Short-term homeostasis of REM sleep assessed in an intermittent REM sleep deprivation protocol in the rat

J. Sleep Res. (2002) 11, 81 89 Short-term homeostasis of REM sleep assessed in an intermittent REM sleep deprivation protocol in the rat ADRIÁN OCAMPO-GARCÉS 1 and ENNIO A. VIVALDI 2 1 Departamento de Neurología y Neurocirugı a, Hospital Clínico Jose Joaquín Aguirre, Universidad de Chile, Santiago, Chile and 2 Programa de Fisiología y Biofísica, Instituto de Ciencias Biomédicas, Facultad de Medicina, Universidad de Chile, Santiago, Chile Accepted in revised form 28 November 2001; received 10 January 2001 SUMMARY An intermittent rapid eye movement (REM) sleep deprivation protocol was applied to determine whether an increase in REM sleep propensity occurs throughout an interval without REM sleep comparable with the spontaneous sleep cycle of the rat. Seven chronically implanted rats under a 12 : 12 light dark schedule were subjected to an intermittent REM sleep deprivation protocol that started at hour 6 after lights-on and lasted for 3 h. It consisted of six instances of a 10-min REM sleep permission window alternating with a 20-min REM sleep deprivation window. REM sleep increased throughout the protocol, so that total REM sleep in the two REM sleep permission windows of the third hour became comparable with that expected in the corresponding baseline hour. Attempted REM sleep transitions were already increased in the second deprivation window. Attempted transitions to REM sleep were more frequent in the second than in the first half of any 20-min deprivation window. From one deprivation window to the next, transitions to REM sleep changed in correspondence to the amount of REM sleep in the permission window in-between. Our results suggest that: (i) REM sleep pressure increases throughout a time segment similar in duration to a spontaneous interval without REM sleep; (ii) it diminishes during REM sleep occurrence; and (iii) that drop is proportional to the intervening amount of REM sleep. These results are consistent with a homeostatic REM sleep regulatory mechanism that operates in the time scale of spontaneous sleep cycle. KEYWORDS rapid eye movement sleep, REM sleep propensity, short-term REM sleep homeostasis, sleep cycle, sleep deprivation, ultradian rhythms INTRODUCTION Studies in rats, cats and humans support the existence of a short-term homeostatic process operating within sleep cycles. That process is reflected by the fact that the duration of an interval without rapid eye movement (REM) sleep is proportional to the duration of the preceding REM sleep episode (Barbato and Wehr 1998; Benington and Heller 1994a; Ocampo et al. 1990; Ursin 1970; Vivaldi et al. 1994a). This suggests the presence of an hourglass type mechanism where REM sleep propensity accumulates in the course of an interval Correspondence: Dr Ennio A. Vivaldi, Programa de Fisiología y Biofísica, Instituto de Ciencias Biomédicas, Facultad de Medicina, Universidad de Chile, Independencia 1027, Casilla 70005, Correo 7, Santiago, Chile. Tel.: +56 2 678 6422; fax: +56 2 777 6916; e-mail: evivaldi@machi.med.uchile.cl without REM sleep and dissipates during the next REM sleep episode. Accordingly, if a REM sleep episode is longer, it dissipates more REM sleep propensity and will take longer to buildup again enough pressure to trigger the next episode. In undisturbed sleep, the duration of a REM sleep episode is not correlated with the duration of the previous interval without REM sleep, suggesting that in spontaneous cycling a REM sleep episode is terminated by factors other than homeostatic (Barbato and Wehr 1998; Benington and Heller 1994a; Ocampo et al. 1990; Ursin 1970; Vivaldi et al. 1994a). The duration of spontaneous sleep cycles in the rat averages about 12 min in the rest-predominant lights-on phase (Trachsel et al. 1991; Vivaldi et al. 1994a). The number of attempts at transitions from non-rapid eye movement (NREM) sleep to REM sleep increases in the absence of REM sleep episodes and may be taken as an index Ó 2002 European Sleep Research Society 81

