## Abstract

Respiratory sinus arrhythmia (RSA) may be associated with improved efficiency of pulmonary gas exchange by matching ventilation to perfusion within each respiratory cycle. Respiration rate, tidal volume, minute ventilation (V˙e), exhaled carbon dioxide (V˙co
_{2}), oxygen consumption (V˙o
_{2}), and heart rate were measured in 10 healthy human volunteers during paced breathing to test the hypothesis that RSA contributes to pulmonary gas exchange efficiency. Cross-spectral analysis of heart rate and respiration was computed to calculate RSA and the coherence and phase between these variables. Pulmonary gas exchange efficiency was measured as the average ventilatory equivalent of CO_{2}(V˙e/V˙co
_{2}) and O_{2}(V˙e/V˙o
_{2}). Across subjects and paced breathing periods, RSA was significantly associated with CO_{2} (partial *r* = −0.53,*P* = 0.002) and O_{2} (partial*r* = −0.49, *P* = 0.005) exchange efficiency after controlling for the effects of age, respiration rate, tidal volume, and average heart rate. Phase between heart rate and respiration was significantly associated with CO_{2} exchange efficiency (partial *r* = 0.40, *P* = 0.03). These results are consistent with previous studies and further support the theory that RSA may improve the efficiency of pulmonary gas exchange.

- heart rate variability
- phase
- ventilatory equivalent

respiratory sinus arrhythmia (RSA), the increase and decrease in heart rate within each respiratory cycle, occurs mainly as a result of fluctuations of parasympathetic output to the heart, although sympathetic outflow also may influence variability (26). During inspiration, impulses originating in stretch receptors in the lungs travel via the vagi to inhibit the cardioinhibitory area in the medulla. The tonic vagal discharge that keeps the heart rate slow decreases, and the heart rate rises. The degree to which this modulation occurs is a function of the level of tonic vagal discharge and blood pressure (9,21). Thus measures of RSA are frequently used as an index of vagal tone (e.g., 2, 12, 19, 20).

Certain features of respiratory mechanics also affect the amplitude of RSA independent of changes in vagal tone. For example, changes in respiration rate and, to a lesser extent, tidal volume (V_{T}) can affect the magnitude of RSA in the absence of any change in tonic vagal activity (8, 11, 15, 23). This slowing of respiration rate and increase in V_{T} are believed to allow more time for the action of acetylcholine on muscarinic receptors at the sinoatrial node during exhalation (7, 25a). Other influences on the genesis and magnitude of RSA include feedback from arterial baroreceptors, a central rhythm generator in the brain stem, and intracardiac reflexes (3).

Although a great deal of research has been published concerning the physiological mechanisms mediating RSA, the question of what function, if any, RSA serves is rarely addressed. That is, does RSA itself serve a physiological role or is it merely an epiphenomenon of respiratory influences on neurocardiovascular control?

Hayano et al. (13) have proposed that RSA serves an active physiological role in improving pulmonary gas exchange efficiency by matching blood perfusion to air flow in the lung during each respiratory cycle. That is, during inspiration, increasing heart rate in combination with increased right ventricular output (due to decreased intrathoracic pressure) increase pulmonary perfusion in time with increasing lung air volume. In support of this theory, Hayano et al. (13) artificially induced RSA and inverse RSA (i.e., bradycardia during inhalation and tachycardia during exhalation) via electric stimulation of the vagus in dogs after surgical elimination of endogenous autonomic efferents. Breathing frequency, V_{T}, minute heart rate, cardiac output, and arterial blood pressure were not different between the conditions; nonetheless artificial RSA (i.e., RSA in phase with respiration) significantly increased pulmonary gas exchange efficiency, as measured by the ratio of physiological dead space to the V_{T} and the fraction of intrapulmonary shunt compared with no-RSA and inverse-RSA conditions. Also, in conscious dogs (27) and humans (22), RSA has been shown to reflexively increase in response to the experimental induction of hypercapnia when breathing frequency and V_{T} are controlled, suggesting that RSA may act as part of an adaptive response to hypercapnia to help restore blood gas homeostasis by increasing the efficiency of pulmonary gas exchange. However, these same investigators found that RSA decreases in response to progressive hypoxia in dogs (28, 29), suggesting that the active role of RSA may be limited to CO_{2} homeostasis.

