Autonomic control of heart rate during exercise studied by heart rate variability spectral analysis YOSHIHARU Department
YAMAMOTO,
RICHARD
L. HUGHSON,
AND JOHN
C. PETERSON
of Kinesiology, University of Waterloo, Waterloo, Ontario NZL 3Gl, Canada
YAMAMOTO,YOSHIHARU,RICHARDL. HUGHSON,ANDJOHN C. PETERSON. Autonomic controZ of heart rate during exercise studied by heart rate variability spectral analysis. J. Appl. Physiol. 71(3): 1136-1142, 1991.-Spectral analysis of heart rate variability (HRV) might provide an index of relative sympathetic (SNS) and parasympathetic nervous system (PNS) activity during exercise. Eight subjects completed six 17-min submaximal exercise tests and one resting measurement in the upright sitting position. During submaximal tests, work rate (WR) was increased for the initial 3 min in a ramp fashion until it reached constant WRs of 20 W, or 30,60,90,100, and 110% of the predetermined ventilatory threshold (Tvent). Ventilatory profile and alveolar gas exchange were monitored breath by breath, and beat-to-beat HRV was measured as R-R intervals of an electrocardiogram. Spectral analysis was applied to the HRV from 7 to 17 min. Low-frequency (O-O.15 Hz) and highfrequency (0.15-1.0 Hz) areas under power spectra (LO and HI, respectively) were calculated. The indicator of PNS activity (HI) decreased dramatically (P < 0.05) when the subjects exercised compared with rest and continued to decrease until the intensity reached 60% Tvent. The indicator of SNS activity (LO/ HI) remained unchanged up to 100% Tvent, whereas it increased abruptly (P < 0.05) at 110% Tvent. The results suggested that (cardiac) PNS activity decreased progressively from rest to a WR equivalent to 60% Tvent, and SNS activity increased only when exercise intensity exceeded Tvent.
Exercise heart rate has recently been studied by spectral analysis of HRV. Arai et al. (3) attempted to compare the results of spectral analyses with data published by several investigators who had employed pharmacological blockade of either or both the PNS and SNS (8, 22). Although the data of Arai et al. clearly showed a reduction in PNS with increasing intensities of exercise, the data failed to show evidence of increased SNS. This failure is difficult to rationalize given the extensive evidence supporting this from studies with P-adrenergic receptor blockade (8, 22) and measurement of plasma catecholamines (9, 10, 17). It was the purpose of this study to examine in more detail the heart rate response to rest and various levels of submaximal exercise. We have done this, first, by studying longer periods of constant load exercise and, second, by applying a new approach to the study of HRV, which we have labeled coarse-graining spectral analysis (30). This method provides improved resolution of short-term HRV spectra and will permit evaluation of the LO/HI ratio as an indicator of SNS activity during increasing intensities of exercise.
autonomic nervous system; sympathetic nervous system; parasympathetic nervous system; 1 lf component; fractal component
Eight healthy individuals (5 males and 3 females) volunteered for the experiments. Their physical characteristics are presented in Table 1. Each subject signed a consent form approved by the Office of Human Research of the University after the test protocol was described. Ail subjects completed preliminary testing in which peak 0, uptake (VO 2peak)and ventilatory threshold (T,,t) were determined (Table 1). From a baseline of 25 W in the upright position on an electrically braked cycle ergometer (Siemens), work rate (WR) increased as a continuous ramp at a rate of 15 W/min. Exhaustion was taken as the point at which subjects could no longer maintain 150 rpm. The v02peak was defined as the highest 0, recorded over a 10-s period during breath-bybreath measurement, as described below. Tvent was taken as the point at which 1) minute ventilation (Tj,) began to increase in a breaking fashion against Vo2, 2) CO, output (VCO~) began to increase in a breaking fashion against VO,, 3) vE/v02 exhibited a systematic increase without a concomitant increase in vE/kO,, and 4) end-tidal 0, pressure exhibited a systematic increase without a concomitant increase in end-tidal CO, pressure (29). At least two researchers determined Tvent on the basis of satisfaction of more than one criterion. If
HEART RATE is controlled
by the balance between parasympathetic (PNS) and sympathetic nervous system (SNS) activity to the sinoatrial node. On a beat-to-beat basis, heart rate is not constant. Rather, there are periodic fluctuations that are indicative of the relative contributions of each of these components of the autonomic nervous system (25). Spectral analysis of heart rate variability (HRV) has shown at least two distinct regions of periodicity in heart rate. In combination with studies of various pharmacological or physiological manipulations, it has been demonstrated that high frequencies (HI, >0.15 Hz) of HRV are associated solely with cardiac PNS activity (1, 2, 5, 19-21, 25, 28). Lower-frequency (LO, co.15 Hz) HRV might be associated with both PNS and SNS activity (1, 2, 5, 25). In addition, the ratio LO/HI might selectively indicate SNS activity (19, 20). Thus studies of HRV could provide noninvasive evaluation of autonomic neural balance in a number of physiological settings. 1136
0161-7567191
$1.50
METHODS
Copyright 0 1991 the American Physiological Society
Downloaded from www.physiology.org/journal/jappl by ${individualUser.givenNames} ${individualUser.surname} (128.151.010.035) on July 29, 2018. Copyright © 1991 American Physiological Society. All rights reserved.
HEART
TABLE
CONTROL
DURING
EXERCISE
1. Physical characteristics of the subjects
Subj No.
Sex
Age, yr
Ht, cm
I 2 3 4 5 6 7 8
M M M M M F F F
22 21 22 28 19 21 22 19
174 176 180 174 173 165 169 160
22 +3
171 -t5
Mean +SD WR, uptake.
