#### INTRODUCTION

The phenomenon of respiratory sinus arrhythmia (RSA) is known as fluctuations of heart rate associated with breathing, with heart rate acceleration during inspiration and heart rate deceleration during expiration (^{1,2}). RSA can be quantified in the time-domain, where the amplitude of heart rate fluctuations related to each breathing cycle is extracted (e.g., the difference between the fastest heart rate during inspiration and the slowest heart rate during expiration, known as the *peak-valley method*) (^{3}). Alternatively, variability of heart rate can be broken down into its frequency components by methods such as spectral analysis, where the spectral power at the frequency of breathing can be extracted as a measure of RSA. Influences on vagal efferent nerve traffic from both the central respiratory rhythm generator and peripheral afferents from the lungs are thought to be the main contributors to this phenomenon (for reviews see, 1,2,4). RSA, or heart rate variability (HRV) at the frequency of breathing, has been linked to cardiac vagal tone in studies using autonomic blockade (^{5–7}).1 Thus, these indices may offer a convenient noninvasive window into the vagal control of cardiac activity. In particular, indices of HRV power spectra have become widely used in basic studies of autonomic regulatory processes and in clinical research on cardiovascular disease and autonomic neuropathy (^{2,8,9}).

However, there is also ample evidence that respiration rate (RR) and tidal volume (V_{T}) have a profound influence on RSA, and this influence is independent from actual changes in cardiac vagal outflow (^{10–15}). Within the range of normal breathing (approximately 6–30 breaths/min) faster RR and lower V_{T} lead to reductions in RSA in an approximately linear fashion. Up to the early 1990s, studies in cardiovascular physiology and psychophysiology have discussed such influences and their implications for interpretation of RSA at length (^{11,13,16,17}). This evidence has also been acknowledged in guidelines on the analysis and interpretation of RSA in 1997 (^{1}). More recently, arguments for a control of respiratory parameters in RSA studies have been reiterated in detail by Grossman and Taylor (^{18}). However, research in psychosomatic medicine has largely lagged behind these developments. An earlier editorial on RSA has focused on other methodologic issues such as stationarity of heart rate and psychometric properties of HRV (^{19}). The present article will mainly deal with issues of respiratory control. In the following, we will briefly review attempts to control for respiratory influences on RSA in studies published in *Psychosomatic Medicine* since 1994, which followed the peak of the methodologic discussion in the early 1990s.

##### Respiratory Control in RSA Studies in *Psychosomatic Medicine*, 1994 to 2004

A total of 39 studies (^{75–113}) that included measures of RSA have been published in *Psychosomatic Medicine* from 1994 to 2004 (approximately 3.5 per year). In the same time period, 171 studies were published that included heart rate (approximately 15.5 per year), and 54 studies that included measures of the respiratory system (approximately 5 per year). Main topics of the RSA studies were cardiovascular psychophysiology/disease (10 studies), psychopathology (eight studies), general autonomic/immune/endocrine psychophysiology (five studies), gastrointestinal disease (four studies), relaxation/breathing/biofeedback intervention (four studies), respiratory disease (two studies), menstruation (two studies), sleep (two studies), fitness (one study), and multiple chemical sensitivity (one study). RSA was more frequently quantified by high-frequency HRV (30 studies) than with time-domain measures such as SD, mean square of successive differences, or the peak-valley index (14 studies; some studies included both measures). Frequency ranges for the high frequency HRV analysis varied across studies, with 0.12 to 0.18 Hz for the lower limit and 0.35 to 0.50 Hz as the upper limit.

Almost half of the studies ignored respiratory influences on RSA completely (Table 1). Authors typically did not reflect on a lack of respiratory control as a potential limitation in discussion of their data. If control for respiration was attempted, it was mostly restricted to RR. Often, measures were taken from one strain gauge (positioned over the thorax, abdomen, or between both), in some cases also extracted from the impedance cardiography signal or estimated from the heart period time series. Four more studies used uncalibrated thoracic and abdominal bands or attempted to estimate volume from the amplitude of a single respiratory band, both of which do not allow a valid determination of V_{T}. A number of studies also paced respiration throughout some or all measurement intervals, using pacing frequencies similar to normal breathing (12–15 breaths/min) or a wider frequency range (6–30 breaths/min), if pacing was central to the aims of the experimental intervention (breathing relaxation, hyperventilation testing). Other strategies were reporting of RR group means and interpreting changes concomitant with RSA changes, or partialling out variance of RR from RSA measures by use of analysis of covariance. Both RR and V_{T} were measured in a valid way in four studies, and only one used this information for a within-individual correction of RSA.

