Automatic classification of activity and apneas using whole body ...

J Appl Physiol 98: 365–370, 2005; doi:10.1152/japplphysiol.00803.2004.

Innovative Methodology

Automatic classification of activity and apneas using whole body plethysmography in newborn mice B. Matrot,1 E. Durand,1 S. Dauger,1,2 G. Vardon,3 C. Gaultier,1,4 and J. Gallego1 1

Institut National de la Sante´ et de la Recherche Me´dicale E9935, 2Service de Pe´diatrie-Re´animation, and 4Service de Physiologie, Hoˆpital Robert-Debre´, Paris; and 3Universite´ de Picardie, Environnement Toxique Pe´rinatal et Adaptations Physiologiques et Comportementales, Amiens, France Submitted 29 July 2004; accepted in final form 23 September 2004

Matrot, B., E. Durand, S. Dauger, G. Vardon, C. Gaultier, and J. Gallego. Automatic classification of activity and apneas using whole body plethysmography in newborn mice. J Appl Physiol 98: 365–370, 2005; doi:10.1152/japplphysiol.00803.2004.—An increasing number of studies in newborn mice are being performed to determine the mechanisms of sleep apnea, which is the hallmark of early breathing disorders. Whole body plethysmography is the method of choice, as it does not require immobilization, which affects behavioral states and breathing. However, activity inside the plethysmograph may disturb the respiratory signal. Visual classification of the respiratory signal into ventilatory activity, activity-related disturbances, or apneas is so time-consuming as to considerably hamper the phenotyping of large pup samples. We propose an automatic classification of activity based on respiratory disturbances and of apneas based on spectral analysis. This method was validated in newborn mice on the day of birth and on postnatal days 2, 5, and 10, under normoxic and hypoxic (5% O2) conditions. For both activity and apneas, visual and automatic scores showed high Pearson’s correlation coefficients (0.92 and 0.98, respectively) and high intraclass correlation coefficients (0.96 – 0.99), supporting strong agreement between the two methods. The present results suggest that breathing disturbances may provide a valid indirect index of activity in freely moving newborn mice and that automatic apnea classification based on spectral analysis may be efficient in terms of precision and of time saved. phenotyping; breathing; Fourier transform; signal processing SLEEP APNEA IS THE HALLMARK of early breathing disorders, especially in preterm infants. Apneas are associated with repeated hypoxemic episodes, which can cause long-term morbidity, including cognitive dysfunction (14, 18). Frequent apneas may be caused by immaturity of central networks controlling breathing, an abnormal fetal environment [e.g., nicotine exposure (23)], or gene mutations (15). An increasing number of respiratory studies in newborn mice have been done to determine the mechanisms of apnea (16). In these studies, early respiratory testing is mandatory because the respiratory phenotype undergoes rapid changes during postnatal development (10, 21). Thus the adult respiratory phenotype may not adequately reflect the respiratory disturbances seen during early development. Furthermore, in genetic studies, homozygous deletion of most of the candidate genes caused death within a few hours after birth (16). However, breathing pattern measurement in newborn mice is difficult due to the small body volume [1–2 ml at 2 days of

Address for reprint requests and other correspondence: J. Gallego, Laboratoire de Neurologie et Physiologie du De´veloppement, INSERM-E9935, Hoˆpital Robert-Debre´, 48 Bd Se´rurier, 75010 Paris, France (E-mail: gallego@ rdebre.inserm.fr). http://www. jap.org

postnatal age (PNA2)] (11) and tidal volume (8 ␮l on PNA2) (1, 25). Currently available methods consist of head-out plethysmography (22) and whole body barometric plethysmography (3, 6, 13, 17, 22). In a whole body plethysmograph, the pups move freely, so that measurements are not influenced by the effects of immobilization on behavioral states and breathing (a crucial point in studies on sleep-related respiratory disorders). On the other hand, pressure changes caused by motor activity may strongly disturb the respiratory signal. Respiratory signal classification into activity-related disturbances, apneas, or ventilatory activity is done visually. This time-consuming task considerably hampers the phenotyping of large pup populations. The aim of this study was to design and to validate a method for automatically classifying respiratory signals obtained by whole body plethysmography in freely moving newborn mice. We recorded breathing patterns in mice from the day of birth to postnatal age 10 days (PNA10), in normoxia and in hypoxia (to cause a ventilatory decline and apneas). First, our software automatically classified activity-related disturbances of the respiratory signal, and we assessed the validity of these disturbances as an indirect index of activity. Second, our software classified apneas using spectral analysis of the respiratory signal. We compared visual and automatic classification of activity and apnea scores using various statistical approaches, analyses of variance, correlation, Bland and Altman analyses (4), and the intraclass correlation coefficient (ICC) to assess agreement between the two methods. METHODS

