178
Electroencephalography and Clinical Neurophysiology, 1978, 4 5 : 1 7 8 - - 1 8 5 © Elsevier/North-Holland Scientific Publishers, Ltd.
RAPID EYE MOVEMENT SLEEP CYCLE, CLOCK TIME AND SLEEP ONSET RICHARD J. McPARTLAND and DAVID J. KUPFER
Department of Psychiatry, Western Psychiatric Institute and Clinic, University of Pittsburgh School of Medicine, Pittsburgh, Pa. (U.S.A.) (Accepted for publication: December 12, 19 7 7 )
Previous research has described the existence of a well-defined 90--110 min sleep rhythm. The predominant feature of this rhythm is the cycling of rapid eye movement (REM) and non-rapid eye movement (NREM) sleep. REM sleep periods (REMPs) are times during which dreaming usually occurs, while NREM sleep is characterized by a relative absence of dreaming. In the past, some investigators have subscribed to the notion that the phase of the REM/NREM sleep cycle is entirely determined by the time of sleep onset; simply stated, periods of REM sleep occur approximately every 90 min after falling asleep. This point of view hypothesizes the REM cycle to be a sleep-dependent rhythm. Globus, in relating the time of the first REMP to the time of onset of daytime nap sleep, concluded that the occurrence of REM sleep varies at least in part as a function of clock time, and therefore timing of REMPs are not totally sleep dependent (Globus 1966). More recently, Schulz has proposed that within a subject, for all but the night's first REMP, there is a night to night interval between corresponding REMPs slightly longer than 24 h, indicating that only the timing of the first REMP is sleep dependent and the timing of all others is entirely sleep independent (Schulz et al. 1975). Because of the internight relationship of the REM sleep cycle, the REM sleep clock is assumed to operate during the waking hours of the day. This is in accordance with Kleitman's hypothesized biological clock (the basic rest activity cycle: BRAC) that continues to operate with the
same periodicity throughout the day and night (Kleitman 1969). In order to account for night to night shifts in the timing, with respect to clock time, of all but the first REMPs, Schulz's study concludes that the BRAC is not a submultiple of 24 h. On the other hand, if the REM cycle is entirely sleep dependent, as others have concluded by an examination of nap sleep data (Moses et al. 1977), then the REM clock runs only when the organism is sleeping and therefore is independent of the BRAC. The present study investigates the hypothesis that the occurrence of REM sleep varies as a function of both clock time and sleep onset time. However, this report differs from previous studies by investigating the possibility that REM sleep tends to occur at the same time of night for all subjects, regardless of time of sleep onset. In addition, the data is from normal nighttime sleep, not from forced daytime napping.
Method Ten normal subjects (7 females and 3 males) with a mean age of 22.6 years (range 20-25 years) were sleep studied for 4 consecutive nights. After retiring, when the subject desired, all-night electroencephalogram, electrooculogram, and electromyogram recordings were obtained by conventional polygraphic methods and scored for REM and other stages of sleep using the standard criteria (Rechtschaffen and Kales 1968). All but two
REM SLEEP CYCLE technically poor sleep records were used. For each of the 38 nights, the times marking the beginning and end of each REMP were logged. The 30 rain rule was used to differentiate REMPs, i.e. REMs belong to different REMPs if they are separated by 30 or more minutes. Across all 38 nights, the REMPs in progress were summed for each minute of clock time from 10 p.m. (EST) to 7 a.m. For each of these minutes, the result was divided by the number of nights comprising the sum. If a subject had n o t y e t fallen asleep for the minute in question, that night was not counted in the respective sum. Similarly, if a subject had awoken and remained awake, that night was n o t included in sums for all subsequent minutes. This resulted in a measure of REM probability as a function of clock time. A similar analysis was done considering just the first REMP of each night, likewise for the second, third and fourth REMPs. After smoothing REM probability data with a 15 min moving window (i.e., the resulting data point for minute n is the average of the u n s m o o t h e d data points for all minutes from n -- 7 to n + 7; this effectively filters out variations of less than 15 min periodicity) and detrending the data (by subtracting a 90 min moving window detrend line) the periodicity of the composite REM probability curve was calculated by fractional harmonic analysis (Vaux 1965), an iterative technique for finding the period corresponding to the o p t i m u m fit of the first term o f the Fourier series expansion ( F ( t ) = A cos(2~rt/p + ~b) + B sin(27rt/p + ¢); t = time in min past 10:00 p.m., p = period of cycle in min, and ¢ (phase angle)= 0 corresponding to 10:00 p.m.) to the data; p varied from 10 to 300 min in 1 min increments. Clock time probability functions were also calculated for each subject independently. Each of these 10 functions, after being s m o o t h e d and detrended as above, was fitted with the first term of the Fourier series corresponding to the o p t i m u m periodicity of the composite REM probability function. In order to determine if peak REM probability time were the
179 same for all subjects, the resulting phase and amplitude information contained in the cosine and sine coefficients for each of the 10 subjects were then compared by a modified Bartel's technique (Armstrong 1973). All nights were then categorized according to time of sleep onset. Those nights where sleep onset occurred prior to 12:46 a.m. were grouped separately from those nights where sleep onset occurred after 12:46 a.m. The time 12:46 a.m. was chosen as the dividing point on the basis of a histogram of sleep onset times showing a bimodal distribution with 12:46 a.m. as the separating time. The clock time REM probability function was calculated as previously described for each group. A second type of REM probability function, indicating the likelihood of REM sleep as a function of elapsed time after sleep onset, was also calculated. The method for finding this function is similar to that for finding clock time REM probability except that each night is aligned according to the time of sleep onset. Across night sums of REMPs in progress are then made for corresponding elapsed times, in minutes, past sleep onset.