##### Control of Respiratory Influences on RSA

Lack of control of respiratory influences on RSA can render results of a study ambiguous and uninformative with respect to a potential vagal origin of the observed effects. For example, one study sought to explore vagal tone in patients with panic disorder, social phobia, and controls during hypo-, normo-, and hyperventilation using paced breathing trials at 6, 12, and 20 breaths/min (^{75}). The findings were similar in all three groups and perfectly reflected the known influences of RR on RSA: the participants, who breathed at a baseline rate of approximately 14 to 16 breaths/min, increased their uncorrected RSA when slowing down to 12 breaths/min and even further (to almost double the size) at 6 breaths/min; when breathing faster at 20 breaths/min, their uncorrected RSA markedly dropped. Before and after these paced breathing trials, RSA remained on a relatively constant level. The observed increases and decreases in RSA were interpreted by the authors as augmentations or attenuations of vagal tone, respectively. However, given a substantial literature demonstrating effects of RR and V_{T}, an interpretation of these findings in terms of changes in vagal tone is clearly not warranted.

A variety of strategies has been devised to account for portions of RSA variance due to changes in respiration. Their aim is to arrive at a “purer” measure of vagal tone changes, which implies that respiratory changes are not linked to vagal tone changes. However, correction for respiration could, under certain conditions, also remove variance due to “true” changes in vagal output. Under these conditions, respiratory correction procedures yield conservative estimates of vagal changes.

In the following, we will discuss issues central to the respiratory correction of RSA that have surfaced in our review of publications in *Psychosomatic Medicine*. Specifically, we will discuss the following questions: a) Should control for respiratory parameters be implemented between or within individuals? b) Should control procedures address RR, V_{T}, or both? c) Is paced breathing is a valid strategy for respiratory control? d) Which instrumentation is required for a valid control of respiratory parameters? And e) does the comparison of RSA with or without respiratory control produce valuable information regarding individual differences in vagal tone? We will also present data to estimate the relative importance of RR and V_{T} influences on RSA under controlled conditions. Finally, we will present the rationale and method of a control procedure for respiratory influences on RSA. A more detailed discussion of other respiratory control procedures is presented by Grossman and Taylor (^{18}).

##### Between—Versus Within—Individual Control

It is sometimes assumed that a lack of changes in RR and V_{T} on the group level implies that measurements of RSA are not influenced by respiration. This argument may hold under conditions in which very little interindividual variation can be expected in respiratory variables. However, if individuals vary in their respiratory adjustments across situations and these variations are cancelled out in the mean of respiratory variables, it cannot be excluded that any observed RSA changes are actually due to these inconspicuous respiratory variations. This is due to a considerable interindividual variation in the extent to which RR modulates RSA (e.g., 14,20). While some individuals show massive increase in RSA when e.g., changing their total respiratory cycle duration (T_{TOT}) from 4 s to 6 s (RR of 15 to 10 breaths/min), others may only show small though reliable increases. If a within-regression analysis of RSA on T_{TOT} would be calculated across a representative range of RRs for both individuals, the former individual would show a very steep and the latter a much shallower positive slope.

We can illustrate the possible consequences on a measure of vagal tone in an extreme example based on typical empirically determined within-individual regression equations of two individuals (see also 20). Assume that individual A has a rather steep slope of RSA/V_{T} on T_{TOT}, with y = −70 + x × 60, and individual B has a shallower slope, with y = −50 + x × 20; in this example we use RSA normalized for V_{T} (see below under *A Method to Control* …), which shows a variation in slopes across individuals similar to RSA. With both individuals breathing at a T_{TOT} of 6 s (10 breaths/min) under baseline condition, their respective RSA/V_{T} values would be 290 and 70 ms/l (*M* = 180 ms/l). If under hypothetical task conditions, individual A would switch to *breathing faster* at a T_{TOT} of 4 s, his or her RSA/V_{T} would be reduced markedly to 170 ms/l due to the steep slope. If individual B under the same experimental condition would switch to *breathing slower* at a T_{TOT} of 8 s, his or her respective RSA/V_{T} would be increased only slightly to 110 ms/l, due to the shallower slope. While the mean T_{TOT} of both individuals again would be 6 s, their group mean RSA/V_{T} would be *M* = 140 ms/l. Thus, despite no change in *mean* T_{TOT}, values of RSA/V_{T} would suggest a decrease in vagal tone on the group level. This effect could materialize independently from any actual change in vagal tone.