To further test the theory that RSA is associated with pulmonary gas exchange efficiency, we measured O_{2} consumption (V˙o
_{2}) and CO_{2} production (V˙co
_{2}) as a function of ventilation in healthy humans during paced breathing. We chose the ventilatory equivalents of CO_{2}(V˙e/V˙co
_{2}) and O_{2} (V˙e/V˙o
_{2}) (where V˙e is minute ventilation) as proxies for pulmonary gas exchange efficiency, because these are generally accepted measures of this construct and can be measured noninvasively. We hypothesized that RSA and the phase between RSA and respiration would contribute to the efficiency of pulmonary gas exchange over and above the variance explained by other measured cardiopulmonary variables, including respiration rate, V_{T}, and mean heart rate.

## METHODS

#### Participants.

Ten healthy nonsmoking volunteers (6 female, 4 male), aged 23–78 yr (means ± SD = 44 ± 16) were studied. None had a history of cardiopulmonary disease. The study was approved by the Human Subjects Review Committee of the University of Washington.

#### Measures.

Respiratory flow, V˙o
_{2},V˙co
_{2}, and pressure of end-tidal CO_{2} (Pet
_{CO2}) were measured using a MedGraphics Cardiopulmonary Exercise System CPX/D (Medical Graphics; St. Paul, MN). Participants wore a nose clip and breathed through a mouthpiece attached by a plastic tube to a flow transducer and gas analyzer. Air temperature, barometric pressure, and relative humidity were measured before each test session to enable off-line conversion of V_{T} to body temperature and pressure, saturated with water vapor (BTPS). The flow transducer and gas analyzer were recalibrated before the study of each participant.

The ECG and abdominal respiratory excursions were detected and recorded using a J&J Engineering I-330-C-2 Physiologic Monitoring System (J&J Engineering; Poulsbo, WA) connected to a notebook PC with a 1-GHz Pentium processor. Ag-AgCl electrodes were placed on the left and right wrists. ECG signals were processed through a 1- to 100-Hz bandpass filter, passed though a 60-Hz notch filter, and then digitized at 1,024 samples/s. Heart rate and interbeat intervals were calculated on-line using an R-wave peak detection algorithm and stored on the PC for further off-line processing. Abdominal respiratory excursions were measured using a pneumograph strain gauge secured around the upper abdomen with a Velcro strap.

#### Procedures.

Participants were seated comfortably in an armchair in front of a table on which was placed a 19” color computer monitor. ECG electrodes and abdominal strain gauge were attached, and participants were instructed to breath normally through the mouthpiece to become habituated to the respiratory apparatus. Participants were then given instructions for following a respiratory-pacing stimulus displayed on the monitor. The pacing stimulus, generated by Physiolab software (J&J Engineering), was a “sawtooth”-shaped line that was colored blue on each upward slope and orange on each downward slope. Each tooth was 2.75 inches high and 2.5 inches wide at its base. A pattern of five peaks was displayed across the monitor. A small green ball traveled along the line from left to right, and participants were instructed to inhale as the ball ascended and exhale as it descended. The speed that the ball traveled along the line varied as a function of the pacing frequency. The pacing stimulus had a 40%:60% inhalation-to-exhalation time ratio and a 15% inhalation and exhalation pause time for each breathing rate. This pacing stimulus was chosen because it produces more consistent respiratory output within and between subjects than simpler methods (e.g., metronome). The order of paced breathing frequencies was randomly determined for each participant from one of two sets of values: 6, 8, 10, and 12 or 5, 7, 9, and 11, which were alternated between sequential participants. Participants breathed at each of the four frequencies for 4 min. A 3-min interval period during which participants removed their mouthpieces and nose clips and breathed at their own spontaneous rate occurred between each paced breathing episode.

#### Data reduction.

Breath-by-breath respiration rate, BTPS-converted V_{T},V˙e, V˙o
_{2},V˙co
_{2}, Pet
_{CO2}, and respiratory quotients (RQ = V˙co
_{2}/V˙o
_{2}) were computed by the MedGraphics CPX/D software. Means of breath-by-breath respiration rate, V_{T},V˙e, V˙o
_{2},V˙co
_{2}, Pet
_{CO2}, RQ,V˙e/V˙o
_{2}, andV˙e/V˙co
_{2} were calculated for 30-s epochs within each 4-min paced breathing period.

Respiratory sinus arrhythmia was computed in accordance with published guidelines (7, 25a). In brief, heart rate time series were manually edited for faulty beat detection^{1} and then sampled at 4 Hz. Equidistant time series were then linear detrended, and spectral analysis was computed by using the fast-Fourier transform method with a Hamming window (3). For each 4-min epoch, RSA was calculated as the spectral density estimate at the respiratory frequency. For example, for a subject pacing at 12 breaths/min, RSA was taken as the amplitude of a 0.20-Hz sine wave fitted to the heart rate time series (see Fig. 1). Cross-spectral analysis was also computed for heart rate and abdominal pneumograph. Squared coherence (*K*
^{2}) and phase between the two variables at the respiratory frequency were calculated. Coherence at the respiratory frequency was ≥0.80 in all cases (means ± SD, 0.95 ± 0.02). Finally, because RSA has been shown to decrease with age (10, 24), values were adjusted for between-subject differences in age. A simple linear regression was computed to determine the slope coefficient *B* for age on RSA for our sample. RSA values were adjusted according to the following equation
where the product of the regression slope coefficient and the difference between the mean group age and subject's age were added to the value of RSA to produce an age-adjusted RSA value.