Results The composite REM probability, as a function of clock time, is shown in Fig. 1. Five peaks are noticed corresponding to approximately 1:30 a.m., 3:15 a.m., 4:30 a.m., 5:45 a.m. and 7:00 a.m. Fractional harmonic analysis revealed an optimum fit for the first term of the Fourier series expansion at p = 90 min. The phase term for this fit indicates peaks at 1:23 a.m., 2:53 a.m., 4:23 a.m., 5:53 a.m. and 7:23 a.m. Fig. 1 also shows the clock time REM probability for each individual REMP. Table I lists the relative amplitude (the actual amplitude in units of REM probability ranging from 0 to 1, equals 2/90 times the relative amplitude), and the cosine and sine
180
R.J. M c P A R T L A N D , D.J. K U P F E R TABLE I
0.5
T h e a m p l i t u d e a n d cosine, sine c o e f f i c i e n t s (A and B r e s p e c t i v e l y ) for fitting t h e first t e r m o f a 90 m i n F o u r i e r series e x p a n s i o n to t h e individual R E M p r o b ability f u n c t i o n s , a n d t h e statistical c a l c u l a t i o n s for t h e significance o f t h e phase c l u s t e r i n g across subjects.
F--
C:) CL :E: W r-K
Subject
Amplitude
A
B
1 2 3 4 5 6 7 8 9 10
59.40 35.22 35.16 23.04 50.96 67.92 7.12 41.67 14.11 23.57
53.86 27.29 --25.15 --12.02 --15.87 58.17 6.46 28.82 10.21 10.23
25.05 22.26 24.58 19.66 48.43 35.06 2.98 30.08 9.74 --21.23
142.01
196.58
14.20
19.66
Total Average
M e a s u r e d average a m p l i t u d e = (~2 + ~ 2 ) 1 / 2 = 2 4 . 2 5 Q u a d r a t i c average a m p l i t u d e = (A 2 + B 2 ~ 2 = 4 0 . 3 6 E x p e c t e d average a m p l i t u d e 4 0 . 3 6 / ~ / 1 0 = 12.73 M e a s u r e d average a m p l i t u d e X = = 1.9, E x p e c t e d average a m p l i t u d e
1 0 PM
1AM
4
7
1 P < e(~'
< 0.027
CLOCK TIME Fig. 1. R E M p r o b a b i l i t y as a f u n c t i o n o f c l o c k t i m e f o r t h e e n t i r e n i g h t ( R E M P s 1, 2, 3 a n d 4) a n d e a c h R E M P s e p a r a t e l y . T i m e s o f p e a k p r o b a b i l i t y are indic a t e d b y vertical lines.
coefficients (A and B respectively) resulting from fitting the first Fourier term o f a 90 min period to the REM probability functions for each subject. The average values o f A and B were computed in order to calculate the measured average amplitude. An estimate of the expected average amplitude, which might result from chance factors alone, was computed in accordance with the rule used to determine the dispersion of the mean, i.e., by dividing the quadratic mean of the individual amplitudes by x/10. A comparison of the
measured average amplitude with the expected average amplitude yields, through the use of standard probability formulae, a statistical measure of significance indicating the likelihood for all subjects to have the same peak probability times for REM sleep. The result of Barrel's test is a ratio of 1.90 corresponding to a significance of P < 0.027. Fig. 2 is a histogram of sleep onset times for all nights studied. A bimodal distribution is observed; 12:46 a.m. is an effective dividing time between the two unimodal distributions, which will be referred to as early sleep onset and late sleep onset nights. Fig. 3 compares the clock time REM probability functions for the total sample (38 nights), the early sleep onset nights (21 nights) and the late sleep
REM SLEEP CYCLE
181
9"
c~
:E
I
ALL NIGHTS
. . . . .
SLEEP ONSET ~ 1 2 : 4 6 A M
.....
SLEEP ONSET~I2:46AM
8-
76 m
Z
ta-
5.
O
4.
.
.
.
.
II !I
:E
I'
Z
l
J
0
I
i0:50 PM
•
ii:i0
Ii:30
,I.....
w
ii:50
|
12:10
TIME
OF
12:30
U ! I •
12:50
SLEEP
w
w
i:i0
1:30
w
1:50
2:10 AM
ONSET
Fig. 2. Histograms s h o w i n g sleep o n s e t t i m e s for all s t u d y nights, nights w i t h sleep onset prior to 1 2 : 4 6 a.m. and nights w i t h sleep o n s e t after 1 2 : 4 6 a.m.
onset nights (17 nights). Fig. 4 shows REM probability as a function of time elapsed since sleep onset.
/
.~ 0
'
t/
All nights
Nights with sleep onset before 12:46 AM
' :VII ; w
IOPM
I
0
Nights with sleep onset a f t e r 12:46 RM
1
i
IAM
4
CLOCK TIME
Fig. 3. REM probabilities as a f u n c t i o n o f c l o c k time for all nights (N = 3 8 ) , nights w i t h sleep o n s e t prior to 1 2 : 4 6 a.m. (N = 2 1 ) , and nights w i t h sleep o n s e t after 1 2 : 4 6 a.m. (N = 17).
|
0
w
•
w
90
180
270
|
360
•
450
ELASPED TIME SINCE SLEEP ONSET
Fig. 4. REM probability as a f u n c t i o n o f elapsed time since sleep onset.