RATE
work
rate;
WR at Tent, W
Tent, ml/min
70 73 95 69 70 57 55 60
190 198 255 208 161 140 115 103
2,370 2,490 3,660 2,780 2,120 1,750 1,540 1,280
321 341 373 324 296 253 207 208
3,850 4,180 4,970 4,250 3,640 2,850 2,280 2,450
69 +12
172 +48
2,250 *710
290 &58
3,560 +890
Wt, kg
Tvent , ventilatory
V02 at
threshold;
Max
WR, W
VO, W,
vozmax,
ml/min
maximal
0,
the difference among determined WRs at Tvent was >5.0 W among investigators, ramp work tests were repeated until it fell within 5.0-W range. Submaximal exercise tests. The subjects underwent in random order six 17-min submaximal exercise tests and one resting measurement in the upright sitting position. During submaximal tests, WR was increased for the initial 3 min in a ramp fashion until it reached the set loads of 20 W or WRs corresponding to 30, 60, 90, 100, and 110% of each subject’s Tvent. The set loads were then maintained for 14 min. Except for the resting measurement and the 20-W exercise, only one test was conducted on 1 day. Subjects 1 and 3 did not complete the resting measurement and the 20-W trial. During the tests, ventilatory profile and alveolar gas exchange were monitored breath by breath and heart rate was measured beat to beat continuously for the entire 17 min. To observe steady-state responses, data during the initial 7 min were discarded, and the remaining lo-min bins (7-17 min) were used for analyses. Data collection. Breath-by-breath measurements of ventilation and alveolar gas exchange were obtained on a computerized system (First Breath) incorporating a respiratory mass spectrometer (MGA-llOOA, PerkinElmer) and a volume turbine (VMM-110, Alpha Technologies) (12). Allowance was made for breath-by-breath changes in nominal functional residual capacity and endtidal fractional gas concentrations to calculate alveolar VO, and VCO, (4). Calibration of the mass spectrometer was checked before each experiment with standard calibration gases. Volume factors were checked with a syringe of known volume; care was taken to use a mean flow rate similar to that of the subject during the exercise test. Temperature and water vapor corrections were made on inspired and expired air for conditions measured near the volume turbine. The surface electrocardiogram (ECG) was measured continuously during the tests from standard bipolar leads with an electrocardiograph (model 7830A, HewlettPackard). Analog output of the ECG meter was differentiated, and the resultant QRS spikes triggered a Schmitt circuit to generate a train of rectangular impulses. The impulse train was processed on a real-time basis with a personal computer via a 12-bit analog-to-digital converter (DAS-16, Metrabyte) at a sampling frequency of
BY
SPECTRAL
ANALYSIS
1137
1,000 Hz. The intervals of the impulses were stored sequentially on a diskette for later analyses. Spectral anaZysis.Before HRV spectra were calculated, lo-min HRV data were searched for the possibility of extra or missing beats that could affect the result of spectral analyses (26). All R-R intervals for which absolute values of beat-to-beat transition in heart rate were >20 beats/min were subjected to filtering. The filtering process either deleted or inserted beats to obtain appropriate intervals for those that had transitions greater than +20 or -20 beats/min, respectively. Thereafter, the unequal intervals of HRV were aligned sequentially to obtain equally spaced samples [interval tachogram (23)]. The resultant data were first divided into three parts, and the linear trend of each part was eliminated by a linear regression method. After passing through the data window, a fast Fourier transform was used to calculate a periodgram of each part. The ensemble average of the periodgrams was subjected to a Hanning-type spectral window and was regarded as an estimate of HRV spectra. Because the HRV spectra calculated in this way had a unit of cycles/beat, we converted it to Hz by dividing by the average R-R interval. We calculated HRV spectra using R-R interval data rather than heart rate; therefore the unit of spectral density in the present study was ms2/Hz. Using almost the same technique, some researchers (14, 26) showed that there was a nonharmonic “fractal” component in HRV data in the frequency band from 0.00003 to 0.1 Hz (a period of -10 h to 10 s). Fourier transforms of such components are known to result in a broad band but nonwhite, or so-called l/f, spectrum (16), in which the power spectral densities are inversely proportional to their frequency. It could be anticipated that the existence of this nonwhite biased noise could affect HI and LO calculations from short-term HRV spectra. Yamamoto and Hughson (30) have described a new approach called coarse-graining spectral analysis (CGSA) for calculating HRV spectra with the elimination of this noise component. In the present study, CGSA was applied to HRV data during exercise, and the results were compared with those derived from the general spectral analysis (GSA) described earlier. The theoretical background and the detailed procedure of CGSA were reported by Yamamoto and Hughson (30). A brief description of the method follows. First, each R-R interval in raw HRV data was repeated so that the length of the data set was twice as long as an original time series (x) used in GSA. Then the “coarse-grained” data set x’ was constructed by taking the first half of the data set before it was resealed by an appropriate scale factor. The objective of CGSA was to eliminate nonharmanic l/f components from HRV autopower spectrum calculated for x by GSA (S,), leaving harmonic components of interest intact. Because of the scale invariant, or self-similar property of the l/f noise, it was shown that the cross-power spectra between x and X’ (SA) preserved l/f components, but SA could not retain any harmonic components. Therefore we subtracted SA from S, to obtain the new spectra, in which only the harmonic components were emphasized. From HRV spectra obtained both from GSA and
Downloaded from www.physiology.org/journal/jappl by ${individualUser.givenNames} ${individualUser.surname} (128.151.010.035) on July 29, 2018. Copyright © 1991 American Physiological Society. All rights reserved.