Animals Mouse pups (n ⫽ 70) were obtained from Swiss female mice (IFFACREDO, L’Arbresle, France) housed at 24°C with a 12:12-h light-dark cycle and were fed ad libitum. In Experiment 1, the mean weights per age group (n ⫽ 10) were 1.66 ⫾ 0.07, 1.88 ⫾ 0.08, 2.92 ⫾ 0.12, and 4.94 ⫾ 0.37 g on postnatal age 0 days (PNA0), PNA2, PNA5, and PNA10, respectively. In Experiment 2, the mean weights per age group (n ⫽ 10) were 2.01 ⫾ 0.10, 2.62 ⫾ 0.28, and 4.80 ⫾ 0.60 g on PNA2, PNA5, and PNA10, respectively. Experimental protocols met the animal research guidelines established by the Institut National de la Sante´ et de la Recherche Me´dicale (French National Institute for Health and Medical Research). Respiratory Signal Acquisition Ventilatory records were obtained by using a previously described whole body flow barometric plethysmograph (13). The plethysmo-

The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

8750-7587/05 $8.00 Copyright © 2005 the American Physiological Society

365

Innovative Methodology 366

AUTOMATIC PROCESSING OF THE RESPIRATORY SIGNAL

Fig. 1. Automatic classification of activity (A) and apneas (B). A, top: respiratory trace in a 5-day-old pup in hypoxia. Activity caused sharp disturbances in the respiratory signal due to confounding effects of movements, positional changes inside the chamber, and actual changes in breathing pattern. The increasing drift of the respiratory signal during apnea was caused by dynamic changes in the balance of pressure between the 2 chambers and had no physiological significance. Apnea detection was insensitive to this effect. Middle: during activity, the disturbance criterion (see text for details) increased above the threshold (arrow, dashed line). Bottom: the apnea criterion (in log scale) remained unchanged. B, top: respiratory trace in a 5-day-old pup during the hypoxic decline. Ventilation was depressed and interrupted by apneas (note the different scales of respiratory traces in A and B). Middle: during apnea, the disturbance criterion remained unchanged. Bottom: the spectral power profile at each point of the time derivative of respiratory signal fell below the threshold (arrow, dashed line; see text for details).

graph was composed of two Plexiglas cylinders serving as the animal (40 ml) and reference (100 ml) chambers, respectively, immersed in a thermoregulated water bath set that maintained their temperature at 32.8°C. The mean temperature obtained by inserting a probe inside several litters of 1-h-old newborn mice in contact with the mother was 32°C (21). Our laboratory previously reported that body temperature in 6-day-old restrained newborn pups exposed to intermittent hypoxia inside the plethysmograph varied within a narrow range (from 33.5 to 33.8°C) (13). A 50 ml/min flow of dry air (Bronkhorst Hi-Tec airflow stabilizer, Uurlo, Holland) was divided into two 25 ml/min flows through the chambers, thus avoiding CO2 and water accumulation. The differential pressure between the two chambers (EFFA transducer, Asnie`res, France; range ⫾0.1 mb) was filtered (bandwidth, 0.05–15 Hz at ⫺3 dB), converted to a digital signal (Instrunet model 200, 14-bit converter, GW-Instruments, Somerville, MA) at a sample rate of 100 Hz, and processed by custom-written software (Software Superscope II, GW-Instruments). The time constant of the pressure decay within the system (2 s) was measured by injecting 2 ␮l into the measurement chamber. The system allowed measurement of breathing frequencies within the 2- to 15-Hz range. Calibration was done before each session by injecting 2 ␮l of air into the animal chamber from a microsyringe (Ito). The pressure rise induced by this injection was of similar magnitude to that induced by a newborn mouse. The respiratory signal was filtered for automatic processing by using 10-data point rectangular smoothing (i.e., moving average) and was normalized by linear transformation so that the entire data point sample had zero mean and unit root mean square. After automatic classification of apnea and activity-related disturbances (see below), we restored the original signal by linear inverse transformation.

