Response of Delta (0-3 Hz) EEG and Eye Movement Density to a Night with 100 Minutes of Sleep

Transcription

1 Sleep 11(5): , Raven Press, Ltd., New York 1988 Association of Professional Sleep Societies Response of Delta (0-3 Hz) EEG and Eye Movement Density to a Night with 100 Minutes of Sleep *,ti. Feinberg, tt. Baker, tr. Leder, and +1. D. March *VA Medical Center, Northport, New York; tstate University of New York (SUNY) at Stony Brook, New York; and idelta Software, San Francisco, California, U.S.A. Summary: In one of a series of experiments aimed at gathering the empirical data required to formulate mathematically our recovery model of sleep, we recently (1) measured the increase in delta electroencephalogram (EEG) following one night of total sleep deprivation (TSD). We found that the delta rebound was confined to the first non-rapid eye movement period (NREM-Pl) of recovery sleep; this unexpected result was documented with direct computer measurement of 3 Hz EEG, as well as with visual scoring of stages 3 and 4. We also found a robust decrease in eye movement density during the second and third REM periods, which we hypothesized to be due to the increased depth of recovery sleep. In the present experiment, we awakened young adult subjects after 100 min of sleep, a duration that includes the first cycle for this age group, and analyzed visual and computer measures of delta and eye movement density during recovery sleep. We again found eye movement density to be significantly reduced in REM-P2 and P3, but to a lesser degree than after total sleep deprivation, a condition that may be presumed to produce a greater increase in sleep depth. Delta increases were again limited to the first cycle, although all subjects completed this cycle on the IOO-min night. The major difference between recovery sleep patterns following the total deprivation and the IOO-min sleep conditions was that 3-Hz wave amplitude increased significantly after the former, but not after the latter. In both studies, recovery sleep showed increased 0-3-Hz wave density. The neurophysiological implications of a response of EEG amplitude as opposed to wave density are briefly considered; separate measurement ofthese variables is more readily accomplished with period-amplitude than with spectral analysis. Our results further illustrate the importance of measuring sleep by physiological units, such as the successive NREMPs and REMPs. They also support other data that indicate that NREM-Pl plays a special role in human sleep: it responds selectively to sleep deprivation, shows the greatest ontogenetic variation across the human lifespan, and is the component of sleep that is most frequently abnormal in psychiatric patients. As we have long argued, it is inappropriate to conceptualize this high priority component of NREM sleep as "REM latency" and as a measure of REM "pressure" exclusively. Key Accepted for publication April Address correspondence and reprint requests to Dr. 1. Feinberg at Psychiatry Service (I16A), V A Medical Center, Northport, NY 11768, U.S.A. 473

2 474 I. FEINBERG ET AL. Words: Sleep-Cycle-Deprivation-EEG-Delta-Computer-Period analysis-rem-eye movement density. It has long been known (2,3) that total sleep deprivation (TSD) increases the amount of visually scored high-amplitude delta electroencephalogram (EEG) (stages 3 and 4, S3 and S4, respectively) during recovery sleep. Recently, we reported (1) that the compensatory increase in delta after TSD was confined to the first period of nonrapid-eye-movement sleep (NREM-Pl). This was true for the high-amplitude delta of S3 and S4 and for total 0-3-Hz EEG measured by computer. On the recovery night after TSD, NREM-PI doubled in length and, as a consequence, total EEG in all frequency bands increased. However, an increased rate of production ofeeg waves (number or integrated amplitude per epoch of time) was found only in frequencies below 4 Hz; above 4 Hz, time/20-s epoch and, hence, integrated amplitude, were significantly decreased. In contrast to NREM, rapid-eye-movement (REM) sleep changed in later cycles after TSD. Eye movement density was substantially reduced in REM periods 2-4, although REM period duration was unaffected. These findings led us to hypothesize that eye movement density is an inverse function of depth of sleep, which is increased by TSD (3). A relation to depth of sleep could also explain the increase in eye movement density across sleep cycles (4) in normal sleep as sleep becomes progressively lighter, the abrupt increase in density in the terminal portion of extended sleep (5,6) when sleep is extremely light in early REM periods in schizophrenics, depressives, and the elderly, and lastly, the depression in eye movement density by sedative-hypnotic drugs (7), which increase sleep depth. Borbely and Wirz-Justice (8) have independently advanced a similar hypothesis, suggesting that eye movement density varies inversely with "propensity for delta." However, the fact that sleep in the later cycles after TSD is sufficiently deep to depress eye movement activity even though delta EEG is unchanged suggests that delta EEG does not adequately reflect sleep depth toward the end of the night. Further consideration of eye movement during REM sleep led us to hypothesize (1) that these movements themselves may be of no functional significance. Their occurrence may simply reflect that fact that in contrast to somatic musculature, activity of the eye muscles does not cause awakening and therefore need not be inhibited. Even if the eye movement of REM sleep can be so simply explained, the fundamental question that remains is: Why do higher motor (and sensory) nuclei show intense neuronal discharge during REM sleep, with an irregular bursting pattern indicating disinhibition and loss of modulating influences? The fact that delta compensation after TSD was limited to the first NREM period led us to wonder what would happen if deprivation of sleep was carried out after the first cycle of sleep had been completed. Perhaps this procedure would force the delta compensation into later cycles. It seemed equally possible that the delta rebound would remain limited to NREM-PI; a more remote possibility was that it might not occur at all. This experiment also provided an opportunity to assess the effects on eye movement density of a milder increase in sleep depth than that following TSD. Sleep depth was not measured operationally (i.e., with arousal thresholds), but it is reasonable to infer that permitting 100 min of sleep, with its high level of deep (stage 3-4) sleep, would Sleep, Vol. 11, No.5, 1988