#### Data analysis.

Repeated measures ANOVAs were computed to assess the stability of respiratory variables during paced breathing episodes, because there was concern that respiration and measuredV˙o
_{2} andV˙co
_{2} might take time to stabilize after the start of each new paced breathing frequency. For each respiratory variable (V_{T}, V˙e,V˙o
_{2},V˙co
_{2}
_{,} RQ, and Pet
_{CO2}), a repeated-measures ANOVA was computed with eight 30-s averages entered as the within-subjects variables and respiration rate as the between-subjects factor. Because repeated physiological measures tend to be highly correlated, inversely proportional to the amount of time between collection points (17), we expected that the within-subjects covariance matrix would be significantly different from zero (i.e., a violation of the sphericity assumption), making univariate *F* tests and associated *P* values invalid. When violations of the sphericity assumption occurred, statistical significance was determined using degrees of freedom corrected by the Huynh-Feldt method (16). When significant time effects were found,*t*-test contrasts between 30-s epochs were computed with significance level adjusted for multiple comparisons.

Descriptive statistics (means ± SD) were computed for each independent variable. In addition, bivariate Pearson product-moment correlations were computed between age-adjusted RSA, phase, and other measured variables.

To check the validity of the manipulation of respiration rate to systematically alter the amplitude of RSA and the phase between respiration and RSA, two additional repeated-measures ANOVAs were computed. In each analysis, RSA and phase at each paced breathing period were entered, respectively, as the within-subject variables. As above, when violations of the sphericity assumption occurred, statistical significance was determined using degrees of freedom corrected by the Huynh-Feldt method (16).

Two linear multiple regression analyses with pooled data across subjects and paced breathing periods were computed for the dependent variables V˙e/V˙o
_{2} andV˙e/V˙co
_{2}. In both models, independent variables were entered hierarchically in two blocks. The first block contained age, respiration rate, V_{T}, heart rate, and eitherV˙co
_{2} orV˙o
_{2}. The second block added RSA and phase. Mean values from each 4-min paced breathing epoch were used in all analyses. Unstandardized and standardized regression coefficients, partial correlation coefficients, and *t* values for independent variables, and *R*
^{2},*R*
^{2} change (for the addition of RSA and phase), and the *F* statistic for *R*
^{2} change for each regression model, were computed.

## RESULTS

Instability of respiratory variables during paced breathing periods was indicated by a significant effect of time in the repeated-measures ANOVAs forV˙e/V˙co
_{2}[*F*(7, 49) = 5.16, *P* < 0.01] and V˙e/V˙o
_{2}[*F*(6.27, 43.89) = 3.24, *P* < 0.05]. No significant Time × Respiration Rate interactions were found. Contrast tests forV˙e/V˙co
_{2} and forV˙e/V˙o
_{2} between all 30-s epochs showed that, for both measures, each of the first three 30-s epochs was significantly different from the final five, suggesting instability of these measures during the first 90 s of each paced breathing period (see Fig. 2). No other significant differences were found. A second set of repeated-measures ANOVAs that excluded the first three 30-s epochs were computed and showed no significant effect of time onV˙e/V˙co
_{2},V˙e/V˙o
_{2}, or any other respiratory measure. As a result of this finding, the first 90 s of each paced breathing period were excluded in calculating the period means used in all subsequent analyses. However, the results of our analyses were not substantially different when the first 90 s were not removed from epoch averages, suggesting that our independent variables covaried with the ventilatory equivalents similarly during stable and unstable periods.

RSA was significantly inversely correlated with age (*r*= −0.44, *P* = 0.005), as has been demonstrated by others (10, 24). Age-adjusted RSA was thus computed, as detailed in methods, and used in subsequent multiple regression analyses.

Mean values of each independent variable across all subjects and paced breathing periods are shown in Table 1. It is of note that subjects tended to hyperventilate during paced breathing on our apparatus (e.g., mean Pe
_{CO2} = 35.7 mmHg), which could account for the relatively high ventilatory equivalent values obtained in this study. Nonetheless, whereas this likely did not represent a long-term equilibrium, it did appear to be a new steady-state value for each paced breathing frequency after an initial 90 s of instability. Ventilatory equivalents and respiratory quotients remained stable over this latter period (see Fig. 2).

