- Email: [email protected]
Reliability and validity of computer scoring of behavioral sleep-wake states in rats and rabbits
Physiology&Behavior.Vol. 54, pp. 269-273, 1993
0031-9384/93 $6.00 + .00 Copyright© 1993 PergamonPressLtd.
Printed in the USA.
Reliability and Validity of Computer Scoring of Behavioral Sleep-Wake State,s in Rats and Rabbits D E B O R A H A. C A R R O L L , *~ V I C T O R H. D E N E N B E R G ~ "2 A N D E V E L Y N B. T H O M A N * t
*Biobehavioral Sciences Graduate Degree Program and tDepartment of Psychology, University o f Connecticut, Storrs, C T 06269-4154 Received l l J a n u a r y 1993 CARROLL, D. A., V. H. DENENBERG AND E. B. THOMAN. Reliabilityand validity of computer scoringof behavioralsleepwake states in rats and rabbits. PHYSIOL BEHAV 54(2) 269-273, 1993.--Previous studies in human infants, rabbits, and rats have shown that states of sleep and wakefulness can be reliably identified from motility signals produced by respiration and body movement. Thoman has described a computer-scoring algorithm for automated scoring of behavioral states from motility signals in human infants. In the present studies, we report the use of the computer scoring program with motility signals obtained from electronic activity monitors. In the newborn rabbit, computer classification of the data into behavioral states agreed with visual scoring of the motility signals. In infant and adult rats, computer scoring across the entire age range was validated against direct behavioral observations, video-taped observations, and visual scoring of the motility signal. This procedure permits prolonged recordings of freely moving animals and eliminates the need for an observer to be present. Active sleep
Quiet sleep
Wake
Sleep-wake transition
STATES of sleep and wakefulness during the neonatal period have been shown to be sensitive indicators of central nervous system function (15,21,22). Not only is the neonate's response to stimuli affected by state, but the organism also changes state in response to stimuli (3,5,9,19,20). Therefore, state has become an important variable both to be studied directly and to be taken into account when conducting other research with developing organisms. States can be measured by direct observation (8,12,25,29), real-time (6,7) or time-lapse (4) video recordings, or EEG (l l, 16). Each procedure has advantages and disadvantages. Direct behavioral observation of state, though reliable, is not feasible over extended periods of time or for large numbers of subjects. Videotaping procedures are similar to direct behavioral observations. However, they are generally not as accurate because small movements such as REMs cannot be clearly identified, and changes in the subject's position may place him or her out of the camera's view. Physiological recordings provide multiple channels of information and permit continuous monitoring of states of sleep and wakefulness. However, human infants must be brought to the laboratory and have electrodes attached for recording EEG waves, eye movements, heart rate, respiration, and muscle
Motility
tonus for states to be identified. In addition, the absence of sleep spindles and the trace alternant pattern makes it difficult to identify quiet sleep in very young (1 week of age) human infants (l 4). When working with animals, they must first be anesthetized in order to implant electrodes for recording of state, and this procedure affects the states (2). Further, it is nearly impossible to record or interpret EEG activity in rats and rabbits during the first few days of life (2,11,16). Because of these difficulties, alternative nonintrusive procedures have been developed for recording states of sleep and wakefulness in developing organisms. The majority of these techniques involve the use of a movement sensor (18); or a pressure-sensitive surface that is placed under the organism (10,13,17,24,26). Thoman and associates have demonstrated that states of sleep and wakefulness can be reliably identified from analog signals produced by respiration and body movement in human infants (23,24), elderly humans (1), and rabbits and rats (8,25-27,29). They have shown that a single-channel analog signal provides distinctly different patterns for the behavioral states of active wake, quiet wake, active sleep, quiet sleep, and sleep--wake transition.
Present address: New York State Psychiatric Institute, Department of Psychiatry, Columbia College of Physicians and Surgeons, New York, NY 10032. 2 Requests for reprints should be addressed to Dr. Victor H. Denenberg.
269
270
CARROLL, DENENBERG AND I H O M A N
Validity studies have been done on the motility signal data. In human infants, rabbits, and rats, visual scoring of the analog signals was validated by comparison to direct observations (27). In adult rabbits, this method of state classification agreed highly with EEG classification of states (30). Acebo et al. (1) have recently demonstrated that motility characterization of periods of sleep, wake, and apneas in elderly humans correlated highly with polysomnography. Another important finding was that the motility patterns that characterize each behavioral state are invariant across species. An experienced state scorer who was familiar with analog signals produced by rats and rabbits could reliably score signals from human babies and vice versa (27). In an extension of this work, Thoman developed a computer program for automated scoring of the motility signals of human infants using a pattern-recognition algorithm (23), adapted from one designed by Vincent (28) for EEG. Prototypical waveforms for each of the behavioral states serve as templates. Each successive epoch of data is compared to the templates and assigned the name of the template that is the least discrepant. The computer-scoring algorithm has previously been successfully used with infant and elderly humans. The purpose of the present studies was to determine whether Thoman's computer algorithm would reliably and validly score behavioral states in rabbits and rats. In the newborn rabbit we applied the computer scoring procedure to motility data on the first day of life. In a parallel fashion we studied behavioral states in the rat from birth through adulthood. Because the motility signals that characterize each state change with development (7), it is necessary to assess reliability and validity across different ages. Therefore, in the rat we recorded behavioral states every day for I h from birth through weaning, and on four separate occasions for 2 h a day during adulthood. STUDY 1: COMPUTER SCORING OF BEHAVIORAL STATES IN THE NEONATAL RABBIT METHOD
Subjects Pregnant multiparous Dutch-belted rabbits were delivered to us on the 26th day of gestation and were housed individually in maternity cages containing wooden nest boxes. They were maintained on a 12-h light/dark cycle, with food and water available ad lib. Twelve pups, from two litters, were used.