3 RESPONSE TO 100 MIN SLEEP 475 produce a smaller increase in the depth of recovery sleep than does total sleep deprivation. This investigation is part of a research program that has been underway since 1975, the goal of which is to gather the empirical data needed to formulate mathematically a recovery model of sleep such as that we put forward in 1974 (4). We first developed effective and reliable methods to analyze EEG waveforms (9,10), and we next applied these methods to measure delta amplitudes and densities across sleep in young and elderly normal subjects (9-11). More recent studies have measured systematically the relations between varied waking durations and subsequent sleep EEG waveforms, with particular emphasis on the quantity and distribution of0--3-hz activity (1,12-15). Several components of our original recovery model were incorporated by Borbely (16), who also added a second (circadian) factor; for a comparison, see Feinberg and Floyd (17). Daan et al. (18) offered a mathematical formulation of the two-process model, which, though of considerable ingenuity, fails to predict a number of experimental results (1,12, and the present study). Our own view (19) is that a predictive model of sleep must take into account at least five factors in addition to the putative recovery function of NREM in relation to waking duration: age (see also, 20), intensity of brain activity during waking, the separate function of REM, circadian mechanisms, and entrainment of physiological systems. There may be considerable overlap in the last two factors. METHODS Subjects Subjects were students from SUNY-Stony Brook (n = 13). Most attended the medical and dental schools. There were 4 men and 9 women, ranging in age from 22 to 26 years (mean 22.9 years). All completed a brief psychiatric interview, and none admitted to abnormal or unusual sleep patterns. Design Subjects were instructed to maintain a bedtime schedule of 11:30 p.m. to 7 a.m. for the 4 nights prior to laboratory study. Their sleep was recorded on this schedule for adaptation and baseline nights. On the third consecutive laboratory night, they were allowed to sleep for 100 min. This duration was chosen so that the intense delta EEG of first NREM period would be completed by most subjects. (We did not wish to require that technicians decide during recording sessions that NREM-Pl was complete, as this decision requires detection of the first REM period, which can be difficult in young adults because of its weak eye movement and less pronounced EEG desynchronization.) After 100 min of sleep, subjects were awakened and remained awake and were observed continuously by laboratory personnel until 11 :30 p.m. the next night, when their sleep was again recorded until 7 a.m. Recording and analysis The C3-A2 EEG and vertical and horizontal electrooculogram (EOG) were recorded. Visual scoring was carried out as described previously (1), with sleep onset determined by the first spindle; the data were grouped into NREM periods and REM periods according to established criteria (21). The EEG was recorded on analog tape (Honeywell 101) and digitized offline at 200 Hz. The digitized signal was saved on magnetic Sleep, Vol. /1, No.5, 1988

