Abstract
Measurement of the acute hypoxic ventilatory response (AHVR) requires careful choice of the hypoxic stimulus. If the stimulus is too brief, the response may be incomplete; if the stimulus is too long, hypoxic ventilatory depression may ensue. The purpose of this study was to compare three different techniques for assessing AHVR, using different hypoxic stimuli, and also to examine the betweenday variability in AHVR. Ten subjects were studied, each on six different occasions, which were ≥1 wk apart. On each occasion, AHVR was assessed using three different protocols: 1) protocol SW, which uses square waves of hypoxia; 2) protocol IS, which uses incremental steps of hypoxia; and 3)protocol RB, which simulates an isocapnic rebreathing test. Mean values for hypoxic sensitivity were 1.02 ± 0.48, 1.15 ± 0.55, and 0.93 ± 0.60 (SD) l ⋅ min^{−} ^{1} ⋅ %^{−} ^{1}for protocols SW, IS, and RB, respectively. These differed significantly (P < 0.01). The coefficients of variation for measurement of AHVR were 20, 23, and 36% for the three protocols, respectively. These were not significantly different. There was a significant physiological variation in AHVR (F _{50,100} = 3.9, P < 0.001), with a coefficient of variation of 26%. We conclude that there was relatively little systematic variation between the three protocols but that AHVR varies physiologically over time.
 acute hypoxic ventilatory response
 ventilation
the ventilatory response of humans to isocapnic hypoxia is triphasic. First, there is a rapid increase in expired minute ventilation (V˙e )known as the acute hypoxic ventilatory response (AHVR), which occurs with a time constant in the order of seconds (6, 16). Second, there is a fall in V˙eknown variously as hypoxic ventilatory depression (HVD), hypoxic ventilatory decline, or ventilatory “rolloff” with a time constant in the order of minutes (4, 16). Third, there is a progressive rise inV˙e with a time constant in the order of hours (10) that appears to be related to ventilatory acclimatization to altitude. A major methodological problem in studies relating to the first component of the response (AHVR) has been obtaining adequate sets of data while avoiding contamination of results from the later phases of the response to hypoxia.
Most techniques used to assess AHVR can loosely be classified into four main groups. Steadystate methods are really those in which the inspired gases are fixed; the experimenter then waits for ventilation and for the endtidal gases to settle before determining the response (2, 14). The development of these techniques predates our current understanding of HVD, and it is essentially inevitable that data that have been collected using such techniques have been “contaminated” by a varying degree of HVD. The second group of methods involves either transient hypoxia or hyperoxia, lasting no more than a few breaths (3, 5, 13). These methods avoid the problems associated with HVD, but the duration of the stimulus is so brief that the methods do not allow the ventilatory response to fully develop (19). Furthermore, because the response only involves a few breaths, the data are very susceptible to random variations inV˙e. The third group of methods involves progressive hypoxia such as can be achieved with a rebreathing circuit (19, 26). By careful choice of the rebreathing volume, these methods are likely to generate a more appropriate total exposure to hypoxia for measuring AHVR than either the steadystate or transient techniques. By careful manipulation of an absorbing circuit, endtidal Pco
_{2}(
In relation to gauging an appropriate amount of hypoxia for assessing AHVR, Mou et al. (15) used an endtidal forcing technique to compare the ventilatory response to ascending and descending steps of hypoxia, each of 50s duration, that were evenly spaced in terms of the saturation of arterial blood (
AHVR varies widely among individuals (7). However, repeat determinations of AHVR within an individual are often also very different. In relation to this, Sahn et al. (22) reported that the betweenday variability in AHVR was greater than the withinday variability in AHVR in seven of eight subjects studied and that this difference was significant in four of the eight subjects. This finding would seem to suggest that there is a genuine slow physiological variation in AHVR within subjects over time. A second aim of this study was to try to confirm the findings of Sahn et al. in relation to our subjects.
METHODS
Subjects
Ten healthy adults (8 men and 2 women), mean age of 23.4 ± 3.9 (SD) yr, mass of 71.3 ± 11.9 kg, and height of 1.77 ± 0.09 m, volunteered to take part in the study. Subjects received both a written and verbal explanation of the experiment before they gave their consent. This study was approved by the Central Oxford Research Ethics Committee.