182
Discussion
The composite clock time REM probability function, shown in Fig. 1, shows that this sample of similar aged normal subjects has a high probability for REM at specific times throughout the night regardless of time of sleep onset. The first peak probability time occurs at approximately 1:30 a.m. and corresponds to a 15% probability for REM; shortly thereafter at 2:15 a.m. the probability has decreased to 7%. There is a rather rapid rise to the next peak of 34% occurring at 3:15 a.m.; this sharp peak is followed by a trough of 10% at 3:50 a.m. The next peak (4:30 a.m.) and trough are n o t well-defined; b u t the fourth peak occurring around 5:45 a.m. and corresponding to 40% REM probability is followed by a sharp fall to a low o f 20% which is followed by a peak probability of 56%. Fig. 1 also shows that after the second REM period there is a shortening of the REM cycle from approximately 105 to less than 90 min, indicating a non-stationary rhythm. The deviation of the experimental peak probability times estimated from Fig. 1 from the predicted (Fourier series) peak probability times are +7, +22, +7, --8, a n d - - 2 3 min for the first, second, third, fourth and fifth peaks, respectively. Experimental and predicted results are close with the exception of the observed 7:00 a.m. peak which falls 23 rain early. The non-stationary nature of the REM rhythm correctly corresponds to the direction of deviation of the earlier peaks. Data were n o t analyzed after 7:00 a.m. (due to the small number of nights that would be included) which could account for the premature observation of the predicted 7:28 a.m. peak. In addition, Fig. 1 shows that the peak clock time REM probability periods are non-specific with respect to the sequentially numbered REMPs. This is seen by observing that the second peak of the curve for the first REMP coincides in clock time with the first peak of the REMP 2 curve, the second peak for curves for REMP 2 and 3 falls closely in time with the first peak of curves for REMP 3 and 4
R.J. McPARTLAND, D.J. KUPFER
respectively, and the third peak of the REMP 3 curve coincides with the second peak of the REMP 4 curve. These results are interpreted as indicating that if a person missed having his first REMP around 1:30 a.m., he would tend to hold off REM until around 3:15 a.m., and so forth. Because it was felt that the bimodal distribution (with distributions displaced by 80 rain) of the time of sleep onset histogram could account for some of the above findings, the nights were grouped into those with sleep onset before and after 12:46 a.m. It is reasoned that if the timing of REM sleep was solely a function of elapsed time from sleep onset, and had no relation to clock time as hypothesized above, the clock time REM probability curve for the post 12:46 a.m. sleep onset nights would look similar to the pre 12:46 a.m. curve except that it would be shifted along the clock time axis by 80 min. However, this is n o t the case; both curves are very similar w i t h o u t any time shift. In addition, both early and late sleep onset time distributions are relatively broad (approximately 80 min} minimizing the probability that the observed clock time REM probability rhythm is an artifact due to clustering of sleep onset times. One does note the absence of a 1:30 a.m. peak in the probability curve for the late sleep onset nights, which is due to the late sleep onset times for this group of nights (mean of 1:30 a.m.). Moreover, there is a pronounced trough in the early sleep onset probability curve separating the first and second REM probability peaks. This trough is n o t present in the late sleep onset curve which instead manifests a steady rise in REM probability after 1:45 a.m., terminating with the 3:15 peak. The d o t t e d line on the late sleep onset curve illustrates the probability expected from the clock time related REM driving force and is similar to that portion of the early sleep onset probability curve. The fact that REM probability is higher than expected between 1:45 a.m. and 2:45 a.m. indicates an additional driving force for REM which may correspond to the tendency for
REM S L E E P C Y C L E
REM sleep to occur approximately 90 min after sleep onset. In this case, the excess REM probability results from the earlier sleep onset nights within the late sleep onset set of nights. This, as does Fig. 4, suggests that times of high probability for REM sleep are certainly n o t related solely to clock time; REM probability is also a very strong function of elapsed time after sleep onset. In light of data presented herein and data presented in other studies cited above, it would seem that, like many other biological phenomena, the phase of the R E M / N R E M cycle is determined by multiple factors. Elapsed time after sleep onset and the time of night are both shown to be strong influences. Other investigators (Webb et al. 1971; Weitzman 1974; Webb and Agnew 1977) studying the sleep of altered sleep-wake regimens have noted a change in REM probability when sleep times are displaced from normal, i.e., an increased probability for REM to occur earlier than they expected (less than 90 min after sleep onset). Perhaps this is a manifestation of the time of day influence. An additional factor, which the above studies of altered sleep-wake cycles address, is the existence of a 24 h REM probability cycle peaking at roughly mid-morning (6:00 a.m.) and having a minim u m at a b o u t 6:00 p.m. This modulation of REM sleep nicely corresponds to the general tendency for REM sleep to increase in probability as the night goes on. Unlike the study of Schulz (Schulz et al. 1975), we did n o t have many consecutive nights o f data on a single subject necessary for investigation of reported night to night shift in the time of corresponding REM periods. However, we have done preliminary analysis of this t y p e on sleep of depressed patients and have noted no shift of this type. In contrast to the data of Schulz, our data support a BRAC that is an integral subharmonic of 24 h. As Table I indicates, all subjects tend to have REM at the same times as each other. This raises the possibility that the driving force which controls the timing of the REM rhythm as related to clock time may be out-
183
side of the person. This being the; case, it is difficult to speculate as to its origin. If it were the light/dark cycle, as is improbable due to its changing over the many months over which this study t o o k place, clock time influence would disappear under conditions of constant or irregular lighting. Moreover, if one accepts the premise that the REM cycle and the BRAC are the same or intimately linked, then the BRAC tends to have a c o m m o n phase across subjects and is also partially controlled by a c o m m o n driving force. Considering most of the published results on the REM rhythm, 3 things seem apparent. First, there is a sleep-independent c o m p o n e n t of the REM rhythm (BRAC); second, there is also a sleep~iependent c o m p o n e n t of the REM rhythm; and finally, whatever phase the 9 0 m i n REM rhythm takes, it seems to be modulated by a 24 h periodicity° This suggests a 2-clock model for REM. One clock is sleep independent running continuously; its period is a submultiple of 24 h. The other runs only during sleep. Both have inputs to a mediating mechanism triggering REM sleep. Neglecting the non-stationary nature of the REM rhythm, and realizing that the period of both clocks is a b o u t 90 min, a beat frequency of 1 c/day will be produced when the average period of the 2 clocks differs by 5 min. This beat frequency corresponds to the circadian periodicity, mentioned above, modulating the ultradian rhythm. This model presents an obvious problem; if the phase of the circadian REM periodicity is fixed from day to day and across subjects, as stated (Webb and Agnew 1977), there must be a fixed phase relationship between the 2 clocks. Since the phase of one clock is stationary and the phase of the other is determined by sleep onset, this seems unlikely without fixing the time of sleep onset (possibly including a third, independent clock for the circadian rhythm would be better). If, however unlikely, there is a fixed phase relationship between the clocks, a change in this relationship could account for the shortened REM latency (time
184
between sleep onset and REM sleep) observed in depressive disease (Kupfer and Foster 1972). On the other hand, if this is n o t the case, shortened REM latencies would seem to be due to a malfunction related to the sleep-dependent clock.