4 476 I. FEINBERG ET AL. tape and was analyzed on a Digital Equipment LSI 11/73 with the same (PAN V) algorithms used previousiy with the PDP i2 (9); these methods are reliabie and vaiid (9, io). However, instead of using an active (Krohn-Hite) filter to center and smooth the EEG signal, a weak, passive, high-pass (0.01 Hz) filter and a one-pole digital high-pass filter set at Hz were employed. We are preparing a detailed comparison of the results generated by the PDP 12 and LSI 11/73 on the same data set; here, the EEG was analyzed with the 11/73 throughout. RESULTS Sleep totals Table 1A gives the total sleep time (TST), total NREM, and total REM for the three conditions, as well as NREM and REM measured by cycle. TST on recovery nights (RN) was slightly but reliably (p < 0.01) greater than on baseline (BN), due in part to the diminished latency (p < 0.02) of recovery sleep. Time in bed did not differ significantly in the two conditions. On the 100-min night, subjects actually averaged 108 min of total sleep, of which 99 min was NREM and 9 min was stage REM. Although the recording technicians aimed at awakening subjects after 100 min of sleep by online visual scoring, the subsequent definitive scoring yielded this slightly different value. REM measures Table 1A shows that on the RN, there was a 10% increase in stage REM. However, total eye movement was reduced, rather than increased, and eye movement density was significantly reduced in REM-P2 and P3, with a near-significant reduction in REM P4. After total sleep deprivation, eye movement density in REM-P2 and P3 was reduced to 49% and 25% of BN, respectively; in the present study, the corresponding reductions were smaller, to 74% and 72%, respectively, of BN. Thus, the density effects were similar in pattern across cycles in the two studies (significant reductions in REM-P2 and P3), but were stronger after total deprivation. NREM measures Table 1A shows that on the truncated night, total NREM sleep was 99.5 min, with S3 plus S4 equaling 40.6 min. The corresponding values in NREM-P1 of BN were and 37.4 min, respectively. The first NREM period was shorter on the 100-min night than on BN and RN. This may have been due to scoring unreliability (the scorers were relatively inexperienced) or to an actual alteration in cycle length, perhaps caused by expectation of being awakened early. This result does not impair the experimental design, which aimed to insure that on the abbreviated night, subjects averaged amounts of dense, high-amplitude delta and a duration of NREM that would equal the levels in NREM-P1 of BN. This goal was nicely achieved. On the RN, NREM-P1 was about 20 min longer and NREM-P2 about 15 min shorter than the corresponding values on BN. These changes, although not statistically significant, suggest some extension of NREM-P1 at the expense of NREM-P2. Stage 4 (S4) duration in NREM-P1 more than doubled on the RN (p < 0.001); there were no effects on S4 in NREM-P2-4 and only slight changes in S3. Computer measures (Table lb) of 3-Hz EEG confirmed that the increase in delta EEG on the RN was confined to NREM-Pl. Total integrated amplitude increased significantly, but this increase was primarily due to the increased length of NREM-P1, as integrated amplitude for the average 20-s NREM epoch did not differ significantly Sleep. Vol. 11. No

5 RESPONSE TO 100 MIN SLEEP 477 from baseline. The size of the average delta wave (average sample amplitude) in NREM-Pl also did not change. Both total time and time/epoch of 0-3-Hz waves increased significantly. The increased time was due mainly to an increase in number of waves/epoch (density), as the frequency of the average delta wave showed a relatively small (statistically insignificant) change; frequency during BN was ± Hz versus frequency during RN of ± Hz. Visually scored high-amplitude, dense delta (S3 + S4) in NREM-Pl on the RN was 175% of BN, whereas total integrated amplitude and total time in 0-3-Hz increased by smaller amounts (to 158%) and 151% ofbn). We observed a similar, but more marked, difference after total sleep deprivation. In that experiment, visually scored S3 + S4 in NREM-Pl increased to 309% ofbn, whereas 0-3-Hz integrated amplitude increased to 245% of BN. These results suggest that 0-3-Hz EEG becomes more concentrated on recovery from sleep deprivation. In considering the data in Table 2, it should be borne in mind that zero-cross analysis does not detect faster frequencies "riding" on larger slow waves that do not reach zero-voltage. This point will be discussed further below. Table 2 compares 0-23-Hz EEG in NREM-Pl of BN and RN by frequency band, presenting the 0-4-Hz EEG by I-Hz bins. The increased time/epoch in 0-3-Hz resulted mainly from an increase in the 1-2-Hz EEG (although the percent increase was greatest in O-I-Hz); the 1-2-Hz band normally contains the lion's share ofthe 0-3-Hz activity of human sleep (22). Table 2 shows that average sample amplitude on the RN did not increase through 23 Hz. However, the increase in time/epoch in 0-2-Hz was associated with significant decreases in time/epoch from 4 to 23 Hz. The decrease in time/epoch in higher frequencies, when time/epoch in delta increases, is a necessary consequence of the fact that epoch length is fixed: more time spent in one band (at the baseline) must result in less time in others. The reduction in time/epoch in the higher frequency bands, with average sample amplitude unchanged, led to significant decreases in their product (integrated amplitude/epoch) in 4-8, 8-12, and Hz bands. Although it remains theoretically possible that after deprivation, the amount of high-frequency waves riding on slow activity increased to compensate for that lost at the baseline, other aspects of our analysis indicate that this did not occur. A reasonably definitive statement will become possible when we complete a collaborative study with Dr. Richard Copola of the National Institute of Mental Health, who is subjecting the digitized data from this and other studies to spectral analysis so that we may compare the two methods. (It is already clear that the two methods give identical results for the delta frequency band.) Figure 1 compares EEG waveform changes in NREM-Pl during recovery from TSD and from the too-min sleep condition. The main difference was that delta wave amplitude did not increase after too-min sleep, whereas it did after TSD. Although the percent increase in time/epoch in 0-3-Hz was about the same under both conditions, the change was statistically more reliable after TSD. That the increased time spent in 0-3-Hz occurred at the expense of the higher frequencies was apparent in both experiments; again, the effects were more reliable after TSD. DISCUSSION Stage REM and eye movement density The proportion of REM sleep increases as sleep progresses both because REM periods become longer (23) and because NREM periods become shorter (4,24). For this Sleep, Vol. n, No.5, 1988

