INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Observed over a period of time, the pattern of breathing can be regarded as consisting of the mean output of the respiratory controller plus the inherent variability of this output. Past analysis of breathing pattern has largely focused on the mean output, but additional insight into the control of breathing can be obtained by analysis of the variational activity of the breath components (1). Employment of mathematical tools, including time series analysis, fast Fourier transformation (FFT), and the multiregressive model developed by Modarreszadeh and colleagues (4), provides powerful means for more closely investigating the nature of the variability of breathing. With these tools, total variational activity can be separated into correlated, oscillatory, and random fractions. Partitioning of the total variability into structured (i.e., correlated and oscillatory) and random fractions is important because different fractions can have different physiological implications (2, 4). While the random variability (white noise) may be a measure of unconstrained freedom of behaviorally (cortically) influenced control (2), the structured fraction may represent the constrained automatic control dictated by vital reflexes (5).

Investigation of the response to external mechanical loads has yielded important insights into the function of the respiratory control system (6). Such physiological perturbations also offer an important means of defining the mechanisms that determine the variational activity of breathing. Recently, we imposed increasing inspiratory elastic loads in healthy volunteers and observed decreases in the random fractions of variability of tidal volume (VT) and expiratory time (TE), and an increase in the random fraction of the variability of inspiratory time (TI) (7). The alterations in random variability were considered to represent behaviorally mediated adaptations of the respiratory controller as it compensated for loads well above the perception threshold. Ensuring a certain level of randomness to minimize discomfort may be one of the goals of higher brain centers in their influence on the respiratory controller (8, 9). Using the same methodology, we now have investigated changes in variational activity with inspiratory resistive loads of 3 and 6 cm H₂O/L/s. The goal of this investigation was to determine the change in the total variational activity of breathing and in its fractions of random and structured variability, brought about by resistive loads of differing magnitude. Changes in random and structured fractions of variational activity can provide insight into load-specific automatic and behavioral influences on the respiratory controller (2, 4).

Resistive loading differs from elastic loading in that the latter is perceived later during the breath cycle than resistive loads (10), and the loads also differ in their perception thresholds and magnitudes of perception (Weber fractions) (10). Nevertheless, we hypothesized that resistive loading would cause an increase in the random fraction of total variability of TI and decreases in that of TE and VTthat is, the same response as observed with elastic loading since the two types of load evoke similar load-compensating mechanisms. The sizes of external elastic (9 and 18 cm H₂O/L) and resistive (3 and 6 cm H₂O/L/s) loads were chosen to double and triple normal human intrinsic elastance and resistance, respectively, although we were aware that compensation for the applied elastic loads would require greater effort than would compensation for the imposed resistive loads (11).

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Subjects

Eighteen nonsmoking, healthy volunteers (7 women, 11 men) with a mean age of 28 yr (range, 20 to 46) participated in this study. All had normal pulmonary function. Informed consent was obtained from all subjects, and the study was approved by the Human Studies Subcommittee of Edward Hines Jr. Veterans Administration Hospital.

Experimental Protocol

Subjects were tested while lying on a bed in a semirecumbent position (upper body elevated to 30° angle) in a quiet room and breathing from an open circuit system. Flow resistors (Model 7100R; Hans Rudolph, Inc., Kansas City, MO) constituted inspiratory resistive loads of 3 and 6 cm H₂O/L/s. The devices provided linear resistances over a flow range of 12 to 120 L/s. The subjects watched a single documentary recording on a video tape player.

A mouthpiece was connected to the resistors through a low-resistive, two-way nonrebreathing valve (Hans Rudolph, Inc., Kansas City, MO). Other attachments to the mouthpiece included a heated, large-diameter pneumotachometer (Model 3813, Hans Rudolph, Inc.), a pressure transducer (Validyne, Northridge, CA), and a capnograph (DATEX, Helsinki, Finland). Mouth pressure, airflow, and its integrated volume signal as well as CO₂ tension were recorded continuously at 50 Hz using a 16-bit analog-to-digital converter (DATAQ Instruments, Akron, OH).

To allow adaptation to the equipment, the subject breathed through the unloaded system for 20 min, and then three loads (0, 3, 6 cm H₂O/ L/s) were applied in a random manner for 1 h each. Data from the first 20 min were considered to represent subject adaptation and were not analyzed. The three different loads could be combined in six different possible sequences which were randomized so that the order of the applied loads did not have a systematic influence. The subjects were continuously supervised to ensure wakefulness throughout the experiment.