Limitations of the between-individual approach also apply to attempts to control for respiratory influences using analysis of covariance: RSA values are adjusted without regard to variation in individual sensitivities of RSA modulation by respiration (for an illustration, see 18). Thus, strategies inferring respiratory influences on RSA from a between-individual analysis of respiratory parameters are not suitable to capture the full extent of possible influences. Interpretation based on mean respiratory changes is less likely to fail if there is reason to expect little interindividual variation in respiratory responses. This is more likely the case in situations of basic biological adaptation such as in physical activity increase but less likely in response to situations of psychological challenge. However, even with a reduced interindividual variability in RRs and V_{T}, in smaller samples the estimate of cardiac vagal tone can be biased considerably, if a few individuals with strong sensitivities of respiratory RSA modulation diverge in their response from the mean group response.

##### Control for RR, V_{T}, or Both?

Continuing uncertainty seems to exist regarding the importance of controlling for RR or both RR and V_{T}. While research has demonstrated that both affect RSA profoundly, the relative importance of both has been debated. Because normally RR and V_{T} are tightly coupled to match metabolic demand, within-individual variations in RR or V_{T} alone may indeed account for most of the variation in RSA under regular conditions. However, voluntary influences on breathing pattern, phasic adaptation to environmental challenges, and central nervous system processes related to stress and anxiety (^{21–23}) often lead to a disruption of brain stem respiratory control and thus the coupling of RR and V_{T}. As a consequence, substantial independent contributions of RR and V_{T} to RSA can be expected.

Using data from a previous study (^{20}), we compared the influence of both parameters on RSA under conditions of baseline measurements (coupling of RR and V_{T}) versus experimental instructions to change the breathing pattern systematically (partial uncoupling of RR and V_{T}). Participants were 25 healthy students (10 women). They breathed at four different RR, with 8, 10.5, 13, and 18 breaths/min (T_{TOT} 7.5, 5.7, 4.6, and 3.3 s), following a pretaped auditory signal with rising pitch of the tone for inspiration and falling tone for expiration. These paced breathing sequences were performed in supine posture with the instruction to a) only follow the tones with the breathing excursions and b) vary V_{T} voluntarily while adhering to the speed of the signal. From continuous measurements of respiration (pneumotachograph with Fleisch No. 2 transducer) and the electrocardiogram, we extracted T_{TOT} and V_{T}, as well as RSA using the peak-valley method (^{3,24}).

As expected, results showed marked differences in the average within-individual correlation between T_{TOT} and V_{T} for the condition with spontaneous breathing versus voluntary variation of V_{T}, with *r* = 0.80 (range: 0.57 to 0.92) versus *r* = 0.37 (range: −0.002 to 0.68), respectively (for direct comparison of both conditions, we analyzed only breaths within the V_{T} range observed under spontaneous V_{T} conditions). Although T_{TOT} and V_{T} were both highly correlated with RSA, the median of the within-individual partial correlations between V_{T} and RSA controlled for T_{TOT} was rather low on average during the spontaneous V_{T} condition (Table 2), due to the high correlation between T_{TOT} and V_{T}. For the condition with varying V_{T}, the partial correlation was higher, indicating an increasing importance of V_{T} for RSA modulation with the smaller correlation between T_{TOT} and V_{T}.

For both conditions, spontaneous and voluntary varying V_{T}, we also calculated within-individual hierarchical linear regressions to estimate the variance of RSA explained by T_{TOT} when entering it as the first predictor and the additional variance explained when entering V_{T} as the second predictor. While under spontaneous V_{T}, most of the variance was indeed explained by T_{TOT}, under conditions of voluntary variation of V_{T}, the contribution of V_{T} rose more than threefold (Figure 1). The increase in explained variance was even greater when the range of V_{T} was not restricted to the range observed under spontaneous V_{T} conditions. (Note that in a number of cases the intercorrelation between T_{TOT} and V_{T} would be viewed as too high for multiple regression (multicollinearity), e.g., *r* > 0.80 in 11 cases for the spontaneous V_{T} condition). When the order of predictors was reversed, V_{T} initially accounted for only slightly less RSA variance under spontaneous breathing, but for similar or slightly greater portions of the variance under voluntary variation of V_{T}. Depending on the condition, approximately 45% to 55% of the total RSA variance was explained by respiratory parameters, and V_{T} contributed up to one third to this prediction after T_{TOT} had been accounted for.

Thus, changes in V_{T} substantially contribute to the variation in RSA. When analyzing data from conditions where RR and V_{T} are tightly coupled, this contribution may not be important and remain mostly unrecognized. However, uncoupling of RR and V_{T} may occur in situations where the individual is behaviorally challenged. Ignoring V_{T} under these conditions may lead to erroneous inferences regarding cardiac vagal tone.