1138
HEART
Rest
l/min Mean Range f, min-’ Mean Range VT, liters Mean Range VO,, l/min Mean Range VC02, llmin Mean Range WR, W Mean Range HR, beats/min Mean Range
RATE
CONTROL
20 w
DURING
EXERCISE
30% xmt
60%
BY SPECTRAL
ANALYSIS
Tvellt
Lit
90%
100%
Tvent
110%
Tvent
7jE,
60.9
8.1 6.6-9.4
17.4 14.0-21.1
24.4 13.4-35.0
33.7 23.5-47.7
46.0 32.4-61.4
52.7 33.3-66.7
36.2-80.7
12.6 10.3-17.0
16.5 11.5-22.1
18.4 13.6-26.6
19.2 12.9-29.4
22.8 15.8-32.3
24.3 17.8-35.7
27.2 20.1-36.3
0.66 0.46-0.82
1.13 0.75-1.84
1.40 0.78-2.12
1.92 1.12-2.96
2.19 1.17-3.51
2.33 1.27-3.70
2.34 1.30-3.71
0.30 0.23-0.37
0.63 0.50-0.74
1.01 0.55-1.70
1.55 1.11-2.54
2.12 1.48-3.35
2.36 1.57-3.63
2.58 1.63-4.05
0.27 0.20-0.32
0.57 0.44-0.62
0.90 0.47-1.49
1.43 1.00-2.22
2.00 1.39-3.16
2.27 1.49-3.47
1.55-3.96
51 31-77
103 62-152
154 93-203
171 103-255
188 113-281
95
114 107-136
135 126-149
145 134-162
155 141-167
20
64 61-66
82 72-92
86-103
2.49
Values are means and ranges for responses obtained between 7 and 17 min of exercise. Tvent, ventilatory threshold; i7E, minute ventilation; f, breathing frequency; VO,, O2 uptake; VCO~, CO, output; WR, work rate; HR, heart rate. Note that the minimum f of 10.3 min-’ or 0.17 Hz was greater than the border frequency used to calculate high- and low-frequency area in heart rate variability spectra (i.e., 0.15 Hz).
CGSA, LO and HI were calculated by integrating spectrum densities for the range of 0.0 < f < 0.15 Hz and 0.15 < f < 1.0 Hz, respectively. Statistical analysis. A Student-Newman-Keuls post hoc test was applied for each parameter when a one-way analysis of variance indicated a significant effect. It would be shown later that the mean values in HI and LO differed very much between rest or light exercise data and moderate-to-high exercise data. The complete interindividual comparisons for all eight subjects were possible for 30-110% Tvent data sets. A two-way analysis of variance followed by the post hoc test was conducted on these data. RESULTS
Steady-state responses of respiratory variables and heart rate obtained between 7 and 17 min increased progressively with WR (Table 2). The mean heart rate at 110% Tvent trials was 155 beats/min, which was only slightly lower than the peak heart rate in the study of Arai et al. (3). It is of note that the minimum breathing frequency of 10.3 breaths/min or 0.17 Hz was higher than the border frequency used to calculate HI and LO (i.e., 0.15 Hz). It was important that breathing frequency was not lower than this border frequency; otherwise the interpretation of calculated LO and HI would become quite complicated because of effects of respiratory sinus ‘arrhythmia
(6, 25, 28). The actual HRV recordings and the spectra calculated by CGSA of subject 6 are shown in Fig. 1. Compared with the recording at rest, HRV decreased considerably during exercise, even at 20 W. The spiky pattern in HRV was still observable at 20 W and 30% Tvent. However, during exercise at intensities higher than 60% Tvent, heart rate
looked almost constant. Reduced HRV at higher exercise intensities could also be observed in the HRV spectra. It should be emphasized that, even in these HRV spectra with low power, there were low- and high-frequency peaks, and the latter occurred at the mean breathing frequencies. Application of the CGSA technique showed that both HI and LO components decreased dramatically (P < 0.05) when the subjects exercised (Table 3) and tended to decrease until the intensity reached 60% Tvent (see Fig. 3). Similar phenomena were also observed in standard deviations (SD) of R-R intervals; i.e., exercise per se caused a significant reduction in HRV until 60% Tvent. In contrast, LO/HI remained unchanged up to a 100% Tvent level (Table 3). During exercise at 110% Tvent, LO/HI increased abruptly (P < 0.05). The PNS indicator (HI) dropped dramatically with exercise per se and continued to decrease up to 60% Tvent, whereas the SNS indicator (LO/HI) increased only when exercise intensity exceeded Tvent. Further analyses were conducted using a two-factor model (exercise intensity X subject) for 30-110% Tvent data sets. HI of 30% Tvent was significantly (P < 0.05) greater than those of the other trials (Fig. 2A), although LO did not show any significant (P > 0.05) change because of the large variance of the data (Fig. 2B). LO/HI remained unchanged from 30-100% Tvent but increased significantly (P < 0.05) at 110% Tvent (Fig. 2C). The same procedures were applied to the data set obtained by GSA (Fig. 2, D-F). The tendencies observed in HI and LO were similar to those when CGSA was used, i.e., significantly (P < 0.05) greater HI was observed at 30% Tvent without any significant (P > 0.05) changes in LO. However, in LO/HI, there were no significant (P > 0.05) effects of exercise intensity. Furthermore, significant (P < 0.05) interindividual differences were ob-
Downloaded from www.physiology.org/journal/jappl by ${individualUser.givenNames} ${individualUser.surname} (128.151.010.035) on July 29, 2018. Copyright © 1991 American Physiological Society. All rights reserved.
HEART RATE CONTROL DURING EXERCISE BY SPECTRAL ANALYSIS
7
12
17
Time (min)
Frequency
1139
FIG. 1. Actual heart rate variability (HRV) recordings (A) and spectra (B) calculated by CGSA of subject 6. HRV spectra were recorded as “amplitude spectra” with units of ms/Hz”-5 , i.e., square root of power spectral density function. Tvent,ventilatory threshold.