82 A. Ocampo-Garce s and E. A. Vivaldi of the REM sleep propensity (Benington and Heller 1994a; Endo et al. 1997; Siegel and Gordon 1965). The increase in the number of REM sleep attempts is affected by the duration of the preceding REM sleep episode, where longer episodes are followed by intervals with less attempts (Vivaldi et al. 1998). In selective REM sleep deprivation protocols the number of interventions required for maintaining the deprivation increases steadily during at least the two first hours of the protocol (Benington et al. 1994; Endo et al. 1997; Ocampo-Garce s et al. 2000). The REM sleep homeostasis and selective REM sleep deprivation have been studied through deprivation protocols in the hours or days scale (Rechtschaffen et al. 1999). The proposition that a short-term homeostatic mechanism is one of the factors that regulate REM sleep occurrence, allows to approach the issue of selective REM sleep deprivation within a time scale in the order of magnitude of the spontaneous REM sleep NREM sleep cycling in the rat. In the present study, we have designed an intermittent selective REM sleep deprivation protocol. Deprivation was performed at a time within the 12 : 12 light dark schedule when the REM sleep expression is highest and slow wave activity in NREM sleep has already declined (Franken et al. 1991a,b; Trachsel et al. 1988; Vivaldi et al. 1994b). Through this protocol we intended to demonstrate that propensity increases throughout an interval without REM sleep that is within the temporal scale of the rat ultradian sleep cycle; and that there is a relationship between REM sleep propensity as assessed by attempts at transitions and the expression of the REM sleep state. METHODS Data acquisition Seven Sprague Dawley rats weighting 300 350 g were implanted under deep chlornembutal 3 ml kg )1 i.p. anesthesia with four epidural and two neck muscle stainless steel electrodes. Cortical electrodes allowed the detection of delta and sigma activities. Two midline epidural electrodes allowed theta activity detection. Rats were housed in a 30 30 25 cm cage, placed within a sound-isolated cube, under a 12 : 12 light dark schedule, ambient temperature of 21 24 C, with water and food ad libitum. The data acquisition program sampled four channels looking for a corresponding relevant element: single delta waves (1 4 Hz), trains of three consecutive sigma waves (11 16 Hz), trains of eight theta waves (4 8 Hz), and muscle spikes or movement artifacts. The software quantified the amount of detected elements in each successive 15-s epoch and built a four column table with that data. An off-line state scoring algorithm assigned each epoch to W, NREM sleep or REM sleep. (Roncagliolo and Vivaldi 1991; Vivaldi et al. 1994a,b). The algorithm favored the scoring of REM sleep over NREM sleep, so that an epoch could be assigned to REM sleep even if less than half of it actually belonged to that state. On the other hand, when the state-by-epoch arrays were processed, two consecutive REM sleep epochs were required to establish a REM sleep episode. Isolated single epochs were considered missed REM sleep transitions. Besides automated scoring, the 3-h experimental and baseline sessions were recorded on paper and visually analyzed by two independent scorers. Selective REM sleep deprivation procedure and deprivation protocols Ten days after surgery, rats were adapted for at least 2 days to the recording environment. Two baseline recordings were then obtained in two consecutive days, to be followed by an experimental recording. Hours are denominated according to the light dark schedule, 0 being the first hour after lights-on. When we refer to a given hour we mean the 1-h interval that starts at that time, e.g. hour 6 refers to the interval from 6:00 to 7:00. Recordings started at hour 3 and continued up to hour 11. Figure 1 illustrates the 3-h REM sleep intermittent deprivation protocol (IRD) that comprised hours 6 8. The experimental protocol consisted in the alternation, for six successive times, of 10 min without interventions (REM sleep permission window), and 20 min of selective REM sleep deprivation (REM sleep deprivation window). We use the term bin to refer to each of these six successive sequences of one permission and one deprivation window (Fig. 1). Each rat was subjected up to two times to the protocol. At least 1 day was left between trials. A total of 11 protocols were successfully run. During deprivation, interventions to interrupt REM sleep were prompted by the appearance of the first unequivocal REM sleep signs (burst of sleep spindles, decreasing amplitude of cortical EEG, sustained midline theta trains, and lowering muscle tonus). The 3 5 s of REM sleep the rats could obtain Figure 1. Diagram of the intermittent rapid eye movement (REM) sleep deprivation protocol (IRD). Each pair of a 10-min permission window and its following 20-min selective REM sleep deprivation window constitutes a half-hour bin numbered from 1 to 6.