Protocols
Each subject was studied on six different occasions, which were separated from one another by at least 1 wk. Subjects were either always studied in the morning, starting at 9:00 AM, or always in the afternoon, starting at 1:00 PM. Female subjects were only studied in the first 2 wk of their menstrual cycle. On each study day, AHVR was measured three times, once using each of three different techniques or protocols. The protocols for determining AHVR were between 12 and 18 min in duration and were undertaken in varied order. The protocols were separated from one another by 60 min during which time the subject sat quietly while breathing room air. This design resulted in a total of 180 separate determinations of AHVR with 60 determinations associated with each of the different protocols. For any given individual, there was a total of 18 measurements of AHVR with 6 measurements associated with each protocol.
The individual protocols for determining AHVR are described below. For each protocol, there was a 5min period immediately preceding it in which the subject's endtidal Po
_{2}
Squarewave protocol (protocol SW).
The hypoxic stimulus associated with this protocol consisted of six square waves in
Incremental step protocol (protocol IS).
The hypoxic stimulus associated with this protocol consisted of seven different levels of
Simulated rebreathing protocol (protocol RB).
The hypoxic stimulus associated with this protocol involved a linear reduction in
Experimental Technique
Before the series of AHVR determinations began, each subject undertook one or two preliminary experiments at the laboratory, which familiarized them with the apparatus and procedures. Subjects were requested to refrain from alcohol and caffeinecontaining beverages starting from the evening before each experimental day. Once at the laboratory, subjects sat quietly for a period of at least 15 min before any measurements were made. After this, the subject's value for
During each determination of AHVR, the subject was seated comfortably in an upright position and breathed through a mouthpiece with the subject's nose occluded by a clip. A pulse oximeter (Ohemeda Biox 3740) was attached to the forefinger to monitor
Once detected by the data acquisition computer, the endtidal gas values were passed to a second computer that controlled a fastresponding gasmixing system. The measured endtidal values were compared with the desired endtidal values; the composition of a new inspiratory gas mixture was then calculated so as to reduce the error for the following breath. The controlling computer adjusted the inspired partial pressures of N_{2}, O_{2}, and CO_{2} through a series of outlet valves connected to a gas supply. The gasmixing system and control scheme have been described in greater detail elsewhere (11, 21).
Data Processing
For each protocol, data from the 4th and 5th min of the 5min period immediately preceding the hypoxic stimulation were used to obtain CO_{2} production (V˙co _{2}) and O_{2}consumption (V˙o _{2}). The calculations associated with obtaining these values have been described elsewhere (17).
Estimation of AHVR from protocol SW.
To quantify AHVR from these data, the responses to the squarewave hypoxia were fitted by a single compartment model of the ventilatory response to hypoxia, as detailed by Clement and Robbins (1). The use of such a dynamic model enables the whole of the data to be used for estimating AHVR, rendering it unnecessary to select portions of the response when V˙e is approximately steady. In this model, a linear steadystate relationship betweenV˙e and hypoxia was obtained by using 100 −
The parameters of the model were estimated from the data by nonlinear regression using the NAG (Numerical Algorithms Group, Oxford, UK) Fortran library routine E04FDF to minimize the sum of squares of the residuals. All the parameters were constrained to be >0, and the dynamic parameters were constrained to be <30 s. The squared residual was weighted by the breath duration to avoid the fit being weighted toward periods of higherfrequency V˙e.
Estimation of AHVR from protocol IS.
V˙e and
Estimation of AHVR from protocol RB.
Breathbybreath values for
Statistical Analysis
Differences in the responses to hypoxia between the protocols were assessed statistically using ANOVA, treating subjects as a random factor and protocol as a fixed factor and including the interactive term. To assess whether there was a physiological variation in the responses over time, the variance was partitioned so as to compare the withinsubject, betweenday variance with the withinsubject, withinday variance. Post hoc comparisons were undertaken with Bonferroni's correction to allow for multiple comparisons. Correlations between variables were assessed using the Pearson correlation coefficient. All statistical analysis was undertaken by using the SPSS statistical software package (Chicago, IL). Differences were considered significant at P < 0.05.
RESULTS
Subjects
All the subjects completed the study, and none reported discomfort from the measurement of AHVR during any of the protocols. Table1 gives the mean values together with standard deviations for each subject for the measurements of initial airbreathing
Quality of Gas Control
Figure 1, A and B, shows breathbybreath plots for
Ventilatory Responses
General.