(Hz)
served, indicating that the patterns of changes in this
(HI) decreased progressively from rest to a WR equiva-
parameter varied very much from person to person. This conclusion about LO/HI with GSA was not altered by
lent to 60% Tvent, and the indicator of SNS activity (LO/ HI) increased only when exercise intensity exceeded Tvent. The suggestion that HI might reflect the PNS activity is in agreement with the findings of many other investigators who have studied exercise (3) or other manipula-
including the resting and 20-W exercise data for six sub-
jects.
Cons such as pharmacological intervention
DISCUSSION
The primary
findings
of the present study of HRV
during exercise were that the indicator of PNS activity
(1, 19-21) or
postural challenge (19,28). The LO/HI has been taken as an indicator of SNS activity (19, 20). Our finding of a significant increase in LO/HI is in contrast to the results
TABLE 3. Parameters of HRV spectral analyses with coarse-graining procedure Work Rest 20 w 30% 60% 90% 100% 110%
Rate
Lnt Tvent Tvent Tvent Tvent
SD, ms
HI, ms2
LO, ms2
LO/HI
90.67*10.93*t 49.50+8.56”r 30.75*3.27-t
4,230,16+598.71* &x.14+380.78 387.05k127.51
2,285.73t901.03* 467.09t216.56 322.85k217.11 31.99*10.40 8.72k2.13
0,58_+0.26
16.00+1.80
12.25&1.60 9.63kl.53 9.OOkO.94
125.4Eb66.52
21.50t8.48 4.69kl.15 l-06&0.37
1.68kO.59 1.58kO.70 0.851tO.49 1.2OkO.55
8.4Ok3.36
1.77t0.46
5.57k2.13
6.19*1.70*
Values are means -t SE; n = 6 for rest and 20 W, n = 8 for other work rates. SD, standard deviation 0.15 Hz) of HRV spectra; LO, low-frequency area (0.15 > f > 0.0 Hz) of HRV spectra. * P -c 0.05 from
of HRV; HI, high-frequency area (1.0 2 f 2 others; t P c 0.05 from 60-110% Tvent trial.
Downloaded from www.physiology.org/journal/jappl by ${individualUser.givenNames} ${individualUser.surname} (128.151.010.035) on July 29, 2018. Copyright © 1991 American Physiological Society. All rights reserved.
1140
HEART
RATE CONTROL
DURING
EXERCISE
BY SPECTRAL
ANALYSIS
D PcO.05 PcO.06
4
t\
i 30
4 80
80
100
4 110
1
110
30
60
80
100
E FIG. 2. Changes in areas under HI and LO regions of power spectral density function and LO/HI in relation to relative exercise intensity. HI (parasympathetic nervous system indicator), LO, and LO/HI (sympathetic nervous system indicator) in A, B, and C, respectively, were calculated by CGSA technique and in D, E, and F, respectively, by general spectral analysis technique.
2400*.
18004~
. 80
30
60
80
100
110
F 17.5
14.0
10.6
7.0
3.6
0.00
o.oG
1 I 30
EXERCISE
1 00
80
INTENSITY
100
110
(%T,,,,)
44 I 30
60
EXERCISE
of Arai et al. (3), who had anticipated such an increase but did not find it. We believe that part of the explanation for this difference in results is a consequence of the CGSA approach to deriving the spectral indicators (30). Methodological considerations. The rationale for using HI and LO/HI as indicators of PNS and SNS activity, respectively, comes largely from studies in which autonomic control of heart rate was manipulated by pharmacological blockade (1, 2, 19, 20) or pathological cardiac vagal nerve damage (6,27,31). More recently, Berger et al. (5) have shown that the sinoatrial node of the dog behaves as a low-pass filter in response to broad-band (i.e., low-to-high frequency) stimulation of the cardiac sympathetic and parasympathetic nerves. Heart rate followed changes in SNS activity only to frequencies ~0.1 Hz. In contrast, heart rate was able to follow PNS activity not only at low frequencies but also at frequencies ~0.1 Hz. These differences in the transfer characteristics to direct neural stimulation were used to differentiate the changes in PNS and SNS activities. However, before this is applied to the study of HRV response to exercise, some methodological problems should be investigated. The magnitude of the respiratory sinus arrhythmia
80
INTENSITY
4 110
100
(%T,,,,)
(RSA) has been taken as an indicator of PNS activity (13). However, unlike spectral analysis, all components in HRV are lumped together in a single parameter, such as the peak-to-peak variation and the SD. It is apparent from Table 3 that, although the SD decreased to 10% of its resting value at the highest work rate studied, HI had decreased to only 0.025% of its resting value. Clearly there are quantitative differences between the SD and HI as indicators of PNS activity. This might be partially explained by the extraction of specific components of variability with HI compared with lumping all components together with the SD. Hirsch and Bishop (11) have suggested that RSA should be normalized for changes in breathing frequency and tidal volume. Their data (Table 5 of Ref. 11) were used to correct HI calculated with the CGSA method (Fig. 3). There was no significant effect of the correction (P > 0.05), probably because of the offsetting effects of increased frequency and tidal volume that occur in combination during exercise. Therefore it does not seem to be important to apply this correction to HRV spectra obtained in exercise. The CGSA approach to spectral analysis was devel-
Downloaded from www.physiology.org/journal/jappl by ${individualUser.givenNames} ${individualUser.surname} (128.151.010.035) on July 29, 2018. Copyright © 1991 American Physiological Society. All rights reserved.
