AN AUTOMATED SYSTEM FOR RECORDING AND ANALYSIS OF SLEEP IN MICE
An Automated System for Recording and Analysis
of Sleep in Mice
Sigrid C. Veasey,,1,2,4 Otto Valladares,3,4 Polina Fenik,1,2,4 David Kapfhamer,3,4 Larry Sanford,5 Joel Benington,6 and Maja Bucan1, 3,4
1Center
for Sleep and Respiratory Neurobiology, 2Department of Medicine, 3Center for Neurobiology of Behavior of the
Department of Psychiatry, 4School of Medicine, 5Department of Animal Biology, School of Veterinary Medicine
6University of Pennsylvania; Department of Biology, St. Bonaventure University
Abstract: Significant differences in many aspects of sleep/wake activity among inbred strains of mice suggest genetic influences
on the control of sleep. A number of genetic techniques, including transgenesis, random and targeted mutagenesis, and analysis
of quantitative trait loci may be used to identify genetic loci. To take full advantage of these genetic approaches in mice, a comprehensive and robust description of behavioral states has been developed. An existing automated sleep scoring algorithm,
designed for sleep analysis in rats, has been examined for acceptability in the analysis of baseline sleep structure and the response
to sleep deprivation in mice. This algorithm was validated in three inbred strains (C57BL/6J, C3HeB/FeJ, 129X1/SvJ) and one
hybrid line (C57BL/6J X C3HeB/FeJ). Overall accuracy rates for behavioral state detection (mean±SE) using this system in mice
were: waking, 98.8%±0.4; NREM sleep, 97.1%±0.5; and REM sleep, 89.7%±1.4. Characterization of sleep has been extended to
include measurements of sleep consolidation and fragmentation, REM sleep latency, and delta density decline with sleep. An experimental protocol is suggested for acquiring baseline sleep data for genetic studies. This sleep recording protocol, scoring, and analysis system is designed to facilitate the understanding of genetic basis of sleep structure.
Key words: Sleep; mice; electroencephalography; genetics; sleep scoring; arousal index; sleep bouts; phenotype
INTRODUCTION
used to understand the genetic control of sleep/wake behavior.
THE AVAILABILITY OF MOUSE STRAINS HARBORING VARIATIONS IN GENES INVOLVED IN SLEEP
CONTROL WILL BE HELPFUL IN UNDERSTANDING
the complexity of sleep at the molecular and biochemical
level. Genetic, physiological and anatomical studies are
providing candidate genes that may influence the control of
sleep. The roles of these genes may be ascertained by the
analysis of aberrant phenotypes of mice with null mutations generated by targeted mutagenesis in embryonic stem
cells.1,2 Alternatively, to discover novel genes, techniques
of forward genetics may be employed. Forward genetics
approaches involve the identification of mouse strains with
unique phenotypic traits, followed by genetic mapping of
the loci controlling the traits. The phenotype-based
approaches include analysis of quantitative trait loci (QTL)
in inbred strains of mice3 and random mutagenesis screens
for single gene mutations.4,5 All of these techniques may be
Sleep states are defined by specific electroencephalographic and electromyographic signals in mice as in other
mammalian species. Rapid-eye-movement sleep (REM
sleep) is characterized by postural muscle atonia, rapideye-movements and predominant theta wave activity in the
hippocampal electroencephalogram (EEG), whereas nonrapid-eye-movement sleep (NREM sleep) is characterized
primarily by lower frequency, higher amplitude waveforms
in the cortical EEG.6,7 Sleep progresses from faster wave
NREM sleep with spindle waves to slower wave NREM
sleep, in which delta waves predominate. After a period of
NREM sleep, REM sleep ensues.6 Unlike in humans, sleep
activity in rodents is more evenly distributed across the 24hour cycle,6 with a preponderance of both NREM sleep and
REM sleep in the period of the 24-hour day with greater
ambient light.7 Moreover, sleep cycle lengths of NREM
sleep and REM sleep are shorter in laboratory rodents.8
While these differences exist, homeostatic9 and circadian10
control of sleep, as well as neurochemical modulation of
sleep,11-18 appear consistent across species. The mouse
provides a good genetic model for the study of sleep state
control and timing, and the study of sleep states in mice
may ultimately provide insight into the control of sleep in
Accepted for publication August 2000
Address Correspondence to: Sigrid Carlen Veasey, MD or Maja Bucan, PhD,
Center for Sleep and Respiratory Neurobiology, University of Pennsylvania
School of Medicine, 972 Maloney Bldg., 3600 Spruce St., Philadelphia, PA
19104-4283. Tel: 215-349-8015; Fax: 215-662-7749;
E-mail: [email protected] or [email protected]
SLEEP, Vol. 23, No. 8, 2000
1
Sleep Analysis in Mice—Veasey et al
Heller,35 was designed for the analysis of sleep in rats, and
has been made freely available (with source code) to the
sleep research community. We have validated the existing
sleep analysis algorithm in mice and have added analyses
to the program extending the characterization of sleep in
mice. In this paper, we determine the accuracy of this program in sleep state detection in mice, provide a protocol for
recording baseline sleep/wake activity, and we describe
sleep structure in three inbred strains and one hybrid strain.
humans.
A number of studies of sleep in different strains of mice
suggest genetic influences on many aspects of behavioral
state control.6-12,14-32 A high degree of variance between
inbred strains of mice has been shown for many objective
parameters of sleep and sleep control.6,7,9-11,21-31 These
include differences in (1) total sleep time within a 24-hour
light-dark cycle,10,21-24 (2) time spent in NREM sleep,9,10
(3) time spent in REM sleep,9,10,21 (4) the diurnal ratio of
wakefulness,10,24 (5) REM sleep as a percentage of total
sleep time,24 (6) sleep state ontogenesis,7,8 (7) the ratio of
NREM sleep to REM sleep with aging,7,25,26 (8) effects of
changes in ambient temperature,27 (9) responses to REM
sleep deprivation,9 (10) effects of infectious challenges,28-30
(11) effects of immune modulation,11,31 and (12) EEG spectral profiles.24 Strains showing differences in one sleep
parameter may not exhibit differences in other sleep parameters, suggesting that the genetic control of sleep will likely involve a multitude of genes.