##### Paced Breathing as a Control Strategy

One approach to control the influence of respiration on RSA is to pace the participants’ respiration at a standard speed throughout the experiment. The goal is to eliminate fluctuations of RR across experimental conditions and thus prohibit a contamination of RSA changes with potential task-induced changes in respiration. Auditory pacing using a metronome or a computer-generated tone pattern has been used for this purpose, as well as visual pacing in which individuals match their breathing excursions to a predetermined template using biofeedback.

However, a number of issues limit the usefulness of paced breathing during experiments. First, auditory pacing can only account for RR. Automatic adjustment of V_{T} to a particular RR can be expected in periods of steady state, but this is not guaranteed under conditions of behavioral demand (see above). Although visual matching of the V_{T} curve using biofeedback would allow an additional tight control for volume aspects, it is impractical in combination with many behavioral tasks, in particular with those requiring visual processing. As a minimum requirement for pacing procedures, a manipulation check with measures of rate and volume would be required throughout the task periods of interest. Second, pacing requires dividing information processing capacities, which can distract from tasks and thus make performance suboptimal. Although the paced breathing pattern is sometimes trained before the experiment, it can be doubted that in complex behavioral tasks (in particular in tasks that change metabolic demand) such learned breathing patterns prevail over the automatic regulation through the respiratory center. Third, psychological and physiologic responding under study can be modified by voluntary alterations in breathing pattern. Paced breathing at a slower speed has been used to modify anxiety and autonomic response to stress (e.g., 25,26). Irregularity in breathing has been observed in clinical populations (^{27,28}), and the imposition of regularity may alter typical response characteristics in these individuals. In addition, voluntary imposition of breathing patterns against metabolic demand has been shown to affect gas exchange unfavorably (^{29}), which in turn could affect autonomic activity. Fourth, given the interindividual variation in basal RR, a particular standard pacing speed may not be acceptable for all subjects under particular task conditions. Theoretically, the most comfortable speed would have to be determined for each individual across the range of conditions to be studied, but this would be impractical for most experimental protocols.

The role of paced breathing as a respiratory control in RSA studies is probably restricted to protocols in which comparisons between repeated baseline measurements are being made. We found that under baseline conditions, pacing at different speeds (eight to 18 breaths per minute) for 2 to 3 minutes is typically tolerated well in healthy participants and asthma patients, with little impact on symptom and mood ratings (^{30}).

##### Additional Control for Carbon Dioxide Partial Pressure (pCO_{2})?

There is evidence from experimental studies that altered pCO_{2} levels can influence RSA (^{31–33}). There is a wide variation of basal pCO_{2} levels in the population (^{34}) making this potentially an additional complication in interpreting RSA differences between individuals. Also, substantial changes in pCO_{2} during stress and anxiety states can be observed within individuals (e.g., 23,35). Because changes in pCO_{2} normally depend closely on changes in ventilation, potential effects of changes in pCO_{2} on RSA will often be masked by effects of RR and V_{T}. Thus, for a given metabolic state, pCO_{2} is inversely proportional to alveolar ventilation, which is the product of RR and (V_{T} minus V_{ds}), where V_{ds} is the volume of anatomical (or physiological, if one wishes to calculate it) dead space. Increases in alveolar ventilation and decreases in pCO_{2} may occur, with constant V_{T} and dead space, by an increase in RR alone, which decreases RSA. Or, increases in alveolar ventilation and decreases in pCO_{2} may occur at constant RR and constant dead space by an increase in V_{T} alone, which would lead to an increase in RSA. Alternatively, an increase in alveolar ventilation and decrease in pCO_{2} may be linked to increases in both RR and V_{T}, which might cancel out respiratory influences on RSA or lead to decreases or increases in RSA, depending on whether RR or V_{T} contribute to a greater degree. Thus, changes in pCO_{2} could be associated with any outcome in (respiration-uncorrected) RSA.

Ventilation (and thus RR and V_{T}) is normally adjusted automatically to maintain or reinstate CO_{2} homeostasis across a variety of internal and environmental challenges. However, under experimental conditions of tightly controlled RR and V_{T} by paced breathing, Sasano et al. (^{33}) observed increase in RSA when the fraction of CO_{2} in the inspired air was increased. The extent to which similar dramatic decoupling might occur under psychophysiologic laboratory conditions remains speculative, and the extent of any influence of pCO_{2} on RSA, in addition to effects of RR and V_{T}, remains to be explored. Manipulations of anatomical dead space at constant RR and V_{T} may be an avenue to explore such influences. For example, under the influence of a potent anticholinergic agent such as inhaled tiotropium bromide, normal individuals manifest a significant increase in dead space (^{36}). Combined with an experimentally fixed RR and V_{T} by paced breathing, this would lead to decreased alveolar ventilation due to increased anatomical dead space and, accordingly, an increase in pCO_{2}. In asthma, provocation with methacholine or cold dry air leads to constriction of the large airways, thereby decreasing anatomical dead space. Alternatively asthmatic patients with airway constriction can be given bronchodilating agents, thereby increasing anatomical dead space. In these cases, maintenance of constant RR and V_{T} will result in decreased pCO_{2} (with provocation) or increased pCO_{2} (with dilation). If pCO_{2} changes should prove to contribute substantial independent effects on RSA under such conditions, then future correction methods may have to incorporate additional adjustments for this factor. It should be noted that some of these conditions are more likely to occur between individuals than within the same individual over time. Also, because considerable voluntary effort must be directed at maintaining a constant breathing pattern in the face of changes in pCO_{2}, it remains to be shown that such experimental manipulations are representative of real-life situations.2