Summary The phase of the REM sleep r h y t h m was studied in 10 normal subjects each of w h o m was sleep studied for 4 consecutive nights. For analysis, each night of sleep was aligned according to clock time and each minute was scored as REM or non-REM. With these data, REM probability was found as a function of clock time. Fractional harmonic analysis indicates a 90 min periodicity. The REM probability curve shows peaks occurring at 1:30 a.m., 3:15 a.m., 4:30 a.m., 5:45 a.m. and 7:00 a.m. Statistical measures comparing the time of REM sleep across subjects suggests that subjects tend to have REM sleep at the same time of the night as each other. The influence of elapsed time after sleep onset on REM sleep is also reestablished. Results indicate that the time of REM sleep is determined by both clock time and time of sleep onset, suggesting t w o clocks, one sleep d e p e n d e n t and the other related to the basic rest activity cycle (BRAC), which are responsible for driving REM sleep. Furthermore, the similarity of REM times across subjects indicates the possible existence of an extra-personae REM driving force linked to clock time and possibly the BRAC.
R~sum~ Cycles de sommeil avec m o u v e m e n t s oculaires rapides, horloge et ddbut du sommeil La phase du r y t h m e de sommeil avec mouvements oculaires rapides (REM) a ~t~ ~tudi~e chez 10 sujets n o r m a u x d o n t chacun a ~t~ enregistr~ pendant 4 nuits cons~cutives.
R.J. McPARTLAND, D.J. KUPFER
Pour l'analyse, chaque nuit de sommeil a ~t~ align~e suivant le temps r~el et chaque minute a ~t~ scor~e c o m m e sommeil avec et sans mouvements oculaires rapides. A partir de ces donn~es, la probabilit~ du sommeil REM s'est r~v~l~e ~tre fonction du temps r~el. Une analyse harmonique fractionn~e indique une p~riodicit~ de 90 min. La courbe de probabilit~ du sommeil REM montre des pics survenant ~: l h 30, 3 h 15, 4 h 30, 5 h 45, et 7 h . Des mesures statistiques comparant le m o m e n t du sommeil REM d'un sujet ~ l'autre sugg~rent que les sujets tendent ~ avoir du sommeil REM aux m~mes m o m e n t s de la nuit. L'influence du temps ~coul~ apr~s l'endormissement sur le sommeil avec mouvements oculaires rapides est ~galement r ~ t a b l i e . Les r~sultats indiquent que les m o m e n t s de sommeil REM sont d~termin~s ~ la fois par le temps r~el et le temps de l'endormissement, sugg~rant que deux horloges, l'une d~pendant du sommeil et l'autre li~e au BRAC, sont responsables de la survenue du sommeil REM. En outre, la similarit~ des temps de REM entre sujets indique l'existence possible d'un facteur extrapersonnel li~ au temps r~el et possiblement au BRAC.
References Armstrong, C.E. Cycle analysis -- a case study, part 25: testing cycles for statistical significance. Cycles, 1973, 24: 231--236. Dewey, E.R. Cycle analysis -- a case study, part 9: the moving lineage. Cycles, 1971, 22: 297--318. Globus, G. Rapid-eye-movement cycle in real time. Arch. gen. Psychiat., 1966, 15: 654--659. Kleitman, N. Basic rest activity cycle in relation to sleep and wakefulness. In: A. Kales (Ed.), Sleep Physiology and Pathology. Lippincott, Philadelphia, Pa., 1969: 33--38. Kupfer, D.J. and Foster, F.G. Interval between onset of sleep and rapid-eye-movement sleep as an indicator of depression. Lancet, 1972, ii: 684-686. Moses, J.M., Lubin, A., Johnson, L.C. and Naitoh, P. The rapid-eye-movement cycle is a sleep dependent rhythm. Nature (Lond.), 1977, 265: 360-361. Rechtschaffen, A. and Kales, A. (Eds.) A Manual of
REM SLEEP CYCLE Standard Terminology, Techniques of Scoring Systems for Sleep Stages of Human Subjects. Government Printing Office, Washington, D.C., 1968. Schulz, H., Dirlich, G. and Zulley, J. Phase shift in the REM sleep rhythm. Pfliigers Arch. gas. Physiol., 1975, 358: 203--212. Vaux, J.E. Technical bulletin 1964---65 fractional harmonic analysis program. Cycles, 1965, 16: 301--310. Webb, W.B. and Agnew, H.W., Jr. Analysis of the
185 sleep stages in sleep-wakefulness regimens of varied lengths. Psychophysiology, 1977, 14: 445--449. Webb, W.B., Agnew, H.W., Jr. and Williams, R.L. Effect on sleep-wake cycles of varied length. Psychophysiology, 1971, 42: 152--155. Weitzman, E., Nogeire, C., Perlow, M., Fukushima, D.S., Assin, J., McGregor, P., Gallagher, T. and Hellman, L. Effects of a prolonged 3 hour sleepwake cycle on sleep stages, plasma cortisol, growth hormone and body temperature in man. J. clin. Endocr., 1974, 38: 1018--1030.