Experiment 2. We exposed the pups to a strong hypoxic stimulus to cause a ventilatory decline and apneas. The pups were tested on PNA2 (n ⫽ 10), PNA5 (n ⫽ 10), and PNA10 (n ⫽ 10). After 1 min of familiarization with the apparatus, baseline ventilation was recorded for 3 min (normoxia). Then the airflow through the plethysmograph was automatically switched to a hypoxic flow (O2, 5%; N2, 95%) for 3 min (hypoxia) and back to air for 6 min (posthypoxic normoxia). Activity-related Disturbances These disturbances were characterized by sharp changes in the baseline respiratory signal (Fig. 1). The following criterion function [disturbance compliance (Cdist)] was calculated on a breath-by-breath basis to detect these changes: Cdist ⫽ 2 䡠

兩VI ⫺ VE兩 共VI ⫹ VE兲

where VI and VE were the magnitude of the inspiratory and expiratory limbs of the volume signal, respectively. This criterion function was

Design Experiment 1. Apneas and movement were studied during spontaneous breathing in normoxic conditions. After 1 min of familiarization with the apparatus, breathing was recorded for 20 min on the day of birth PNA0 (n ⫽ 10), PNA2 (n ⫽ 10), PNA5 (n ⫽ 10), and PNA10 (n ⫽ 10). Pup behavior was video-recorded through the wall of the plethysmograph by using a Sony Hi-8 digital camera for off-line determination of activity periods. J Appl Physiol • VOL

Fig. 2. Activity scores (expressed as percent recording time) estimated from video recordings by 2 independent scorers and automatically from the respiratory signal (Experiment 1). Values are means ⫾ SE; in each age group, n ⫽ 10. Visual and automatic scores were very similar at all study ages.

98 • JANUARY 2005 •

www.jap.org

Innovative Methodology AUTOMATIC PROCESSING OF THE RESPIRATORY SIGNAL

367

Fig. 3. A: scatterplot of automatic vs. visual activity scores in newborn mice on postnatal age 0 days (PNA0), 2 days (PNA2), 5 days (PNA5), and 10 days (PNA10). Each symbol represents 1 pup (n ⫽ 10 per age group). The 2 visual scores were averaged. Pearson’s product-moment correlation over the whole sample (n ⫽ 40): 0.92. Solid line, regression line; dotted line, identity line. B: Bland and Altman plot (difference against mean). The mean ⫾ SD difference between automatic and visual activity scores (continuous line) was ⫺0.21 ⫾ 3.80. The lower and upper limits of agreement (dotted lines) were ⫺7.77 (95% confidence interval, ⫺9.86 to ⫺5.70) and 7.35 (95% confidence interval, 5.25 to 9.43), respectively.

then filtered by using 10-data point rectangular smoothing and interpolated (Matlab Interp1 procedure, Mathworks, Natick, MA) to match the sample rate of the respiratory signal. Activity periods were detected whenever Cdist was above a threshold calculated for each pup. To calculate this threshold, we determined the frequency value histogram of Cdist values over the entire respiratory recording, and we fitted this histogram by using first-order decreasing exponential regression. The Cdist threshold was empirically defined as the time constant of this exponential (corresponding to a 63% fall in frequency distribution). The total activity time was expressed as the percentage of the total respiratory recording time (and hereafter is called the automatic activity score). For automatic and visual classification, we discarded disturbance periods shorter than 5 s to avoid contaminating the activity scores with motor twitches, external pressure disturbances, or gas switches (in Experiment 2). Visual Activity Scores Two experienced scorers independently examined the video recordings of all pups (Experiment 1) and determined the onset and the

