Abstract
Concomitant sympathetic and vagal activation can occur in various physiological conditions, but there is limited information on heart rate (HR) behavior during the accentuated sympathovagal antagonism. Beattobeat HR and blood pressure were recorded during intravenous infusion of incremental doses of norepinephrine in 18 healthy male volunteers (mean age 23 ± 5 yr). HR and blood pressure spectra and twodimensional Poincaré plots were generated from the baseline recordings and from the recordings at different doses of norepinephrine. The mean blood pressure increased (from 90 ± 7 to 120 ± 9 mmHg,P < 0.001), HR decreased (from 60 ± 9 to 48 ± 7 beats/min, P < 0.001), and the highfrequency spectral component of HR variability increased (P < 0.001) during the norepinephrine infusion as evidence of accentuated sympathovagal interaction. Abrupt aperiodic changes in sinus intervals that were not related to respiratory cycles or changes in blood pressure occurred in 14 of 18 subjects during the norepinephrine infusions. These fluctuations in sinus intervals resulted in a complex or parabolashaped structure of the Poincaré plots of successive RR intervals and a widening of the highfrequency spectral peak. In four subjects, the abrupt fluctuations in sinus intervals were followed by a sudden onset of fixed RR interval dynamics with a loss of respiratory modulation of HR, resulting in a torpedoshaped structure of the Poincaré plots. These data show that HR behavior becomes remarkably unstable during accentuated sympathovagal interaction, resembling stochastic dynamics or deterministic chaotic behavior. These features of HR dynamics can be better identified by dynamic analysis of beattobeat behavior of RR intervals than by traditional analysis techniques of HR variability.
 heart rate variability
 cardiovascular regulation
power spectral analysis of heart rate (HR) variability is a commonly used method in the measurement of sympathovagal interaction on sinus node (2, 24). Because traditional analysis techniques are insensitive to abrupt, aperiodic changes in HR dynamics, beattobeat analysis techniques and other dynamic methods have been developed to uncover stochastic or nonlinear features in HR behavior (9, 14, 20, 25).
Reciprocal changes in sympathetic and vagal activity have been observed in some physiological conditions, such as during passive tilt, which can be detected by typical changes in spectral components of HR variability (22, 26). In other physiological and pathological states, concomitant sympathetic and vagal activation can occur, leading to accentuated sympathovagal antagonism (6, 8, 16, 30). Acetylcholine and norepinephrine have a complex interaction at the level of the sinus node, resulting in a typical condition favoring the occurrence of complex HR dynamics (17, 18). The present research was designed to study the HR behavior during accentuated sympathovagal interaction by generating power spectra and Poincaré plots of successive RR intervals at the baseline and during infusion of incremental doses of norepinephrine in young healthy males.
METHODS
Subjects and study protocol.
HR dynamics were studied in 18 healthy male volunteers (mean age 23 ± 5 yr) at rest in the supine position under quiet baseline conditions and during incremental doses of intravenous norepinephrine. Subjects with atrial or ventricular ectopic beats or those with episodes of nodal rhythm during the experiment were excluded. The design was approved by the ethics committee of the institution, and all subjects gave their informed consent. All the tests were performed between 10:00 AM and 4:00 PM, and vigorous exercise, alcohol intake, or smoking were forbidden for 48 h before the testing days. The subjects lay in a supine position in a quiet room for 30 min before the data collection and became accustomed to breathing at a constant metronomeguided rate of 0.25 Hz for the duration of the experiment. The beattobeat RR intervals were recorded with a wireless HR monitoring system having a sampling frequency of 1,000 Hz (Polar Electro, Kempele, Finland) (27). A continuous surface electrocardiogram was also recorded during the experiment to confirm the sinus origin of the beats. Beattobeat arterial blood pressure was measured by the Finapres fingercuff method, the respiration was measured with a disposable screentype flow transducer, and the data were stored by means of menudriven software packages (1, 29). HR, blood pressure, and respiration signals were fed into an analogtodigital converter and stored in a microcomputer for further analysis. The protocol included baseline recordings for 15 min and, during infusion of norepinephrine at constant rates of 50, 100, and 150 ng ⋅ kg^{−1} ⋅ min^{−1}, recordings of 15 min at each concentration. The doses of norepinephrine were based on previous studies (3) that have shown these doses to result in plasma concentrations of norepinephrine observed under various physiological conditions. If the blood pressure increased >180/110 mmHg or the subjects complained of any uncomfortable symptoms during the norepinephrine infusion, the infusion was stopped (6 subjects, see Table 1).