Figure 1
C shows breathbybreath plots forV˙e against time from a single experiment in one subject (1091) for each of the three protocols ( protocols SW, IS, and RB). Forprotocol SW, there was a clear variation in ventilation caused by the square waves in
The individual values obtained for Gp and V˙s for each subject on each test day were plotted in Fig.2. From Fig. 2, it is clear that different subjects possess different mean values. This finding was highly significant for both Gp and V˙s for all three protocols (P < 0.001, ANOVA). Also apparent is the considerable daytoday variation in the measurements obtained. This had no significant relationship with the sequential number of the test (ANOVA).
Comparison between protocols.
Table 2 lists the mean values for Gp andV˙s for each subject, together with the overall mean values for each of the three protocols. There were significant differences between protocols for Gp (P < 0.01) but not forV˙s. Post hoc analysis revealed that the ∼20% difference in value for Gp between protocols IS and RBwas significant but that the intermediate value for Gp obtained fromprotocol SW did not differ significantly from the values obtained in either protocol IS or RB.
For protocols IS and RB, although Gp andV˙s were determined by linear regression without modelling the dynamics, there is no intrinsic reason why the model used to determine Gp and V˙s from protocol SW cannot also be used with these data. Mean values for Gp obtained in this manner were 1.19 ± 0.58 (SD) l ⋅ min^{−} ^{1} ⋅ %^{−} ^{1}and 1.18 ± 0.76 l ⋅ min^{−} ^{1} ⋅ %^{−} ^{1}for protocols IS and RB, respectively, and mean values for V˙s were 15.1 ± 4.0 l/min and 15.2 ± 3.7 l/min, respectively. For protocol IS, the difference in results between the two techniques was small. For protocol RB, the differences between the two techniques were much more substantial. With the use of the dynamic model, the values obtained for Gp were ∼27% higher (P < 0.005) and those for V˙s were ∼5% lower (P < 0.001) than those obtained using linear regression. ANOVA revealed no significant differences between the three protocols for either Gp or V˙s when all parameters had been obtained using the model that included the dynamics.
An idea of the inherent variability associated with each protocol for determining Gp and V˙s can be obtained from the mean squared error remaining after the betweensubject variability is removed. This variance may be thought of as the sum of variance associated with the betweentest variation and the betweenday variability of subjects. The latter should be approximately constant across methods and may be estimated as 0.07 (l ⋅ min^{−} ^{1} ⋅ %^{−} ^{1})^{2}for Gp and 8.9 (l/min)^{2} for V˙s (see next section). For Gp, the mean squared error remaining for protocol SW was 0.11 (l ⋅ min^{−} ^{1} ⋅ %^{−} ^{1})^{2};forprotocol IS, it was 0.14 (l ⋅ min^{−} ^{1} ⋅ %^{−} ^{1})^{2}; and, for protocol RB, it was 0.18 (l ⋅ min^{−} ^{1} ⋅ %^{−} ^{1})^{2}. Although these results suggest that protocol SW yielded the most repeatable results for Gp and protocol RB the least, the differences did not quite reach significance. By subtracting the betweenday variance for subjects, estimates for the actual betweentest variance may be obtained as 0.04, 0.07, and 0.11 (l ⋅ min^{−} ^{1} ⋅ %^{−} ^{1})^{2}for protocols SW, IS, and RB, respectively. These yield coefficients of variation of 20, 23, and 36% for the three protocols, respectively.
For V˙s, the mean squared error remaining forprotocol SW was 10.0 (l/min)^{2}; for protocol IS, it was 16.0 (l/min)^{2}; and, for protocol RB, it was 21.5 (l/min)^{2}. For this variable, the mean squared error associated with protocol SW was significantly smaller than with either protocol IS (P < 0.05) orprotocol RB (P < 0.01). By subtracting the estimate for the variance associated with the betweenday variability of 8.9 (l/min)^{2} (see next section), estimates for the actual betweentest variance may be obtained as 1.1, 7.1, and 12.6 (l/min)^{2} for protocols SW, IS, andRB, respectively. These yield coefficients of variation of 7, 18, and 22% for the three protocols, respectively.