Intermittent REM sleep deprivation in the rat 83 before REM sleep was interrupted were not included in the analysis. The intervention to interrupt REM sleep was the gentle movement of the cage that was standing over a foam cushion, by means of a manually operated mechanism connected to a lever outside the isolation box. Interventions were documented on polygraphic paper. In less than 20% of cases REM sleep ended spontaneously before the intervention, an event designated as a missed REM sleep transition. Rapid eye movement sleep transitions were counted as such whether they were spontaneously or externally interrupted. They had to be separated by at least 15 s from one another to be counted as distinct events. Any ongoing REM sleep episode occurring at the time in which a deprivation window had to start, was immediately interrupted. Data analysis and statistics The values of several variables were calculated, as stated below, for hours or for bins or for halves of deprivation windows. All rats had two baselines that are averaged to obtain a single baseline value. Four rats had two experimental protocols and three rats had one experimental protocol. In the former case, results are treated similarly to baselines, i.e. the two experimental protocols are averaged to obtain a single experimental value. In the latter case the single value is used. The hourly amount of REM sleep and NREM sleep throughout the recording hours (3 11 after lights-on) were compared by ANOVAs with the factors hour (nine levels), protocol (two levels: baseline and experimental) and the interaction of hour and protocol (hour protocol). Posthoc analysis within protocols were performed using the Tukey multiple range test. As in the experimental protocol IRD took place during hours 6 8, the time interval analyzed corresponded to the 3 h occurring before, during and after IRD. In the 20-min deprivation windows within IRD the following variables were analyzed: (a) amount of NREM sleep, (b) number of REM sleep transitions and (c) REM sleep transition index. The REM sleep transition index was quantified as the number of transitions per 10 min of sleep. In the 10-min permission windows within IRD the following variables were analyzed: (a) amount of NREM sleep, (b) amount of REM sleep, (c) fraction of total sleep time (TST) occupied by REM sleep (REM/TST), (d) number of REM sleep episodes, and (e) duration of REM sleep episodes. Transition index and REM/ TST were calculated only if TST amounted to at least 10% of the time window. In order to compare the amount of NREM sleep and of REM sleep transitions between baseline and deprivation windows, the actual baseline values in the 30-min period were multiplied by 2/3. Analogously, to compare amount of NREM sleep, the amount of REM sleep and number of REM sleep episodes between baseline and permission windows, the actual baseline values were multiplied by 1/3. In the cases of number and duration of REM sleep episodes, data were aggregated in two-bin groups. All variables were subjected to ANOVAs with either factor bin (six levels) or two bin aggregate (three levels), factor protocol (two levels: baseline and experimental) and the interaction of bin or two bin aggregate and protocol (bin protocol). Post-hoc analysis within protocols were performed using the Tukey multiple range test. When the ANOVA showed a significant effect for factor protocol or interaction, paired t-tests were performed between corresponding baseline and experimental values. The REM sleep transition index was calculated for the 10-min halves of the deprivation windows. To be included in the analysis both halves had to contain at least 1 min of NREM sleep. An ANOVA was made with factor bin (six levels), half (two levels) and interaction of bin and half (bin half). The average transition index for the first and second half of the deprivation windows in each rat was also calculated and a comparison was made by a paired t-test. A relationship between amount of REM sleep in a permission window and change in REM sleep propensity was sought by regressing the difference in transition index between two consecutive deprivation windows on the amount of REM sleep in the intervening permission window. Ordinary linear and robust regression (Hamilton 1998) methods were used. RESULTS REM and NREM sleep before, during and after the intermittent deprivation protocol The time courses of REM and NREM sleep in hours 3 11 are summarized in Fig. 2 with a 1-h resolution. IRD occurs in hours 6 8 in the experimental protocol. The amount of REM sleep during IRD is the sum of their corresponding two 10-min REM sleep permission windows. The two-way ANOVA for response variable REM sleep content showed significant effects for factors hour (F ¼ 2.91, P ¼ 0.0057) and hour protocol (F ¼ 3.47, P ¼ 0.0014), but not for factor protocol (F ¼ 0.17, P ¼ 0.6795). The average hourly amount of REM sleep was 6.3 min for baseline and 6.1 min for the experimental protocol. Post-hoc analysis within the experimental condition showed that hour 9 (i.e. the first hour after IRD) differed significantly from all others. Hour 6 (i.e. the first hour of IRD) differed significantly from all others except 7 and 3 (with the latter the significance reached P < 0.08). The only other significant pairwise comparison was between hours 7 and 10. In the baseline condition no pairwise comparison reached significance. The paired t-tests showed significant differences for hour 6(P< 0.02) and 9 (P < 0.001), and in hour 7 it resulted in P ¼ 0.065. The two-way ANOVA for response variable NREM sleep content showed a significant effect only for factor hour (F ¼ 3.79, P ¼ 0.0006), not for factors protocol nor hour protocol. The average hourly amount of NREM sleep was 29.1 min for baseline and 28.4 min for the experimental protocol. This significance of the factor hour