HEART
RATE CONTROL
DURING
EXERCISE
BY SPECTRAL
ANALYSIS
1141
RSA mediated via the PNS (1,2, 18,20,21). The contribution of RSA to HRV is evident in Fig. lB, where the M uncorrected 4000 peak of the HI component tracks to higher frequencies as m-----a corrected the breathing frequency increases with exercise intensity. The HRV at frequencies
YAMAMOTO,
RICHARD
L. HUGHSON,
AND JOHN
C. PETERSON
of Kinesiology, University of Waterloo, Waterloo, Ontario NZL 3Gl, Canada
YAMAMOTO,YOSHIHARU,RICHARDL. HUGHSON,ANDJOHN C. PETERSON. Autonomic controZ of heart rate during exercise studied by heart rate variability spectral analysis. J. Appl. Physiol. 71(3): 1136-1142, 1991.-Spectral analysis of heart rate variability (HRV) might provide an index of relative sympathetic (SNS) and parasympathetic nervous system (PNS) activity during exercise. Eight subjects completed six 17-min submaximal exercise tests and one resting measurement in the upright sitting position. During submaximal tests, work rate (WR) was increased for the initial 3 min in a ramp fashion until it reached constant WRs of 20 W, or 30,60,90,100, and 110% of the predetermined ventilatory threshold (Tvent). Ventilatory profile and alveolar gas exchange were monitored breath by breath, and beat-to-beat HRV was measured as R-R intervals of an electrocardiogram. Spectral analysis was applied to the HRV from 7 to 17 min. Low-frequency (O-O.15 Hz) and highfrequency (0.15-1.0 Hz) areas under power spectra (LO and HI, respectively) were calculated. The indicator of PNS activity (HI) decreased dramatically (P < 0.05) when the subjects exercised compared with rest and continued to decrease until the intensity reached 60% Tvent. The indicator of SNS activity (LO/ HI) remained unchanged up to 100% Tvent, whereas it increased abruptly (P < 0.05) at 110% Tvent. The results suggested that (cardiac) PNS activity decreased progressively from rest to a WR equivalent to 60% Tvent, and SNS activity increased only when exercise intensity exceeded Tvent.
Exercise heart rate has recently been studied by spectral analysis of HRV. Arai et al. (3) attempted to compare the results of spectral analyses with data published by several investigators who had employed pharmacological blockade of either or both the PNS and SNS (8, 22). Although the data of Arai et al. clearly showed a reduction in PNS with increasing intensities of exercise, the data failed to show evidence of increased SNS. This failure is difficult to rationalize given the extensive evidence supporting this from studies with P-adrenergic receptor blockade (8, 22) and measurement of plasma catecholamines (9, 10, 17). It was the purpose of this study to examine in more detail the heart rate response to rest and various levels of submaximal exercise. We have done this, first, by studying longer periods of constant load exercise and, second, by applying a new approach to the study of HRV, which we have labeled coarse-graining spectral analysis (30). This method provides improved resolution of short-term HRV spectra and will permit evaluation of the LO/HI ratio as an indicator of SNS activity during increasing intensities of exercise.
autonomic nervous system; sympathetic nervous system; parasympathetic nervous system; 1 lf component; fractal component
Eight healthy individuals (5 males and 3 females) volunteered for the experiments. Their physical characteristics are presented in Table 1. Each subject signed a consent form approved by the Office of Human Research of the University after the test protocol was described. Ail subjects completed preliminary testing in which peak 0, uptake (VO 2peak)and ventilatory threshold (T,,t) were determined (Table 1). From a baseline of 25 W in the upright position on an electrically braked cycle ergometer (Siemens), work rate (WR) increased as a continuous ramp at a rate of 15 W/min. Exhaustion was taken as the point at which subjects could no longer maintain 150 rpm. The v02peak was defined as the highest 0, recorded over a 10-s period during breath-bybreath measurement, as described below. Tvent was taken as the point at which 1) minute ventilation (Tj,) began to increase in a breaking fashion against Vo2, 2) CO, output (VCO~) began to increase in a breaking fashion against VO,, 3) vE/v02 exhibited a systematic increase without a concomitant increase in vE/kO,, and 4) end-tidal 0, pressure exhibited a systematic increase without a concomitant increase in end-tidal CO, pressure (29). At least two researchers determined Tvent on the basis of satisfaction of more than one criterion. If
HEART RATE is controlled
by the balance between parasympathetic (PNS) and sympathetic nervous system (SNS) activity to the sinoatrial node. On a beat-to-beat basis, heart rate is not constant. Rather, there are periodic fluctuations that are indicative of the relative contributions of each of these components of the autonomic nervous system (25). Spectral analysis of heart rate variability (HRV) has shown at least two distinct regions of periodicity in heart rate. In combination with studies of various pharmacological or physiological manipulations, it has been demonstrated that high frequencies (HI, >0.15 Hz) of HRV are associated solely with cardiac PNS activity (1, 2, 5, 19-21, 25, 28). Lower-frequency (LO, co.15 Hz) HRV might be associated with both PNS and SNS activity (1, 2, 5, 25). In addition, the ratio LO/HI might selectively indicate SNS activity (19, 20). Thus studies of HRV could provide noninvasive evaluation of autonomic neural balance in a number of physiological settings. 1136
0161-7567191
$1.50
METHODS
Copyright 0 1991 the American Physiological Society
Downloaded from www.physiology.org/journal/jappl by ${individualUser.givenNames} ${individualUser.surname} (128.151.010.035) on July 29, 2018. Copyright © 1991 American Physiological Society. All rights reserved.
HEART
TABLE
CONTROL
DURING
EXERCISE
1. Physical characteristics of the subjects
Subj No.
Sex
Age, yr
Ht, cm
I 2 3 4 5 6 7 8
M M M M M F F F
22 21 22 28 19 21 22 19
174 176 180 174 173 165 169 160
22 +3
171 -t5
Mean +SD WR, uptake.