A recent important study highlights the tremendous
value of the mouse as a model organism for studies of
genetics of sleep/wake control.20 A targeted null mutation
of the prepeptide orexin was studied for its effect on
sleep/wake behavior in the mouse. These mice exhibited
behavioral and electrophysiological anomalies similar to
narcolepsy in humans: increased sleep in the active period,
shortened REM sleep latencies, and cataplexy-like behavior.
A complete phenotypic description of sleep in inbred
strains of mice is necessary for the genetic analysis of
sleep/wake control. The characterization of sleep/wake
behavior in large numbers of mice can be performed more
quickly using an automated sleep analysis system. Over
the past three decades, numerous sleep recording and automated sleep analysis systems have been developed and
modified9,10, 32-40, 45-52 for the study of sleep in laboratory
rodents. The accuracy of each of the validated programs,
relative to human scoring, is excellent.31,33,38,39 There are,
however, potential limitations to the existing programs.
Not all of the programs are readily available, and there are
differences among the automated programs in both the
parameters used to describe sleep and in the electrophysiological characteristics used to identify behavioral states. A
phenotypic assay of sleep/wake activity should identify not
only baseline behavioral state distribution, but also should
measure parameters used in the analysis of human sleep:
sleep fragmentation, REM sleep latency, sleep bout duration, the robustness of slow-wave sleep early in the sleep
period, and sleep cycle durations.
To implement an automated sleep analysis system that
can provide a complete phenotypic characterization of
sleep structure in mice, we have adopted and modified an
existing software system. This software, ACQ/SSSSS version 3.4© (ACQ 3.4), developed by Benington and
SLEEP, Vol. 23, No. 8, 2000
METHODS
Mouse Strains
Adult male mice, aged 13 and 19 weeks at the time of
behavioral state recordings, were used for these studies.
All mice were obtained from the Jackson Laboratory, Bar
Harbor, ME. Mice weighed 25 to 30 grams on the day of
surgical implantation. Three inbred strains of mice were
implanted with EEG and EMG electrodes for chronic
recordings (C3HeB/FeJ, n=17; C57BL/6J, n=15;
129X1/SvJ, n=10), as well as a hybrid strain (F2
C3HeB/FeJ X C57BL/6J, n=13). The Institutional Animal
Care and Use Committee of the University of Pennsylvania
approved the methods and study protocols in full.
Surgical Implantation of Electrodes
Recording electrodes were made as follows: Tefloncoated silver wire (Medwire, Mt. Vernon, NY) was used to
make ball electrodes for the three electroencephalographic
(EEG) electrodes. These were fashioned by shaving 4 mm
of insulation from the distal tip of the electrode wire and
then heating the exposed wire under flame until a 1 mm
ball was formed. Initially, a stainless steel screw (size 00,
Morris Precision) was used as the ground electrode, and in
later animals, this was replaced by a fourth ball electrode.
The silver wire was also used to fashion two neck electromyography (EMG) electrodes. For these, 8 mm at one
end was stripped to create an electrically conductive loop
2-mm in diameter. All six electrodes were soldered onto a
shielded multi-stranded cable (6-wire 30 gauge SH; Cooner
Wire, Chatsworth, CA). The end of the cable proximal to
recording equipment was soldered onto a 6-pin plug (363
plug, Plastics One, Roanoke, VA).
Prior to surgery, mice were anesthetized with 70-85
mg/kg ketamine and 9 mg/kg xylazine administered
intraperitoneally. In all four strains, ketamine and xylazine
provided 60 to 90 minutes of general anesthesia. Using
aseptic techniques, the scalp hair was shaved, and a midline
incision of 2 cm was made to expose the dorsal cranium
and the neck musculature. The dorsal surface of the skull
was completely cleaned of the periosteum. A trace amount
of etching gel (P-10 gel, 3M Dental, Irvine, CA) was
2
Sleep Analysis in Mice—Veasey et al
placed over the cranium to increase the skull surface area.
After five minutes, the skull was thoroughly rinsed with
sterile saline and allowed to dry. Using a 21-gauge needle,
four holes through the cranium were made for EEG and
ground electrodes at 2 mm lateral to Bregma and 1 mm rostral (for two rostral EEG electrodes) and 2 mm lateral to
Bregma/ 3 mm caudal to Bregma (for the ground and a caudal EEG electrode). The caudal electrode position was
found optimal for the detection of hippocampal theta, after
testing both 1 mm rostral and 1-2 mm caudal to this site.
EEG electrodes and the ground electrode were implanted 1
mm below the skull’s surface (on the dura) and secured first
with dental paste (Resin Bonded Ceramic, 3M) and then
with dental acrylic (P-10 Acrylic, 3M). The neck EMG
loops were sutured to dorsal neck musculature with 5-0 silk
ties. Cranial electrodes were secured. The incision was
then closed with a suture. Each mouse received 0.005
mg/kg gentamycin intramuscularly and was placed in a
clean cage in the recording chamber. The cable was then
connected to a fully rotating commutator (SLC 6, Plastics
One, Roanoke, VA) supported by a counterweight. Three
days after surgery, animals received a second injection of
gentamycin (same dose).
2000
1600
1200
EMG
800
400
0
0
51
153
204
255
EMG Amplitude
255
204
153
S*T
102
51
0
0
1
2
3
4
5
Delta Power Density
Recording Procedures
Figure 1—Sleep state scoring algorithm. Plotted in the upper panel scattergram
are values for every 10-second epoch for the product of sigma and theta densities (S*T) versus the amplitude of the integrated EMG (EMG) for a 24-hour period. Plotted epochs fall into two general clusters: waking and sleep. A line may
be placed separating the two clusters. This line is distinguished by its x-intercept
and its slope. Typically, epochs with significant movement artifact will have high
EMG values, so that these epochs will be correctly scored as waking. The lower
panel scattergram plots EMG vs. delta/theta densities. This generally results in
three loose clusters, where the lower left hand cluster represents REM sleep, the
upper cluster represents waking and the lower right represents NREM sleep. An
X-intercept just beyond the REM sleep cluster is used to distinguish REM sleep
from NREM sleep. Both intercepts and the slope may be adjusted to balance
overcalls and undercalls for a given behavioral state.