Electroencephalography and Clinical Neurophysiology, 1978, 4 5 : 1 7 8 - - 1 8 5 © Elsevier/North-Holland Scientific Publishers, Ltd.
RAPID EYE MOVEMENT SLEEP CYCLE, CLOCK TIME AND SLEEP ONSET RICHARD J. McPARTLAND and DAVID J. KUPFER
Department of Psychiatry, Western Psychiatric Institute and Clinic, University of Pittsburgh School of Medicine, Pittsburgh, Pa. (U.S.A.) (Accepted for publication: December 12, 19 7 7 )
Previous research has described the existence of a well-defined 90--110 min sleep rhythm. The predominant feature of this rhythm is the cycling of rapid eye movement (REM) and non-rapid eye movement (NREM) sleep. REM sleep periods (REMPs) are times during which dreaming usually occurs, while NREM sleep is characterized by a relative absence of dreaming. In the past, some investigators have subscribed to the notion that the phase of the REM/NREM sleep cycle is entirely determined by the time of sleep onset; simply stated, periods of REM sleep occur approximately every 90 min after falling asleep. This point of view hypothesizes the REM cycle to be a sleep-dependent rhythm. Globus, in relating the time of the first REMP to the time of onset of daytime nap sleep, concluded that the occurrence of REM sleep varies at least in part as a function of clock time, and therefore timing of REMPs are not totally sleep dependent (Globus 1966). More recently, Schulz has proposed that within a subject, for all but the night's first REMP, there is a night to night interval between corresponding REMPs slightly longer than 24 h, indicating that only the timing of the first REMP is sleep dependent and the timing of all others is entirely sleep independent (Schulz et al. 1975). Because of the internight relationship of the REM sleep cycle, the REM sleep clock is assumed to operate during the waking hours of the day. This is in accordance with Kleitman's hypothesized biological clock (the basic rest activity cycle: BRAC) that continues to operate with the
same periodicity throughout the day and night (Kleitman 1969). In order to account for night to night shifts in the timing, with respect to clock time, of all but the first REMPs, Schulz's study concludes that the BRAC is not a submultiple of 24 h. On the other hand, if the REM cycle is entirely sleep dependent, as others have concluded by an examination of nap sleep data (Moses et al. 1977), then the REM clock runs only when the organism is sleeping and therefore is independent of the BRAC. The present study investigates the hypothesis that the occurrence of REM sleep varies as a function of both clock time and sleep onset time. However, this report differs from previous studies by investigating the possibility that REM sleep tends to occur at the same time of night for all subjects, regardless of time of sleep onset. In addition, the data is from normal nighttime sleep, not from forced daytime napping.
Method Ten normal subjects (7 females and 3 males) with a mean age of 22.6 years (range 20-25 years) were sleep studied for 4 consecutive nights. After retiring, when the subject desired, all-night electroencephalogram, electrooculogram, and electromyogram recordings were obtained by conventional polygraphic methods and scored for REM and other stages of sleep using the standard criteria (Rechtschaffen and Kales 1968). All but two
REM SLEEP CYCLE technically poor sleep records were used. For each of the 38 nights, the times marking the beginning and end of each REMP were logged. The 30 rain rule was used to differentiate REMPs, i.e. REMs belong to different REMPs if they are separated by 30 or more minutes. Across all 38 nights, the REMPs in progress were summed for each minute of clock time from 10 p.m. (EST) to 7 a.m. For each of these minutes, the result was divided by the number of nights comprising the sum. If a subject had n o t y e t fallen asleep for the minute in question, that night was not counted in the respective sum. Similarly, if a subject had awoken and remained awake, that night was n o t included in sums for all subsequent minutes. This resulted in a measure of REM probability as a function of clock time. A similar analysis was done considering just the first REMP of each night, likewise for the second, third and fourth REMPs. After smoothing REM probability data with a 15 min moving window (i.e., the resulting data point for minute n is the average of the u n s m o o t h e d data points for all minutes from n -- 7 to n + 7; this effectively filters out variations of less than 15 min periodicity) and detrending the data (by subtracting a 90 min moving window detrend line) the periodicity of the composite REM probability curve was calculated by fractional harmonic analysis (Vaux 1965), an iterative technique for finding the period corresponding to the o p t i m u m fit of the first term o f the Fourier series expansion ( F ( t ) = A cos(2~rt/p + ~b) + B sin(27rt/p + ¢); t = time in min past 10:00 p.m., p = period of cycle in min, and ¢ (phase angle)= 0 corresponding to 10:00 p.m.) to the data; p varied from 10 to 300 min in 1 min increments. Clock time probability functions were also calculated for each subject independently. Each of these 10 functions, after being s m o o t h e d and detrended as above, was fitted with the first term of the Fourier series corresponding to the o p t i m u m periodicity of the composite REM probability function. In order to determine if peak REM probability time were the
179 same for all subjects, the resulting phase and amplitude information contained in the cosine and sine coefficients for each of the 10 subjects were then compared by a modified Bartel's technique (Armstrong 1973). All nights were then categorized according to time of sleep onset. Those nights where sleep onset occurred prior to 12:46 a.m. were grouped separately from those nights where sleep onset occurred after 12:46 a.m. The time 12:46 a.m. was chosen as the dividing point on the basis of a histogram of sleep onset times showing a bimodal distribution with 12:46 a.m. as the separating time. The clock time REM probability function was calculated as previously described for each group. A second type of REM probability function, indicating the likelihood of REM sleep as a function of elapsed time after sleep onset, was also calculated. The method for finding this function is similar to that for finding clock time REM probability except that each night is aligned according to the time of sleep onset. Across night sums of REMPs in progress are then made for corresponding elapsed times, in minutes, past sleep onset.