RATE
work
rate;
WR at Tent, W
Tent, ml/min
70 73 95 69 70 57 55 60
190 198 255 208 161 140 115 103
2,370 2,490 3,660 2,780 2,120 1,750 1,540 1,280
321 341 373 324 296 253 207 208
3,850 4,180 4,970 4,250 3,640 2,850 2,280 2,450
69 +12
172 +48
2,250 *710
290 &58
3,560 +890
Wt, kg
Tvent , ventilatory
V02 at
threshold;
Max
WR, W
VO, W,
vozmax,
ml/min
maximal
0,
the difference among determined WRs at Tvent was >5.0 W among investigators, ramp work tests were repeated until it fell within 5.0-W range. Submaximal exercise tests. The subjects underwent in random order six 17-min submaximal exercise tests and one resting measurement in the upright sitting position. During submaximal tests, WR was increased for the initial 3 min in a ramp fashion until it reached the set loads of 20 W or WRs corresponding to 30, 60, 90, 100, and 110% of each subject’s Tvent. The set loads were then maintained for 14 min. Except for the resting measurement and the 20-W exercise, only one test was conducted on 1 day. Subjects 1 and 3 did not complete the resting measurement and the 20-W trial. During the tests, ventilatory profile and alveolar gas exchange were monitored breath by breath and heart rate was measured beat to beat continuously for the entire 17 min. To observe steady-state responses, data during the initial 7 min were discarded, and the remaining lo-min bins (7-17 min) were used for analyses. Data collection. Breath-by-breath measurements of ventilation and alveolar gas exchange were obtained on a computerized system (First Breath) incorporating a respiratory mass spectrometer (MGA-llOOA, PerkinElmer) and a volume turbine (VMM-110, Alpha Technologies) (12). Allowance was made for breath-by-breath changes in nominal functional residual capacity and endtidal fractional gas concentrations to calculate alveolar VO, and VCO, (4). Calibration of the mass spectrometer was checked before each experiment with standard calibration gases. Volume factors were checked with a syringe of known volume; care was taken to use a mean flow rate similar to that of the subject during the exercise test. Temperature and water vapor corrections were made on inspired and expired air for conditions measured near the volume turbine. The surface electrocardiogram (ECG) was measured continuously during the tests from standard bipolar leads with an electrocardiograph (model 7830A, HewlettPackard). Analog output of the ECG meter was differentiated, and the resultant QRS spikes triggered a Schmitt circuit to generate a train of rectangular impulses. The impulse train was processed on a real-time basis with a personal computer via a 12-bit analog-to-digital converter (DAS-16, Metrabyte) at a sampling frequency of
BY
SPECTRAL
ANALYSIS
1137
1,000 Hz. The intervals of the impulses were stored sequentially on a diskette for later analyses. Spectral anaZysis.Before HRV spectra were calculated, lo-min HRV data were searched for the possibility of extra or missing beats that could affect the result of spectral analyses (26). All R-R intervals for which absolute values of beat-to-beat transition in heart rate were >20 beats/min were subjected to filtering. The filtering process either deleted or inserted beats to obtain appropriate intervals for those that had transitions greater than +20 or -20 beats/min, respectively. Thereafter, the unequal intervals of HRV were aligned sequentially to obtain equally spaced samples [interval tachogram (23)]. The resultant data were first divided into three parts, and the linear trend of each part was eliminated by a linear regression method. After passing through the data window, a fast Fourier transform was used to calculate a periodgram of each part. The ensemble average of the periodgrams was subjected to a Hanning-type spectral window and was regarded as an estimate of HRV spectra. Because the HRV spectra calculated in this way had a unit of cycles/beat, we converted it to Hz by dividing by the average R-R interval. We calculated HRV spectra using R-R interval data rather than heart rate; therefore the unit of spectral density in the present study was ms2/Hz. Using almost the same technique, some researchers (14, 26) showed that there was a nonharmonic “fractal” component in HRV data in the frequency band from 0.00003 to 0.1 Hz (a period of -10 h to 10 s). Fourier transforms of such components are known to result in a broad band but nonwhite, or so-called l/f, spectrum (16), in which the power spectral densities are inversely proportional to their frequency. It could be anticipated that the existence of this nonwhite biased noise could affect HI and LO calculations from short-term HRV spectra. Yamamoto and Hughson (30) have described a new approach called coarse-graining spectral analysis (CGSA) for calculating HRV spectra with the elimination of this noise component. In the present study, CGSA was applied to HRV data during exercise, and the results were compared with those derived from the general spectral analysis (GSA) described earlier. The theoretical background and the detailed procedure of CGSA were reported by Yamamoto and Hughson (30). A brief description of the method follows. First, each R-R interval in raw HRV data was repeated so that the length of the data set was twice as long as an original time series (x) used in GSA. Then the “coarse-grained” data set x’ was constructed by taking the first half of the data set before it was resealed by an appropriate scale factor. The objective of CGSA was to eliminate nonharmanic l/f components from HRV autopower spectrum calculated for x by GSA (S,), leaving harmonic components of interest intact. Because of the scale invariant, or self-similar property of the l/f noise, it was shown that the cross-power spectra between x and X’ (SA) preserved l/f components, but SA could not retain any harmonic components. Therefore we subtracted SA from S, to obtain the new spectra, in which only the harmonic components were emphasized. From HRV spectra obtained both from GSA and
Downloaded from www.physiology.org/journal/jappl by ${individualUser.givenNames} ${individualUser.surname} (128.151.010.035) on July 29, 2018. Copyright © 1991 American Physiological Society. All rights reserved.