84 A. Ocampo-Garce s and E. A. Vivaldi a experimental protocol (IRD), a 36% decrease (P < 0.05 in a paired t-test). In hours 9 11 REM sleep average was 17.7 min during baseline and 25.4 min during the experimental protocol, a 44% increase (P < 0.005). For NREM sleep in hours 6 8 the values were, respectively, 98.1 and 90.1, an 8% decrease (P > 0.05). In hours 9 11 the values were, respectively, 72.6 and 74.1, a 2% increase (P > 0.05). NREM sleep, REM sleep, REM sleep transitions and transition index throughout the IRD protocol b Figure 2. Hourly mean values (±SEM) of rapid eye movement (REM) sleep (a) and non-rapid eye movement (NREM) sleep (b) before, during (horizontal hatched bar in the abscissa) and after the IRD protocol (filled symbols), and throughout the corresponding 9 h of baseline (open symbols). The amount of REM sleep during IRD (hours 6 8 in the experimental protocol) is the sum of the two 10-min REM sleep permission windows corresponding to that hour. Analysis of variance (ANOVA) tests showed significant effects for factors hour and hour protocol in the case of variable REM sleep and in factor hour in the case of NREM sleep. When a significant ANOVA for factor protocol or hour protocol occurred, paired t-tests between baseline and experimental values were performed (results are indicated by * ¼ 0.05 and ** ¼ 0.01). See text for a summary of statistical analysis. for NREM sleep is mainly as a result of lesser amount observed in the last 3 h in both protocols. The amount of REM sleep and the amount of NREM sleep were added within 3-h groups (hours 3 5, 6 8 and 9 11) for baseline and experimental conditions. The ANOVA for REM sleep was not significant for factor 3-h nor factor protocol, but it was significant for the interaction factor (P < 0.01). The ANOVA for NREM sleep was significant for factor 3-h (P < 0.001) and it was not significant for factor protocol nor the interaction. The amount of REM sleep in groups 6 8 was 20.3 min during baseline and 13.0 min during the Figure 3 displays the time course of six variables through the bins that constitute the IRD. Figure 3a confirms that NREM sleep is not affected throughout the six REM sleep deprivation windows of IRD, the ANOVA not being significant for factor protocol (F ¼ 1.33, P ¼ 0.2524). Figure 3(b,c) shows the temporal courses of REM sleep transitions and transition index. Factor protocol differs significantly in both variables (F ¼ 47.35, P < 0.00005 in the former and F ¼ 45.83, P < 0.00005 in the latter). Pairwise post-hoc analysis for both variables show that in IRD bin 1 differs significantly from bins 2 6 and that there are no significant differences among these last five bins. This indicates that both variables increase from the first to the second deprivation window and afterwards they stay at that higher than baseline level without undergoing further increase. The paired t-tests showed significant differences in bins 2, 4, 5 and 6 both for variables REM sleep transitions and transition index. Figure 3(d f) shows that throughout the permission windows of the experimental protocol the amount of NREM sleep tends to decrease while the amount of REM sleep, and hence REM/TST follows an opposite course. ANOVAs with model bin, protocol and bin protocol are highly significant for all three variables. Factor protocol is significant for NREM sleep (F ¼ 19.44, P < 0.00005), REM sleep (F ¼ 17.26, P ¼ 0.0001) and REM/TST (F ¼ 51.27, P < 0.00005). For variable REM/TST factors bin (F ¼ 6.25, P ¼ 0.0001) and bin protocol (F ¼ 4.45, P ¼ 0.0014) are also significant. For this same variable REM/TST, pairwise comparisons are significant for bin 1 vs. 5 and 6, for bin 2 vs. 4, 5 and 6, for bin 3 vs. 6 and for bin 4 vs. 6. The paired t-tests showed significant differences in bins 1 and 6 for variable NREM sleep; in bin 6 for variable REM sleep; and in bins 1, 5 and 6 for variable REM/TST. Putting all variables in context, the increase in REM sleep transitions and transition index in the deprivation window is the first significant change to be assessed, already evident at bin 2, followed by the change in REM/TST in the permission window. Number and duration of REM sleep episodes throughout the IRD protocol Figure 4a shows the incidence of REM sleep episodes throughout the IRD protocol. Results are aggregated in two-bin groups, and the number of episodes is standardized to a 20-min base to compare experimental and baseline results. The graphs