We calculated respiratory timing from the airflow tracings: inspiratory time was measured as the period of inspiratory flow, and expiratory time was taken as the remaining duration of the breath cycle.

Data Analysis

Any time series for a breath component typically displays breath-to-breath variability in the magnitude of that component. This variability is composed of a fixed part, namely the mean value for the entire time series, and a variable part, which is the deviation of the magnitude of each breath from the mean. The variable breath-to-breath variation from the mean may be considered to have, in turn, a nonrandom (correlated and/or oscillatory) component and a random component. Autocorrelation and spectral analysis enable the determination of the relative magnitudes of these components. These approaches to data analysis have been described in depth in our previous publications (7, 12), and the reader is referred to the appendices of these reports for the mathematical details.

Gross variability. The standard deviation (SD), i.e., the square root of the variance, for each breath component was calculated in each subject during rest and loaded breathing as a measure of gross breath-to-breath variability. The coefficient of variation (CV = SD/mean) of each breath component was taken as a measure of relative variability.

Autocorrelation analysis. Autocorrelation analysis was employed to determine the fraction of variational activity that was correlated on a breath-to-breath basis (3, 7). Because autocorrelation analysis requires stationary (time-invariant) data, three segments during rest and loaded breathing that displayed the least deviation from mean minute ventilation (I) on visual inspection were chosen for analysis. For each breath component, the data strings containing 550 to 700 breaths were first mathematically detrended before autocorrelation analysis was performed (3). By its ability to extract correlated activity from data obscured by random noise, autocorrelation analysis can determine if there is a strong relationship between one breath and another at some interval (or lag) away. It also determines the relative strength of "short-term memory" for each breath component (11). "Short-term memory" refers to the number of consecutive lags, starting at a lag of 1 breath, that display autocorrelation coefficients that are statistically different from zero at p < 0.01 levels (3). While memory is the term used to describe this statistical relationship, it does not necessarily signify that the consecutive serial autocorrelation coefficients have a neural origin (3, 13, 14).

Spectral analysis. Power spectral analysis can also be used to quantitate breath-to-breath variability in a breath component (7, 15). The power spectrum expresses the variance of a signal as a function of frequency. The presence of a significant peak (7) in the spectrum indicates that some of the variance in the data is caused by a periodic oscillation with a period equal to the inverse of the frequency of the peak (16). The area inscribed by the peak (amount of power) reflects the degree of variability in the signal resulting from fluctuations at that frequency (17).

Although spectral analysis and autocorrelation analysis are mathematically related (18), periodic oscillations as detected by spectral analysis can represent physiological mechanisms other than autoregressive behavior (4); in particular, spectral analysis can reveal low-frequency (slow) oscillations of breath components that might be missed by autocorrelation analysis; it has also been shown (4) that both types of analysis are necessary for correct partitioning of the total variability of breathing without corrupting the autocorrelation coefficients.

Fractionation of variational activity. The variational activity of breathing was partitioned into autoregressive, periodic, and white noise fractions employing the multilinear regression model described and validated by Modarreszadeh and coworkers (4, 7). This model enables the quantification of each fraction and its contribution to the entire variational activity of breathing. Total variational activity is modeled as a compound consisting of both random and nonrandom fractions. The correlated and oscillatory fractions are quantified using autocorrelation and spectral analysis, and the random (white-noise) fraction is derived as the remainder of the total variance (4). Fractionation of variational activity allows the influence of a load on the composition of the total variational activity to be investigated.

General statistical methods. For each breath component, data during unloaded breathing and the two different loads were compared using an analysis of variance (ANOVA) with repeated measures. If ANOVA was statistically significant, a multiple-range test (i.e., Newman-Keuls' test), with appropriate correction for multiple comparisons, was performed to determine which variable was different during a load of 3 or 6 cm H₂O/L/s and rest. To ensure that the data approximated a Gaussian distribution, they were logarithmically transformed if appropriate before statistical testing with ANOVA was performed. Because distribution of sample values for autocorrelation coefficients (r) is restricted to values between 1.0 and +1.0, r values underwent a Fisher's z transformation to approximate a normal distribution before parametric statistical testing was performed (19).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Mean Changes in Breath Components

Values of the breath components during rest and loaded breathing are shown in Table 1. Compared with resting breathing and a resistive load of 3 cm H₂O/L/s, a resistive load of 6 cm H₂O/L/s caused an increase in inspiratory time (TI) (p < 0.01 in both instances) and in tidal volume (VT) (p < 0.05 in both instances). Compared with a load of 3 cm H₂O/L/s, the load of 6 cm H₂O/L/s decreased respiratory frequency (f) (p < 0.05). Compared with rest, neither TE nor I changed significantly during breathing with either of the resistive loads.