##### Instrumentation Issues

Often, the only respiratory measure included in studies is RR derived from measurements with one strain gauge or from secondary processing of the interbeat-interval time series or the impedance cardiography signal. This approach does not allow for an estimation of V_{T} and thus generates rather limited information about the respiratory system. In particular, when studying the respiratory autonomic relationship, measurements of V_{T} are needed, which allow the derivation of multiple indices of interest (^{23}). V_{T} can be estimated indirectly using thoracic and abdominal sensors such as inductance plethysmography bands, pneumatic belts, or piezoelectric strain gauges, or directly using spirometry or pneumotachography (^{22,37}). Interpretation of indirect measurements requires a precalibration using instrumentation for direct volume measurement or isovolume maneuvers (^{38}). It should be noted that costs for such equipment and computational requirements in data processing are rather modest compared with more demanding techniques commonly used in psychophysiology (^{39}).

##### A Note on the Comparison of Absolute RSA Levels

It has become common to compare absolute levels of RSA or HRV between groups. Comparisons of absolute levels of RSA between groups was reported in 24 of the 39 studies published in *Psychosomatic Medicine* in the 11-year period. In cardiology, a host of studies has demonstrated improved clinical outcome in patients with myocardial infarct who have a higher cardiac interbeat-interval variability (for reviews, see 2,8). Although impressive on a practical level, the interpretation of these findings in terms of cardiac vagal outflow is tentative at best. The assumption that the absolute level of cardiac vagal outflow can be estimated by RSA and can be compared between individuals rests on animal experiments, which have shown that the vagus is silent or almost silent during inspiration, whereas its excitation is maximal during expiration (for a review, see 17). However, in humans a residual vagal tone is observed in inspiration to maximal inspiratory level, and this residual vagal activity is not related to RSA (^{40}). With this added uncertainty, an interpretation of RSA as a between-individual measure of cardiac vagal tone is not well justified. Even controlling for respiratory parameters cannot alleviate the problem. Findings with complete vagal blockade suggest that the prediction can be improved by a linear combination of heart rate and RSA, but this awaits evaluation in larger samples (^{17,40}). Additional problems arise from a strong dependency of RSA on physical activity levels (^{41}). Because the immediate autonomic adjustment to physical activation is a vagal withdrawal, varying levels of activation or deactivation, body movements, or static tension between individuals during baseline measurements can introduce substantial error when individuals are being compared regarding their “basal” vagal tone. Given these doubts, differences in RSA values between individuals would be more cautiously interpreted as evidence for respiratory modulation of heart rate.

##### A Method to Control for RR and Volume Effects on the Time-Domain Index of RSA

Procedures have been proposed to control for RR and V_{T} in the frequency domain measure of high frequency HRV (^{42}) and in the time-domain peak-valley index (^{20}). Without respiratory correction, high-frequency HRV and the peak-valley index yield almost identical results (^{3}), but data comparing the respiration-corrected indices of both methods are still missing. Frequency- and time-domain measures both require unique precautions in quantification, have unique advantages and disadvantages for addressing different research questions, and vary their suitability for different experimental designs and settings (for a detailed comparison, see 43). Here, we will focus on a brief introduction of a correction procedure for the time-domain method. This method has the advantage of yielding a breath-by-breath index of corrected RSA, which also allows an estimation of short-term, phasic changes in cardiac vagal tone. Frequency domain indices typically require uninterrupted measurement epochs of at least 1 minute to yield valid estimates (^{2}). In the following, we briefly present the rationale and method of this correction procedure, followed by examples of its application.