Intermittent REM sleep deprivation in the rat 85 a d b e c f Figure 3. Time course of non-rapid eye movement (NREM) sleep (a and d), number of REM sleep transitions (b), transition index (c), rapid eye movement (REM) sleep (e) and REM/total sleep time (TST) (f), during IRD protocol (filled symbols), and corresponding baseline values (open symbols). IRD values in (a) (c) were obtained during the 20-min REM sleep deprivation windows. IRD values in (d) (f) were obtained during the 10-min REM sleep permission windows. As explained in the Methods section, corresponding 30-min baseline values are multiplied by 2/3 in (a) and (b) and by 1/3 in (d) and (e) to allow for comparison with the respective IRD window. The number of observations in each bin that qualified for analysis is shown over the abscissa. Analysis of variance (ANOVA) tests showed significant effects for factor protocol in all cases except the NREM sleep in deprivation window shown in panel (a). When a significant ANOVA for factor protocol or hour protocol occurred, paired t-tests between baseline and experimental values were performed (results are indicated by * ¼ 0.05 and ** ¼ 0.01). See text for a summary of statistical analysis. suggest that the increase in the amount of REM sleep observed in Fig. 3e is caused by an increase in the duration rather than in the number of episodes. For the latter variable the ANOVA is not significant, whereas for the duration of episodes factors bin group (F ¼ 9.99, P ¼ 0.0004) and protocol (F ¼ 13.61, P ¼ 0.0008) are both significant. Pairwise comparisons are significant for bin groups 1, 2 vs. both 3, 4 and 5, 6. The paired t-tests showed significant differences in 2-bin groups 3 + 4 and 5 + 6 for variable REM episode duration. REM sleep transitions within deprivation windows The time course of attempted transitions to REM sleep within deprivation windows was evaluated by comparing the transition index in the two halves. A requirement of at least 1.0 min of NREM sleep in both 10-min halves was established. Figure 5 shows that in all bins a higher transition index is observed in the second half of the deprivation window. The ANOVA was significant for factors half (F ¼ 17.19,

86 A. Ocampo-Garce s and E. A. Vivaldi a b Figure 4. Rapid eye movement (REM) sleep episode number (a) and REM sleep episode duration (b) in groups of two REM sleep permission windows (mean ± SEM) throughout IRD (filled symbols) and corresponding baseline hours (open symbols). As explained in the Methods section, in (a) the corresponding 1-h baseline values are multiplied by 1/3 to allow for comparison with the respective IRD data. The number of observations in each bin that qualified for analysis are shown over the abscissa. ANOVA tests showed significant effects for factor protocol in the case of variable episode duration. When a significant ANOVA for factor protocol or hour protocol occurred, paired t-tests between baseline and experimental values were performed (results are indicated by * ¼ 0.05 and ** ¼ 0.01). See text for a summary of statistical analysis. P ¼ 0.0001) and bin (F ¼ 2.75, P ¼ 0.0259). The average transition indexes of the first and second halves within each animal were also calculated and subjected to a paired t-test that was also significant (n ¼ 7, t ¼ 4.261, P ¼ 0.0053). REM sleep transition index in consecutive deprivation windows as a function of REM sleep in the intervening permission window Figure 6 shows the change in REM sleep transition index between two consecutive REM sleep deprivation windows as a function of the REM sleep present in the intervening REM sleep permission window. A negative slope was obtained with ordinary linear regression (n ¼ 42, slope ¼ )0.16, R 2 ¼ 0.15, P < 0.01) and with robust regression (n ¼ 42, slope ¼ )0.18, P < 0.001). The analysis was repeated leaving aside the observations with coordinates (0.75, 13.3), (3.75, 9.57) and (9.00, )3.33). The ordinary linear regression remained significant (n ¼ 39, P ¼ 0.011). Figure 5. Transitions indexes (mean ± SEM) in the first (upward triangles) and second (downward triangles) halves of each 20-min deprivation window are shown. Analysis of variance (ANOVA) tests were significant for factors half and bin. An average value within each rat for each half window was obtained, their mean values (±SEM) being represented by filled circles at the right of the graph. These average values were subjected to a paired t-test that resulted significant. The number of pairs of observations in each bin that qualified for analysis is shown over the abscissa. See text for a summary of statistical analysis. DISCUSSION The present report explored whether short-term REM sleep homeostasis in the rat can be evidenced by means of selective deprivation of REM sleep in a protocol performed within the order of magnitude of the spontaneous sleep cycle in the rat. The protocol lasted 3 h and consisted of six alternations of a 10-min REM sleep permission and a 20-min REM sleep deprivation window. The main goal of this study was to test two predictions that would derive from the existence of a short-term homeostatic mechanism: (1) that REM sleep pressure increases throughout a time interval comparable with the length of the interval without REM sleep in the spontaneous sleep cycle of the rat, and (2) that during a REM sleep episode a significant reduction of the REM sleep pressure occurs. During deprivation, the homeostatic pressure to REM sleep was operationally assessed as the transition rate, or attempts to enter REM sleep relative to the amount of NREM sleep. Selectivity, effectiveness and rebound of REM sleep deprivation The assessment of REM sleep requires a definition of a threshold duration that will distinguish an episode from an aborted attempt to enter the state. These criteria are relevant for variables such as the total amount of REM sleep, the average length of REM sleep cycles and the mean duration of episodes (Trachsel et al. 1991). As explained in Methods, we