Procedure On the morning a litter was found, litter size was culled to six pups. Each pup was weighed and then placed inside a bottomless Plexiglas cylinder on one of six Stoelting Electronic Activity Monitors (Model 31408). Some nesting material from the home cage was placed inside each cylinder as a familiar odor. Body temperature was maintained by a 100 W red light bulb suspended above each Plexiglas cylinder. The air temperature at the platform floor was 27°C. The analog signals were amplified through a Grass Polygraph (Model 78B), sampled at a rate of 10 samples per second, digitized (Tecmar LabTender A-D Board), and stored on a 30 mB Seagate hard drive in an IBM PC XT computer. Each rabbit pup's motility signals were recorded for 2 h. Data from these records were then used to correlate computer scoring with visual scoring. Visual signal scoring. The 2-h analog signal record was divided into 30-s epochs. An experienced state scorer viewed the signals on a computer screen and assigned a specific state category to each epoch. The following four states were scored: wake, active
sleep, quiet sleep, and sleep-wake transition. These states have been defined in detail, and prototypical analog motility signals have been presented (26,27,30). Briefly, the states can be characterized as follows. Wake: locomotion or other long-lastinggross body movements predominate. Active sleep: there is low level activity with sudden spikes reflecting twitches and jerks. Quiet sleep: the major features here are little or no movement with low-level regular respiration (the monitors easily pick up the respiration of a newborn rat). Sleep-wake transition: behaviors indicative of both sleep and wake are present, including twitches and jerks, interspersed with low level activity. Computer scoring. Thoman's computer program for automated scoring of the motility signals of human infants was used (23). The program uses a pattern-recognition algorithm. A data profile for each successive epoch of data is compared with data profiles from waveform regions chosen as prototypical templates of each of the four states. Multiple templates are chosen for each state because the analog signals for any state are not homogeneous. The numerical discrepancy between each digitized sample of waveform region and the waveform patterns of the prototypical templates is computed using the following four parameters obtained from the waveforms of each epoch. Amplitude: the amount above or below the zero line. Zero crossing: a measure of the time interval between the first positive and the first negative amplitude (or vice versa). Interval between direction change: the time during which the amplitude curve shows no slope change. Total change in amplitude: the amount of amplitude change during the interval when the slope is monotonic. The four parameters of each successive epoch of data were compared to the parameters of the templates, and the epoch was assigned the name of the template that was the least discrepant. For more detailed information, see Thoman and Glazier (23). Motility signals prototypical of each behavioral state were chosen as templates from each animal's 2-h recording. The complete record was then computer scored with the individualized templates. Data from the computer scoring of the signal record were was then compared with the data from visual scoring by the experienced state scorer. RESULTS The correlations between observer scoring and computer scoring of the same analog signals for the four behavioral states are shown in Table 1. This table also includes the age of observation, the n per group, the observation length in hours, and the epoch length in seconds. The correlations ranged from 0.49 to 0.94. Table 2 presents the percent time spent in each state as determined by computer scoring. STUDY 2: COMPUTER SCORING OF BEHAVIORAL STATES IN THE RAT FROM BIRTH THROUGH ADULTHOOD Study 1 established the validity of computer scoring in the infant rabbit. Our next objective was to do the same with the rat between birth and adulthood. To do so, data were collected by direct observation, from video recording and from the activity monitor. METHOD
Subjects Purdue-Wistar rats from our closed colony were bred. The pregnant females were housed individually in maternity cages containing wood shavings. Food and water were available ad
B E H A V I O R A L S T A T E IN R A T S A N D RABBITS
271
TABLE 1 CORRELATIONS BETWEEN DIFFERENT BEHAVIORALSTATE SCORING METHODS IN RABBITS AND RATS Species
Age (days)
Obser. (h)
Epoch (s)
Scoring Method
Active Wake
Quiet Wake
Rabbit [12] Rat [28] Rat [28] Rat [31 ] Rat [12]
1 1-11 1-12 14-21 Adult
2 1 1 1 2
30 10 10 30 30
Visual/computer DO/video Video/computer Video/computer Visual/computer
0.94* 0.98* 0.75* 0.94* 0.99*
NS NS NS 0.85* 0.99*
Sleelr-Wake Transition 0.49i" 0.95* 0.66* 0.80* 0.81'
Active Sleep
Quiet Sleep
0.92* 0.98* 0.83* 0.97* 0.89*
0.84* 0.96* 0.80* 0.95* 0.93*
Number in brackets = number of animals. 130 = direct observation. NS = not scored. *p < 0.01. "t p < 0.05.
lib. On the morning a litter was found (day 1), litter size was culled to six pups, each with three males and three females. Thirty-six animals, from six litters, were used. Throughout the study, the animals were maintained on a 12h light/dark cycle with ad lib food and water. Behavioral state was recorded during the light hours.
Procedure On day 1 the six pups from one litter were placed singly into bottomless Plexiglas cylinders, with a diameter at least twice the animal's length, on the Stoelting Monitors. Some shavings from the maternity cage were put in each cylinder. A 100 W red light bulb suspended over each Plexiglas cylinder kept the air temperature inside the cylinder at 27°C. Analog signals produced by respiration and body movements were recorded for 1 h. At the same time, video-taped recordings of one animal per litter were obtained. On days 1-11, one pup per litter was also observed directly and behavioral state was recorded. Three of the litters were tested on odd numbered days (1, 3 . . . . 15); the other three litters were tested on even numbered days (2, 4 . . . . 14). The data for day 13 were lost due to equipment failure. After eye opening (days 14-16) the animals continued to be tested as above. However, no direct observations were made because the presence of an observer would disrupt normal state behaviors. Video-taped recordings were still obtained for one animal from each litter per day. This testing protocol continued until the animals were weaned and group housed at 21 days of age. At 40 days the animals were housed individually in clear
plastic tubs with hardwood shavings. Two-hour motility recordings were obtained on days 68, 75, 118, and 146.
Scoring of State Behavioral state was scored by 1. 2. 3. 4.
direct observation, from video-taped recordings, from visual inspection of the waveforms, and by the computer algorithm.