Four individual mouse cages were housed inside a
sound-attenuated, shielded and well-ventilated chamber
(EPC-010, BRS/LVE, Laurel MD), with a 12-hour lightson (07:00 h to 19:00 h) and 12 h lights-off schedule. The
ambient lighting was 75 lux for the lights-on period and <1
lux for lights-out. The temperature in the recording chambers was held constant at 25°C±2°C, as measured daily
during recordings.
EEG signals were filtered at 0.3 and 35 Hz (1/2 max,
6dB/octave), and EMG signals were filtered at 1 and 100
Hz, and both signals were amplified (12A5, AC amp,
Astro-Med, West Warwick, RI). Various combinations of
the rostral-caudal EEG electrodes were viewed polygraphically (Grass 12-32-35S, Neurodata Acquisition, AstroMed) using an electrode selector board (12 PB_36
Electrode Selector, Astro-Med) to find optimal EEG combinations for detecting active waking, theta activity in the
EEG, and slow wave sleep. Either caudal electrode could
function as the ground or EEG electrode. Signals were sent
to an A/D board (Converter 4801A, ADAC, Woburn, MA)
by way of a mini-phone plug (CAB-20333, Astro-Med)
connected to a male BNC cable. Prior to beginning recordings, the EEG signals for each mouse were calibrated using
a 100 mV calibration signal from the polygraph.
The behavioral state acquisition and analysis program
used for these studies was ACQ 3.4 from Dr. Joel
Benington. This recording and analysis system was originally developed for behavioral state study in rats.38 The
SLEEP, Vol. 23, No. 8, 2000
102
optimal EEG and EMG signals for four mice, per computer, were each digitized at 100 Hz. EEG was recorded as
raw signals. The program performs Fourier analysis on 10second epochs of the EEG signals and then sums spectral
power for three frequency bands: delta (0.5-4.0 Hz), sigma
(10-14 Hz) and theta (6-9 Hz). The program rectifies the
EMG signal and then records a moving average amplitude
for each 10-second epoch.
Behavioral states are determined based on three variables: (1) the product of sigma density and theta density
(sigma*theta), (2) delta density divided by theta density
(delta/theta), and (3) the EMG amplitude. High EMG and
low sigma*theta is characteristic of waking, while low
EMG and high sigma*theta is characteristic of sleep. High
delta/theta is characteristic of NREM sleep, while low
delta/theta is characteristics of REM sleep. User-defined
3
Sleep Analysis in Mice—Veasey et al
thresholds for these variables are determined based on
visual examination of scatter plots displaying sigma*theta
vs. EMG (see Fig.1, upper panel) and delta/theta vs. EMG
(see Fig. 1, bottom panel). Waking was distinguished from
sleep by defining slope and intercept values for the
sigma*theta vs. EMG plot, and NREM sleep was distinguished from REM sleep by defining a delta/theta threshold.
and REM sleep was defined as a high frequency EEG with
a significant predominance (>75%) of theta waves and a
very low EMG amplitude. Figure 2 illustrates samples of
the electrographic signals as they appear on the computer
screen during off-line analysis of each 10-second epoch. If
signals for a given mouse could not be used to visually
determine sleep/wake activity because of noise, poor signal
quality or movement artifact, the data for that mouse were
not analyzed further.
For all mice with visually recognizable sleep states,
human state scores were compared with the computer’s
interpretation of state, and the algorithm’s intercepts and
slope were adjusted to optimize the accuracy of the computer-determined behavioral states and to balance overcalls
and undercalls for a given state (see Fig. 1). An accuracy
value for each state for 360 consecutive epochs was calculated in each mouse. A second one-hour sample from a different day was then used for validation purposes. This was
typically selected as an hour within the middle third of the
light period in an effort to increase REM sleep epochs within the one hour sample. A third sample was studied for
accuracy during a random hour of the dark period and compared with the mid-day light period hour in six mice for
each strain. In cases where sleep deprivation was performed and recordings were acceptable after sleep deprivation, a third or fourth validation was performed to determine the stability of sleep analysis accuracy after sleep
deprivation by gentle handling. With practice, this entire
process required one person-hour per mouse to validate,
determine the most accurate settings, and perform the analysis. Mice for which we were unable to obtain an accuracy
for the computerized algorithm of at least 90% for waking
and NREM sleep and at least 75% for REM sleep, were
included in the validation analysis but rejected from sleep
structure analysis.
Recording Protocol
After the seven to nine day post-operative recovery period, baseline sleep activity was recorded for five to seven
days. Sleep deprivation was then performed continuously
throughout one 12-hour lights-on period, by way of gentle
handling (stroking the head, vibrissae and back with a
feather duster). Animals were stroked whenever they were
observed to be quiescent for more than 10 seconds, or if
EEG activity showed higher amplitude or slower waveforms. EEG and EMG activity was recorded during all but
the initial 30 minutes of the sleep deprivation period, when
baseline signals were downloaded. During the period in
which EEG signals could not be viewed, mice were stroked
on faces and vibrissae when immobile. Two hours prior to
the end of the 12-hour deprivation, water bottles were
filled, the bedding was changed and a week’s supply of
food was placed in the cage. Sleep deprivation continued
until the lights were turned off at the usual time, and animals remained in the recording chambers for one week following sleep deprivation, during which sleep/wake activity
was continuously recorded.