Table 1. Agreement between visual and automatic activity scores in newborn mice Agreement Between Activity Scores Age, days

Pearson’s r

ICC

0 2 5 10 Overall sample

0.78 0.78 0.88 0.97 0.92

0.88 (0.54–0.97) 0.86 (0.49–0.96) 0.90 (0.64–0.97) 0.98 (0.89–0.99) 0.96 (0.92–0.98)

n ⫽ 10 Mice per age group. ICC, intraclass correlation coefficient (with 95% confidence interval). See Fig. 2 for mean activity scores per group and Fig. 3 for individual data. J Appl Physiol • VOL

end of each activity period. The total duration of disturbances was expressed as the percentage of the total recording time (and hereafter is called the visual activity score). Visual and Automatic Classification of Apneas Apnea was defined as two consecutive missed breaths, i.e., as absence of the respiratory signal for at least twice the mean breath duration in baseline normoxic conditions. The same two experienced scorers scrolled each respiratory recording from each pup to detect apneas visually. No attempt was made to distinguish apneas from hypopneas (amplitude changes ⬍25% of baseline tidal volume were classified as apneas). The total apnea duration was expressed as the percentage of the total recording time (visual apnea score). We used short-time Fourier transform (STFT; Matlab, Wavelab toolbox, Statistics Department, Stanford University) to calculate the spectral power profile at each point of the time derivative of the plethysmographic signal. The window size and displacement were set at 0.32 and 0.04 s, respectively; this allowed frequency analysis from 0 to 50 Hz, which encompasses the range of breathing frequencies in newborn mice. STFT maps the signal from the time domain into a joint time-frequency plane. Apneas were characterized by large bandwidths of the STFT spectrogram. The STFT spectrogram point-by-point variance was calculated, and the value of the upper threshold of variance was empirically set at 5. Between-subject differences in respiratory signal amplitude did not affect this threshold because the signal was previously normalized (see above). The total apnea duration was expressed as the percentage of the total recording time (automatic apnea score). Agreement Between Visual and Automatic Scores We compared the visual and the automatic scores (for activity and apneas) using the following statistical tests. 1) ANOVAs, with the scores as repeated measures and with the scoring method (visual or automatic) and age as independent variables, were used. In Experiment 2, we also considered the experimental phase (air, hypoxia,

98 • JANUARY 2005 •

www.jap.org

Innovative Methodology 368

AUTOMATIC PROCESSING OF THE RESPIRATORY SIGNAL

Fig. 4. Apnea scores (expressed as percent recording time, Experiment 2). A: PNA2; B: PNA5; C: PNA10. Values are means ⫾ SE; n ⫽ 10 in each age group. See text for analyses. Note the different scales at different PNAs.

posthypoxia) as an independent factor. We used Dunn-Bonferroni post hoc tests for partial comparisons. 2) Pearson product-moment correlation was used to assess the linear relationship between visual and automatic scores. 3) ICC [Pearson’s r and ICC are two indepen-

Table 2. Agreement between visual and automatic apnea scores in newborn mice Agreement Between Apnea Scores

Experiment 1 PNA0 PNA2 PNA5 PNA10 Overall sample Experiment 2 PNA2 PNA5 PNA10 Overall sample Experiments 1 ⫹ 2

Pearson’s r

ICC

0.99 0.99 0.84 0.96 0.99

0.98 (0.84–0.99) 0.94 (0.94–0.99) 0.91 (0.67–0.98) 0.97 (0.88–0.99) 0.96 (0.92–0.98)

0.94 0.99 0.99 0.98 0.98

0.97 (0.87–0.99) 0.99 (0.98–0.99) 0.99 (0.98–0.99) 0.99 (0.98–0.99) 0.99 (0.98–0.99)

PNA0, PNA2, PNA5, PNA10: postnatal age 0, 2, 5, and 10 days, respectively. Experiment 1: 20 min in air; Experiment 2: 3 min in air, 3 min in 5% O2 in N2, 6 min in air. n ⫽ 10 Mice per age group. J Appl Physiol • VOL