6 '" "... ; t::.. TABLE 1. NREM and REM measures for the first 4 sleep cycles on baseline (BN), 100 minute (100 min) and recovery nights (RN) (N = 13*, values are mean ± SEM. Total sleep times: BN = ± 2.75,100 min = ± 1.28, and RN = ± 1.84) Cycle All Cycles a A. Visually Scored Measures NREMP duration (min) BN ± ± ± ± ± min 99.5 ± ± ± 3.78 RN ± ± ± ± ± 2.85 Stage 3 (min) BN 29.8 ± ± ± ± ± min 13.4 ± ± ± 0.64 RN 22.6 ± ± ± ± ± O.13 d Stage 4 (min) BN 46.2 ± ± ± ± ± min 27.2 ± ± ± 1.5 RN 72.1 ± 7.45 e 53.8 ± 7.0(/ 9.8 ± ± ± 0.08 REMP duration (min) BN ± ± ± ± ± min 9.1 ± ± 1.79 RN ± 4.88 d 11.2 ± ± ± ± 5.67 Eye movement densityb BN 0.24 ± ± ± 0.Q ± ± 0.04 RN 0.16 ± O.OY 0.14 ± ± 0.02 e 0.15 ± 0.02d 0.13 ± 0.03 S2 t:x:l tl:i """l t-;

7 B. Computer measures of delta (0-3.0 Hz) Time/IOO (sec) BN 82.6 ± ± ± min 29.7 ± ± ± 1.21 RN 93.8 ± ± 5.10d 18.8 ± 1.49 Time/epoch (sec/20 sec) BN 9.1 ± ± min 11.5 ± ± 0.78 RN 9.8 ± d 9.9 ± 0.63 Average sample amplitude C (fj-v) BN 22.4 ± ± min 24.4 ± ± 1.64 RN 21.9 ± ± 1.47 Integrated amplitude (fj-v x sec/1000) BN ± ± min 73.3 ± ± 2.61 RN ± e 39.7 ± 5.32 Integrated amp/epoch (fj-v x sec/20 sec) BN ± ± min ± ± RN ± ± *For BN and RN, cycle 4, N = 12; for 100 min night, cycle 2, N = 9. a includes data for all NREM and REM sleep (including cycles 5, 6 and any incomplete cycles) per night. b 4 sec epochs with EM/4 sec epochs stage REM. C Average sample amplitude is integrated amplitude divided by time. d p < 0.05, differs from BN, paired 2-tailed t tests. " P < I P < ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ::tl a t; Cl '- <: [il '"ti '" " - '"" -'0 gg -t:...

8 '" i TABLE 2. EEG characteristics by frequency band in NREMP 1 on baseline (BN) and recovery night (RN) after sleep deprivation. Average sample Frequency band Integrated ampl. Integrated ampl.lepoch amplitude a Time Time/epoch (Hz) (myb sec) (uyb sec/20 sec) (uy) (sec) (sec/20 sec) 0-1 BN 14.1 ± ± ± ± ± 0.19 RN 23.7 ± 4.25 b 65.4 ± ± ± l.37 h 2.1 ± BN 39.1 ± ± ± ± ± 0.33 RN 63.4 ± 7.85 c ± ± ± 2.61b 6.9 ± 0.36 b 2-3 BN 15.1 ± ± ± ± ± 0.13 RN 21.0 ± 2.60 b 62.9 ± ± ± ± 0.12 til 3-4 BN 6.9 ± ± ± ± ± 0.07 b::i RN 8.3 ± ± ± ± ± C) BN 8.8 ± ± ± ± ± 0.15 RN 9.6 ± ± 1.89 b 10.4 ± ± ±0.12c..., 8-12 BN 3.1 ± ± ± ± ± 0.08 t-< RN 3.1 ± ± 0.91b 7.6 ± ± ± 0.08 d BN 1.0 ± ± ± ± ± 0.07 RN 0.8 ± ± 0.29 C 5.6 ± ± ± 0.03 b BN 1.0 ± ± ± ± ± 0.06 RN 0.8 ± ± ± ± ± 0.06 c N = 13, values are mean ± SEM. a average sample amplitude is integrated amplitude divided by time. b p < cp<o.oi. d p < 0.001, differs from BN, paired 2-tailed t tests.