The following states were scored between birth and eye opening: wake, sleep-wake transition, active sleep, and quiet sleep. After eye opening, the wake state was separated into active wake and quiet wake. The state definitions for the rat are the same as for the rabbit. The distinction between active and quiet wake after eye opening is as follows. Active wake: locomotion and gross motor activity predominate. Quiet wake: there is lack of activity; the animal maintains an erect posture with eyes open and alert. See Thoman, Zeidner, and Denenberg (27) for analog signals prototypical of these states. Prior to eye opening, behavioral state was scored every 10 s. After eye opening state, changes did not occur as frequently, and state was scored every 30 s. Direct observation. An experienced observer sat about 18 inches from the subject. The observer code-recorded a state category for each 10-s epoch. An electronic timer signaled, through an earphone, every t0 s with a unique tone at 1-min intervals. Video-tape scoring. The video-taped recording included the audio electronic timer signals at 10-s and 1-min intervals. At
TABLE 2 PERCENT OF TIME SPENT IN EACH STATE AS DETERMINED BY COMPUTER SCORING Species
Age
Rabbit [12] Rat [28] Rat [31] Rat [12]
Day 1 Days 1-12 Days 14-21 Adulthood
Number in brackets = number of animals. Mean _ SE. NS = not scored.
ActiveWake
Quiet Wake
Sleep-Wake Transition
Active Sleep
Quiet Sleep
30.8 7.0 5.6 40.0
NS NS 24.7 _+ 1.3 24.1 _+ 6.1
11.5 + 0.4 7.0 ___0.8 4.9 ___0.5 2.0 + 0.4
57.3 _ 1.4 85.6 _+ 1.8 49.8 _+ 2.0 14.7 ___3.9
0.3 __+0.1 0.4 _+ 0.1 14.9 ___ 1.2 19.1 + 4.6
___1.6 + 1.3 _+0.5 _+ 6.2
272
CARROLL, DENENBERG AND [HOMAN
least 30 days after the recording was obtained, an experienced observer watched each tape and assigned state categories. For the first 12 days, a 10-s epoch was used. After that, the epoch was 30 s long. Visual signal scoring. The analog data were divided into 30s epochs. At least 30 days after behavioral state recordings were obtained, an experienced state scorer viewed the signals on a computer screen and assigned a specific state category to each 30-s epoch. Computer scoring. Three sets of templates representative of the five behavioral states were compiled from analog signals from several different animals. One set consisted of signals from days 1-5 and was used to score data from days 1-12 in 10-s epochs. The second set of templates consisted of analog signals from days 17-21, and was used to score data from days 14-21 in 30s epochs. The third set was taken from animals that were 68, 75, 118, and 146 days old and was used to score data from adult animals in 30-s epochs. After eye opening (i.e, days 14-16), a smoothing procedure was imposed upon the computer-scored data. The following rules were used: 1. Isolated epochs of active wake surrounded by active sleep or quiet sleep were changed to transition. 2. Isolated epochs of transition surrounded by active wake or quiet wake were changed to the prevailing state. 3. If more than three transitions occurred in succession, beginning with the fourth epoch, they were changed to active wake. 4. Isolated epochs of either sleep state surrounded by another state were changed to the prevailing state. 5. Isolated epochs of quiet wake surrounded by active sleep, quiet sleep, or active wake were changed to the prevailing state. Inspection of the state record before and after the smoothing consistently revealed that less than 10% of the epochs were changed. RESULTS
Birth to Eye Opening (Days 1-12) Correlations between direct observation and video-tape scoring for the four states summed across days 1-11 ranged from 0.95 to 0.98 (Table 1). Correlations between the video-tape scoring and computer scoring for each of the four states summed over days 1-12 ranged from 0.66 to 0.83. The percent time spent in each state over this age range is in Table 2.
Eye Opening to Weaning (Days 14-21) The correlations between video-tape and computer scoring, summed over days 14-21, for the five behavioral states ranged from 0.80 to 0.97. The percent time spent in each state over this age range is given in Table 2.
Adulthood The correlations between visual scoring of the analog signals by an observer and computer scoring of the same signals. summed across the four adult ages, ranged from 0.81 to 0.99. The percent time spent in each state over this age range is given in Table 2. GENERAL DISCUSSION In both the rabbit and the rat, the correlation coefficients for computer scoring of the wake states, active sleep, and quiet sleep were very high. Corelations for sleep--wake transition were lower, though still acceptable. This state is the most difficult to identify regardless of the measurement procedure. These correlations establish that the electronic activity monitor is a valid instrument to record behavioral states. Thoman and Glazier (23) validated computer-scoring of motility signals against direct observations of behavioral states in human infants. The present results show that computerized template matching is equally valid for categorizing behavioral states in newborn rabbits, and infant, and adult rats. For behavioral state scoring of sleep in the rabbit, individualized sets of templates were chosen from each rabbit's analog record. In the rat, it is important to note that standardized sets of templates, chosen from six different animals, were used for all subjects in the study. This procedure removes subjectivity in scoring and suggests an invariance in motility patterns across individuals (27). Three sets of templates were necessary to validly characterize behavioral states from birth through adulthood. The motility signals changed qualitatively at eye opening (days 1416) and again at about the time of puberty (between days 4050). Qualitative changes are also found in EEG parameters that characterize sleep states at about the same ages (11,16). Quantitatively, the changes in the amount of time that rats spent in each of the states from birth through adulthood indicate the sensitivity of the recording and scoring procedures to developmental effects. Though we only used this procedure with rabbits on day 1 of life, we have shown that direct behavioral observations correlate highly with a judge's scoring of the analog signals from the activity monitor over the age range of 7-19 days (26,27). Further, when analog signal recordings and hippocampal EEG recordings of two adult rabbits were compared, using 10-s epochs as the unit of measurement, there was 85.4% agreement (30), Therefore, the activity monitoring system is applicable for both the rat and the rabbit from birth through adulthood. The motility recording procedure permits prolonged recordings of behavioral states in the freely moving animal. Computerized recording and scoring is more objective and economical than visual scoring, and eliminates interference caused by observational procedures after eye opening.