Results The composite REM probability, as a function of clock time, is shown in Fig. 1. Five peaks are noticed corresponding to approximately 1:30 a.m., 3:15 a.m., 4:30 a.m., 5:45 a.m. and 7:00 a.m. Fractional harmonic analysis revealed an optimum fit for the first term of the Fourier series expansion at p = 90 min. The phase term for this fit indicates peaks at 1:23 a.m., 2:53 a.m., 4:23 a.m., 5:53 a.m. and 7:23 a.m. Fig. 1 also shows the clock time REM probability for each individual REMP. Table I lists the relative amplitude (the actual amplitude in units of REM probability ranging from 0 to 1, equals 2/90 times the relative amplitude), and the cosine and sine
180
R.J. M c P A R T L A N D , D.J. K U P F E R TABLE I
0.5
T h e a m p l i t u d e a n d cosine, sine c o e f f i c i e n t s (A and B r e s p e c t i v e l y ) for fitting t h e first t e r m o f a 90 m i n F o u r i e r series e x p a n s i o n to t h e individual R E M p r o b ability f u n c t i o n s , a n d t h e statistical c a l c u l a t i o n s for t h e significance o f t h e phase c l u s t e r i n g across subjects.
F--
C:) CL :E: W r-K
Subject
Amplitude
A
B
1 2 3 4 5 6 7 8 9 10
59.40 35.22 35.16 23.04 50.96 67.92 7.12 41.67 14.11 23.57
53.86 27.29 --25.15 --12.02 --15.87 58.17 6.46 28.82 10.21 10.23
25.05 22.26 24.58 19.66 48.43 35.06 2.98 30.08 9.74 --21.23
142.01
196.58
14.20
19.66
Total Average
M e a s u r e d average a m p l i t u d e = (~2 + ~ 2 ) 1 / 2 = 2 4 . 2 5 Q u a d r a t i c average a m p l i t u d e = (A 2 + B 2 ~ 2 = 4 0 . 3 6 E x p e c t e d average a m p l i t u d e 4 0 . 3 6 / ~ / 1 0 = 12.73 M e a s u r e d average a m p l i t u d e X = = 1.9, E x p e c t e d average a m p l i t u d e
1 0 PM
1AM
4
7
1 P < e(~'
< 0.027
CLOCK TIME Fig. 1. R E M p r o b a b i l i t y as a f u n c t i o n o f c l o c k t i m e f o r t h e e n t i r e n i g h t ( R E M P s 1, 2, 3 a n d 4) a n d e a c h R E M P s e p a r a t e l y . T i m e s o f p e a k p r o b a b i l i t y are indic a t e d b y vertical lines.
coefficients (A and B respectively) resulting from fitting the first Fourier term o f a 90 min period to the REM probability functions for each subject. The average values o f A and B were computed in order to calculate the measured average amplitude. An estimate of the expected average amplitude, which might result from chance factors alone, was computed in accordance with the rule used to determine the dispersion of the mean, i.e., by dividing the quadratic mean of the individual amplitudes by x/10. A comparison of the
measured average amplitude with the expected average amplitude yields, through the use of standard probability formulae, a statistical measure of significance indicating the likelihood for all subjects to have the same peak probability times for REM sleep. The result of Barrel's test is a ratio of 1.90 corresponding to a significance of P < 0.027. Fig. 2 is a histogram of sleep onset times for all nights studied. A bimodal distribution is observed; 12:46 a.m. is an effective dividing time between the two unimodal distributions, which will be referred to as early sleep onset and late sleep onset nights. Fig. 3 compares the clock time REM probability functions for the total sample (38 nights), the early sleep onset nights (21 nights) and the late sleep
REM SLEEP CYCLE
181
9"
c~
:E
I
ALL NIGHTS
. . . . .
SLEEP ONSET ~ 1 2 : 4 6 A M
.....
SLEEP ONSET~I2:46AM
8-
76 m
Z
ta-
5.
O
4.
.
.
.
.
II !I
:E
I'
Z
l
J
0
I
i0:50 PM
•
ii:i0
Ii:30
,I.....
w
ii:50
|
12:10
TIME
OF
12:30
U ! I •
12:50
SLEEP
w
w
i:i0
1:30
w
1:50
2:10 AM
ONSET
Fig. 2. Histograms s h o w i n g sleep o n s e t t i m e s for all s t u d y nights, nights w i t h sleep onset prior to 1 2 : 4 6 a.m. and nights w i t h sleep o n s e t after 1 2 : 4 6 a.m.
onset nights (17 nights). Fig. 4 shows REM probability as a function of time elapsed since sleep onset.
/
.~ 0
'
t/
All nights
Nights with sleep onset before 12:46 AM
' :VII ; w
IOPM
I
0
Nights with sleep onset a f t e r 12:46 RM
1
i
IAM
4
CLOCK TIME
Fig. 3. REM probabilities as a f u n c t i o n o f c l o c k time for all nights (N = 3 8 ) , nights w i t h sleep o n s e t prior to 1 2 : 4 6 a.m. (N = 2 1 ) , and nights w i t h sleep o n s e t after 1 2 : 4 6 a.m. (N = 17).
|
0
w
•
w
90
180
270
|
360
•
450
ELASPED TIME SINCE SLEEP ONSET
Fig. 4. REM probability as a f u n c t i o n o f elapsed time since sleep onset.