9 RESPONSE TO 100 MIN SLEEP 481 Frequency Band (Hz) Min Sleep Total Sleep Deprivation Average Sample Amplitude (Integ Amp/Time) - + ** * *** Time/Epoch 8-12 * * ""_r_---r r--;-.--- I *.* 3-4 Integ Amp/Epoch 4-8 * * * * ** rh----r,--.--r o 20 Percent Change in Computer Measures from BN to RN in NREMPI «BN-RN/BN -100) p S 0.05 * P 0.01 * p So FIG. 1. Comparison of percent change from baseline levels in NREM-Pl for three EEG waveform measures during the recovery sleep following a night of total sleep deprivation and that following a night with 100 min of sleep. The amplitude of the average delta wave increased after total deprivation, but not after 100-min sleep. Delta wave density increased after both deprivation conditions; the increased delta waves/time was necessarily associated with a decreased density of waves in the higher frequencies, as the epoch length was fixed (see text for some limitations on this conclusion). The pattern shown by integrated amplitude is the consequence of the changes in the first two measures. reason, deprivation of the later portion of the night's sleep disproportionately decreases stage REM as opposed to NREM. Some would predict, therefore, that the partial deprivation of the loo-min sleep condition would produce a compensatory increase of stage REM duration on recovery, and a 10% increase (p <.05) was in fact observed_ However, as we discussed at length elsewhere (4), REM compensation is not a reliable phenomenon, even after much more severe deprivation. The findings with respect to eye movement activity during stage REM are consistent with our hypothesis that the occurrence of these movements is an inverse function of depth of sleep. Recovery from the milder deprivation of the 100-min sleep condition produced a smaller decrease in eye movement density in REM-P2 and P3 than was found after total sleep deprivation. We would interpret the fact that the reductions were Sleep, Vol. 11, No.5, 1988

10 482 I. FEINBERG ET AL. statistically significant only in REM-P2 and P3 in both experiments as follows. There is a "floor" effect in REM-Pi, which aiready has iow eye movement density due to the high sleep depth in the early part of the night (25). However, REM-P2 and P3 can reflect the increased sleep depth produced by partial deprivation, both because of their higher baseline level of eye movement activity and because sleep depth is no longer so high (i.e., because of initial value effects). By the time REM-P4 occurs, the increased sleep depth produced by these two deprivation conditions has largely (but not completely) dissipated as a result of the recovery processes. Eye movement density in REM-P4 was lower in both studies than in baseline, but the differences were not statistically significant. It seems likely that they would become so with a larger number of subjects. It is of interest that in both experiments, the reduction in eye movement density during recovery sleep occurred in the absence of an increase in either visual or computer measures of delta EEG during the same sleep cycles. This parallels our findings after total deprivation and again indicates, if our hypothesis is correct, that delta EEG by itself is not an adequate measure of sleep depth. According to our further hypothesis that eye movements occur during REM sleep simply because there is no need to inhibit them 0), and that eye movement density indicates the degree of disinhibition of higher motor systems during REM, one would predict that motor neuron discharge rate should vary inversely with the amount of prior sleep deprivation, a readily testable corollary. Whether or not the eye movement density-sleep depth hypothesis is correct, the present findings argue against any "need" for eye movement: in spite of the great loss of eye movement activity during the deprivation conditions, eye movement density decreased rather than increased during recovery sleep. I Delta loss and make up It is interesting to compare the degree to which 0--3-Hz EEG lost in the TSD and loo-min sleep conditions is "made up" on recovery nights. Such comparisons contain major uncertainties, as we shall show. Nevertheless, we present an example of the type of analysis that we hope will eventually shed light on the regulation of delta sleep. We will use for this analysis total integrated amplitude (in m V x s), as it reflects both average wave amplitude and time spent in delta. First, we calculate the delta made up as a proportion of that lost, using the totals across NREM-Pl-4 for each night. Baseline night Deprivation night Lost on deprivation night Recovery night Made up (over BN) Percent of loss made up TSD % 100 min % Clearly, despite greater loss of delta under TSD, a higher proportion of that loss was made up in recovery than in recovery from loo-min sleep. We believe that this result stems from the fact that sleep at the beginning of the night is of higher priority than that at the end [see Feinberg (4) and below1. Taking this into account, we next assume, for the purpose of this example, that all 0--3-Hz EEG lost from NREM-Pl is made up during recovery sleep and that the remainder of the delta made up comes from NREM P2. (We have evidence, soon to be presented, that loss ofnrem sleep after the second Sleep, Vol. 11, No.5, 1988