REFERENCES 1. Acebo, C.; Watson, R. K.; Bakos, L.; Thoman, E. B. Sleep and apnea in the elderly: Reliability and validity of 24-hour recordings in the home. Sleep 14:56-64; 1991. 2. Alfoldi, P.; Tobler, I.; Borbely, A. A. Sleep regulation in rats during early development. Am. J. Physiol. 258:R634-R654; 1990. 3. Anders, T. F.; Chalemian, R. J. The effectsof circumcision on sleep-wake states in human neonates. Psychosom. Med. 36:174-179; 1974. 4. Anders, T.; Sostek-Miller, A. The use of time lapse video recording of sleep-wake behavior in human infants. Psychophysiology13:155158; 1976.
5. Brackbill, Y. Continuous stimulation reduces arousal level: Stability of the effect over time. Child Dev. 44:43-46; 1973. 6. Carroll, D. A.; Thoman, E. B.; Denenberg, V. H. Computer scoring of respiration and motility patterns for behavioral states in the neonatal rabbit. Poster presented at the Northeastern Sleep Society 3rd annual meeting, White Plains, NY; March 31-April 1: 1989. 7. Carroll, D. A.; Thoman, E. B.; Denenberg, V. H. Development of sleep and wake states of rats. Poster p~ented at the Northeastern Sleep Society 4th annual meeting, Cambridge, MA; March 3 l-April 1: 1990.
B E H A V I O R A L S T A T E IN R A T S A N D RABBITS
8. Denenberg, V. H.; Zeidner, L. P.; Thoman, E. B.; Kramer, P.; Rowe, J. C.; Philipps, A. F.; Raye, J. R. Effects oftheophylline on behavioral state development in the newborn rabbit. J. Pharmacol. Exp. Ther. 221:604-608; 1982. 9. Emde, R. N.; Harmon, R. J.; Metcalf, D.; Koenig, K. L.; Chalemian, R. J. Stress and neonatal sleep. Psychosom. Med. 33:491--497; 1971. 10. Erkinjuntti, M.; Kero, P.; Halonen, J. P.; Miloka, H.; Sainio, K. SCSB method compared to EEG-based polygraphy in sleep state scoring of newborn infants. Acta Paediatr. Scand. 79:274-279; 1990. 11. Gramsbergen, A. The development of the EEG in the rat. Dev. Psychobiol. 9:501-515; 1976. 12. Gramsbergen, A.; Schwartze, P.; Prechtl, H. F. R. The postnatal development of behavioral states in the rat. Dev. Psychobiol. 3:267280; 1970. 13. Hilakivi, L. A.; Hilakivi, I. T. Sleep-wake behavior of newborn rats recorded with movement sensitive method. Behav. Brain Res. 19: 241-248; 1986. 14. Hoppenbrouwers, T.; Hodgman, J.; Arakawa, K.; Geidel, S. A.; Sterman, M. B. Sleep and waking states in infancy: Normative studies. Sleep 11:387-401; 1988. 15. loffe, S.; Chernick, V. Prediction of subsequent motor and mental retardation in newborn infants exposed to alcohol in utero by computerized EEG analysis. Neuropediatrics 21:11-17; 1990. 16. Jouvet-Mounier, D.; Astic, L.; Lacote, D. Ontogenesis of the states of sleep in rat, cat, and guinea pig during the first postnatal month. Dev. Psychobiol. 2:216-239; 1970. 17. Keefe, M. R.; Kotzer, A. M.; Reuss, J. L.; Sander, L. W. Development of a system for monitoring infant state behavior. Nurs. Res. 38:344347; 1989. 18. Sadeh, A.; Lavie, P.; Scher, A.; Tirosh, E.; Epstein, R. Actigraphic home-monitoring sleep-disturbed and control infants and young children: A new method for pediatric assessment of sleep--wake patterns. Pediatrics 87:494-499; 1991. 19. Sander, L.; Julia, H.; Stechler, G.; Burns, P. Continuous 24-hour
273
20. 21. 22.
23. 24. 25. 26. 27. 28. 29. 30.
interactional monitoring in infants reared in two caretaking environments. Psychosom. Med. 34:270-282; 1972. Thoman, E. B. Sleeping and waking states in infants: A functional perspective. Neurosci. Biobehav. Rev. 14:93-1071; 1990. Thoman, E. B.; Acebo, C.; Becker, P. T. Infant crying and stability in the mother-infant relationship: A systems analysis. Child Dev. 54:653-659; 1983. Thoman, E. B.; Denenberg, V. H.; Sievel, J.; Zeidner, L. P.; Becker, P. State organization in neonates: Developmental inconsistency indicates risk for developmental dysfunction. Neuropediatrics 12:4554; 1981. Thoman, E. B.; Glazier, R. C. Computer scoring of motility patterns for states of sleep and wakefulness: Human infants. Sleep 10:122129; 1987. Thoman, E. B.; Tynan, W. D. Sleep states and wakefulness in human infants: Profiles from motility monitoring. Physiol. Behav. 23:519525; 1979. Thoman, E. B.; Waite, S. P.; DeSantis, D. T.; Denenberg, V. H. Ontogeny of sleep and wake states in the rabbit. Anim. Behav. 27: 95-106; 1979. Thoman, E. B.; Zeidner, L. P. Sleep--wake states in infant rabbits: Profiles from motility monitoring. Physiol. Behav. 22:1049-1054; 1979. Thoman, E. B.; Zeidner, L. P.; Denenberg, V. H. Cross-species invariance in state-related motility patterns. Am. J. Physiol. 10:R312R315; 1990. Vincent, F. A. SQUEEK--A computer algorithm for real time automatic sleep state scoring of mouse electroencephalogram. Thesis. Stanford University, Stanford, CA.; 1976. Waite, S. P.; DeSantis, D. S.; Thoman, E. B.; Denenberg, V. H. The predictive validity of early sleep development on later behavioral characteristics in the rabbit. Biol. Behav. 2:249-261; 1977. Zeidner, L. P.; Denenberg, V. H.; Thoman, E. B.; Weyand, T. Comparisons of behavioral, motoric and electrical criteria for assessment of sleep-wake states in the rabbit. Physiol. Behav. 31:273-278; 1983.