Intermittent REM sleep deprivation in the rat 87 deprivation windows would add up to less than 10% of the average amount of REM sleep in permission windows. We did not include this REM sleep expression in our analysis and graphs. To be truly selective, a REM sleep deprivation procedure should not affect NREM sleep. In our study, when baseline and experimental values during IRD are compared, REM sleep is reduced from 20.3 to 13 min and NREM sleep from 98.1 to 90.1 min. Although these changes are similar in absolute terms, they constitute a 36% decrease in the case of REM sleep but only an 8% decrease in the case of NREM sleep, pairwise comparisons being significant only in the former case. Furthermore, we can anecdotally indicate that in two of the seven rats the absolute value of NREM sleep is higher in IRD than in baseline. Effect of REM sleep absence on REM sleep pressure Figure 6. Scatter plot of changes in transition rate between two consecutive REM sleep deprivation windows (following window minus preceding window) as a function of the REM sleep cumulated in the intervening REM sleep permission window. Regression analysis was significant with ordinary linear regression (slope ¼ )0.16, P < 0.01), with robust regression (slope ¼ )0.18, P < 0.001). It remained significant if the observations with coordinates (0.75, 13.3), (3.75, 9.57) and (9.00, )3.33) were not considered. See text for a summary of statistical analysis. detected missed REM sleep transitions and scored states using an epoch length of 15 s. The intermittent deprivation protocol was always started at the same phase of the light dark schedule, 6 h after lights-on. This phase was chosen because it is the time when REM sleep is attaining its highest expression (Franken et al. 1991a; Trachsel et al. 1988; Vivaldi et al. 1994a) and NREM sleep homeostatic pressure is already yielding (Franken et al. 1991b; Trachsel et al. 1988). The expression of REM sleep is much more strongly linked to the circadian oscillator than that of NREM sleep (Dijk and Czeisler 1995). Al least some aspects of REM sleep homeostasis appear to be under a circadian modulation. After a 2-h selective REM sleep deprivation in Wistar rats at two different phases within the light period, only the REM sleep deprivation performed late in the light period induced a homeostatic rebound (Benington et al. 1994). A study based on 24-h REM sleep deprivation ending at different circadian phases concludes that the attempts to enter REM sleep during deprivation, but not the amount of compensatory REM sleep expression, varies according to circadian phase (Wurts and Edgar 2000). Ideally, a REM sleep deprivation protocol should be complete and selective. Concerning the completeness of the deprivation, each time the animals attempted a REM sleep episode in a deprivation window they were not allowed to show signs of REM sleep for more than 3 s, usually theta activity without full cortical desynchronization. According to this estimation, the number of interrupted transitions in The ultradian sleep cycle is operationally defined as a REM sleep episode and the adjacent interval without REM sleep. The interval can be occupied exclusively by NREM sleep, as it is usually the case for humans, or by variable proportions of NREM sleep and wakefulness, as it occurs in the fragmented sleep of most rodents (Zepelin 1994). Failed attempts to enter REM sleep can be detected within the interval. The probability of occurrence of those aborted episodes increases throughout the interval (Benington and Heller 1994a), a fact that can be likened to the increase of attempts to enter REM sleep as the REM sleep deprivation protocol progresses. This analogy suggests that in baseline conditions during a normal interval REM sleep propensity augments as implied by short-term homeostasis. This implication is supported by our results. In fact, the length of our 20-min deprivation window is not far from the scale of spontaneous intervals, and when its 10-min halves are compared, attempts to enter REM sleep are much more frequent in the second half. Effect of REM sleep occurrence on REM sleep pressure The existence of a REM sleep homeostatic mechanism assumes that REM sleep pressure builds up during REM sleep deprivation and dissipates itself by the actual occurrence of REM sleep. The relationship between the magnitude of the debt acquired during deprivation and the compensatory rebound (Rechtschaffen et al. 1999) would indicate that REM sleep rebound ceases once the accumulated pressure has been dissipated (Parmeggiani et al. 1980). At the core of the short-term REM sleep homeostasis hypothesis is the concept that during a REM sleep episode REM sleep pressure is dissipated so that the longer the episode, the less REM sleep pressure remains and the longer the interval to the next REM sleep episode. Observations in the cat (Ursin 1970), rat (Benington and Heller 1994a; Vivaldi et al. 1994b) and human (Barbato and Wehr 1998) are consistent with this hypothesis. Schuller et al. (1999) have communicated that diminution of REM sleep propensity after a 30 or 60 min recovery following