Validation and Inclusion of Data
Sleep-wake states were scored visually by two human
scorers for a one-hour sample of raw EEG and integrated
EMG signals. For human scoring, waking was defined as
a predominance (within a 10-second epoch) of fast, desynchronized waves and high EMG amplitude. NREM sleep
was defined as a predominance of higher amplitude EEG
waves, consistent with either sigma or delta waveforms,
Parameters of Sleep and Their Definitions
The measures of sleep timing analyzed were: (1) time
spent asleep in minutes (overall, in the light period and in
the dark period), (2) the diurnal ratio of wakefulness, (3)
sleep efficiency (overall, in the light period, and in the dark
period), (4) percentage of recording time spent in NREM
sleep (overall, in the light period, and in the dark period),
(5) mean EEG delta density in NREM sleep, (6) the magnitude of the decrease in delta density over the course of the
light period, (7) mean EEG sigma density in NREM sleep,
(8) percentage of recording time spent in REM sleep (overall, in the light period, and in the dark period), (9) mean
EEG theta density in REM sleep, (10) the diurnal ratio of
REM sleep, (11) mean bout lengths for sleep states (minutes per total sleep bout, minutes per bout of NREM sleep,
and minutes per bout of REM sleep), (12) arousal index,
and (13) REM latency. Detailed descriptions of how each
of these variables was defined and calculated are below.
Figure 2—Digitized electroencephalographic recordings of 10-second epochs of
waking (W), NREM sleep (N), and REM sleep (R) as seen on the recording system along with histograms of the average EMG amplitudes for each epoch.
REM sleep is scored when >75% of the EEG waves in a 10-second epoch are
theta frequency and the EMG amplitude is reduced.
SLEEP, Vol. 23, No. 8, 2000
4
Sleep Analysis in Mice—Veasey et al
as waking following combinations of three or more epochs
scored as NREM sleep or two or more epochs of REM
sleep. This was expressed as the frequency of occurrence
per hour of sleep. Using a similar approach, REM sleep
latency was quantified. This was determined as the duration from the onset of three or more consecutive epochs of
NREM sleep to two consecutive epochs of REM sleep. If
three or more consecutive waking epochs occurred
between the onset and offset of REM sleep latency, REM
sleep latency was not measured.
The homeostatic response to short-term sleep loss was
characterized using parameters: the percentage change
from baseline in times spent in NREM sleep and REM
sleep for the first and second six hours of the recovery period, the percentage change from baseline in delta density for
NREM sleep for the same time periods, the percentage
change in delta energy (the product of delta density and
NREM sleep time) for the same periods, the change in
NREM sleep bout length relative to baseline, and the
change in the arousal index from baseline.
The time spent asleep was defined as the total time for
all NREM sleep and REM sleep epochs. This was calculated for the 24-hour cycle, the 12-hour light period and the
12-hour dark period, and expressed as a mean±standard
error (SE) for each strain. The diurnal ratio of wakefulness
was calculated as the ratio of all time spent in wakefulness
epochs in the light period relative to time spent in wakefulness epochs in the dark period. This also was expressed as
the mean±SE for each strain. Sleep efficiency was measured as the time spent in both, NREM sleep or REM sleep
epochs as a percentage of the total time (24-hour or 12-light
or dark period).
The sleep percentage of NREM was calculated as the
percentage of recording time during which epochs were
scored as NREM sleep. Delta density for NREM sleep was
calculated as the average delta density for all NREM sleep
epochs per animal, expressed as the mean±SE for each
strain. To characterize how slow-wave activity declines
across the rest period, we measured, for each baseline day,
the density for 0.5–4 Hz in comparison to the remaining
frequencies studied (4–20 Hz), expressed each hour’s value
as a percentage relative to the last hour of the light period,
and determined the slope of the best fit line for delta density change across the light period (delta decline).23 Mean
sigma density was calculated for all NREM sleep epochs in
each animal and expressed as the mean±SE for each strain.
To characterize REM sleep, we measured the percentage
of time spent in REM sleep (24-hour, 12-hour light and 12hour dark periods). These data were expressed as
mean±SE’s. Mean EEG theta density was calculated for all
epochs scored as REM sleep and expressed as mean±SE for
each strain. The diurnal ratio of REM sleep was calculated
as the time spent in REM sleep during the light period relative to the dark period.
Analysis of sleep bout lengths, arousal index, and REM
sleep latency required additional programming. Total sleep
bouts were defined as periods of three or more consecutive
epochs scored as either NREM or REM sleep. NREM
sleep bouts were defined as periods of three or more consecutive epochs scored as NREM sleep, and REM sleep
bouts were defined as periods or two or more epochs scored
as REM sleep. The termination of a sleep bout was defined
by the presence of three or more consecutive wake epochs.
To measure average duration of sleep bout lengths, data
were transferred from the Statout files of ACQ 3.4 onto an
Excel spreadsheet. A custom script was written in PERL
to detect combinations of consecutive epochs meeting criteria for the various types of sleep bouts. Data were calculated for each animal as the average duration in minutes.
The arousal index (the number of times per hour that sleep
is disrupted by one or more waking epoch) was measured
in the following manner. Data were transferred from the
Statout files of ACQ 3.4 onto an Excel spreadsheet, and a
custom script was written in PERL to detect epochs scored
SLEEP, Vol. 23, No. 8, 2000
Statistical Analysis
Components of sleep structure during baseline activity
were analyzed for differences between strains using factorial Analysis of Variance (ANOVA), and where applicable,
the Tukey post-hoc tests were applied. Sleep state comparisons within mice were drawn using repeated-measures
ANOVA, with Scheffe’s F-test when applicable. Paired ttests with Bonferroni correction for the number of parameters analyzed were used to analyze the response to sleep
deprivation. Statistical significance was achieved for probabilities of the null hypothesis at p<0.05.
RESULTS
Validation of the ACQ 3.4 Algorithm in Mice
The sleep analysis algorithm was validated for the light
and dark periods at baseline in 24 mice (6 in each strain)
and after 12 hours of sleep deprivation in 23 mice.
Analysis of baseline sleep parameters was performed on all
mice with stable baseline recordings and accurate computerized scoring using the ACQ 3.4 algorithm. Following
baseline recordings, mice were deprived of sleep for the 12
hour light period. Recovery sleep was recorded for 24
hours following sleep deprivation. Baseline sleep parameters were analyzed in 36 of the 55 implanted mice. Those
not analyzed included six that died within the peri- or postoperative period; in eight mice behavioral states could not
readily be determined by human scorers, and in five mice
behavioral states were not steady over the baseline period.
Over the course of this study, the success rates for stable,
accurate recordings increased steadily.