0031-9384/93 $6.00 + .00 Copyright© 1993 PergamonPressLtd.
Printed in the USA.
Reliability and Validity of Computer Scoring of Behavioral Sleep-Wake State,s in Rats and Rabbits D E B O R A H A. C A R R O L L , *~ V I C T O R H. D E N E N B E R G ~ "2 A N D E V E L Y N B. T H O M A N * t
*Biobehavioral Sciences Graduate Degree Program and tDepartment of Psychology, University o f Connecticut, Storrs, C T 06269-4154 Received l l J a n u a r y 1993 CARROLL, D. A., V. H. DENENBERG AND E. B. THOMAN. Reliabilityand validity of computer scoringof behavioralsleepwake states in rats and rabbits. PHYSIOL BEHAV 54(2) 269-273, 1993.--Previous studies in human infants, rabbits, and rats have shown that states of sleep and wakefulness can be reliably identified from motility signals produced by respiration and body movement. Thoman has described a computer-scoring algorithm for automated scoring of behavioral states from motility signals in human infants. In the present studies, we report the use of the computer scoring program with motility signals obtained from electronic activity monitors. In the newborn rabbit, computer classification of the data into behavioral states agreed with visual scoring of the motility signals. In infant and adult rats, computer scoring across the entire age range was validated against direct behavioral observations, video-taped observations, and visual scoring of the motility signal. This procedure permits prolonged recordings of freely moving animals and eliminates the need for an observer to be present. Active sleep
Quiet sleep
Wake
Sleep-wake transition
STATES of sleep and wakefulness during the neonatal period have been shown to be sensitive indicators of central nervous system function (15,21,22). Not only is the neonate's response to stimuli affected by state, but the organism also changes state in response to stimuli (3,5,9,19,20). Therefore, state has become an important variable both to be studied directly and to be taken into account when conducting other research with developing organisms. States can be measured by direct observation (8,12,25,29), real-time (6,7) or time-lapse (4) video recordings, or EEG (l l, 16). Each procedure has advantages and disadvantages. Direct behavioral observation of state, though reliable, is not feasible over extended periods of time or for large numbers of subjects. Videotaping procedures are similar to direct behavioral observations. However, they are generally not as accurate because small movements such as REMs cannot be clearly identified, and changes in the subject's position may place him or her out of the camera's view. Physiological recordings provide multiple channels of information and permit continuous monitoring of states of sleep and wakefulness. However, human infants must be brought to the laboratory and have electrodes attached for recording EEG waves, eye movements, heart rate, respiration, and muscle
Motility
tonus for states to be identified. In addition, the absence of sleep spindles and the trace alternant pattern makes it difficult to identify quiet sleep in very young (1 week of age) human infants (l 4). When working with animals, they must first be anesthetized in order to implant electrodes for recording of state, and this procedure affects the states (2). Further, it is nearly impossible to record or interpret EEG activity in rats and rabbits during the first few days of life (2,11,16). Because of these difficulties, alternative nonintrusive procedures have been developed for recording states of sleep and wakefulness in developing organisms. The majority of these techniques involve the use of a movement sensor (18); or a pressure-sensitive surface that is placed under the organism (10,13,17,24,26). Thoman and associates have demonstrated that states of sleep and wakefulness can be reliably identified from analog signals produced by respiration and body movement in human infants (23,24), elderly humans (1), and rabbits and rats (8,25-27,29). They have shown that a single-channel analog signal provides distinctly different patterns for the behavioral states of active wake, quiet wake, active sleep, quiet sleep, and sleep--wake transition.
Present address: New York State Psychiatric Institute, Department of Psychiatry, Columbia College of Physicians and Surgeons, New York, NY 10032. 2 Requests for reprints should be addressed to Dr. Victor H. Denenberg.
269
270
CARROLL, DENENBERG AND I H O M A N
Validity studies have been done on the motility signal data. In human infants, rabbits, and rats, visual scoring of the analog signals was validated by comparison to direct observations (27). In adult rabbits, this method of state classification agreed highly with EEG classification of states (30). Acebo et al. (1) have recently demonstrated that motility characterization of periods of sleep, wake, and apneas in elderly humans correlated highly with polysomnography. Another important finding was that the motility patterns that characterize each behavioral state are invariant across species. An experienced state scorer who was familiar with analog signals produced by rats and rabbits could reliably score signals from human babies and vice versa (27). In an extension of this work, Thoman developed a computer program for automated scoring of the motility signals of human infants using a pattern-recognition algorithm (23), adapted from one designed by Vincent (28) for EEG. Prototypical waveforms for each of the behavioral states serve as templates. Each successive epoch of data is compared to the templates and assigned the name of the template that is the least discrepant. The computer-scoring algorithm has previously been successfully used with infant and elderly humans. The purpose of the present studies was to determine whether Thoman's computer algorithm would reliably and validly score behavioral states in rabbits and rats. In the newborn rabbit we applied the computer scoring procedure to motility data on the first day of life. In a parallel fashion we studied behavioral states in the rat from birth through adulthood. Because the motility signals that characterize each state change with development (7), it is necessary to assess reliability and validity across different ages. Therefore, in the rat we recorded behavioral states every day for I h from birth through weaning, and on four separate occasions for 2 h a day during adulthood. STUDY 1: COMPUTER SCORING OF BEHAVIORAL STATES IN THE NEONATAL RABBIT METHOD
Subjects Pregnant multiparous Dutch-belted rabbits were delivered to us on the 26th day of gestation and were housed individually in maternity cages containing wooden nest boxes. They were maintained on a 12-h light/dark cycle, with food and water available ad lib. Twelve pups, from two litters, were used.