Gross Variability of Breath Components

Figure 1 shows consecutive values of TI and end-tidal CO₂ tension (PET_CO2) in a representative subject; the scatter of TI increased with a resistive load of 6 cm H₂O/L/s. Standard deviations (SDs) of VT, TI, TE, f, and I for each subject during rest and loaded breathing are shown in Figure 2. Compared with resting breathing, a resistive load of 6 cm H₂O/L/s increased the SDs of TI and TE (p < 0.01 in both instances) and a load of 3 cm H₂O/L/s decreased the SDs of VT and I (p < 0.05 in both instances). Compared with a load of 3 cm H₂O/ L/s, the load of 6 cm H₂O/L/s also increased the SDs of VT, TI, TE, and I (p < 0.01 in the first two instances, p < 0.05 in the last two instances). The SDs for f did not change significantly during loaded breathing compared with rest.

View larger version (32K): [in this window] [in a new window]   Figure 1.   Time plot of end-tidal CO2 (PETCO2) and inspiratory time (TI) in a subject who was randomized to breathe with no load during the first hour, with an inspiratory resistive load of 6 cm H2O/L/s during the second hour, and with a load of 3 cm H2O/L/s during the third hour. The scatter in the values of TI increased with the highest load compared with rest.

View larger version (17K): [in this window] [in a new window]   Figure 2.   Standard deviations (SDs) of VT (L), TI (s), TE (s), f (breaths/min), and I (L/min) during resting breathing (L0) and resitive loads of 3 cm H2O/L/s (L3) and 6 cm H2O/L/s (L6) in 18 subjects. Compared with L0, L3 decreased SD of VT and I. Compared with L0, L6 increased SD of TI and TE. Compared with L3, L6 increased SDs of VT, TI, TE, and I. Circles and bars represent group mean ± SD. (*p < 0.05 for L3 versus L0; **p < 0.01 for L6 versus L0; p < 0.05 for L3 versus L6; p < 0.01 for L3 versus L6.)

The mean CVs for all breath components are shown in Table 1. Compared with resting breathing, the load of 3 cm H₂O/L/s decreased the CVs of I (p < 0.01) and VT (p < 0.05). Compared with a load of 3 cm H₂O/L/s, the load of 6 cm H₂O/L/s increased the CVs of I (p < 0.01), TI, and f (p < 0.05 in both instances).

End-tidal CO₂

PET_CO2 was 40.9 ± 2.5 mm Hg during resting breathing, 41.2 ± 1.8 mm Hg with a resistive load of 3 cm H₂O/L/s (p = NS), and 40.5 ± 2.6 mm Hg with a load of 6 cm H₂O/L/s (p = 0.22; ANOVA for the group). Incidentally, CO₂ was not added to the open circuit system.

Autocorrelation Analysis

The mean autocorrelation coefficients at a lag of one breath for the group during rest and loaded breathing are shown in Table 1. Compared with rest, a resistive load of 3 cm H₂O/L/s increased the autocorrelation coefficient of VT (p < 0.05) while a load of 6 cm H₂O/L/s decreased the autocorrelation coefficient of I (p < 0.05). The autocorrelation coefficients of TI, TE, and f did not change significantly during loaded breathing.

Figure 3 shows autocorrelograms of TI during resting breathing and with a load of 6 cm H₂O/L/s in a representative subject. The average number of consecutive breath lags displaying significant (nonzero, p < 0.01) autocorrelation coefficients for the breath components of the group during rest and loaded breathing are listed in Table 1. Compared with resting breathing, both the loads of 3 and 6 cm H₂O/L increased the number of breath lags displaying significant serial correlations of TI (p < 0.05 in both instances). VT, TE, f, and I showed no significant changes in the number of serially correlated breath lags.

View larger version (16K): [in this window] [in a new window]   Figure 3.   Autocorrelogram of TI in a subject during rest (load 0) and with a resistive load of 6 cm H2O/L/s. The number of breath lags with significant serial correlations (i.e., "short-term memory") was greater with loading (n = 7) than at rest (n = 0). On the autocorrelograms, points lying outside the inner pair of isopleths are significantly different than zero, at p < 0.05, and outside the outer pair of isopleths, at p < 0.01.