11 RESPONSE TO 100 MIN SLEEP 483 cycle produces no increase in delta on recovery, perhaps because it does not consistently eliminate cyclic peaks.) Baseline (NREM-Pl) Deprivation (NREM-Pl) Lost from NREM-Pl Made up, ofnrem-plloss Baseline (NREM-P2) Deprivation (NREM-P2) Lost from NREM-P2 Made up, of NREM-P2 loss Percent NREM-P2 loss made up TSD % 100 Min (100%, by assumption) (all non-<:ycle-l make up, i.e., ) 43.2% Thus, a deprivation recovery model requiring the make up of all NREM-Pl 0-3-Hz loss and about one-third of the NREM-P2 loss would be consistent with these data. However, these preliminary calculations are uncertain in several respects. First, restriction of the analysis to NREM-PI-4 may distort the model, as many subjects had NREM sleep after NREM-P4 on BN in both experiments. The delta loss during deprivation may therefore be underestimated for these subjects. However, we would expect a similar effect on the data for both experiments, as BN time in bed and total sleep time were quite similar. More serious is the fact that during recovery from TSD, NREM-Pl duration was twice the baseline level, whereas it increased only to 121% of BN after 100-min sleep. Thus, the sleep times being compared in NREM-Pl on the recovery nights of the two studies are unequal. Finally, we note that this analysis does not take into account recovery effects that would result from increased total sleep duration (i.e., an increased number of cycles), as is likely to occur in naturalistic situations. The measurement of sleep length as a function of prior wakefulness requires a quite different experimental design; for one approach see Akerstedt and Gillberg (26). Such measurements will eventually be required before a complete model of sleep can be constructed. Selective delta increase in NREM-Pl Before attempting to interpret the selective increase of delta in NREM-Pl, it may be useful to review briefly the phenomenology of delta EEG during NREM sleep. When the integrated amplitude or spectral density of delta across the night is plotted, a series of peaks (which would be scored visually as S3 and S4) is observed. These peaks typically (but not reliably) become smaller across the night, in a pattern that resembles a "damped sinusoid" (27). The peaks usually coincide with visually scored NREM periods, but the agreement is not perfect, as more than one peak may occur in an NREM period. It seems reasonable to consider the peaks themselves as physiological units (28), but there are as yet unresolved problems in distinguishing genuine nadirs from transient partial arousals (such as those caused by environmental noise or bodily discomfort) that are not intrinsic to sleep. We previously measured and analyzed statistically the amounts and rates of change of 0-3-Hz EEG across visually defined NREM periods in young and elderly normal subjects (9-11). Recently, we began analysis of the 0-3-Hz peaks in the same groups, using, as one approach, the exploratory smoothing techniques devised by Tukey (29). The cyclical nature of sleep raises fundamental biological questions (30). If we need two qualitatively different kinds of sleep (NREM and REM), why must they occur in Sleep, Vol. 11, No.5, 1988

12 484 I. FEINBERG ET AL. alternation rather than as single blocks? If we require a certain quantity of delta EEG to reverse the effects of waking, why does it occur repetitiveiy (usuauy, but not aiways, punctuated by REM) rather than in a single concentrated burst? One possibility is that the delta in successive NREM periods is a correlate of sequential, qualitatively different metabolic processes (31). However, if this were the case, one would expect that loss of delta in the later cycles would lead to delta augmentation in these same cycles in recovery sleep. The present results argue against that interpretation, but do not rule it out. Certainly, the eye movement density effects indicate important changes in sleep in other cycles. An alternative hypothesis is one we proposed in 1974 (4). According to this view, it is NREM sleep that reverses certain effects of plastic brain activities during waking. The reversal process is assumed to be metabolic. It consumes a substrate that accumulates in proportion to (A) the duration of waking and (B) the intensity of waking brain activity (roughly indexed by brain metabolic rate). Intensity of this reversal or recovery metabolic process is correlated with production of dense, high-amplitude delta, i.e., stages 2, 3, and 4 represent increasing rates or intensities. (Note that the delta waves are assumed to be a correlate of, or a permissive state for, the process, but not the process itself.) Rate or intensity is highest early in sleep (largest delta peak in NREM PI) because the concentration of substrate is then highest. As with other metabolic processes, rate of activity declines exponentially as substrate is consumed. According to our model, periodic REM sleep acts to produce a cofactor or neuromodulator that is required by NREM processes to proceed to completion. Noting certain similarities between REM and waking, we speculated that REM might produce some of the same substrates as waking without consuming what waking might consume, for example, in memory consolidation. Interpreted in this light, both the 100-min sleep condition and TSD increase the total substrate for sleep and thereby prolong, to different extents, the intense, initial phase of the hypothetical metabolic process. However, the mechanisms of delta augmentation differ in the two conditions: after total deprivation, both delta wave density and amplitude increase, whereas after a night with 100 min of sleep, only density increases. We can offer no explanation of the different effects of strong and weaker deprivation on wave amplitudes and densities, but it may be useful to consider the mechanisms that might come into play. EEG waves are believed to be generated by synchronous inhibitory and excitatory postsynaptic potentials in cortical neurons and dendrites that are driven by subcortical pacemakers (32). An increase in amplitude of delta waves could therefore result from an increase in the number of synchronously driven neurons (with the same average potential change/neuron) or an increase in the average potential change. Average potential change could vary because of the "state" of the neurons (e.g., degree of "depletion" after sleep deprivation) or from altered intensity of the pacemaker stimulus. Changes in wave density might be expected to reflect pacemaker output more directly; however, feedback mechanisms are so pervasive in the central nervous system that pacemaker output may be largely determined by its neuronal targets. We make this brief neurophysiological excursion simply to point out that by measuring wave amplitudes and durations separately under a range of experimental conditions, it may become possible to elicit clues to the underlying biology. Period-amplitude analysis, which distinguishes effects on amplitude from those on density of waves more Sleep, Vol. 11, No.5, 1988