Procedure On the morning a litter was found, litter size was culled to six pups. Each pup was weighed and then placed inside a bottomless Plexiglas cylinder on one of six Stoelting Electronic Activity Monitors (Model 31408). Some nesting material from the home cage was placed inside each cylinder as a familiar odor. Body temperature was maintained by a 100 W red light bulb suspended above each Plexiglas cylinder. The air temperature at the platform floor was 27°C. The analog signals were amplified through a Grass Polygraph (Model 78B), sampled at a rate of 10 samples per second, digitized (Tecmar LabTender A-D Board), and stored on a 30 mB Seagate hard drive in an IBM PC XT computer. Each rabbit pup's motility signals were recorded for 2 h. Data from these records were then used to correlate computer scoring with visual scoring. Visual signal scoring. The 2-h analog signal record was divided into 30-s epochs. An experienced state scorer viewed the signals on a computer screen and assigned a specific state category to each epoch. The following four states were scored: wake, active
sleep, quiet sleep, and sleep-wake transition. These states have been defined in detail, and prototypical analog motility signals have been presented (26,27,30). Briefly, the states can be characterized as follows. Wake: locomotion or other long-lastinggross body movements predominate. Active sleep: there is low level activity with sudden spikes reflecting twitches and jerks. Quiet sleep: the major features here are little or no movement with low-level regular respiration (the monitors easily pick up the respiration of a newborn rat). Sleep-wake transition: behaviors indicative of both sleep and wake are present, including twitches and jerks, interspersed with low level activity. Computer scoring. Thoman's computer program for automated scoring of the motility signals of human infants was used (23). The program uses a pattern-recognition algorithm. A data profile for each successive epoch of data is compared with data profiles from waveform regions chosen as prototypical templates of each of the four states. Multiple templates are chosen for each state because the analog signals for any state are not homogeneous. The numerical discrepancy between each digitized sample of waveform region and the waveform patterns of the prototypical templates is computed using the following four parameters obtained from the waveforms of each epoch. Amplitude: the amount above or below the zero line. Zero crossing: a measure of the time interval between the first positive and the first negative amplitude (or vice versa). Interval between direction change: the time during which the amplitude curve shows no slope change. Total change in amplitude: the amount of amplitude change during the interval when the slope is monotonic. The four parameters of each successive epoch of data were compared to the parameters of the templates, and the epoch was assigned the name of the template that was the least discrepant. For more detailed information, see Thoman and Glazier (23). Motility signals prototypical of each behavioral state were chosen as templates from each animal's 2-h recording. The complete record was then computer scored with the individualized templates. Data from the computer scoring of the signal record were was then compared with the data from visual scoring by the experienced state scorer. RESULTS The correlations between observer scoring and computer scoring of the same analog signals for the four behavioral states are shown in Table 1. This table also includes the age of observation, the n per group, the observation length in hours, and the epoch length in seconds. The correlations ranged from 0.49 to 0.94. Table 2 presents the percent time spent in each state as determined by computer scoring. STUDY 2: COMPUTER SCORING OF BEHAVIORAL STATES IN THE RAT FROM BIRTH THROUGH ADULTHOOD Study 1 established the validity of computer scoring in the infant rabbit. Our next objective was to do the same with the rat between birth and adulthood. To do so, data were collected by direct observation, from video recording and from the activity monitor. METHOD
Subjects Purdue-Wistar rats from our closed colony were bred. The pregnant females were housed individually in maternity cages containing wood shavings. Food and water were available ad
B E H A V I O R A L S T A T E IN R A T S A N D RABBITS
271
TABLE 1 CORRELATIONS BETWEEN DIFFERENT BEHAVIORALSTATE SCORING METHODS IN RABBITS AND RATS Species
Age (days)
Obser. (h)
Epoch (s)
Scoring Method
Active Wake
Quiet Wake
Rabbit [12] Rat [28] Rat [28] Rat [31 ] Rat [12]
1 1-11 1-12 14-21 Adult
2 1 1 1 2
30 10 10 30 30
Visual/computer DO/video Video/computer Video/computer Visual/computer
0.94* 0.98* 0.75* 0.94* 0.99*
NS NS NS 0.85* 0.99*
Sleelr-Wake Transition 0.49i" 0.95* 0.66* 0.80* 0.81'
Active Sleep
Quiet Sleep
0.92* 0.98* 0.83* 0.97* 0.89*
0.84* 0.96* 0.80* 0.95* 0.93*
Number in brackets = number of animals. 130 = direct observation. NS = not scored. *p < 0.01. "t p < 0.05.
lib. On the morning a litter was found (day 1), litter size was culled to six pups, each with three males and three females. Thirty-six animals, from six litters, were used. Throughout the study, the animals were maintained on a 12h light/dark cycle with ad lib food and water. Behavioral state was recorded during the light hours.
Procedure On day 1 the six pups from one litter were placed singly into bottomless Plexiglas cylinders, with a diameter at least twice the animal's length, on the Stoelting Monitors. Some shavings from the maternity cage were put in each cylinder. A 100 W red light bulb suspended over each Plexiglas cylinder kept the air temperature inside the cylinder at 27°C. Analog signals produced by respiration and body movements were recorded for 1 h. At the same time, video-taped recordings of one animal per litter were obtained. On days 1-11, one pup per litter was also observed directly and behavioral state was recorded. Three of the litters were tested on odd numbered days (1, 3 . . . . 15); the other three litters were tested on even numbered days (2, 4 . . . . 14). The data for day 13 were lost due to equipment failure. After eye opening (days 14-16) the animals continued to be tested as above. However, no direct observations were made because the presence of an observer would disrupt normal state behaviors. Video-taped recordings were still obtained for one animal from each litter per day. This testing protocol continued until the animals were weaned and group housed at 21 days of age. At 40 days the animals were housed individually in clear
plastic tubs with hardwood shavings. Two-hour motility recordings were obtained on days 68, 75, 118, and 146.
Scoring of State Behavioral state was scored by 1. 2. 3. 4.
direct observation, from video-taped recordings, from visual inspection of the waveforms, and by the computer algorithm.