Spectral Analysis

The centroid frequencies, i.e., the mathematically weighted median frequencies of the entire spectrum (0.0-0.5 cycle/ breath), for each of the three principal breath components did not change significantly between rest and loaded breathing. The centroid frequency of VT was 0.19 ± 0.04 cycle/breath during resting breathing, 0.18 ± 0.04 cycle/breath with a resistive load of 3 cm H₂O/L/s, and 0.19 ± 0.04 cycle/breath with a load of 6 cm H₂O/L/s (p = 0.21, ANOVA for the group); the respective values for TI were 0.19 ± 0.04, 0.17 ± 0.04, and 0.19 ± 0.04 cycle/breath (p = 0.14); the respective values for TE were 0.20 ± 0.04, 0.19 ± 0.04, and 0.20 ± 0.04 cycle/breath (p = 0.39); the respective values for f were 0.17 ± 0.04, 0.16 ± 0.04, and 0.17 ± 0.05 (p = 0.44); and the respective values for I were 0.21 ± 0.05, 0.20 ± 0.04, and 0.22 ± 0.04 (p = 0.20).

Neither the number of subjects who displayed a significant peak, nor the corresponding frequency or power of the significant peaks, for each of the three breath components showed a significant change with the resistive loads of 3 or 6 cm H₂O/L/s.

Fractionation of Variational Activity of Breathing

The fractions of variational activity of breathing secondary to oscillatory behavior, correlated behavior, and uncorrelated random [w(n)] behavior for each breath component during rest, a resistive load of 3 cm H₂O/L/s, and a resistive load of 6 cm H₂O/L/s are shown in Table 2 and Figure 4. During both rest and loaded breathing, uncorrelated random [w(n)] behavior constituted > 88% of the variance of each breath component, correlated behavior represented 3 to 11%, and oscillatory behavior represented < 0.4%.

View larger version (54K): [in this window] [in a new window]   Figure 4.   Partitioning of total variances of VT (L2), TI (s2), TE (s2), and I ([L/min]2) into fractions of oscillatory (black columns), correlated (white columns), and uncorrelated random (stippled columns) behavior during resting breathing and with resistive loads of 3 and 6 cm H2O/L/s. Compared with rest, the load of 6 cm H2O/L/s increased the random fraction of TI (p < 0.01) and the correlated fraction of VT (p < 0.05). Compared with rest, the load of 3 cm H2O/L/s decreased the random fractions of TE and I (p < 0.05 in both instances). Compared with a load of 3 cm H2O/L/s, the load of 6 cm H2O/L/s increased the random fractions of VT, TI, I (p < 0.01 in all three instances), and TE (p < 0.05).

Compared with resting breathing, a load of 6 cm H₂O/L/s increased the total variational activity of TI owing to an increase in its uncorrelated random [w(n)] fraction (p < 0.01 in both instances), and it increased the correlated fraction of VT (p < 0.05) without altering its total variability. Compared with resting breathing, a resistive load of 3 cm H₂O/L/s decreased the total variational activity of TE and I due to a decrease in [w(n)] (p < 0.05 in all instances). Compared with the load of 3 cm H₂O/L/s, the load of 6 cm H₂O/L/s increased the total variational activity of TI, TE, VT, and I through an increase in their random behavior (p < 0.01 for TI, VT, and I; p < 0.05 for TE), while the correlated behavior did not change (Figure 4 and Table 2). Neither total variational activity of f, nor its subfractions, changed significantly during loaded breathing.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Compared with resting breathing, a resistive load of 3 cm H₂O/L/s caused a decrease in total variational activity of TE and I, while a load of 6 cm H₂O/L/s caused an increase in total variational activity of TI. Compared with the load of 3 cm H₂O/L/s, the load of 6 cm H₂O/L/s increased total variational activity of VT, TI, TE, and I. Partitioning of the variational activity revealed that the alterations in gross variability resulted from changes in the random uncorrelated fraction (Figure 4 and Table 2).