88 A. Ocampo-Garce s and E. A. Vivaldi a 2-h selective REM sleep deprivation is determined by the time spent in REM sleep during the recovery interval. Our results indicate that the REM sleep transition index tends to diminish or increase from one deprivation window to the next according to the amount of REM sleep present in the permission window in-between. Effect of selective REM sleep deprivation on the duration of REM sleep episodes The increase in REM sleep during permission windows is explained by an increase in episode duration. In baseline conditions short-term homeostasis is not evidenced by a regulation of REM sleep episode length by the duration of its preceding interval (Barbato and Wehr 1998; Benington and Heller 1994a; Ursin 1970; Vivaldi et al. 1994a). In our intermittent REM sleep deprivation protocol we observed that REM sleep transition rate is already increased in the first hour of the protocol while REM/TST and REM sleep episode durations become significantly augmented at the second hour. These two results suggest that REM sleep episode duration responds homeostatically to the preceding interval without REM sleep only if the latter is long enough for sufficient pressure to accumulate. Organization of the sleep cycle: short-term homeostasis vs. ultradian oscillator Comparative studies of adult mammals indicate that the duration of the sleep cycle is related to constitutive variables such as body weight, metabolic rate and brain size (Zepelin 1994). This fact suggests that the ultradian sleep cycle is linked to metabolic and energetic requirements (Jouvet 1994; Parmeggiani 1990) that must somehow limit the duration of intervals without REM sleep. The periodic signal that prompts the occurrence of a REM sleep episode might be the output of an ultradian pacemaker as originally proposed by Kleitmann (for review, see Kleitman 1982). Rapid eye movement sleep alternation might also result from a REM sleep oscillator system that emerges from the interaction of REM-on and REM-off neurons (Hobson et al. 1975; McCarley and Massaquoi 1986). The issue of an oscillatory vs. an homeostatic control of REM sleep expression has been considered within the context of a discussion on whether the function of REM sleep is related to waking or to NREM sleep (Benington and Heller 1994b). As discussed elsewhere (Vivaldi et al. 1994a), a short-term homeostatic regulation is easier to reconcile with explanations of REM sleep periodic occurrence that assign to REM sleep itself a role in its cyclicity, either because the REM sleep generator may reciprocally interact with other state generators or because there is some functional property specific to REM sleep that may be translated into a feedback loop. The basic tenant of short-term homeostasis states that there must be some mechanism by which REM sleep pressure builds up when REM sleep is absent propitiating a new REM sleep episode, and by which REM sleep pressure dissipates when REM sleep occurs allowing thus for a new interval without the state. The rebound that follows REM sleep deprivation speaks eloquently in favor of a homeostatic mechanism that must be taken into account by any model aimed at explaining REM sleep periodicity. In conclusion, the results of this intermittent REM sleep deprivation protocol indicate that, within a time scale close to that of spontaneous sleep cycles, first, REM sleep pressure increases throughout a time segment not much longer in duration than that of a spontaneously occurring interval without REM sleep; secondly, REM sleep pressure diminishes during the REM sleep episode; and, thirdly, that decrease is related to the amount of REM sleep expressed. These results support the postulate that a short-term REM sleep homeostatic mechanism plays a role in the ultradian organization of the sleep cycle. ACKNOWLEDGEMENTS We thank Carolina Llanos and Alberto Rodríguez for their help in this study and Rodrigo Villegas for his advice in the statistical analysis. This work was supported by grants FONDECYT 1961227, 2970026 and 4980007. Dr Adrián Ocampo-Garce s was supported by a fellowship from the National Council of Science and Technology (CONICYT Chile). REFERENCES Barbato, G. and Wehr, T. A. Homeostatic regulation of REM sleep in humans during extended sleep. Sleep, 1998, 21: 267 276. Benington, J. H. and Heller, H. C. Does the function of REM sleep concern non-rem sleep or waking? Prog. Neurobiol., 1994b, 44: 433 449. Benington, J. H. and Heller, H. C. REM-sleep timing is controlled homeostatically by accumulation of REM-sleep propensity in non- REM sleep. Am. J. Physiol., 1994a, 266: R1992 R2000. Benington, J. H., Woudenberg, M. and Heller, H. C. REM-sleep propensity accumulates during 2-h REM-sleep deprivation in the rest period in rats. Neurosc. Lett., 1994, 180: 76 80. Dijk, D. J. and Czeisler, C. Contribution of the circadian pacemaker and the sleep homeostat to sleep propensity, sleep structure, electroencephalographic slow waves, and sleep spindle activity in humans. J. Neurosci., 1995, 15: 3526 3538. Endo, T., Schwierin, B., Borbély, A. A. and Tobler, I. Selective and total sleep deprivation: effect on the sleep EEG in the rat. Psychiatry Res., 1997, 66: 97 110. Franken, P., Dijk, D. J., Tobler, I. and Borbély, A. A. Sleep deprivation in rats: effects on EEG power spectra, vigilance states, and cortical temperature. Am. J. Physiol., 1991a, 261: R198 R208. Franken, P., Tobler, I. and Borbély, A. A. Sleep homeostasis in the rat. simulation of the time course of EEG slow-wave activity. Neurosci. Lett., 1991b, 130: 141 144. Hamilton, L. C. Statistics with Stata 5. Duxbury Press, Pacific Grove, USA, 1998. Hobson, J. A., McCarley, R. W. and Wyzinski, P. W. Sleep cycle oscillation: reciprocal discharge by two brainstem neuronal groups. Science, 1975, 189: 55 58. Jouvet, M. Paradoxical sleep mechanisms. Sleep, 1994, 17 (Suppl. 1): S77 S83.