5
Sleep Analysis in Mice—Veasey et al
Table 1—Accuracy of the sleep scoring algorithm compared to human scoring: Initial Accuracy (IN AC), Baseline validation (BL Val),
and After Sleep Deprivation Validation (SD Val).
Mouse Strain
Waking Accuracy
NREM Sleep Accuracy
REM Sleep
Accuracy
C3HeB/FeJ
IN AC=99.1%±0.8
IN AC =96.3%±1.0
IN AC =87.3%±2.5
(n=8)
BL Val=97.8%±1.1
BL Val=96.9%±1.1
BL Val=93.4%±1.8
SD Val=99.6%±1.1
SD Val=97.6%±0.4
SD Val=88.8%±1.8
C57BL/6J
IN AC=98.1%±0.8
IN AC=98.6%±0.7
IN AC =89.2%±2.4
(n=8)
BL Val=99.3%±0.3
BL Val=97.7%±0.8
BL Val=88.8%±2.4
SD Val=97.2%±2.3
SD Val=94.6%±1.9
SD Val=80.0%±1.3
C3HeB/FeJ X
IN AC=95.1%±2.6
IN AC=95.3%±1.0
IN AC=85.2%±2.1
C57BL/6J
BL Val=94.3%±0.6
BL Val=97.4%±0.5
BL Val=90.1%±2.3
(n=7)
SD Val=90.0%±3.3
SD Val=90.4%±1.2
SD Val=75.2%±1.1
129X1/SvJ
IN AC=98.3%±0.9
IN AC=98.6%±0.4
IN AC=90.5%±4.0
(n=5)
BL Val=97.7%±0.3
BL Val=95.5%±1.6
BL Val=81.4%±3.2
SD Val=99.4%±0.3
SD Val=98.3%±0.2
SD Val=81.6%±4.5
The use of this algorithm was validated for mice by
comparing algorithmic state scores with state scores determined by visual inspection of electrographic parameters for
a) one hour of baseline in the light period (n=28), b) one
hour in the dark period (n=24) and c) one hour of recovery
sleep in the light period (n=23). Table 1 summarizes the
accuracy and validation data for each mouse strain and for
each state for the light period. Overall agreement between
the two scoring methods was 95.2%±0.7, with a range by
strain and state of 81.4% to 99.3%, during baseline recordings and a range of 75.2% to 99.6% after sleep deprivation.
There was no main effect of strain on the validity of the
program to accurately score any of the behavioral states,
except that the accuracy with which the program detected
REM sleep was higher in the C3H mice than in other
strains (p<0.01 for each other strain in comparison).
Overcalls and undercalls for each state were balanced in all
strains. As a group, the accuracy of the program in determining REM sleep was lower than its accuracy in determining waking or NREM sleep (p<0.05).
For each state, there was no significant difference in
accuracy between the first hour of scored epochs used to
compare with human scoring and the sample used in the
baseline validations procedure, as shown in Table 1. The
accuracy of the program in determining sleep/wake behavior during the dark period, relative to the light period was
also determined, and no significant difference was found
between daytime and nighttime validations (Table 2). In
addition, there was no difference between the baseline
accuracy and the post-sleep deprivation accuracy for waking and NREM sleep. There was, however, a decline in the
accuracy of scoring REM sleep after sleep-deprivation
from 89.7%±1.4 for baseline to 85.3%±2.2 after sleep loss
(p<0.05).
Table 2—Accuracy of the algorithm: light vs. dark period
Mouse Strain
Light
Period
W=99%±1
N=95%±1
R=87%±3
Dark
Period
W=99%±1
N=97%±1
R=90%±3
P value
C57BL/6J (n=6)
W=96%±1
N=96%±1
R=84%±3
W=98%±1
N=96%±1
R=85%±6
P=0.08
P=0.65
P=0.81
C3HeB/FeJ* C57BL/6J (n=6)
W=90%±3
N=95%±1
R=85%±3
W=100%±0
N=97%±1
R=92%±2
P=0.08
P=0.45
P=0.06
W=997%±1
N=96%±2
R=91%±3
W=98%±1
N=97%±1
R=87%±5
P=0.72
P=0.76
P=0.73
C3HeB/FeJ (n=6)
129X1/SvJ (n=6)
SLEEP, Vol. 23, No. 8, 2000
P=1
P=0.27
P=0.39
Graphical Display of Sleep Parameters
We have developed two graphical displays of sleep
states to examine on-going data obtained from individual
animals and to identify technical problems or phenotypic
outliners. Figure 3 illustrates the two ways to display data.
6
Sleep Analysis in Mice—Veasey et al
3a.
3b.
Figure 3—Graphical displays used to rapidly find outliers in genetic sleep studies. Two versions are presented. On the left is a display of hourly percentages of each
behavioral state. This example shows a steady-state pattern for all three behavioral states and provides general information on ultradian cycle length, ultradian amplitudes, diurnal ratio amplitude, and estimates of NREM sleep time and REM sleep time. The graphical displays to the right are double-raster plots as used for wheelrunning activity in screening for mutant circadian rhythms. Sensitivities may be adjusted to distinguish three bin-sizes for various percentages of time spent in each
behavioral state. These raster plots reveal the higher levels of waking in the early dark period, and the increase in REM sleep time within the second portion of the
lights-on period.
4b.
4a.
Figure 4—Individual hourly percentage distributions for each behavioral state are shown for two different mice. Fig 4a. shows behavioral state percentages for a
C57BL/6J mouse. The upper panel displays NREM sleep hourly percentages, the middle panel displays REM sleep percentages, and the lower panel shows waking percentages. Notice the apparent decline in REM sleep midway through the baseline recordings. Review of raw signals revealed deteriorating electrical signals
in this animal. Fig 4b. shows hourly percentages for a 129X1/SvJ mouse with blunted ultradian and circadian fluctuation, typical for that substrain of mice. Notice at
130 hours into recording baseline, there was an unexplained and significant increase in wakefulness lasting approximately 12 hours, followed by a sleep rebound.
This was unexplained, and not seen in other mice within the same recording chamber.