Analysis of RR interval dynamics.
Data analyses were performed as described in detail previously (13,29). An autoregressive model was used to estimate the power spectrum densities of HR and blood pressure variabilities (see ). The computer program automatically calculates the autoregressive coefficients to define the power spectrum density. The power spectra were quantified by measuring the area under two frequency bands: lowfrequency power from 0.04 to 0.15 Hz, and highfrequency power from 0.15 to 0.4 Hz.
Twodimensional return maps or Poincaré plots were generated by plotting each RR interval as a function of its previous RR interval and each systolic blood pressure value as a function of its previous systolic blood pressure value, respectively, obtained at the baseline and with different levels of norepinephrine infusion. Twodimensional vector analysis was used to quantify the shape of the plots as described previously (13, 29). In this quantitative method, shortterm (SD1) and longterm RR interval variability (SD2) and the ellipse area of the plot are separately quantified. The shapes of Poincaréplots were classified as 1) a normal, cometshaped plot, in which increasing beattobeat RR interval dispersion is observed with increasing RR intervals (SD1/SD2 >0.15); 2) a torpedoshaped plot with small overall beattobeat dispersion (SD1) and without increasing dispersion at longer RR intervals (SD1/SD2 <0.15); or3) a complex or parabolalike plot, in which two or more distinctive limbs are separated from the main body of the plot, with at least three points included in each limb.
Statistical methods.
Analysis of variance for repeated measurements was used to compare the changes in HR, blood pressure, and normally distributed HR variability measures during the norepinephrine infusion. Normal Gaussian distribution of the data was verified by the KolmogorovSmirnov goodnessoffit test. Whenever the data were not normally distributed (z value > 1.0 for all spectral components of HR variability), Friedman’s randomized block analysis of variance followed by post hoc analysis (Wilcoxon test) was used. Differences in baseline data between the subjects with different RR interval dynamics during the norepinephrine infusion were analyzed by a MannWhitney U test.
RESULTS
The mean blood pressure increased and mean HR decreased progressively with increasing doses of norepinephrine (Table 1). The highfrequency spectral component of HR variability increased during the initial dose of 50 ng ⋅ kg^{−1} ⋅ min^{−1}, but no additional increase occurred with higher doses. No significant changes were observed in the lowfrequency component of HR variability during norepinephrine infusion.
Recordings of electrocardiogram, blood pressure, and respiration and corresponding RR interval tachograms for one subject with typical HR dynamics at the baseline and with incremental doses of norepinephrine are presented in Fig. 1. Respiratory modulation of RR intervals and blood pressure were observed under the baseline conditions with lengthening of RR intervals and an increase in blood pressure during expiration and shortening of the RR intervals and a decrease of blood pressure during inspiration, resulting in a discrete highfrequency spectral peak at 0.25 Hz. During the small dose of norepinephrine (50 ng ⋅ kg^{−1} ⋅ min^{−1}), a pattern with abrupt lengthening of the RR intervals followed by gradual shortening to the baseline level (Fig.1 B) was observed without any concomitant abrupt changes in blood pressure. These sudden changes in HR occurred aperiodically and were not related to the frequency or depth of respiration. At a medium dose (100 ng ⋅ kg^{−1} ⋅ min^{−1}), abrupt shortenings of RR intervals were observed that again occurred aperiodically and were not related to the respiration cycles (Fig.1 C). At a high dose of norepinephrine (150 ng ⋅ kg^{−1} ⋅ min^{−1}), a periodic respiratory modulation of the RR intervals reappeared at a slower HR without any abrupt changes in RR intervals.