Intermittent REM sleep deprivation in the rat 89 Kleitman, N. Basic rest-activity cycle 22 years later. Sleep, 1982, 5: 311 317. McCarley, R. W. and Massaquoi, S. G. A limit cycle mathematical model of the REM sleep oscillator system. Am. J. Physiol., 1986, 251: R1011 R1029. Ocampo, A., Wyneken, U. and Vivaldi, E. A. Active sleep and homeostatic regulation of sleep wake cycle. Arch. Biol. Med. Exp., 1990, 23 (Suppl. 2): R218. Ocampo-Garcés, A., Molina, E., Rodríguez, A. and Vivaldi, E. A. Homeostasis of REM sleep after total and selective sleep deprivation in the rat. J. Neurophysiol., 2000, 84: 2699 2702. Parmeggiani, P. L., Cianci, L., Calasso, M., Zamboni, G. and Perez, E. Quantitative analysis of short term deprivation and recovery of desynchronized sleep in cats. Electroencephalogr. Clin. Neurophysiol., 1980, 50: 293 302. Parmeggiani, P. L. Homeostatic function of the hypotalamus and control of the wake-sleep cycle. In: M. Mancia (Ed) The Diencephalon and Sleep. Raven Press, New York, 1990: 133 145. Rechtschaffen, A., Bergmann, B. M., Gilliland, M. A. and Bauer, K. Effect of method, duration, and sleep stage on rebounds from sleep deprivation in the rat. Sleep, 1999, 22: 11 31. Roncagliolo, M. and Vivaldi, E. A. Time course of rat sleep variables assessed by a microcomputer-generated data base. Brain Res. Bull., 1991, 27: 573 580. Schuller, J., Borbély, A. A. and Tobler, I. REM sleep regulation in the rat. Sleep Res. Online, 1999, 2 (Suppl. 1): 556. Siegel, J. and Gordon, T. P. Paradoxical sleep: deprivation in the cat. Science, 1965, 148: 978 980. Trachsel, L., Tobler, I. and Borbély, A. A. Electroencephalogram analysis of non-rapid eye movement sleep in rats. Am. J. Physiol., 1988, 255: R27 R37. Trachsel, L., Tobler, I., Achermann, P. and Borbély, A. A. Sleep continuity and the REM-non REM cycle in the rat under baseline conditions and after sleep deprivation. Physiol. Behav., 1991, 49: 575 580. Ursin, R. Sleep stage relation within the sleep cycles in the cat. Brain Res., 1970, 11: 91 97. Vivaldi, E. A., Ocampo, A., Roncagliolo, M., Wyneken, U. and Zapata, A. M. Short-term homeostasis of active sleep and the architecture of sleep in the rat. J. Neurophysiol., 1994a, 72: 1745 1755. Vivaldi, E. A., Wyneken, U., Roncagliolo, M., Ocampo, A. and Zapata, A. M. Measures of location and dispersion of sleep state distributions within the circular frame of a 12: 12 light: dark schedule in the rat. Sleep, 1994b, 17: 208 219. Vivaldi, E. A., Ocampo, A. and Zapata, A. M. Distribution of missed REM transitions throughout the intervals between REM episodes in the rat. J. Sleep Res., 1998, 7 (Suppl. 2): 297. Wurts, S. W. and Edgar, D. M. Circadian and homeostatic control of rapid eye movement (REM) sleep: promotion of REM tendency by the suprachiasmatic nucleus. J. Neurosci., 2000, 20: 4300 4310. Zepelin, H. Mammalian sleep. In: M. Kryger, T. Roth and W. C. Dement (Eds) Principles and Practice of Sleep Medicine. W.B. Saunders Company, Philadelphia, 1994, 69 80.