Fig. 5. A: scatter plot of automatic vs. visual apnea scores in newborn mice on PNA0, PNA2, PNA5, and PNA10. Visual scores were estimated from the respiratory recordings either visually by 2 independent scorers or automatically. The 2 visual scores were averaged. Each symbol represents 1 pup (n ⫽ 10 per age group). Top: Experiment 1, 20 min in air, Pearson’s r ⫽ 0.99. Bottom: Experiment 2, 3 min air, 3 min 5% O2 in N2, 6 min air, Pearson’s r ⫽ 0.98. Each symbol represents 1 pup. Solid line, regression line; dotted line, identity line. B: Bland and Altman plot (difference against mean). The data from Experiment 1 and Experiment 2 were pooled. The mean ⫾ SD difference between automatic and visual apnea scores (continuous line) was 0.13 ⫾ 0.72. The lower and upper limits of agreement (dotted lines) were ⫺1.31 (95% confidence interval, ⫺1.60 to ⫺1.02) and 1.57 (95% confidence interval, 1.27 to 1.87), respectively.

dent statistics (24)] and 4) Bland and Altman 95% limits of agreement method (4) were used to assess the degree of absolute agreement between the scoring methods. The scores found by the two observers were averaged for Pearson’s correlation, Bland and Altman analyses, and ICC. ICC is the proportion of the total explainable variance in scores due to differences among animals (5, 19, 24). ICC values theoretically range from 0 to 1; values close to 1 indicate that the variation between the methods is small compared with the total variation. We evaluated absolute agreement (rather than consistency) because systematic differences between scoring methods were considered relevant (19). The ICC appropriate for defining absolute agreement (i.e., average score ICC for two-way fixed models) was determined following McGraw and Wong’s recommendations (19). Statistical analyses were done by using SPSS 11 software.

98 • JANUARY 2005 •

www.jap.org

Innovative Methodology AUTOMATIC PROCESSING OF THE RESPIRATORY SIGNAL RESULTS

Our results suggest that defective arousal and ventilatory responses may be simultaneously and automatically analyzed from the plethysmographic signal.

Comparison of Automatic and Visual Activity Scores Visual and automatic scores were similar at all study ages (effect of scoring method, not significant, Fig. 2). Age significantly influenced activity (P ⬍ 0.003). We found a strong correlation between visual and automatic activity scores (the two visual scores were averaged for this analysis, Fig. 3A and Table 1). Bland and Altman analysis (Fig. 3B) and ICC values indicated close agreement between visual and automatic activity scores (Table 1). Comparison Between Visual and Automatic Apnea Scores Apnea scores were small during normoxic periods for both experiments, but they were considerably larger during posthypoxic normoxia in all age groups (Experiment 2, Fig. 4). These scores (two visual and one automatic) showed significant differences, although the automatic scores were generally intermediate between the two visual scores (Fig. 4). We found a strong correlation between visual and automatic apnea scores (Fig. 5 and Table 2). Finally, Bland and Altman analysis (Fig. 5B) and ICC values (Table 2) indicated very close agreement between visual and automatic apnea scores. DISCUSSION

We compared visual and automatic scoring of activity and apnea periods from plethysmographic signals in newborn mice. We found that both scores were reliably determined by automatic processing. Furthermore, the results showed that breathing disturbances provided a valid indirect index of activity in freely moving newborn mice. Agreement Between Visually and Automatically Detected Activity Visual and automatic activity scores were compared by using ANOVAs, correlation coefficients, and ICC. A strong linear correlation between visual and automatic activity scores was observed within each age group and in the overall sample. The data showed a relatively high level of agreement between visual and automatic scoring of activity. This was due to the fact that the threshold of disturbance criterion was calculated individually, so that classification into activity categories was dependent on the regularity of each individual breathing pattern. This is particularly important on PNA0, which is characterized by highly irregular breathing patterns. However, complete agreement seems difficult to achieve, as none of the scoring methods used here provide a fully accurate activity score. In fact, small movements (especially in the axis of the video camera) were not readily detected by visual scoring. This is a limitation of the present validation study. Furthermore, movements do not always disturb the breathing pattern, and the breathing pattern may be disturbed by cognitive activity without associated motor activity (12). Despite above-mentioned limitations, the ICC between automatic and visual activity scores remained relatively high. This result supports the use of the breathing pattern as an indirect index of activity score, as previously proposed in neonatal mice (9) and adult mice (7). This is particularly relevant to the study of arousal, which is characterized by a stereotyped overt behavioral response in newborn mice (8). J Appl Physiol • VOL