Effect of Resistive Loading on Mean Changes and Gross Variability of Breathing

Compared with rest and with a load of 3 cm H₂O/L/s, the resistive load of 6 cm H₂O/L/s increased mean TI and VT. The higher load decreased mean f compared with the smaller load. I and TE did not change with either load (Table 1). Our results are similar to the findings of Im Hof and coworkers (20) and Iber and coworkers (21), although these investigators used inspiratory resistive loads of 8.5 and 17 cm H₂O/L/s, respectively, which were applied for a much shorter time (up to 5 min). Prolongation of TI offers an energetically advantageous strategy of compensating for the load, in that it minimizes the flow-resistive component of the applied inspiratory pressure (22). During wakefulness, however, investigators (20, 21) demonstrated that the prolongation of TI with resistive loading could not be explained on the basis of optimization of respiratory energetics but instead reflected behavioral (cortical) adaptation. Employing psychophysical analysis, Killian and coworkers found that the role of peak airway pressure (P) in the perception of the magnitude () of a resistive load was 2.5-fold greater than that of TI ( = k · P^1.5 · TI^0.6), indicating the advantage of compensating for a resistive load by modulating TI rather than by altering P (23). During NREM-sleep (21, 24) and anesthesia (25), the lack of prolongation of TI with resistive loading further emphasizes the behavioral origin of modifications in TI.

We quantitated gross variability in terms of standard deviations (SDs) and coefficients of variation (CVs). The SDs of TI and TE increased with a load of 6 cm H₂O/L/s compared with both rest and a load of 3 cm H₂O/L/s. The SDs and CVs of VT and I decreased with a load of 3 cm H₂O/L/s compared with rest, but increased with a load of 6 compared with a load of 3 cm H₂O/L/s. We cannot compare our findings with the literature, because exposure to loaded breathing in previous studies was too brief to yield sufficient data for quantification of variability. Daubenspeck (26) employed a resistive load of 2.98 cm H₂O/L/s and measured the gross variability in breath components over 300 consecutive breaths, but the load was applied during both inspiration and expiration whereas we applied only inspiratory loads. He found a decrease of SDs of f and I with loading, while the SD of TI did not change.

Effect of Resistive Loading on Variational Activity of Breathing

Compared with rest and a load of 3 cm H₂O/L/s, a load of 6 cm H₂O/L/s increased the total variational activity of TI due to an increase in the random, unstructured fraction (Figure 4 and Table 2). As discussed earlier, TI appears to be mainly influenced by behavioral (cortical) control during loaded breathing in the awake state. Rafferty and Gardner (27) showed that subjects asked to track different respiratory patterns at a fixed level of CO₂ displayed greater freedom in the ability to voluntarily alter respiratory timing, whereas the control of VT was more tightly constrained.

It is axiomatic that behavioral control can be exerted only when loads are cortically perceived. The response to the first loaded breath, or to loads applied randomly, is less variable when a load is subliminal than if it is above the perception threshold (28, 29), suggesting that the variability of the response may act as a measure of behavioral influences on the control of breathing. In our previous study (7), two elastic loads clearly above the perception threshold produced increases in total and random variability of TI of similar magnitude to the changes caused by a resistive load of 6 cm H₂O/L/s in the present study. We considered that the changes observed during elastic loading represented behavioral attempts to minimize respiratory discomfort (7). Because even the highest resistive load, 6 cm H₂O/L/s, evoked a smaller compensatory peak airway pressure than either of the elastic loads of 9 and 18 cm H₂O/L (7), and because the level of dyspnea did not exceed moderate (3 on a modified Borg scale of 1 to 10), respiratory discomfort may have exerted a less important influence on the control of breathing in the present study.

The increase in the random variability of TI may reflect the overriding influence of behavioral factors over the automatic control of respiration resulting in a new breathing pattern under loaded conditions (7); this new pattern can be modulated by anticipatory influences (30), personality (31), and genetic trait (32). Because behavioral influences on respiratory control are multifactorial, they may be reflected by the random (unpredictable) fraction of variational activity, whereas the automatic regulation of respiration may be reflected by the correlated (predictable) fraction, in that predictability is a marker of automatic neural reflexes dictated by vital needs (2, 5). The load of 3 cm H₂O/L/s may not have evoked behavioral influences since it is close to the perception threshold. When subjects were randomly presented with resistive loads of 3 cm H₂O/L/s over 372 breaths, up to seven training sessions were necessary for subjects to reliably perceive a load of this magnitude (29), and even loads of ~ 4-5 cm H₂O/L/s are detected only about 80% of the time (10). Since loading was applied for 1 h in the present study, the feedback system that perceives mechanical loading could have undergone adaptational changes. In the presence of a sustained load (background load) the ability to perceive a further superimposed load is impaired (Weber's law) (10, 33); this further suggests that the inspiratory load of 3 cm H₂O/L/s, being applied for 1 h, was truly subliminal. Thus, automatic control of respiration may have been dominant with the load of 3 cm H₂O/L/s.