The following states were scored between birth and eye opening: wake, sleep-wake transition, active sleep, and quiet sleep. After eye opening, the wake state was separated into active wake and quiet wake. The state definitions for the rat are the same as for the rabbit. The distinction between active and quiet wake after eye opening is as follows. Active wake: locomotion and gross motor activity predominate. Quiet wake: there is lack of activity; the animal maintains an erect posture with eyes open and alert. See Thoman, Zeidner, and Denenberg (27) for analog signals prototypical of these states. Prior to eye opening, behavioral state was scored every 10 s. After eye opening state, changes did not occur as frequently, and state was scored every 30 s. Direct observation. An experienced observer sat about 18 inches from the subject. The observer code-recorded a state category for each 10-s epoch. An electronic timer signaled, through an earphone, every t0 s with a unique tone at 1-min intervals. Video-tape scoring. The video-taped recording included the audio electronic timer signals at 10-s and 1-min intervals. At
TABLE 2 PERCENT OF TIME SPENT IN EACH STATE AS DETERMINED BY COMPUTER SCORING Species
Age
Rabbit [12] Rat [28] Rat [31] Rat [12]
Day 1 Days 1-12 Days 14-21 Adulthood
Number in brackets = number of animals. Mean _ SE. NS = not scored.
ActiveWake
Quiet Wake
Sleep-Wake Transition
Active Sleep
Quiet Sleep
30.8 7.0 5.6 40.0
NS NS 24.7 _+ 1.3 24.1 _+ 6.1
11.5 + 0.4 7.0 ___0.8 4.9 ___0.5 2.0 + 0.4
57.3 _ 1.4 85.6 _+ 1.8 49.8 _+ 2.0 14.7 ___3.9
0.3 __+0.1 0.4 _+ 0.1 14.9 ___ 1.2 19.1 + 4.6
___1.6 + 1.3 _+0.5 _+ 6.2
272
CARROLL, DENENBERG AND [HOMAN
least 30 days after the recording was obtained, an experienced observer watched each tape and assigned state categories. For the first 12 days, a 10-s epoch was used. After that, the epoch was 30 s long. Visual signal scoring. The analog data were divided into 30s epochs. At least 30 days after behavioral state recordings were obtained, an experienced state scorer viewed the signals on a computer screen and assigned a specific state category to each 30-s epoch. Computer scoring. Three sets of templates representative of the five behavioral states were compiled from analog signals from several different animals. One set consisted of signals from days 1-5 and was used to score data from days 1-12 in 10-s epochs. The second set of templates consisted of analog signals from days 17-21, and was used to score data from days 14-21 in 30s epochs. The third set was taken from animals that were 68, 75, 118, and 146 days old and was used to score data from adult animals in 30-s epochs. After eye opening (i.e, days 14-16), a smoothing procedure was imposed upon the computer-scored data. The following rules were used: 1. Isolated epochs of active wake surrounded by active sleep or quiet sleep were changed to transition. 2. Isolated epochs of transition surrounded by active wake or quiet wake were changed to the prevailing state. 3. If more than three transitions occurred in succession, beginning with the fourth epoch, they were changed to active wake. 4. Isolated epochs of either sleep state surrounded by another state were changed to the prevailing state. 5. Isolated epochs of quiet wake surrounded by active sleep, quiet sleep, or active wake were changed to the prevailing state. Inspection of the state record before and after the smoothing consistently revealed that less than 10% of the epochs were changed. RESULTS
Birth to Eye Opening (Days 1-12) Correlations between direct observation and video-tape scoring for the four states summed across days 1-11 ranged from 0.95 to 0.98 (Table 1). Correlations between the video-tape scoring and computer scoring for each of the four states summed over days 1-12 ranged from 0.66 to 0.83. The percent time spent in each state over this age range is in Table 2.
Eye Opening to Weaning (Days 14-21) The correlations between video-tape and computer scoring, summed over days 14-21, for the five behavioral states ranged from 0.80 to 0.97. The percent time spent in each state over this age range is given in Table 2.
Adulthood The correlations between visual scoring of the analog signals by an observer and computer scoring of the same signals. summed across the four adult ages, ranged from 0.81 to 0.99. The percent time spent in each state over this age range is given in Table 2. GENERAL DISCUSSION In both the rabbit and the rat, the correlation coefficients for computer scoring of the wake states, active sleep, and quiet sleep were very high. Corelations for sleep--wake transition were lower, though still acceptable. This state is the most difficult to identify regardless of the measurement procedure. These correlations establish that the electronic activity monitor is a valid instrument to record behavioral states. Thoman and Glazier (23) validated computer-scoring of motility signals against direct observations of behavioral states in human infants. The present results show that computerized template matching is equally valid for categorizing behavioral states in newborn rabbits, and infant, and adult rats. For behavioral state scoring of sleep in the rabbit, individualized sets of templates were chosen from each rabbit's analog record. In the rat, it is important to note that standardized sets of templates, chosen from six different animals, were used for all subjects in the study. This procedure removes subjectivity in scoring and suggests an invariance in motility patterns across individuals (27). Three sets of templates were necessary to validly characterize behavioral states from birth through adulthood. The motility signals changed qualitatively at eye opening (days 1416) and again at about the time of puberty (between days 4050). Qualitative changes are also found in EEG parameters that characterize sleep states at about the same ages (11,16). Quantitatively, the changes in the amount of time that rats spent in each of the states from birth through adulthood indicate the sensitivity of the recording and scoring procedures to developmental effects. Though we only used this procedure with rabbits on day 1 of life, we have shown that direct behavioral observations correlate highly with a judge's scoring of the analog signals from the activity monitor over the age range of 7-19 days (26,27). Further, when analog signal recordings and hippocampal EEG recordings of two adult rabbits were compared, using 10-s epochs as the unit of measurement, there was 85.4% agreement (30), Therefore, the activity monitoring system is applicable for both the rat and the rabbit from birth through adulthood. The motility recording procedure permits prolonged recordings of behavioral states in the freely moving animal. Computerized recording and scoring is more objective and economical than visual scoring, and eliminates interference caused by observational procedures after eye opening.