369

Agreement Between Visual and Automatic Scores of Apneas We found a strong level of agreement between visual and automatic scores of apnea, in all age groups and in all conditions. Visual detection of apneas resulted in fairly high interscorer variability, as shown clearly by the ANOVA results. However, the automatic scores were usually intermediate values, suggesting that apnea/hypopnea classification from spectral analysis advantageously reduced the subjective component of visual classification of apneas. The strong linear relationship between visual and automatic scores (whether within age groups or over the entire sample) suggested that our apnea classification based on spectral analysis was not affected by major changes in the number or durations of apnea due to age or to the hypoxic decline. This relationship showed that the threshold used for automatic apnea classification was robust to these changes and that individual determination of this threshold was unnecessary. However, individual determination may prove useful when wider age ranges are considered (e.g., from birth to adulthood). Therefore, the automatic classification of apneas proved valid for respiratory phenotyping in newborn mice. This method may be particularly well suited to the analysis of genetic factors involved in early disturbances of respiratory control in mice, because an increased apnea rate is a feature shared by several gene mutations affecting the development of respiratory-related structures (e.g., Refs. 1–3, 6, 17, 20). Conclusion Respiratory and behavioral phenotyping in newborn mice is a promising approach for determining the pathophysiological mechanisms of early disturbances in respiratory control and, in particular, the function of genes in these processes. However, phenotyping newborn mice raises specific difficulties that have rarely been addressed. The novel method for automatic classification of activity and apneas proposed here was both precise and timesaving, and it may facilitate respiratory and behavioral phenotyping in large samples of newborn mice. GRANTS This study was supported by the Institut National de la Sante´ et de la Recherche Me´dicale and by the Universite´ Paris VII (Legs Poix). REFERENCES 1. Aizenfisz S, Dauger S, Durand E, Vardon G, Levacher B, Simonneau M, Pachnis V, Gaultier C, and Gallego J. Ventilatory responses to hypercapnia and hypoxia in heterozygous c-ret newborn mice. Respir Physiol Neurobiol 131: 213–222, 2002. 2. Berry GT, Wu S, Buccafusca R, Ren J, Gonzales LW, Ballard PL, Golden JA, Stevens MJ, and Greer JJ. Loss of murine Na⫹/myoinositol cotransporter leads to brain myo-inositol depletion and central apnea. J Biol Chem 278: 18297–18302, 2003. 3. Blanchi B, Kelly LM, Viemari JC, Lafon I, Burnet H, Bevengut M, Tillmanns S, Daniel L, Graf T, Hilaire G, and Sieweke MH. MafB deficiency causes defective respiratory rhythmogenesis and fatal central apnea at birth. Nat Neurosci 6: 1091–1100, 2003. 4. Bland JM and Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1: 307–310, 1986.