##### Basic Rationale and Method

The time-domain correction method is based on observations that the quotient of RSA per liter V_{T} shows a systematic decline across a range of increasing RR (^{13,14}). The decline of RSA/VT is fairly linear from approximately 6 to 30 breaths/min, which covers a great part of the spectrum of naturally occurring RRs in healthy human individuals (note: For more clarity we will only refer to T_{TOT} in the following, which is the inverse of RR). A change in vagal tone within an individual affects the intercept of the regression equation (elevation of the regression line) of RSA/V_{T} on T_{TOT} but not the slope (^{15,17,44,45}) (Figure 2A). Grossman and colleagues (^{13}) introduced this idea to psychophysiology and showed that RSA/V_{T} corrected for T_{TOT} is closely correlated within individuals with changes in β-blocked heart rate across a range of experimental tasks. We utilized this concept for a within-individual correction procedure of the peak-valley RSA measure for T_{TOT} and V_{T} (^{20,46}). The procedure requires a baseline calibration procedure to determine each individual’s unique modulation of RSA/V_{T} by T_{TOT} using three to four epochs of paced breathing across a representative range of T_{TOT} (e.g., 7.5, 5, and 3.3 s, equals 8, 12, and 18 breaths/min). RSA/V_{T} during the main experiment is then expressed as a deviation from RSA/V_{T} observed under baseline conditions at a particular T_{TOT}. For each respiratory cycle during the main experiment, the index is calculated as the deviation from the value predicted on the basis of the preexperimental paced breathing trials (see Figure 2B; Appendix 1 for computational procedures).3 A similar method has been employed for estimation of additional heart rate, with preexperimental calibration trials for metabolically justified heart rate using stepwise increases in exercise workload (^{47}).

##### Empirical Validation

An important aspect of the correction procedure is that it is performed on a within-individual level because of the variation in slopes of the regression of RSA/V_{T} on T_{TOT} between individuals. Slopes and intercepts of the model are calculated for each participant separately (Figure 2B). The validity of the correction procedure rests on the assumption that the slope of the model is a stable characteristic of the individual, while the intercept changes with manipulations of vagal tone.

In support of the basic assumption of the model, we found that the slope parameter was sufficiently stable across months (^{46,48}) and across voluntary variations in V_{T} (^{20}). As predicted, for variations in posture (standing versus supine), which typically have a pronounced effect on vagal tone, we found strong effects only on the intercept and considerable stability of the slope parameter on the group level (^{20}). However, stability of the individual differences in slopes was low between postures; thus, determination of the slope parameter for within-individual correction should take into account the posture at which measurements in the main experiment are undertaken.

Accurate estimation of the slope parameter is dependent on a stable level of vagal tone throughout the paced breathing trials. Changes in the level of vagal tone during paced breathing would lead to steeper or flatter slopes or a reduced precision in determining the slope parameter of the individual, depending on which breathing frequencies were affected. There has been some disagreement on the stability of cardiac vagal tone during paced breathing (^{49,50}). In our study (^{20}), mean heart rate (used as an estimator of vagal tone under these conditions) was relatively stable across the four paced breathing trials from 8 to 18 breaths/min during the supine condition (maximum difference between conditions 0.6 beats/min), but showed significant increases (though minimal, with 0.9–2.8 beats/min) from slower to faster breathing frequencies for conditions standing and supine with variable V_{T}. Substantial decreases in heart rate were observed during supine conditions for two participants and during standing conditions for six individual participants. Excluding these individuals did not greatly affect the findings regarding the stability of the slope parameter.

In further analyses, we addressed the argument that the slope parameter may only reflect aspects of the individual’s habitual breathing pattern. It could be speculated that the main determinant of the slope of RSA/V_{T} against T_{TOT} is the individual’s inherent tendency to vary V_{T} in a characteristic fashion across breathing cycles of different duration. Thus, individuals with very high volumes at longer cycles and very small volumes at shorter cycles would have a reduced steepness of their RSA/V_{T} slope (e.g., individual ID 2 in Figure 2B) compared with individuals with only moderately high volumes at longer cycles and moderately small volumes at shorter cycles, who would have a steeper RSA/V_{T} slope (e.g., individual ID 1 in Figure 2B) (in this example identical RSA values are assumed for both individuals). We tested this assumption by correlating the slopes of V_{T} on T_{TOT} with the slope of RSA/V_{T} on T_{TOT} in three samples (total *N* = 129 (^{51})) and observed only low to moderate correlations between these slope parameters. Thus, individual tendencies to modulate V_{T} across the T_{TOT} range are only a small source of variance in the RSA/V_{T} change. These findings also demonstrate that adjustment of RSA for both V_{T} and T_{TOT} is necessary.