Abrupt changes in RR intervals resulted in a significant widening of the highfrequency spectral peak toward a very high frequency area (see Fig. 3). The twodimensional return maps showed a typical cometshaped plot for the baseline RR intervals (Fig.2 A). Abrupt episodes of HR slowing during norepinephrine infusion resulted in a complex or an inverted parabolalike shape for the Poincaréplot (Fig. 2 B). The aperiodic shortenings of the RR intervals typically resulted in a horseshoe or parabolalike structure of the Poincaré plot (Fig.2 C). A cometshaped or a circlelike Poincaré plot of RR intervals reappeared during the high dose (150 ng ⋅ kg^{−1} ⋅ min^{−1}) of norepinephrine. The shape of the Poincaré plots of the blood pressure remained similar during the different doses of norepinephrine infusion (Fig. 2).
A cometshaped Poincaré plot of RR intervals was observed in 14 of the 18 subjects at the baseline, whereas 4 subjects already had a parabolalike structure before the norepinephrine infusion. This latter group had slower mean HR (51 ± 9 vs. 63 ± 9 beats/min,P < 0.05) and higher highfrequency spectral component (4,892 ± 1,209 vs. 1,277 ± 838 ms^{2},P < 0.01) than subjects with a cometshaped plot at the baseline. During low or medium doses of norepinephrine infusion, a parabolalike plot was observed in 12 subjects, and in 2 subjects the parabolalike structure occurred during a high dose of norepinephrine. Only four subjects had a normal cometshaped plot at the baseline and during all phases of norepinephrine infusion. When the baseline HR and blood pressure data were compared between those subjects with a parabolalike plot and those with a cometshaped plot during the norepinephrine infusion, no significant differences were observed in any of these data (Table2), nor did changes in mean HR or blood pressure differ between these subjects during the norepinephrine infusion.
In four subjects, the abrupt changes in RR interval dynamics were followed by a sudden change into fixed RR interval dynamics, resulting in a torpedoshaped Poincaré plot (Fig.3). No respiratory modulation of RR intervals was observed during the fixed dynamics.
Reproducibility.
When HR and blood pressure dynamics were assessed twice under similar baseline conditions in four subjects with a parabolalike Poincaré plot at the baseline, all of them showed similar RR interval dynamics during the second recording. However, in one subject who underwent four recording sessions at 1wk intervals under similar external conditions, the parabolalike structure was not repeated during the last experiment, when completely different HR dynamics were observed (Fig. 4).
When norepinephrine was repeatedly infused in four subjects with occurrence of parabolalike structure during the first experiment, similar dynamics were repeated in two subjects, but in two cases the parabolalike structure did not reappear. In three cases without recurrence of nonlinear dynamics during the repeated test (one at the baseline and two during the norepinephrine infusion), the baseline measures of HR variability differed significantly (>20% difference for all measures) between the first and the repeated experiment (see e.g., Fig. 4), whereas in cases in which RR interval dynamics showed similar behavior during the repeated experiments, the baseline measures of HR variability were almost identical (<20% difference in all measures).
DISCUSSION
HR dynamics during accentuated sympathovagal interaction.
The observations of this study show that atypical, abrupt changes occur in HR dynamics during norepinephrine infusion in young healthy males. These findings suggest that the HR may become remarkably unstable during stressful situations that result in accentuated sympathovagal antagonism.
Unstable behavior of HR may be explained by complex interaction of acetylcholine and norepinephrine at the presynaptic and postsynaptic level of sinus node (17, 18). Norepinephrine infusion causes baroreceptormediated vagal activation, resulting in accentuated sympathovagal interaction, as evidenced here by an increase in blood pressure, a decrease in HR, and an increase in highfrequency spectral component of HR variability. Norepinephrine and acetylcholine have different temporal influences on the basic RR interval length; vagal effects on RR intervals occur more rapidly than sympathetic influences (17, 18), and the beattobeat fluctuations in RR intervals depend on summation and timing of the opposing effects of norepinephrine and acetylcholine on the sinus node. Abrupt changes in RR intervals are most likely a result of sudden vagal bursts or withdrawals, respectively, during high sympathetic influences on sinus node firing. The physiological background for onset of fixed RR interval dynamics and abrupt disappearance of respiratory modulation of HR may be explained by the saturation of the respiratory vagal modulation of sinus node during a very high tonic vagal activity (19, 21).