13 RESPONSE TO 100 MIN SLEEP 485 readily than spectral analysis (33), should prove to be particularly useful for this purpose. Whatever the underlying mechanisms, the findings in these two deprivation studies suggest that NREM-Pl plays a uniquely plastic role in human sleep. Ontogenetic data further support this possibility; NREM-Pl shows the greatest changes with age of any sleep cycle component, both in total duration and in its high-amplitude delta EEG (4). Studies in psychiatric patients also point to the importance ofnrem-pl. Thus, Hiatt and coworkers (34) reported that abnormalities in visually scored stage 4 and computer measures of 0--3-Hz waveforms in schizophrenia are concentrated in NREM-Pl. Unfortunately, in much of the clinical sleep literature, NREM-Pl has been conceptualized as a measure of REM sleep, or REM "pressure." We have long argued against this illogical view. Twenty-one years ago, we pointed out (24) that NREM-Pl duration was correlated with the amount of stage 4 EEG in normal subjects. Two years later (35), we suggested that reduced stage 4 EEG might underlie "early onset of the first REM period and an increased duration of this period" in schizophrenic patients, hypothesizing "that these latter findings, rather than representing primary alterations in REM mechanisms, are secondary to decreased pressure of NREM sleep" (p 264). We reiterated this proposal on several occasions. For example, Feinberg and Hiatt (36) again noted this possibility and went on to suggest that" it is probable that REM and NREM sleep are in dynamic interaction, and that for this reason the two kinds of sleep ought to be considered together rather than separately" (p 206). This point of view has gained recent acceptance (8,37), which should lead to a more balanced view of sleep. However, it is necessary to point out that the hypothesis that deficient NREM processes underlie the early REM onset (short NREM-PI) in psychiatric patients long antedates the application of the "two-process model" to sleep in depression by Borbely and Wirz-Justice. Moreover, the sleep abnormalities in depression are not, as these authors assert, pathognomonic, but overlap extensively or completely with those in schizophrenia (34,35,38,39). Lastly, we emphasize that the analyses carried out in our studies use physiological units of sleep, the successive NREM and REM periods. The value of this approach seemed intuitively obvious when we established it for our laboratory more than two decades ago (24). However, such analyses are not yet standard, and investigators who rely entirely on arbitrary units such as hours or fractions of the sleep period risk losing vital information. Acknowledgment: This research was supported by V A Research Funds, by NIMH Grant 7ROIMH41592, and by BRSG funds from SUNY-Stony Brook. We thank W.-J. Chern for invaluable programming assistance. REFERENCES 1. Feinberg I, Floyd Te, March JD. Effects of sleep loss on delta (0.3-3 Hz) EEG and eye movement density: new observations and hypotheses. Electroencephalogr Clin NeurophysioI1987;67: Berger RJ, Oswald I. Effects of sleep deprivation on behavior, subsequent sleep, and dreaming. J Ment Sci 1962;108: Williams HL, Hammack JT, Daly RL, Dement we, Lubin A. Responses to auditory stimulation, sleep loss and the EEG stages of sleep. Electroencephalogr Clin Neurophysiol 1964;16: Sleep, Vol. Jl, No.5, 1988