The repeated-measures ANOVAs computed as a manipulation check for the effect of respiration rate on the amplitude of RSA and the phase between respiration and RSA showed significant effects of paced breathing period on both variables [RSA:*F*(3) = 22.19, *P* < 0.001; phase: *F*(3) = 15.62, *P* < 0.001]. Graphs of RSA and phase as a function of respiration rate are shown in Fig. 3. Phase between respiration and RSA crosses from positive (i.e., respiration leads RSA) to negative (i.e., RSA leading respiration) at ∼6 breaths/min.

Output from both linear regression analyses is shown in Table2. Age-adjusted RSA was significantly associated with the ventilatory equivalents for both CO_{2}(partial *r* = −0.53, *t* = −3.36,*P* = 0.002) and O_{2} (partial*r* = −0.49, *t* = 3.03, *P*= 0.005), and phase between RSA and respiration was significantly associated with the ventilatory equivalent for CO_{2} (partial*r* = 0.40, *t* = 2.33, *P* = 0.03) after statistically controlling for the effects of age, respiration rate, V_{T}, and heart rate. Age-adjusted RSA and the phase between RSA and respiration contributed an additional 10% to the total explained variance inV˙e/V˙co
_{2}(*R*
^{2} = 0.99 vs. 0.89) and an additional 7% to that for V˙e/V˙o
_{2}(*R*
^{2} = 0.95 vs. 0.88). Figures4 and 5 show scatter plots between age-adjusted RSA and ventilatory equivalents and phase and ventilatory equivalents for CO_{2}and O_{2}, respectively. In each plot, individual subjects' data points are connected by lines to show within-subject associations and changes. Bivariate correlations among RSA, phase, and other measures are shown in Table3. Neither RSA nor phase was significantly associated withV˙co
_{2}
_{,}V˙o
_{2}, or Pet
_{CO2}.

## DISCUSSION

This study tested the hypothesis that respiratory sinus arrhythmia is independently associated with the efficiency of pulmonary gas exchange during paced breathing after controlling for the effects of age, respiration rate, V_{T}, oxygen uptake, CO_{2}output, and mean heart rate. We found that across a range of breathing frequencies, RSA was independently associated with gas exchange efficiency, as indexed by the ventilatory equivalents for O_{2} and CO_{2}, and that the phase between heart rate and respiration was significantly associated with the ventilatory equivalent for CO_{2}. Combined, these two variables contributed an additional 7% and 10% to the explained variance in the efficiency of pulmonary gas exchange for O_{2} and CO_{2}, respectively, over and above that accounted for by age, respiration rate, V_{T},V˙co
_{2}
_{,} andV˙o
_{2}. Importantly, RSA and phase were not significantly associated withV˙co
_{2}
_{,}V˙o
_{2}, or Pet
_{CO2}, but rather with the uptake of oxygen and output of CO_{2} as a function of ventilation, i.e., gas exchange efficiency.

Our findings are consistent with the theory advanced by Hayano et al. (13) that RSA may serve an active physiological role in increasing the efficiency of pulmonary gas exchange and circulation by matching perfusion to ventilation from moment to moment within each breathing cycle. Also, consistent with the findings by these investigators and others that the amplitude of RSA reflexively increases in response to induced hypercapnia (22, 27) but not hypoxia (29), we found that the phase between RSA and respiration was associated with the ventilatory equivalent for CO_{2} but not O_{2} (see Table 2).

Slow, deep breathing has long been taught to patients with chronic respiratory disease as a means to improve respiratory function, presumably via the mechanical effects these maneuvers impart for improved ventilation. However, the amplitude of RSA also increases substantially with decreases in respiration rate and increases in V_{T} (15). Thus supported by our results, the advantage of slow, deep breathing may also be due, in part, to improvements in pulmonary gas exchange efficiency mediated by increases in RSA. Interestingly, data from this and other studies (e.g., 8, 11, 15) show that RSA amplitude reaches a maximum at ∼6 breaths/min. Our data show, in addition, that RSA also slightly precedes the respiratory phase at this breathing frequency, which may be optimal from a perfusion-ventilation matching perspective. It is noteworthy that breathing training based on certain yoga and meditation traditions instructs practitioners to breath at ∼6 breaths/min (1, 6, 14,25).