SLEEP, Vol. 23, No. 8, 2000
7
Sleep Analysis in Mice—Veasey et al
Expression of each state can be displayed as a percentage
of all 10 second epochs scored as a given behavioral state
within each hour of total recording time across multiple
days (Figure 3a). Alternatively, the diurnal distribution of
each state can be displayed in a double raster plot, as is
commonly used in studies of circadian rhythms (Fig. 3b).
more of the behavioral states, particularly REM sleep (Fig.
4a), persisting for several days or longer, or an unexplained
sleep disruption for several hours or more in an individual
mouse (3 of 41 mice, as in Fig. 4b). From these hourly percentage graphs, a sample of three to five days of baseline
recordings was selected for further analysis.
Establishing steady-state baseline sleep
Baseline Sleep Parameters in Each of the Studied Strains
Prior to calculating baseline sleep/wake parameters in
the mice, hourly percentage distributions were plotted for
REM sleep, NREM sleep and waking for each hour of the
baseline recordings (Figure 4). This display of the baseline
sleep:wake states was monitored for steady-state pattern
over the five to ten day period. Occasionally (in 2 of 41
mice), we detected a gradual increase or decrease in one or
Total sleep: One-hour samples of EEG and EMG signals were reviewed for the 36 [C3HeB/FeJ (n=10),
C57BL/6J (n=10), C3HxC57 (n=9) and 129X1/SvJ (n=7)]
mice with steady-state baselines and validated computerized sleep state scoring. Baseline sleep state distribution
timing data are presented in Table 3. There was an overall
difference across strains of mice in time spent asleep across
Table 3—Baseline sleep parameters for commonly-used strains of mice
C3HeB/FeJ
(n=10)
C57BL/6J
(n=10)
C3HeB/FeJ*
C57BL/6J
(n=9)
129X1/SvJ
(n=7)
Time Spent Asleep
in 24 Hours (min)
T:613.6±24.7
L:410.8±12.7
D:202.9±15.9
T:665.11±20.7
L:449.6±7.9
D:215.5±13.8
T:562.7±25.2
L:344.5±14.3*
D:218.2±28.9
T:857.1±33.7*
L:440.2±13.6
D:416.9±26.4*
Wake Diurnal ratio
(Light:Dark)
0.60±0.02
0.54±0.01
0.78±0.07
1.01±0.08*
Sleep Efficiency
(% of 24 hour
cycle asleep)
T:42.6±1.7
L:57.0%±1.8
D:28.2%±2.2
T:48.2%±2.4
L:62.7%±1.0
D:33.7%±4.1
T:39.1%±1.7
L:47.8%±2.0*
D:30.3%±4.0
T:51.7%±4.8*
L:56.1%±3.7
D:47.3%±6.0*
Average sleep
cycle duration
(min)
T: 13.5±0.8
N:10.4±0.8
R:4.8±0.6
T:14.8±0.8
N:12.7±1.1
R:4.1±0.6
T:12.6±0.7
N:11.3±0.5
R:4.0±0.2
T:15.1±2.9
N:12.6±1.8
R:2.9±0.7*
Percentage of
time spent in
NREMS
T:37.7%±1.7
L:49.4%±1.8
D:25.9%±2.1
T:42.3%±1.4
L:56.9%±1.6
D:27.6%±1.8
T:35.6%±1.9
L:47.6±1.7
D:23.6%±2.8
T:50.7%±4.3*
L:54.4%±3.4
D:47.0%±5.4*
Delta density of
NREMS
55.8±11.9
64.0±16.3
79.3±22.4
77.7±12.2
Delta density
decline for light
period
-5.1±0.7
-7.2±2.3
-4.8±0.7
-2.6±0.7*
Sigma density of
NREMS
15.6±4.3
8.2±2.2*
15.2±6.2
13.0±3.3
Percentage of
time spent in
REMS
T:5.0%±0.2*
L:7.7%±0.4
D:2.3%±0.3
T:3.6%±0.5
L:5.4%±0.8
D:1.8%±0.3
T:3.7%±0.4
L:5.6%±0.6
D:1.7%±0.2
T:3.8%±0.5
L:4.6%±0.5
D:3.0%±0.5
Theta density of
REMS
60.2±23.5
23.9±8.1*
38.9±15.6
36.6±12.9
REM sleep Latency (min)
7.5±0.7
9.6±1.6
8.0±0.7
9.5±1.1
REMS
Diurnal Ratio
3.7±0.5
3.3±0.4
3.5±0.4
1.6±0.1*
Sleep Bout
length
(min)
T:12.9±0.9*
N:10.5±0.7
R:4.1±0.1
T:14.1±1.5
N:12.3±1.1
R:4.16±0.1
T:13.6±2.3
N:12.6±1.4
R:4.0±0.2
T:15.1±2.3
N:12.6±1.5
R:2.9±0.2
SLEEP, Vol. 23, No. 8, 2000
8
Sleep Analysis in Mice—Veasey et al
129SvJ
129SvJ
C3H/B6J
C3H/B6J
C3H/B6J
C3H/B6J
C57BL/6J
C57BL/6J
C57BL/6J
C57BL/6J
C3HeB/FeJ
C3HeB/FeJ
C3HeB/FeJ
C3HeB/FeJ
129SvJ
*
0
250
500
750
1000
0
100
200
300
400
500
0
100
TST-light
Total Sleep Time for 24 hours (in minutes)
129SvJ
*
200
300
400
500
0
TST-dark
129SvJ
C3H/B6J
C3H/B6J
C3H/B6J
C3H/B6J
C57BL/6J
C57BL/6J
C57BL/6J
C57BL/6J
C3HeB/FeJ
C3HeB/FeJ
C3HeB/FeJ
C3HeB/FeJ
0
*
10
20
30
NREMS
40
50
60
0
10
20
30
NREMS%
%24
40
50
60
0
10
20
30
NREMS%
40
50
60
0
dark
129SvJ
129SvJ
C3H/B6J
C3H/B6J
C3H/B6J
C3H/B6J
C57BL/6J
C57BL/6J
C57BL/6J
C57BL/6J
C3HeB/FeJ
C3HeB/FeJ
C3HeB/FeJ
C3HeB/FeJ
0
1
2
3
4
REMS%-24
5
6
0
2.5
5
7.5
0
10
*
1
2
3
REMS%-dark
REMS%-light
1
1.25
5
10
15
REMS time % total sleep
129SvJ
*
0.25 0.5 0.75
129SvJ
*
light
*
Waking Diurnal Ratio
129SvJ
129SvJ
*
129SvJ
4
0
*
1
2
3
4
REMS diurnal ratio
Figure 5—Histogram displays for parameters of sleep architecture. Data are expressed as means±standard errors for each strain studied. Overall values for
C3HeB/FeJ and C57BL/6J mice are similar for most parameters, except REM sleep time for the light period, and are therefore reflected in the total 24-hour period as
well. The strain showing larger differences for most parameters was the 129X1/SvJ strain.