Daytoday variability in values for Gp and V˙s within subjects.
To examine the physiological variability in Gp and V˙s over time, we first compared the withinsubject, betweenday mean squared error with the withinsubject, withinday mean squared error. This was done by first removing the variability associated with subjects and protocols and the interactive term between these two factors and then partitioning the remaining error into betweenday and withinday components. The withinsubject, betweenday mean squared error for Gp was 0.283 (l ⋅ min^{−} ^{1} ⋅ %^{−} ^{1})^{2}and that for V˙s was 33.7 (l/min)^{2}. The withinsubject, withinday mean squared error for Gp was 0.073 (l ⋅ min^{−} ^{1} ⋅ %^{−} ^{1})^{2}and that for V˙s was 6.9 (l/min)^{2}. For both Gp and V˙s, the withinsubject, betweenday mean squared error was significantly larger than the withinsubject withinday mean squared error (Gp: F _{50,100} = 3.9,P < 0.001; V˙s: F _{50,100} = 4.9, P < 0.001). The expectation for the withinsubject, betweenday mean squared error is that it should equal three times the physiological betweenday variance (because there are three observations in a day) plus the residual mean squared error [0.073 (l ⋅ min^{−} ^{1} ⋅ %^{−} ^{1})^{2}for Gp and 6.9 (l/min)^{2} for V˙s]. Thus the physiological betweenday variance for Gp may be calculated as 0.070 (l ⋅ min^{−} ^{1} ⋅ %^{−} ^{1})^{2}and that for V˙s may be calculated as 8.9 (l/min)^{2}. These variances are equivalent to coefficients of variation for the physiological betweenday variability of 26% and 19% for Gp and V˙s, respectively.
An alternative, more graphic representation of the betweenday variability is given in Fig. 3. Figure 3shows the daytoday deviations around the mean values for Gp andV˙s for each subject, plotted so that the deviations from different protocols on the same day can be compared. The data from all plots for both Gp and V˙s show a positive correlation, such that large values for either Gp orV˙s obtained on a given day from one protocol tend to be associated with large values for either Gp or V˙s obtained on the same day from the other two protocols and vice versa.
Relationship between Gp and V˙s.
For single determinations of Gp and V˙s fromprotocol SW, the correlation matrix between model parameters obtained as part of the fitting process did not reveal any consistent relation between the value returned for Gp and the value returned forV˙s (R = −0.06 ± 0.23, mean ± SD). For single determinations of Gp and V˙s fromprotocols IS and RB, the correlation between Gp andV˙s was significantly negative (R = −0.88 ± 0.00 and R = −0.84 ± 0.00, respectively). These results suggest that the model and the process of fitting the model to the data introduce either no correlation or else quite a strong negative correlation between the values for Gp andV˙s.
The results from a single fit of the model to the data contrast with those obtained when the variation in Gp around the subject mean on different days is compared with the variation in V˙s around the subject mean on different days (Fig.4). For protocol SW, the variation in value for Gp around the mean for each subject was very significantly correlated (R = 0.65, P < 0.001) with the variation in value for V˙s around the mean for each subject.For protocols IS and RB, the strong negative correlation that existed between values for Gp and V˙s from individual determinations was lost. For protocol IS, no correlation was detected (R = 0.00), and for protocol RB there was instead a weak, positive correlation (R = 0.28, P < 0.05). Because, for individual fits of the model to the data, the correlation between Gp and V˙s was either absent (protocol SW) or strongly negative (protocols IS and RB), these results suggest a physiological effect such that on days when the value for Gp was higher then so was the value for V˙s.
Relationship of daytoday variability in Gp and V˙s to daytoday variability in
P ET CO 2
and metabolic rate.
Values for Gp and V˙s obtained on the same day were first averaged across the three protocols. The variations in both Gp and V˙s around their mean values (separate mean values were calculated for each subject) did not correlate significantly with the variations around the subject means for
DISCUSSION
This study provides two main findings. The first is that there was relatively little systematic variation between the three protocols for determining AHVR, and that which does occur seems to be related to data handling rather than to real physiological differences. Protocol ISgave the highest mean value for AHVR, which may suggest that this form of hypoxic exposure is least affected by the problems associated with the hypoxic exposure being too short or too long, both of which would tend to reduce the measured value for AHVR. Protocol SW was associated with the lowest variance or greatest repeatability for the measurements. The second main finding is that there was real physiological variation over time in the values for both Gp and V˙s.