14 486 I. FEINBERG ET AL. 4. Feinberg I. Changes in sleep cycle patterns with age. J Psychiatr Res 1974;10: Aserinsky E. The maximal capacity for sleep: rapid eye movement de.nsity as an index of sleep satiety. Bioi Psychiatry 1969;1: Feinberg I, Fein G, Floyd TC. EEG patterns during and following extended sleep in young adults. Electroencephalogr Clin Neurophysiol 1980;50: Feinberg I, Koegler A. Hypnotics and the elderly: clinical and basic science issues. In: Eisdorfer C, Fann WE, eds. Treatment of psychopathology in the aging. New York: Springer, 1982: Borbely AA, Wirz-Justice A. Sleep, sleep deprivation and depression. Hum NeurobioI1982;1: Feinberg I, March JD, Fein G, Floyd TC, Walker JM, Price L. Period and amplitude analysis of c/sec activity in non-rem sleep of young adults. Electroencephalogr Clin NeurophysioI1978;44: Feinberg I, Fein G, Floyd TC. Period and amplitude analysis of non-rem EEG in sleep: repeatability of results in young adults. Electroencephaiogr Clin Neurophysioi 1980;48: Feinberg I, Fein G, Floyd TC, AminoffMJ. Delta (.5-3 Hz) EEG waveforms during sleep in young and elderly normal subjects. In: Chase MH, Weitzman ED, eds. Sleep disorders: basic and clinical research. New York: Spectrum, 1983: Feinberg I, March JD, Floyd Te, Jimison R, Bossom-Demitrack L, Katz PH. Homeostatic changes during postnap sleep maintain baseline levels of delta EEG. Electroencephalogr Clin Neurophysiol 1985;61: Feinberg I, Fein G, Floyd TC. Computer-detected patterns of electroencephalographic delta activity before and after extended sleep. Science 1982;215: Feinberg I, March JD, Floyd TC, Fein G, Aminoff MJ. Log amplitude is a linear function oflog frequency in NREM sleep of young and elderly subjects. Electroencephalogr Clin NeurophysioI1985;58: Fein G, Floyd TC, Feinberg I. Computer measures of sleep EEG reliably sort visual stage 2 epochs by NREM period of origin. Psychophysiology 1981;18: Borbely A. A two-process model of sleep regulation. Hum NeurobioI1982;1: Feinberg I, Floyd TC. A comparison of two models of sleep. In: Koella WP, Ruther E, Schulz H, eds. Sleep '84. New York: Gustav Fischer, 1985: Daan S, Beersma DGM, Borbely A. Timing of human sleep: recovery processes gated by a circadian pacemaker. Am J Physiol 1984;246 (Reg Integ. Compo Physiol 15):RI Feinberg I, Floyd TC. The regulation of human sleep. Hum NeurobioI1982;1: Webb WB, Agnew HW. Stage 4 sleep: influence of time course variables. Science 1971 ;174: Feinberg I, Floyd Te. Systematic trends across the night in human sleep cycles. Psychophysiology 1979;16: Church MW, March JD, Hibi S, Benson K, Cavness C, Feinberg I. Changes in frequency and amplitude of delta activity during sleep. Electroencephalogr Clin NeurophysioI1975;39: Dement W, Kleitman N. Cyclic variations in EEG during sleep and their relation to eye movements, body motility, and dreaming. Electroencephalogr Clin Neurophysiol 1957;9: Feinberg I, Koresko RL, Heller N. EEG sleep patterns as a function of normal and pathological aging in man. J Psychiatr Res 1967;5: Rechtschaffen A, Hauri P, Zeitlin M. Auditory awakening thresholds in REM and non-rem sleep stages. Percept Motor Skills 1966;22: Akerstedt T, Gillberg M. Sleep duration and the power spectral density of the EEG. Electroencephalogr Clin Neurophysiol 1986;64: Lubin A, Nute C, Naitoh P, Martin WB. EEG delta activity during human sleep as a damped ultradian rhythm. Psychophysiology 1973;10: Haustein W, Pilcher J, Klink J, Schulz H. Automatic analysis overcomes limitations of sleep stage scoring. Electroencephalogr Clin Neurophysiol 1986;64: Velleman PF, Hoaglin DC. Applications, basics, and computing of exploratory data analysis. Boston: Doxbury Press, 1981: Feinberg I, Evarts EV. Changing concepts of the function of sleep: discovery of intense brain activity during sleep calls for revision of hypotheses as to its function. Bioi Psychiatr 1969;1: Guiditta A, Ambrosini MV, Sadile A, Gironi Carnevale UA. Role of synchronized sleep (SS) in brain information processing. Paper presented at International Symposium: Current Trends in Slow Wave Sleep, June 25-27, 1987, Janssen Pharmaceutica, Beerse, Belgium. 32. Elul R. The genesis of the EEG. In: Pfeiffer CC, Smythies JR, eds. International review of neurobiology, vol 15. Orlando, FL: Academic Press, 1972: Ktonas PY, Gosalia AP. Spectral analysis vs. period-amplitude analysis of narrow band EEG activity: A comparison based on the sleep delta-frequency band. Sleep 1981 ;4: Hiatt JF, Floyd TC, Katz PH, Feinberg I. Further evidence of abnormal non-rapid-eye-movement sleep in schizophrenia. Arch Gen Psychiatr 1985;42: Feinberg I, Braun M, Koresko RL, Gottlieb F. Stage 4 sleep in schizophrenia. Arch Gen Psychiatry 1969;21 : Sleep, Vol. 11, No.5, 1988

15 RESPONSE TO 100 MIN SLEEP Feinberg I, Hiatt JF. Sleep patterns in schizophrenia: a selective review. In: Williams RL, Karacan I, eds. Sleep disorders diagnosis and treatment. New York: John Wiley and Sons, 1978: Reynolds ef, Kupfer DJ. Sleep research in affective illness: state of the art circa Sleep 1987;10: Zarcone VP, Benson KL, Berger PA. Abnormal rapid eye movement latencies in schizophrenia. Arch Gen Psychiatry 1987;44: Guazzelli M, Maggini e, Landini G, Floyd Te, Feinberg I. Similarity of non-rem abnormalities in schizophrenia and depression (letter). Arch Gen Psychiatry 1985;42: Sleep, Vol. 11, No, 5, 1988