degeneration,42 light entrainment appears normal as determined from graphical display records of individual mice
(Fig 4b).
NREM sleep: The differences between strains for
NREM sleep percentages paralleled the strain differences
observed in total sleep time. The 129X1/SvJ mice spent a
greater percentage of time in NREM sleep for the 24-hour
cycle than the other three strains (129X1/SvJ vs.
C3HeB/FeJ, p=0.008; 129X1/SvJ vs. C57BL/6J, p=0.05;
129X1/SvJ vs. C3H X B6, p=0.002). The 24-hour percentage of N-REM sleep time in C3H X B6 was similar to
C3HeB/FeJ mice. NREM sleep time was statistically less
for the hybrid strain than the percentage of time for
C57BL/6J mice (p=0.04). There were very small differences in NREM sleep percentage time in the light period
overall (p=0.008). Specifically, C57BL/6J mice spent a
slightly higher percentage of the light period in NREM
sleep than C3H X B6, p=0.02). Reflecting the pattern seen
with differences in total sleep times, the percentage of the
a 24-hour cycle (ANOVA, p<0.0001). Total sleep time
within the 24-hour cycle was similar for C57BL/6J,
C3HeB/FeJ and C3H X B6 mice (Scheffe’s post-hoc, N.S.).
In contrast, the 129X1/SvJ mice spent significantly more
time asleep than each of the other strains (Scheffe’s,
p<0.001). This increase in total time was directly related to
a two-fold increase (relative to each other of the strains) in
total sleep time during the dark period (p<0.0001).
Waking diurnal ratios: The diurnal ratios were higher
in C3H X B6 than C57BL/6J (p=0.005), and the diurnal
ratio was highest in the 129X1/SvJ’s (129X1/SvJ vs.
C3HeB/FeJ, p=0.0004; 129X1/SvJ vs. C57BL/6J, p<0.001;
129X1/SvJ vs. C3H X B6, p=0.008). In this substrain of
129 mice, the diurnal ratio of wakefulness time approached
1, and by repeated-measures ANOVA, there were no differences in waking activity for the light vs. the dark periods
(p=0.74). In contrast, a diurnal rhythm in this substrain
was present for REM sleep activity (light:dark = 1.75,
p<0.001). Although this substrain is susceptible to retinal
SLEEP, Vol. 23, No. 8, 2000
9
Sleep Analysis in Mice—Veasey et al
phasic sleep period with minimal sleep at lights-out, a gradual increase in sleep until through the first third of the
lights-on period, followed by a decline in sleep in lightsout. The hybrid group shows an intermediate pattern.
Hourly patterns also revealed less pronounced diurnal
rhythms for both NREM and REM sleep in the 129X1/SvJ
mice, although the rhythm was more appreciable for REM
sleep.
REM sleep latency: REM sleep latencies during baseline sleep were similar for the three inbred strains and the
one hybrid strain, with an overall mean latency from the
onset of a NREM sleep bout to REM sleep of 9.5±1.1 minutes.
Percentage of Each Hour
Spent in NREMS
80
60
40
20
0
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48
ZeitgeberTime
Percentage of Each Hour
Spent in REMS
12.5
10
7.5
Sleep Fragmentation, Consolidation, and Delta Decline
Across the Light Period
5
2.5
Sleep fragmentation (arousal index) did not differ
among these four strains. Total sleep bout lengths were
also similar in the four strains. REM sleep bouts, in contrast, were shorter in the 129X1/SvJ mice (p<0.05). The
decline in delta density across the light period was most
pronounced in the C57BL/6J mice (p<0.01) and was least
pronounced in the 129X1/SvJ mice (the strain with more
0
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48
ZeitgeberTime
C57BL/6J
C3HeB/FeJ
C3HxC57
129X1/SvJ
Figure 6—Hourly distributions of NREM sleep (top panel) and REM sleep (lower
panel) are double-plotted for each strain of mice studied. The diurnal ratio is significantly blunted in 129X1/SvJ mice (closed triangle) showing marked increases
in NREM sleep in the dark period. In contrast, this strain demonstrates a more
typical diurnal ratio for REM sleep (lower panel). State distributions for the hybrid
strain (closed square) fall just at or between the parent strains, C57BL/6J (closed
circles) and C3HeB/FeJ (open circles).
200
C3HeB/FeJ
150
100
dark period spent in NREM sleep was two-fold higher in
the 129X1/SvJ’s than in the other groups (129X1/SvJ vs.
C3HeB/FeJ, p=0.0007; 129X1/SvJ vs. C57BL/6J, p=0.003;
129X1/SvJ vs. C3H X B6, p=0.0002).
REM sleep: The percentages of time spent in REM
sleep were more variable within strains than percentages of
time spent in NREM sleep. The between-group differences
were of similar magnitude to the between-group differences for NREM sleep (a two-fold increase in the dark period for 129X1/SvJ mice, relative to other strains), but were
not statistically significant overall. In contrast, REM sleep
percentages during the light period were higher in the
C3HeB/FeJ mice compared to 129X1/SvJ mice (p=0.01),
while REM sleep percentages in the dark period were higher in the 129X1/SvJ mice compared to other groups
(p<0.001). Despite differences in REM sleep times, the
ratios of NREM sleep time to REM sleep time for the 24hour period were similar for the four groups (p>0.6 in all
cases).