Comparison Between Protocols
Overall, the mean values obtained from the different protocols for both Gp and V˙s did not differ very greatly. As outlined in the introduction, a major problem with measuring AHVR is that the hypoxic exposure associated with the measurement may itself affect the sensitivity. In relation to protocol IS, a separate study showed that the results obtained are not excessively influenced either by an inadequate period of time for the response to develop or by an excessive exposure to hypoxia (15). Thus the similarity in results between protocol IS and protocols SW and RB of the present study also suggests that protocols SW andRB do not suffer from either an inadequate or excessive exposure to hypoxia.
The one significant difference detected was that the value for Gp fromprotocol RB was ∼20% below that associated with protocol IS. This difference vanished when Gp was estimated fromprotocol RB via the dynamic model associated with protocol SW rather than via simple linear regression. In contrast, values obtained for Gp from protocol IS were affected very little by whether they were estimated via linear regression or through the dynamic model. These findings led us to hypothesize that determinations of Gp via linear regression of V˙e against desaturation were more likely to be influenced by the dynamics of the ventilatory response when it was curvilinear against time (as inprotocol RB) than when the response was linear against time (as in protocol IS). To test this hypothesis, we simulated the noisefree ventilatory response to protocols RB and ISusing the dynamic model together with the mean parameter values obtained from protocol SW. We then attempted to recover the value for Gp using linear regression of V˙eagainst desaturation. For protocol IS, the value for Gp that was recovered was within 1% of the value used to generate the data, but, for protocol RB, the value that was recovered was 14% lower than the value used to generate the data.
One question that does arise is whether any of the three protocols may be considered to have been optimized in relation to the dynamics of the hypoxic exposure. In relation to protocol IS, a previous study using an identical sequence in
Overall, we consider that all three protocols were reasonably satisfactory for determining Gp. Protocol RB did appear to underestimate Gp slightly, unless the dynamics of the response were taken into account, and it was also associated with the largest variance. However, it is also probably the easiest function to generate without access to the technique of dynamic endtidal forcing.
BetweenDay Variability in Gp and V˙s
Sahn et al. (22) previously reported that variance for AHVR was greater when determined from a set of measurements on different days compared with measurements made on the same day, and our results confirm this observation. They found no significant relationship between AHVR andV˙o _{2},V˙co _{2}, temperature, Pco _{2} , or bicarbonate levels. They did find a significant inverse relationship between AHVR and an “arterialized” venous pH, which in turn correlated with Pco _{2}. They considered that this might provide a partial explanation for the daytoday variability in AHVR.
As part of the protocol for the present study, we decided we should assess
Both Gp and V˙s are likely to vary with metabolic rate. Assessments of V˙co
_{2} andV˙o
_{2} were obtained during the control periods of breathing for each of the protocols. There was a weak correlation of Gp withV˙co
_{2} but notV˙o
_{2}. V˙s correlated with neither V˙co
_{2}nor V˙o
_{2}. As for the variations in
Gp and V˙s both increase considerably after exposure to hypoxia (9, 25), and recent unpublished observations from our laboratory suggest that more moderate levels of hypoxia or even hyperoxia can influence these values. Thus a further possibility is that the variations in Gp and V˙s arose through variations in
In summary, we found no large differences in measured values for AHVR among the three different techniques studied. We have, however, confirmed the presence of substantial betweenday variation in the respiratory controller. The origins of this variation remain to be determined.
Acknowledgments
S. Zhang thanks Professor T. Stapleton, who provided financial support. We thank Dr. M. Fatemian for assistance with programming, Dr. F. Marriott for help with statistical analyses, and D. F. O'Connor for expert technical assistance.
Footnotes

Address for reprint requests and other correspondence: P. A. Robbins, Univ. Laboratory of Physiology, Parks Road, Oxford OX1 3PT, UK (Email:peter.robbins{at}physiol.ox.ac.uk).

This study was supported by the Wellcome Trust.

Original submission in response to a special call for papers on “Hypoxia Influence on Gene Expression.”

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. §1734 solely to indicate this fact.
 Copyright © 2000 the American Physiological Society