This study has several limitations. First, we used the ventilatory equivalents for CO_{2} and O_{2} as a surrogate of pulmonary gas exchange efficiency. Whereas this is a generally accepted method for representing ventilatory efficiency, it is only one of several ways to define and measure this construct. To strengthen our interpretations based on the use of this proxy measure, we also showed that RSA and phase were not associated withV˙co
_{2}
_{,} orV˙o
_{2}, thus decreasing the likelihood of an alternative interpretation that our findings were an artifact of the choice of outcome measure. Nonetheless, several tonic and phasic influences may affect the efficiency of pulmonary gas exchange, and the choice of measures to include in our multiple regression model was not exhaustive. However, our model did account for 99% and 95% of the variance in the ventilatory equivalents for CO_{2} and O_{2}, respectively. Still, our findings are associations only, and thus provide little insight regarding mechanisms of action. There are both global and local changes in ventilation-perfusion matching that can affect pulmonary gas exchange, particularly if there is a preferential redistribution of flow, which may also occur with local vascular changes. Whereas we propose that increases in efficiency may be partly due to changes in the ratio of physiological dead space to V_{T}, as has been suggested in one previous study (13), such a conclusion is limited by the measures used in this study. Furthermore, it cannot be stated from our data whether changes in RSA per se produced physiologically significant improvements in pulmonary gas exchange efficiency. It is possible that a different unmeasured variable accounts for the apparent contribution of RSA.

A second limitation of this study is that cardiac output, a potentially important contributor to pulmonary gas exchange efficiency, was not fully accounted for by the measurement and inclusion of tonic heart rate. Stroke volume would also need to be measured to fully assess cardiac output, but this was not done in our study. However, phasic changes in stroke volume, if these were to occur, would be most likely to covary with V_{T}, which was accounted for in our model (via changes in intrathoracic pressure) rather than with the variability in heart rate occurring at the frequency of respiration (i.e., RSA). In addition, evidence suggests that preferential redistribution of flow, which could result in improved efficiency, does not occur with changes in cardiac output (4, 18). Nonetheless, it is conceivable that phasic changes in stroke volume could account for the variance in ventilatory equivalents explained by RSA in our study. Related to this potential limitation, subjects were not recorded at the same time of day or after a period of controlled fasting. This may have resulted in hemodynamic confounds caused, for example, by changes in blood volume. Third, whereas we statistically accounted for the effects of changes in V_{T} on the ventilatory equivalents for CO_{2} and O_{2}, a more methodologically rigorous approach would have been to have subjects maintain a constant V_{T} as they did with respiration rate. Finally, we used the experimental manipulation of respiration rate as a means of altering the magnitude of RSA in our study. This manipulation may have produced a state of hyperventilation in some subjects. We were able to demonstrate that this was a steady state within our window of measurement and analysis, but it is unlikely that this represented a long-term steady state in respiration. Also, whereas we statistically controlled for the effect of respiration rate on the ventilatory equivalents for CO_{2} and O_{2}, a more methodologically rigorous approach would have been to alter RSA via mechanisms that would themselves have less of an impact on pulmonary gas exchange efficiency. One way to accomplish this would be through the use of pharmacological agents (e.g., atropine and β-blockers) that alter RSA while holding respiratory parameters constant. Future studies should include these methodological improvements.

In summary, we found that respiratory sinus arrhythmia was independently associated with the ventilatory equivalents for CO_{2} and O_{2}, and that the phase between heart rate and respiration was significantly associated with the ventilatory equivalent for CO_{2}, after statistically controlling for the effects of age, respiration rate, V_{T},V˙o
_{2}, V˙co
_{2}, and mean heart rate. Our findings are consistent with the theory that RSA serves a physiological role in improving the efficiency of pulmonary gas exchange by matching perfusion to ventilation within each respiratory cycle. Further studies should include experimental control of both breathing frequency and V_{T}, and the manipulation of RSA amplitude by means that minimally impact pulmonary gas exchange efficiency.

## Acknowledgments

We thank Marjorie Anderson, Bruce Culver, H. Thomas Robertson, and Robert Schoene for helpful comments during this study and paper preparation.

## Footnotes

Address for reprint requests and other correspondence: N. D. Giardino, Dept. of Rehabilitation Medicine, Univ. of Washington, Box 356490, 1959 NE Pacific St., Seattle, WA (E-mail: giardino{at}u.washington.edu).

↵1 Accurate determination of R-R intervals was not possible for >10% of the ECG record for two of the paced breathing periods due to movement artifact. These epochs were therefore excluded from our final analyses. However, results from analyses that included variables derived from the clean portion of these epochs were not different from those presented.

The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked “

*advertisement*” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.First published January 23, 2003;10.1152/ajpheart.00893.2002

- Copyright © 2003 the American Physiological Society