182
Discussion
The composite clock time REM probability function, shown in Fig. 1, shows that this sample of similar aged normal subjects has a high probability for REM at specific times throughout the night regardless of time of sleep onset. The first peak probability time occurs at approximately 1:30 a.m. and corresponds to a 15% probability for REM; shortly thereafter at 2:15 a.m. the probability has decreased to 7%. There is a rather rapid rise to the next peak of 34% occurring at 3:15 a.m.; this sharp peak is followed by a trough of 10% at 3:50 a.m. The next peak (4:30 a.m.) and trough are n o t well-defined; b u t the fourth peak occurring around 5:45 a.m. and corresponding to 40% REM probability is followed by a sharp fall to a low o f 20% which is followed by a peak probability of 56%. Fig. 1 also shows that after the second REM period there is a shortening of the REM cycle from approximately 105 to less than 90 min, indicating a non-stationary rhythm. The deviation of the experimental peak probability times estimated from Fig. 1 from the predicted (Fourier series) peak probability times are +7, +22, +7, --8, a n d - - 2 3 min for the first, second, third, fourth and fifth peaks, respectively. Experimental and predicted results are close with the exception of the observed 7:00 a.m. peak which falls 23 rain early. The non-stationary nature of the REM rhythm correctly corresponds to the direction of deviation of the earlier peaks. Data were n o t analyzed after 7:00 a.m. (due to the small number of nights that would be included) which could account for the premature observation of the predicted 7:28 a.m. peak. In addition, Fig. 1 shows that the peak clock time REM probability periods are non-specific with respect to the sequentially numbered REMPs. This is seen by observing that the second peak of the curve for the first REMP coincides in clock time with the first peak of the REMP 2 curve, the second peak for curves for REMP 2 and 3 falls closely in time with the first peak of curves for REMP 3 and 4
R.J. McPARTLAND, D.J. KUPFER
respectively, and the third peak of the REMP 3 curve coincides with the second peak of the REMP 4 curve. These results are interpreted as indicating that if a person missed having his first REMP around 1:30 a.m., he would tend to hold off REM until around 3:15 a.m., and so forth. Because it was felt that the bimodal distribution (with distributions displaced by 80 rain) of the time of sleep onset histogram could account for some of the above findings, the nights were grouped into those with sleep onset before and after 12:46 a.m. It is reasoned that if the timing of REM sleep was solely a function of elapsed time from sleep onset, and had no relation to clock time as hypothesized above, the clock time REM probability curve for the post 12:46 a.m. sleep onset nights would look similar to the pre 12:46 a.m. curve except that it would be shifted along the clock time axis by 80 min. However, this is n o t the case; both curves are very similar w i t h o u t any time shift. In addition, both early and late sleep onset time distributions are relatively broad (approximately 80 min} minimizing the probability that the observed clock time REM probability rhythm is an artifact due to clustering of sleep onset times. One does note the absence of a 1:30 a.m. peak in the probability curve for the late sleep onset nights, which is due to the late sleep onset times for this group of nights (mean of 1:30 a.m.). Moreover, there is a pronounced trough in the early sleep onset probability curve separating the first and second REM probability peaks. This trough is n o t present in the late sleep onset curve which instead manifests a steady rise in REM probability after 1:45 a.m., terminating with the 3:15 peak. The d o t t e d line on the late sleep onset curve illustrates the probability expected from the clock time related REM driving force and is similar to that portion of the early sleep onset probability curve. The fact that REM probability is higher than expected between 1:45 a.m. and 2:45 a.m. indicates an additional driving force for REM which may correspond to the tendency for
REM S L E E P C Y C L E
REM sleep to occur approximately 90 min after sleep onset. In this case, the excess REM probability results from the earlier sleep onset nights within the late sleep onset set of nights. This, as does Fig. 4, suggests that times of high probability for REM sleep are certainly n o t related solely to clock time; REM probability is also a very strong function of elapsed time after sleep onset. In light of data presented herein and data presented in other studies cited above, it would seem that, like many other biological phenomena, the phase of the R E M / N R E M cycle is determined by multiple factors. Elapsed time after sleep onset and the time of night are both shown to be strong influences. Other investigators (Webb et al. 1971; Weitzman 1974; Webb and Agnew 1977) studying the sleep of altered sleep-wake regimens have noted a change in REM probability when sleep times are displaced from normal, i.e., an increased probability for REM to occur earlier than they expected (less than 90 min after sleep onset). Perhaps this is a manifestation of the time of day influence. An additional factor, which the above studies of altered sleep-wake cycles address, is the existence of a 24 h REM probability cycle peaking at roughly mid-morning (6:00 a.m.) and having a minim u m at a b o u t 6:00 p.m. This modulation of REM sleep nicely corresponds to the general tendency for REM sleep to increase in probability as the night goes on. Unlike the study of Schulz (Schulz et al. 1975), we did n o t have many consecutive nights o f data on a single subject necessary for investigation of reported night to night shift in the time of corresponding REM periods. However, we have done preliminary analysis of this t y p e on sleep of depressed patients and have noted no shift of this type. In contrast to the data of Schulz, our data support a BRAC that is an integral subharmonic of 24 h. As Table I indicates, all subjects tend to have REM at the same times as each other. This raises the possibility that the driving force which controls the timing of the REM rhythm as related to clock time may be out-
183
side of the person. This being the; case, it is difficult to speculate as to its origin. If it were the light/dark cycle, as is improbable due to its changing over the many months over which this study t o o k place, clock time influence would disappear under conditions of constant or irregular lighting. Moreover, if one accepts the premise that the REM cycle and the BRAC are the same or intimately linked, then the BRAC tends to have a c o m m o n phase across subjects and is also partially controlled by a c o m m o n driving force. Considering most of the published results on the REM rhythm, 3 things seem apparent. First, there is a sleep-independent c o m p o n e n t of the REM rhythm (BRAC); second, there is also a sleep~iependent c o m p o n e n t of the REM rhythm; and finally, whatever phase the 9 0 m i n REM rhythm takes, it seems to be modulated by a 24 h periodicity° This suggests a 2-clock model for REM. One clock is sleep independent running continuously; its period is a submultiple of 24 h. The other runs only during sleep. Both have inputs to a mediating mechanism triggering REM sleep. Neglecting the non-stationary nature of the REM rhythm, and realizing that the period of both clocks is a b o u t 90 min, a beat frequency of 1 c/day will be produced when the average period of the 2 clocks differs by 5 min. This beat frequency corresponds to the circadian periodicity, mentioned above, modulating the ultradian rhythm. This model presents an obvious problem; if the phase of the circadian REM periodicity is fixed from day to day and across subjects, as stated (Webb and Agnew 1977), there must be a fixed phase relationship between the 2 clocks. Since the phase of one clock is stationary and the phase of the other is determined by sleep onset, this seems unlikely without fixing the time of sleep onset (possibly including a third, independent clock for the circadian rhythm would be better). If, however unlikely, there is a fixed phase relationship between the clocks, a change in this relationship could account for the shortened REM latency (time
184
between sleep onset and REM sleep) observed in depressive disease (Kupfer and Foster 1972). On the other hand, if this is n o t the case, shortened REM latencies would seem to be due to a malfunction related to the sleep-dependent clock.