REFERENCES 1. Acebo, C.; Watson, R. K.; Bakos, L.; Thoman, E. B. Sleep and apnea in the elderly: Reliability and validity of 24-hour recordings in the home. Sleep 14:56-64; 1991. 2. Alfoldi, P.; Tobler, I.; Borbely, A. A. Sleep regulation in rats during early development. Am. J. Physiol. 258:R634-R654; 1990. 3. Anders, T. F.; Chalemian, R. J. The effectsof circumcision on sleep-wake states in human neonates. Psychosom. Med. 36:174-179; 1974. 4. Anders, T.; Sostek-Miller, A. The use of time lapse video recording of sleep-wake behavior in human infants. Psychophysiology13:155158; 1976.
5. Brackbill, Y. Continuous stimulation reduces arousal level: Stability of the effect over time. Child Dev. 44:43-46; 1973. 6. Carroll, D. A.; Thoman, E. B.; Denenberg, V. H. Computer scoring of respiration and motility patterns for behavioral states in the neonatal rabbit. Poster presented at the Northeastern Sleep Society 3rd annual meeting, White Plains, NY; March 31-April 1: 1989. 7. Carroll, D. A.; Thoman, E. B.; Denenberg, V. H. Development of sleep and wake states of rats. Poster p~ented at the Northeastern Sleep Society 4th annual meeting, Cambridge, MA; March 3 l-April 1: 1990.
B E H A V I O R A L S T A T E IN R A T S A N D RABBITS
8. Denenberg, V. H.; Zeidner, L. P.; Thoman, E. B.; Kramer, P.; Rowe, J. C.; Philipps, A. F.; Raye, J. R. Effects oftheophylline on behavioral state development in the newborn rabbit. J. Pharmacol. Exp. Ther. 221:604-608; 1982. 9. Emde, R. N.; Harmon, R. J.; Metcalf, D.; Koenig, K. L.; Chalemian, R. J. Stress and neonatal sleep. Psychosom. Med. 33:491--497; 1971. 10. Erkinjuntti, M.; Kero, P.; Halonen, J. P.; Miloka, H.; Sainio, K. SCSB method compared to EEG-based polygraphy in sleep state scoring of newborn infants. Acta Paediatr. Scand. 79:274-279; 1990. 11. Gramsbergen, A. The development of the EEG in the rat. Dev. Psychobiol. 9:501-515; 1976. 12. Gramsbergen, A.; Schwartze, P.; Prechtl, H. F. R. The postnatal development of behavioral states in the rat. Dev. Psychobiol. 3:267280; 1970. 13. Hilakivi, L. A.; Hilakivi, I. T. Sleep-wake behavior of newborn rats recorded with movement sensitive method. Behav. Brain Res. 19: 241-248; 1986. 14. Hoppenbrouwers, T.; Hodgman, J.; Arakawa, K.; Geidel, S. A.; Sterman, M. B. Sleep and waking states in infancy: Normative studies. Sleep 11:387-401; 1988. 15. loffe, S.; Chernick, V. Prediction of subsequent motor and mental retardation in newborn infants exposed to alcohol in utero by computerized EEG analysis. Neuropediatrics 21:11-17; 1990. 16. Jouvet-Mounier, D.; Astic, L.; Lacote, D. Ontogenesis of the states of sleep in rat, cat, and guinea pig during the first postnatal month. Dev. Psychobiol. 2:216-239; 1970. 17. Keefe, M. R.; Kotzer, A. M.; Reuss, J. L.; Sander, L. W. Development of a system for monitoring infant state behavior. Nurs. Res. 38:344347; 1989. 18. Sadeh, A.; Lavie, P.; Scher, A.; Tirosh, E.; Epstein, R. Actigraphic home-monitoring sleep-disturbed and control infants and young children: A new method for pediatric assessment of sleep--wake patterns. Pediatrics 87:494-499; 1991. 19. Sander, L.; Julia, H.; Stechler, G.; Burns, P. Continuous 24-hour
273
20. 21. 22.
23. 24. 25. 26. 27. 28. 29. 30.
interactional monitoring in infants reared in two caretaking environments. Psychosom. Med. 34:270-282; 1972. Thoman, E. B. Sleeping and waking states in infants: A functional perspective. Neurosci. Biobehav. Rev. 14:93-1071; 1990. Thoman, E. B.; Acebo, C.; Becker, P. T. Infant crying and stability in the mother-infant relationship: A systems analysis. Child Dev. 54:653-659; 1983. Thoman, E. B.; Denenberg, V. H.; Sievel, J.; Zeidner, L. P.; Becker, P. State organization in neonates: Developmental inconsistency indicates risk for developmental dysfunction. Neuropediatrics 12:4554; 1981. Thoman, E. B.; Glazier, R. C. Computer scoring of motility patterns for states of sleep and wakefulness: Human infants. Sleep 10:122129; 1987. Thoman, E. B.; Tynan, W. D. Sleep states and wakefulness in human infants: Profiles from motility monitoring. Physiol. Behav. 23:519525; 1979. Thoman, E. B.; Waite, S. P.; DeSantis, D. T.; Denenberg, V. H. Ontogeny of sleep and wake states in the rabbit. Anim. Behav. 27: 95-106; 1979. Thoman, E. B.; Zeidner, L. P. Sleep--wake states in infant rabbits: Profiles from motility monitoring. Physiol. Behav. 22:1049-1054; 1979. Thoman, E. B.; Zeidner, L. P.; Denenberg, V. H. Cross-species invariance in state-related motility patterns. Am. J. Physiol. 10:R312R315; 1990. Vincent, F. A. SQUEEK--A computer algorithm for real time automatic sleep state scoring of mouse electroencephalogram. Thesis. Stanford University, Stanford, CA.; 1976. Waite, S. P.; DeSantis, D. S.; Thoman, E. B.; Denenberg, V. H. The predictive validity of early sleep development on later behavioral characteristics in the rabbit. Biol. Behav. 2:249-261; 1977. Zeidner, L. P.; Denenberg, V. H.; Thoman, E. B.; Weyand, T. Comparisons of behavioral, motoric and electrical criteria for assessment of sleep-wake states in the rabbit. Physiol. Behav. 31:273-278; 1983.