Correlated variability can be reported in three ways, each providing complementary information on breath-to-breath regulation: the autocorrelation coefficient at a lag of one breath describes the relationship between a breath and its immediate predecessor, the number of significantly correlated breath lags provides a measure of "short-term memory," and the employment of a multiregressive model indicates the fraction of total variational activity due to correlated behavior (3, 4). Both the resistive loads of 3 and 6 cm H₂O/L/s increased the number of breath lags displaying significant serial correlations ("short-term memory") for TI (Figure 3 and Table 1). Compared with rest, the load of 3 cm H₂O/L/s increased the autocorrelation coefficient at a lag of one breath for VT and the load of 6 cm H₂O/L/s increased the correlated fraction of variational activity of VT. That is, correlated variability increased with resistive loads above and below the perception threshold, and this response may reflect the influence of automatic (subcortical) influences on the respiratory controller dictated by the vital need to compensate for the load (34).

The proximity of the smaller resistive load to the perception threshold can explain the apparent paradox that the smaller resistive load decreased the random and total variational activity of two breath components (TE and I) compared with rest, whereas the higher load increased the random and total variational activity of four breath components (VT, TI, TE, and I) compared with the smaller load (Figure 4 and Table 2). Respiratory control during resting wakefulness is accepted as being conjointly influenced by automatic and behavioral factors. A load below or around the perception threshold would be expected to emphasize automatic regulation and to decrease random variability, while eventually increasing correlated variability. A suprathreshold load necessarily activates behavioral (cortical) control, and by recruiting load-related behavioral influences, may have been responsible for the increased random variability of the four breath components with a load of 6 cm H₂O/L/s. Increasing behavioral control with an increasing load is supported by Steven's law, which defines the relationship between the magnitude of perception and the size of a load as a power function: as a load increases, the influence of perception, and thus cortical control, increases in an exponential rather than linear fashion (10, 23, 33). This power function has been shown to also hold true for electrical potentials evoked in the cerebral cortex in response to resistive loading (35).

Compared with rest, a resistive load of 3 cm H₂O/L/s decreased total variational activity of TE and I due to a decrease in the random fraction. The reduction in the random fraction of the variational activity of TE and I with a load of 3 cm H₂O/L/s supports the view that automatic respiratory control is more dominant with subliminal loads than with resting wakefulness. Wakefulness constitutes a tonic cortical input to the controller and this wakefulness drive may be reflected by the random fraction of variational activity at rest; in contrast, a subliminal load could partly supersede this wakefulness input by engaging vital automatic reflexes that override this wakefulness input through the neural mechanisms of gating and presynaptic inhibition (14). Thus, a load of 3 cm H₂O/L/s could reduce the wakefulness input, leading to a decrease in the random fraction of variability compared with rest.

In summary, a resistive load of 6 cm H₂O/L/s increased the total variational activity of all breath components compared with a resistive load of 3 cm H₂O/L/s. Compared with rest, the smaller load decreased the total variational activity of TE and I, whereas the greater load increased the total variational activity of TI. For each breath component, changes in total variational activity were mediated through the random fractions. The apparent paradox of a smaller load decreasing random variability and a higher load increasing random variability can be explained by differences in the perception of the two loads. A sustained resistive load of 3 cm H₂O/L/s is near the perception threshold, whereas a load of 6 cm H₂O/L/s is perceived during every inspiration. With both loads, the number of breath lags exhibiting significant serial correlations ("short-term memory") for TI increased. Compared with rest, the smaller load increased the autocorrelation coefficient at a lag of 1 breath for VT and the higher load increased the correlated fraction of VT. Thus, both loads similarly increased correlated behavior that may represent the automatic (subcortical) influences on the respiratory controller which are independent of load perception. Subliminal loading is likely to emphasize the automatic regulation of breathing, whereas behavioral (cortical) influences prevail with suprathreshold loads. We speculate that the fractions of variational activity have different physiologic implications: unstructured random variability may be a measure of behavioral influences, whereas the structured correlated fraction may represent automatic influences on respiratory control.