Summary The phase of the REM sleep r h y t h m was studied in 10 normal subjects each of w h o m was sleep studied for 4 consecutive nights. For analysis, each night of sleep was aligned according to clock time and each minute was scored as REM or non-REM. With these data, REM probability was found as a function of clock time. Fractional harmonic analysis indicates a 90 min periodicity. The REM probability curve shows peaks occurring at 1:30 a.m., 3:15 a.m., 4:30 a.m., 5:45 a.m. and 7:00 a.m. Statistical measures comparing the time of REM sleep across subjects suggests that subjects tend to have REM sleep at the same time of the night as each other. The influence of elapsed time after sleep onset on REM sleep is also reestablished. Results indicate that the time of REM sleep is determined by both clock time and time of sleep onset, suggesting t w o clocks, one sleep d e p e n d e n t and the other related to the basic rest activity cycle (BRAC), which are responsible for driving REM sleep. Furthermore, the similarity of REM times across subjects indicates the possible existence of an extra-personae REM driving force linked to clock time and possibly the BRAC.
R~sum~ Cycles de sommeil avec m o u v e m e n t s oculaires rapides, horloge et ddbut du sommeil La phase du r y t h m e de sommeil avec mouvements oculaires rapides (REM) a ~t~ ~tudi~e chez 10 sujets n o r m a u x d o n t chacun a ~t~ enregistr~ pendant 4 nuits cons~cutives.
R.J. McPARTLAND, D.J. KUPFER
Pour l'analyse, chaque nuit de sommeil a ~t~ align~e suivant le temps r~el et chaque minute a ~t~ scor~e c o m m e sommeil avec et sans mouvements oculaires rapides. A partir de ces donn~es, la probabilit~ du sommeil REM s'est r~v~l~e ~tre fonction du temps r~el. Une analyse harmonique fractionn~e indique une p~riodicit~ de 90 min. La courbe de probabilit~ du sommeil REM montre des pics survenant ~: l h 30, 3 h 15, 4 h 30, 5 h 45, et 7 h . Des mesures statistiques comparant le m o m e n t du sommeil REM d'un sujet ~ l'autre sugg~rent que les sujets tendent ~ avoir du sommeil REM aux m~mes m o m e n t s de la nuit. L'influence du temps ~coul~ apr~s l'endormissement sur le sommeil avec mouvements oculaires rapides est ~galement r ~ t a b l i e . Les r~sultats indiquent que les m o m e n t s de sommeil REM sont d~termin~s ~ la fois par le temps r~el et le temps de l'endormissement, sugg~rant que deux horloges, l'une d~pendant du sommeil et l'autre li~e au BRAC, sont responsables de la survenue du sommeil REM. En outre, la similarit~ des temps de REM entre sujets indique l'existence possible d'un facteur extrapersonnel li~ au temps r~el et possiblement au BRAC.
References Armstrong, C.E. Cycle analysis -- a case study, part 25: testing cycles for statistical significance. Cycles, 1973, 24: 231--236. Dewey, E.R. Cycle analysis -- a case study, part 9: the moving lineage. Cycles, 1971, 22: 297--318. Globus, G. Rapid-eye-movement cycle in real time. Arch. gen. Psychiat., 1966, 15: 654--659. Kleitman, N. Basic rest activity cycle in relation to sleep and wakefulness. In: A. Kales (Ed.), Sleep Physiology and Pathology. Lippincott, Philadelphia, Pa., 1969: 33--38. Kupfer, D.J. and Foster, F.G. Interval between onset of sleep and rapid-eye-movement sleep as an indicator of depression. Lancet, 1972, ii: 684-686. Moses, J.M., Lubin, A., Johnson, L.C. and Naitoh, P. The rapid-eye-movement cycle is a sleep dependent rhythm. Nature (Lond.), 1977, 265: 360-361. Rechtschaffen, A. and Kales, A. (Eds.) A Manual of
REM SLEEP CYCLE Standard Terminology, Techniques of Scoring Systems for Sleep Stages of Human Subjects. Government Printing Office, Washington, D.C., 1968. Schulz, H., Dirlich, G. and Zulley, J. Phase shift in the REM sleep rhythm. Pfliigers Arch. gas. Physiol., 1975, 358: 203--212. Vaux, J.E. Technical bulletin 1964---65 fractional harmonic analysis program. Cycles, 1965, 16: 301--310. Webb, W.B. and Agnew, H.W., Jr. Analysis of the
185 sleep stages in sleep-wakefulness regimens of varied lengths. Psychophysiology, 1977, 14: 445--449. Webb, W.B., Agnew, H.W., Jr. and Williams, R.L. Effect on sleep-wake cycles of varied length. Psychophysiology, 1971, 42: 152--155. Weitzman, E., Nogeire, C., Perlow, M., Fukushima, D.S., Assin, J., McGregor, P., Gallagher, T. and Hellman, L. Effects of a prolonged 3 hour sleepwake cycle on sleep stages, plasma cortisol, growth hormone and body temperature in man. J. clin. Endocr., 1974, 38: 1018--1030.