98 • JANUARY 2005 •

www.jap.org

Innovative Methodology 370

AUTOMATIC PROCESSING OF THE RESPIRATORY SIGNAL

5. Carrasco JL and Jover L. Estimating the generalized concordance correlation coefficient through variance components. Biometrics 59: 849 – 858, 2003. 6. Chatonnet F, Boudinot E, Chatonnet A, Taysse L, Daulon S, Champagnat J, and Foutz AS. Respiratory survival mechanisms in acetylcholinesterase knockout mouse. Eur J Neurosci 18: 1419 –1427, 2003. 7. Cohen G, Han ZY, Grailhe R, Gallego J, Gaultier C, Changeux JP, and Lagercrantz H. Beta 2 nicotinic acetylcholine receptor subunit modulates protective responses to stress: a receptor basis for sleep-disordered breathing after nicotine exposure. Proc Natl Acad Sci USA 99: 13272–13277, 2002. 8. Dauger S, Aizenfisz S, Renolleau S, Durand E, Vardon G, Gaultier C, and Gallego J. Arousal response to hypoxia in newborn mice. Respir Physiol 128: 235–240, 2001. 9. Dauger S, Durand E, Cohen G, Lagercrantz H, Changeux JP, Gaultier C, and Gallego J. Control of breathing in newborn mice lacking the beta-2 nAChR subunit. Acta Physiol Scand 182: 205–212, 2004. 10. Dauger S, Pattyn A, Lofaso F, Gaultier C, Goridis C, Gallego J, and Brunet JF. Phox2b controls the development of peripheral chemoreceptors and afferent visceral pathways. Development 130: 6635– 6642, 2003. 11. Drobac E, Durand E, Laudenbach V, Mantz J, and Gallego J. A simple method for short-term controlled anesthesia in newborn mice. Physiol Behav 82: 279 –283, 2004. 12. Durand E, Dauger S, Vardon G, Gressens P, Gaultier C, De Schonen S, and Gallego J. Classical conditioning of breathing pattern after two acquisition trials in 2-day-old mice. J Appl Physiol 94: 812– 818, 2003. 13. Durand E, Lofaso F, Dauger S, Vardon G, Gaultier C, and Gallego J. Intermittent hypoxia induces transient arousal delay in newborn mice. J Appl Physiol 96: 1216 –1222; discussion 1196, 2004. 14. Gauda EB, McLemore GL, Tolosa J, Marston-Nelson J, and Kwak D. Maturation of peripheral arterial chemoreceptors in relation to neonatal apnoea. Semin Neonatol 9: 181–194, 2004. 15. Gaultier C, Amiel J, Dauger S, Trang H, Lyonnet S, Gallego J, and Simonneau M. Genetics and early disturbances of breathing control. Pediatr Res 55: 729 –733, 2004.

J Appl Physiol • VOL

16. Gaultier C, Dauger S, Simonneau M, and Gallego J. Genes modulating chemical breathing control: lessons from mutant animals. Respir Physiol Neurobiol 136: 105–114, 2003. 17. Kuwana S, Okada Y, Sugawara Y, Tsunekawa N, and Obata K. Disturbance of neural respiratory control in neonatal mice lacking GABA synthesizing enzyme 67-kDa isoform of glutamic acid decarboxylase. Neuroscience 120: 861– 870, 2003. 18. Martin RJ, Abu-Shaweesh JM, and Baird TM. Apnoea of prematurity. Paediatr Respir Rev 5, Suppl A: S377–S382, 2004. 19. McGraw KO and Wong SP. Forming inferences about some intraclass correlation coefficients. Psychol Med 1: 30 – 46, 1996. 20. Nsegbe E, Wallen-Mackenzie A, Dauger S, Roux JC, Shvarev Y, Lagercrantz H, Perlmann T, and Herlenius E. Congenital hypoventilation and impaired hypoxic response in Nurr1 mutant mice. J Physiol 556: 43–59, 2004. 21. Renolleau S, Dauger S, Vardon G, Levacher B, Simonneau M, Yanagisawa M, Gaultier C, and Gallego J. Impaired ventilatory responses to hypoxia in mice deficient in endothelin-converting-enzyme-1. Pediatr Res 49: 705–712, 2001. 22. Robinson DM, Kwok H, Adams BM, Peebles KC, and Funk GD. Development of the ventilatory response to hypoxia in Swiss CD-1 mice. J Appl Physiol 88: 1907–1914, 2000. 23. Robinson DM, Peebles KC, Kwok H, Adams BM, Clarke LL, Woollard GA, and Funk GD. Prenatal nicotine exposure increases apnoea and reduces nicotinic potentiation of hypoglossal inspiratory output in mice. J Physiol 538: 957–973, 2002. 24. Streiner DL. Learning how to differ: agreement and reliability statistics in psychiatry. Can J Psychiatry 40: 60 – 66, 1995. 25. Viemari JC, Burnet H, Bevengut M, and Hilaire G. Perinatal maturation of the mouse respiratory rhythm-generator: in vivo and in vitro studies. Eur J Neurosci 17: 1233–1244, 2003.

98 • JANUARY 2005 •

www.jap.org