1138
HEART
Rest
l/min Mean Range f, min-’ Mean Range VT, liters Mean Range VO,, l/min Mean Range VC02, llmin Mean Range WR, W Mean Range HR, beats/min Mean Range
RATE
CONTROL
20 w
DURING
EXERCISE
30% xmt
60%
BY SPECTRAL
ANALYSIS
Tvellt
Lit
90%
100%
Tvent
110%
Tvent
7jE,
60.9
8.1 6.6-9.4
17.4 14.0-21.1
24.4 13.4-35.0
33.7 23.5-47.7
46.0 32.4-61.4
52.7 33.3-66.7
36.2-80.7
12.6 10.3-17.0
16.5 11.5-22.1
18.4 13.6-26.6
19.2 12.9-29.4
22.8 15.8-32.3
24.3 17.8-35.7
27.2 20.1-36.3
0.66 0.46-0.82
1.13 0.75-1.84
1.40 0.78-2.12
1.92 1.12-2.96
2.19 1.17-3.51
2.33 1.27-3.70
2.34 1.30-3.71
0.30 0.23-0.37
0.63 0.50-0.74
1.01 0.55-1.70
1.55 1.11-2.54
2.12 1.48-3.35
2.36 1.57-3.63
2.58 1.63-4.05
0.27 0.20-0.32
0.57 0.44-0.62
0.90 0.47-1.49
1.43 1.00-2.22
2.00 1.39-3.16
2.27 1.49-3.47
1.55-3.96
51 31-77
103 62-152
154 93-203
171 103-255
188 113-281
95
114 107-136
135 126-149
145 134-162
155 141-167
20
64 61-66
82 72-92
86-103
2.49
Values are means and ranges for responses obtained between 7 and 17 min of exercise. Tvent, ventilatory threshold; i7E, minute ventilation; f, breathing frequency; VO,, O2 uptake; VCO~, CO, output; WR, work rate; HR, heart rate. Note that the minimum f of 10.3 min-’ or 0.17 Hz was greater than the border frequency used to calculate high- and low-frequency area in heart rate variability spectra (i.e., 0.15 Hz).
CGSA, LO and HI were calculated by integrating spectrum densities for the range of 0.0 < f < 0.15 Hz and 0.15 < f < 1.0 Hz, respectively. Statistical analysis. A Student-Newman-Keuls post hoc test was applied for each parameter when a one-way analysis of variance indicated a significant effect. It would be shown later that the mean values in HI and LO differed very much between rest or light exercise data and moderate-to-high exercise data. The complete interindividual comparisons for all eight subjects were possible for 30-110% Tvent data sets. A two-way analysis of variance followed by the post hoc test was conducted on these data. RESULTS
Steady-state responses of respiratory variables and heart rate obtained between 7 and 17 min increased progressively with WR (Table 2). The mean heart rate at 110% Tvent trials was 155 beats/min, which was only slightly lower than the peak heart rate in the study of Arai et al. (3). It is of note that the minimum breathing frequency of 10.3 breaths/min or 0.17 Hz was higher than the border frequency used to calculate HI and LO (i.e., 0.15 Hz). It was important that breathing frequency was not lower than this border frequency; otherwise the interpretation of calculated LO and HI would become quite complicated because of effects of respiratory sinus ‘arrhythmia
(6, 25, 28). The actual HRV recordings and the spectra calculated by CGSA of subject 6 are shown in Fig. 1. Compared with the recording at rest, HRV decreased considerably during exercise, even at 20 W. The spiky pattern in HRV was still observable at 20 W and 30% Tvent. However, during exercise at intensities higher than 60% Tvent, heart rate
looked almost constant. Reduced HRV at higher exercise intensities could also be observed in the HRV spectra. It should be emphasized that, even in these HRV spectra with low power, there were low- and high-frequency peaks, and the latter occurred at the mean breathing frequencies. Application of the CGSA technique showed that both HI and LO components decreased dramatically (P < 0.05) when the subjects exercised (Table 3) and tended to decrease until the intensity reached 60% Tvent (see Fig. 3). Similar phenomena were also observed in standard deviations (SD) of R-R intervals; i.e., exercise per se caused a significant reduction in HRV until 60% Tvent. In contrast, LO/HI remained unchanged up to a 100% Tvent level (Table 3). During exercise at 110% Tvent, LO/HI increased abruptly (P < 0.05). The PNS indicator (HI) dropped dramatically with exercise per se and continued to decrease up to 60% Tvent, whereas the SNS indicator (LO/HI) increased only when exercise intensity exceeded Tvent. Further analyses were conducted using a two-factor model (exercise intensity X subject) for 30-110% Tvent data sets. HI of 30% Tvent was significantly (P < 0.05) greater than those of the other trials (Fig. 2A), although LO did not show any significant (P > 0.05) change because of the large variance of the data (Fig. 2B). LO/HI remained unchanged from 30-100% Tvent but increased significantly (P < 0.05) at 110% Tvent (Fig. 2C). The same procedures were applied to the data set obtained by GSA (Fig. 2, D-F). The tendencies observed in HI and LO were similar to those when CGSA was used, i.e., significantly (P < 0.05) greater HI was observed at 30% Tvent without any significant (P > 0.05) changes in LO. However, in LO/HI, there were no significant (P > 0.05) effects of exercise intensity. Furthermore, significant (P < 0.05) interindividual differences were ob-
Downloaded from www.physiology.org/journal/jappl by ${individualUser.givenNames} ${individualUser.surname} (128.151.010.035) on July 29, 2018. Copyright © 1991 American Physiological Society. All rights reserved.
HEART RATE CONTROL DURING EXERCISE BY SPECTRAL ANALYSIS
7
12
17
Time (min)
Frequency
1139
FIG. 1. Actual heart rate variability (HRV) recordings (A) and spectra (B) calculated by CGSA of subject 6. HRV spectra were recorded as “amplitude spectra” with units of ms/Hz”-5 , i.e., square root of power spectral density function. Tvent,ventilatory threshold.
