

Biography

Michinari Hieda graduated from a pharmaceutical science course (2001) and a master degree course (2003) at Nagasaki University and the School of Medicine at Shimane University (2007). He worked at Otowa Hospital (Resident, 2007–2009) in Kyoto, the Kokura Memorial Hospital (Fellow, 2009–2012) in Kokura, and the National Cerebral and Cardiovascular Centre (Senior Fellow, 2012–2016) in Osaka, Japan. Currently, he is working at the Institute for Exercise and Environmental Medicine as a Senior Visiting Researcher (2016 to present). His research focuses on haemodynamics in patients with heart failure based on pressure–volume loop theory with a special interest in left ventricular energetics.