Experience comparing corrected and uncorrected indices shows that controlling for respiratory effects can uncover effects on vagal tone that are predicted but are masked by concomitant changes in respiration. For example, vagal withdrawal can be expected as a consequence of static skeletal muscle activation (^{52}). At the same time, increases in V_{T} due to muscle activation will increase RSA and thus mask the actual effects on cardiac vagal tone in this parameter (which would be expected to materialize in a decrease of RSA) to a certain extent. In a study with brief contractions of facial and forearm muscles, we found little effect of the experimental protocol on uncorrected RSA but clear reductions in the respiration-corrected measure of RSA (^{53}). A comparison of indices showed that using RSA/V_{T} uncorrected for RR yielded the most favorable results, whereas additional correction for the reduction of RSA/V_{T} across RR yielded slightly less powerful results (^{46}). This can be expected as skeletal muscle tension is also expected to increase RR, thus reducing RSA and playing in favor of the expected direction of changes in RSA by vagal withdrawal.

On the other hand, in situations where vagal activation is expected, correction for concomitant changes has slightly attenuated the effects on RSA but not abolished them altogether. For example during facial cold stimulation, which is known to produce vagal slowing of heart rate (^{54,55}), we observed increases in both RSA and RSA/V_{T} corrected for T_{TOT}; however, only the former were statistically significant, while the latter were reduced to a trend (^{46}). In another study that included viewing of specifically selected categories of affective pictures (^{56,57}), robust increases in RSA were seen during erotic pictures that were also accompanied by increases in T_{TOT}. Using RSA/V_{T} corrected for T_{TOT} did not change the findings of an increase in vagal tone, thus confirming assumptions that in early stages of erotic tension, parasympathetic excitation is prominent (^{58}).

In other studies using experimental emotion induction, the uncorrected peak-valley RSA index suggested increases in cardiac vagal tone, but these findings were not retained after correction for respiratory parameters. In one study, participants viewed a mixed selection of positive, negative, and neutral pictures (^{59}), whereas in the other study they were presented with homogeneous blocks of happy and depressing pictures and self-referring statements (^{60}). In interpreting these findings, we preferred to retain the more conservative null hypothesis of no changes in cardiac vagal tone, given the limited empirical evidence on emotion-induced vagal excitation, which includes a lack of corroborating evidence from animal and pharmacologic blockade studies. Also, in these and other studies using the correction procedure, the results in the corrected RSA measure were often not predicted from patterns of group mean changes in RR and/or V_{T}, which can be due to the within-individual nature of the correction procedure. Findings of emotion induction studies vary with regard to observed changes in respiratory parameters (e.g., 61–63), which could indicate a considerable individual response specificity or situation by individual response specificity in the activation of this system. Especially the former in combination with interindividual differences in the sensitivity of RSA to respiratory changes would be likely to produce spurious findings in the estimation of vagal activation.

Finally, we used corrected and uncorrected RSA indices in analyses addressing a clinical hypothesis on depression and asthma. It has been assumed for some time that depressive states are particularly potent in eliciting asthmatic airway obstruction due to vagal excitation (^{64,65}). Vagal activity is known to be a potent constrictor of the airways (^{66}). In two studies with asthma patients (^{56,67}), we found positive correlations between changes in RSA/V_{T} corrected for T_{TOT} and the degree of airway obstruction during or after emotion induction with depressing pictures or self-referring statements. Although a tight relationship between cardiac and airway vagal tone is not necessarily guaranteed (^{68}) and the correlations were calculated between individuals, the findings are consistent with the hypothesis concerning depression and asthma, and with the long-held assumption that the vagus is the pathway for psychologically induced airway obstruction (^{69,70}). Most important in the present context, this relationship was not seen with uncorrected RSA.

In summary, in a number of studies corrected RSA was superior over uncorrected RSA by supporting plausible physiological (^{53}) and clinical (^{56,67}) hypotheses. It is likely that in at least some of these instances respiratory parameters have worked against the directions of changes in RSA, such as increases in RSA during muscle tension due to increases in V_{T}. Without correction for respiratory parameters, the estimation of vagal changes would have been too conservative in these cases. In other studies, respiratory correction did not lead to greatly diverging results (^{46,56}). These could be instances of strong vagal excitation and/or little accompanying changes of respiratory parameters working in the directions of RSA changes. Finally, in studies in which respiratory correction abolished changes seen in uncorrected RSA (^{59,60}), respiratory parameters either may have worked progressively toward producing RSA changes, or interindividual variations in respiratory changes in combination with interindividual variation in the sensitivity of RSA to respiratory changes may have existed. These were also mostly instances in which no clear a priori hypothesis on vagal changes existed.