Present observations also show that incremental doses of norepinephrine infusion seldom result in a linear slowing of HR but that the HR behavior can be described as stochastic increases or decreases in RR intervals followed by return to control. There are also some features of deterministic chaos in HR dynamics during this experimental condition. A parabolalike or ringlike structure rather than a random distribution of the successive RR intervals was observed in the Poincaré plots during unstable HR behavior. This type of specific structure in the return maps of successive data points has been considered to provide evidence for deterministic chaos in the experimental animal models (5, 7, 10, 12, 15, 28). Also, the occurrence of fixed beattobeat RR interval dynamics after abrupt fluctuations on RR intervals represents another feature of deterministic chaos, i.e., abrupt temporal changes as cascades to fixed dynamics (7, 11,15). Finally, completely different RR interval dynamics were observed in the same subject in the similar external conditions when the baseline HR dynamics were different. Dependence of system dynamics on initial conditions is also one of the typical features of deterministic chaos (7, 15). The present analysis of data may not provide definite evidence of whether the observed HR dynamics can be better described as stochastic or as having characteristics of deterministic chaos, but it nevertheless emphasizes the need for analysis of HR behavior with dynamic methods in addition to methods based on moment statistics.
Analysis methods of HR dynamics during sympathovagal interaction.
Spectral analysis techniques have been most commonly used in assessment of the effects of sympathovagal balance on sinus node (2, 23, 24, 26). These analysis methods are based on the assumption that reciprocal changes occur in sympathetic and vagal activity under various physiological conditions (23, 24). Present findings demonstrate that traditional measures of HR variability are not specific for measurement of accentuated sympathovagal interaction. Abrupt changes in RR intervals resulted in a widening of the highfrequency spectral peak without consistent changes in any numerical measure of spectral components. These changes in HR dynamics could be accurately described not by quantitative twodimensional analysis of the Poincaréplots but only by visual inspection of the plots, suggesting that the visual interpretation of the shape of the Poincaré plot is more reliable than numerical methods in revealing the atypical HR behavior during accentuated autonomic interaction. These findings support the concept that beattobeat dynamic analysis methods may give important physiological information on HR behavior that cannot be detected by traditional methods of HR variability based on moment statistics. From a methodological point of view, it is also important to note that abrupt changes in sinus intervals can occur in various physiological and pathological states (4, 13, 30). These changes in RR intervals may become deleted as artifacts or ectopic beats in automatic and visual editing of RR interval tachograms and in analysis of HR variability by some geometric methods (22).
In conclusion, the results of this study show that unstable stochastic dynamics or deterministic chaos is involved in the genesis of HR variability during accentuated sympathovagal interaction. These features of HR behavior can be better observed by dynamic beattobeat analysis of RR intervals than by traditional nonspectral and spectral HR variability methods. Dynamic analysis methods may importantly increase our understanding of the physiological background for the complex behavior of HR in various conditions.
Acknowledgments
This work is supported by grants from the Foundation for Cardiovascular Research (Helsinki, Finland) and from the Instrumentarium Research Foundation (Helsinki, Finland).
Footnotes

Address for reprint requests: H. V. Huikuri, Div. of Cardiology, Dept. of Medicine, Oulu Univ. Central Hospital, Kajaanintie 50, 90220 Oulu, Finland.
 Copyright © 1998 the American Physiological Society
Appendix
Spectrum estimation with autoregressive modeling of time series.
The most popular of the timeseries modeling approaches to spectral estimation is the autoregressive (AR) spectral estimation (15a). Other names by which the AR spectral estimator is known are the maximum entropy spectral estimator and the linear prediction spectral estimator.
In AR(p) modeling of time series, it is assumed that a time series can be predicted with a linear combination of past p samples
Many methods have been developed for solving for the YuleWalker equations, e.g., autocorrelation, covariance, and modified covariance methods. We apply the Burg method, which estimates reflection coefficients first and then uses the Levinson recursion to obtain the AR parameter estimates. The reflection coefficients are estimated by minimizing the average of the estimates of the forward and backward prediction error powers. The Burg estimate is the only one of a large class of estimates that maintains the minimumphase property. A drawback is that line splitting in spectrum may occur if too large a model order p is used in the AR(p) model. It is therefore recommended that model order p should not exceed onethird of the data size. However,p should be at least twice the number of distinct frequency components in the time series.
After the data series is modeled, power spectral density can be computed straightforwardly from the model