(Hz)
served, indicating that the patterns of changes in this
(HI) decreased progressively from rest to a WR equiva-
parameter varied very much from person to person. This conclusion about LO/HI with GSA was not altered by
lent to 60% Tvent, and the indicator of SNS activity (LO/ HI) increased only when exercise intensity exceeded Tvent. The suggestion that HI might reflect the PNS activity is in agreement with the findings of many other investigators who have studied exercise (3) or other manipula-
including the resting and 20-W exercise data for six sub-
jects.
Cons such as pharmacological intervention
DISCUSSION
The primary
findings
of the present study of HRV
during exercise were that the indicator of PNS activity
(1, 19-21) or
postural challenge (19,28). The LO/HI has been taken as an indicator of SNS activity (19, 20). Our finding of a significant increase in LO/HI is in contrast to the results
TABLE 3. Parameters of HRV spectral analyses with coarse-graining procedure Work Rest 20 w 30% 60% 90% 100% 110%
Rate
Lnt Tvent Tvent Tvent Tvent
SD, ms
HI, ms2
LO, ms2
LO/HI
90.67*10.93*t 49.50+8.56”r 30.75*3.27-t
4,230,16+598.71* &x.14+380.78 387.05k127.51
2,285.73t901.03* 467.09t216.56 322.85k217.11 31.99*10.40 8.72k2.13
0,58_+0.26
16.00+1.80
12.25&1.60 9.63kl.53 9.OOkO.94
125.4Eb66.52
21.50t8.48 4.69kl.15 l-06&0.37
1.68kO.59 1.58kO.70 0.851tO.49 1.2OkO.55
8.4Ok3.36
1.77t0.46
5.57k2.13
6.19*1.70*
Values are means -t SE; n = 6 for rest and 20 W, n = 8 for other work rates. SD, standard deviation 0.15 Hz) of HRV spectra; LO, low-frequency area (0.15 > f > 0.0 Hz) of HRV spectra. * P -c 0.05 from
of HRV; HI, high-frequency area (1.0 2 f 2 others; t P c 0.05 from 60-110% Tvent trial.
Downloaded from www.physiology.org/journal/jappl by ${individualUser.givenNames} ${individualUser.surname} (128.151.010.035) on July 29, 2018. Copyright © 1991 American Physiological Society. All rights reserved.
1140
HEART
RATE CONTROL
DURING
EXERCISE
BY SPECTRAL
ANALYSIS
D PcO.05 PcO.06
4
t\
i 30
4 80
80
100
4 110
1
110
30
60
80
100
E FIG. 2. Changes in areas under HI and LO regions of power spectral density function and LO/HI in relation to relative exercise intensity. HI (parasympathetic nervous system indicator), LO, and LO/HI (sympathetic nervous system indicator) in A, B, and C, respectively, were calculated by CGSA technique and in D, E, and F, respectively, by general spectral analysis technique.
2400*.
18004~
. 80
30
60
80
100
110
F 17.5
14.0
10.6
7.0
3.6
0.00
o.oG
1 I 30
EXERCISE
1 00
80
INTENSITY
100
110
(%T,,,,)
44 I 30
60
EXERCISE
of Arai et al. (3), who had anticipated such an increase but did not find it. We believe that part of the explanation for this difference in results is a consequence of the CGSA approach to deriving the spectral indicators (30). Methodological considerations. The rationale for using HI and LO/HI as indicators of PNS and SNS activity, respectively, comes largely from studies in which autonomic control of heart rate was manipulated by pharmacological blockade (1, 2, 19, 20) or pathological cardiac vagal nerve damage (6,27,31). More recently, Berger et al. (5) have shown that the sinoatrial node of the dog behaves as a low-pass filter in response to broad-band (i.e., low-to-high frequency) stimulation of the cardiac sympathetic and parasympathetic nerves. Heart rate followed changes in SNS activity only to frequencies ~0.1 Hz. In contrast, heart rate was able to follow PNS activity not only at low frequencies but also at frequencies ~0.1 Hz. These differences in the transfer characteristics to direct neural stimulation were used to differentiate the changes in PNS and SNS activities. However, before this is applied to the study of HRV response to exercise, some methodological problems should be investigated. The magnitude of the respiratory sinus arrhythmia
80
INTENSITY
4 110
100
(%T,,,,)
(RSA) has been taken as an indicator of PNS activity (13). However, unlike spectral analysis, all components in HRV are lumped together in a single parameter, such as the peak-to-peak variation and the SD. It is apparent from Table 3 that, although the SD decreased to 10% of its resting value at the highest work rate studied, HI had decreased to only 0.025% of its resting value. Clearly there are quantitative differences between the SD and HI as indicators of PNS activity. This might be partially explained by the extraction of specific components of variability with HI compared with lumping all components together with the SD. Hirsch and Bishop (11) have suggested that RSA should be normalized for changes in breathing frequency and tidal volume. Their data (Table 5 of Ref. 11) were used to correct HI calculated with the CGSA method (Fig. 3). There was no significant effect of the correction (P > 0.05), probably because of the offsetting effects of increased frequency and tidal volume that occur in combination during exercise. Therefore it does not seem to be important to apply this correction to HRV spectra obtained in exercise. The CGSA approach to spectral analysis was devel-
Downloaded from www.physiology.org/journal/jappl by ${individualUser.givenNames} ${individualUser.surname} (128.151.010.035) on July 29, 2018. Copyright © 1991 American Physiological Society. All rights reserved.
HEART
RATE CONTROL
DURING
EXERCISE
BY SPECTRAL
ANALYSIS
1141
RSA mediated via the PNS (1,2, 18,20,21). The contribution of RSA to HRV is evident in Fig. lB, where the M uncorrected 4000 peak of the HI component tracks to higher frequencies as m-----a corrected the breathing frequency increases with exercise intensity. The HRV at frequencies