#### CONCLUSION

The study of respiration-related HRV will continue to offer unique insights into the autonomic regulation of the heart. However, a greater degree of complexity regarding these measures has to be conceded if psychosomatic research aims to make progress in this area. Controlling for respiratory factors, which constitute the most significant contributors to RSA, is among the greatest concerns. The number of situations in which correction of respiratory influences on RSA is of secondary importance, such as strong vagal changes with very little respiratory changes (on the group level and in terms of interindividual variation in the direction of changes), is probably very limited. The small increment in procedural and computational efforts involved in respiratory control is far outweighed by the advantage of a more stringent estimation of cardiac vagal tone. To allow for comparison among the substantial body of literature that, thus far, has primarily reported uncontrolled indices, future studies should report both findings with controlled and uncontrolled indices.

It should be noted that the concerns raised about respiratory control are relevant for the interpretation of RSA changes within individuals (and for between-individual comparison of reactivity scores, e.g., difference scores of task minus baseline). Respiratory control is important when respiration-related HRV should be interpreted with respect to underlying autonomic mechanisms, such as vagal activity. This is often the case in experiments that test hypotheses about disease-relevant psychobiological models. In the between-individual comparison of absolute RSA values, substantial uncertainty remains about the interpretation in terms of autonomic activity. This is particularly due to the residual vagal activity during inspiration in humans, which has been shown to attenuate substantially the between-individual relationship of vagal tone and absolute RSA values. Nevertheless, research on the predictive power of absolute HRV values for cardiovascular health is valuable on a practical level and can be important in shaping behavioral prevention and intervention techniques.

Available procedures for respiratory control show progress over uncontrolled practice, but these controlled measures of RSA are by no means perfect. As with any other research area, research on respiration-related perturbations of autonomic outflow is constantly evolving in its exploration of the origins of RSA, its physiological mechanisms, and factors influencing it. It is now known that potential autonomic determinants of RSA are not restricted to the vagal system and contributions of adrenergic and peptidergic systems require further exploration and possibly future precautions in study design and interpretation (^{18}). These and other research directions, such the potential function of RSA for optimizing gas exchange (^{71,72}), links of RSA with central nervous system pathways (^{73}), and the genetic determination (^{74}) of RSA, will both profit from respiratory control procedures and enable us to refine these techniques in the future.

We wish to thank Michael D. Goldman, Paul Grossman, Andreas von Leupoldt, Stefan M. Schulz, and three anonymous reviewers for valuable comments.

##### APPENDIX

Computational steps and procedures involved in determining respiration-corrected RSA as an indicator of cardiac vagal output.

First step: Measuring the shortest and the longest heart period (HP_{min} and HP_{max}) within each breathing cycle.

Second step: Calculation of (uncorrected) peak-valley RSA for each breathing cycle: HPmax − HP_{min} (under the condition that HP_{min} precede HP_{max} in time).

Third step: Normalization of RSA: RSA divided by V_{T} for each breath: RSA/V_{T}.

Fourth step: Repeating steps 1 through 3 for each paced breathing epoch from the baseline calibration procedure and the experimental data.

Fifth step: Calculation of the within-individual regression of RSA/V_{T} on T_{TOT} (measured in seconds: s; T_{TOT} = 1/RR × 60) across the whole paced breathing task (all paced breathing epochs from the baseline calibration procedure concatenated): y′ = a + bx, with a = M_{y} − b × M_{x}, b = cov (x,y)/s^{2}, and x = (T_{TOT})_{calibration}. M_{x}, M_{y} are the means, s_{x}^{2} the variance of the x values, cov (x,y) = 1/N Σ_{i} (x − M_{x}) (y − M_{y}). More precisely, x, y, y′ and y_{residual} must be stated: x_{i}, y_{i,} y^{′}_{i} y_{residuali} with i = 1 …. *N*.

Using the SPSS routine REGRESSION you will get the means (M_{x},M_{y}), variance of the x values (s^{2x}), the covariance cov (x,y), the intercept a and the slope b, the predicted y′, and the residuals, which are our wanted (RSA/V_{T})_{corrected} values. To get these parameters, you have to apply the following SPSS syntax:

REGRESSION

/MISSING LISTWISE

/STATISTICS COEFF OUTS BCOV

/CRITERIA = PIN(0.05) POUT(0.10)

/NOORIGIN

/DEPENDENT RSADIVVT

/METHOD = ENTER TTOT

Sixth step: Calculating the corrected RSA index for each respiratory cycle during the main experiment [y_{residual} = (RSA/V_{T})_{corrected}] as the deviation from the value predicted on the basis of the preexperimental paced breathing trials: y_{residual} = y − y′, with y = (RSA/V_{T})_{experiment}, and y′ = (RSA/V_{T})_{predicted} by taking the regression coefficients a and b from the concatenated paced breathing trials.

In SPSS, this is realized by the following COMPUTE command:

COMPUTE YRESEXP = Y − (A + B × TTOTEXP). Cited Here...