Behavioral state distribution: Mean hourly distributions of NREM sleep and REM sleep (±SE) are shown in
Figure 6 for the four strains tested. As previously reported,23 C57BL/6J mice have a biphasic sleep pattern with a
second sleep-predominant time in the second half of the
dark period. In contrast, C3HeB/FeJ mice show a monoSLEEP, Vol. 23, No. 8, 2000
50
0
0 1 2 3 4 5 6 7 8 9 1011
Zeitgeber Time
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour into Recovery Period
300
250
C57BL/6J
200
150
100
50
0
0 1 2 3 4 5 6 7 8 9 1011
Zeitgeber Time
1 2 3 4 5 6 7 8 9 101112131415 161718192021222324
Hour into Recovery Period
250
129X1/SvJ
200
150
100
50
0 1 2 3 4 5 6 7 8 9 1011
Zeitgeber Time
1 2 3 4 5 6 7 8 9 10111213 1415161718 19202122 2324
Hour into Recovery Period
Figure 7—Delta power expressed as a percentage of the delta power relative to
the last hour of the lights-on period. The three strains differ significantly in the
delta decline and in the relative delta power following sleep deprivation (The
arrow indicated the onset of sleep recovery).
10
Sleep Analysis in Mice—Veasey et al
Table 4—Response to 12-hour total sleep loss on recovery sleep
Recovery Period
Parameter:
∆ in Total Sleep
(24 hours)
∆ in NREM sleep
(first dark period)
C3HeB/FeJ
n=9
47.74 min ±43.94
C57BL/6J
n=6
190.50 min ±46.15*
p<0.05
141.45 min ±35.21*
p<0.01
129X1/SvJ
N=n6
-126.28 min ±165.59
∆ in NREM sleep
(light period)
∆ in REM sleep
(dark period)
∆in REM sleep
(light period)
∆ Delta Density
(dark period)
∆ Delta Density
(light period)
% Baseline Delta Energy
(dark period)
% Baseline Delta Energy
(light period)
∆ in NREM Sleep Bout
Length (24-hour recovery)
∆ in arousal index
(arousals/hour)
76.88 min ±25.83
25.75 min ±26.71
-43.78 min ±78.85
-2.08 min ±2.31
5.78 min ±7.99
0.85 min ±2.97
21.32 min ±8.53*
p<0.02
9.73 min ±9.32
-5.50 min ±7.11
20.68±17.22
22.00±12.69
92.56±52.23
-9.24±3.85
-37.67±15.68
68.92±49.31
183.43%±28.51*
p<0.05
94.25%±12.02
90.33%±33.36
220.14%±66.91
64.75%±13.51
2.1min±0.3*
p<0.05
-1.5±0.2*
p<0.05
3.0min±.6
215.00%±79.26*
p<0.01
0.5min±.3
-1.7±1.0
1.1±1.2
23.76 min ±25.45
50
-32.73 min ±81.46
50
*
c3h
40
40
b6
30
30
20
20
10
10
c3hxb6
129
0
0
<0.5
0.5-1
1-2
2-4
4-6
6-8
8-10
10-12
>12
11-20
21-40
LengthofAllTotalSleepPeriodsatBaseline
(Minutes)
41-60
61-80
81-100 101-120 121-140 141-160 161-180 181-200 >200
LengthofAllREMSleepPeriodsatBaseline
(Seconds)
50
50
40
40
*
30
30
*
20
20
10
10
0
0
<0.5
0.5-1
1-2
2-4
4-6
6-8
8-10
10-12
11-20
>12
LengthofAllTotalSleepPeriodsAfterSleepDeprivation
(Minutes)
21-40
41-60
61-80
81-100 101-120 121-140 141-160 161-180 181-200 >200
LengthofAllREMSleepPeriodsAfterSleepDeprivation
(Seconds)
Figure 8—Frequency distributions for sleep bout lengths (total sleep and REM sleep bouts) at baseline (upper panel) and after sleep deprivation (lower panels). Bout
lengths are similar in the four groups C3HeB/FeJ (white), C57BL/6J (black), C3HeB/FeJ X C57BL/6J (gray), and 129X1/SvJ (speckled). Total sleep bout lengths
increased following sleep deprivation, particularly for bouts >12 minutes (asterisk, p<0.05).
SLEEP, Vol. 23, No. 8, 2000
11
Sleep Analysis in Mice—Veasey et al
sleep in the active period) relative to other groups (p < 0.05,
see Figure 7).
sleep was expanded to include measures of sleep bout
length, sleep fragmentation, the homeostatic response to
sleep loss and REM sleep latency. In addition, we have presented in detail: (1) a protocol for recording behavioral
states in mice, (2) descriptions of behavioral states and definitions of sleep parameters, and (3) reference baseline
sleep parameters and sleep response patterns after 12 hour
sleep deprivation in four strains of mice commonly used in
genetics experiments.
Response to Sleep Deprivation Across the 12 Hour Light
Period
Twelve hours of total sleep deprivation during the light
period, ending at the onset of the dark period, produced
highly variable recovery responses within strains, as well
as across strains. Sleep deprivation was performed on just
three C3HXB6 mice. These mice, therefore, are not included in the recovery sleep analysis. Comparisons were made
between baseline sleep and the first 24 hours following the
termination of sleep deprivation. C57BL/6J mice demonstrated an increase from baseline in both NREM sleep
(141.70 minutes ± 33.20, p<0.05) and in REM sleep
(21.32±8.53 minutes, p<0.01) during the first dark period
post SD. Delta density, however, was below baseline, and
these mice did not demonstrate an increase in delta energy.
The 129X1/SvJ mice did not increase sleep times post SD,
but significantly increased delta density and therefore, this
group demonstrated an increase in the product of delta density and NREM sleep time: f (p<0.01). The most striking
finding was a persistent increase in delta density for each
hour of the 24-hour recovery period in the 129X1/SvJ
mice, while the other two strains showed a persistent
decline in delta density. C3HeB/FeJ mice showed highly
variable changes in total sleep time in the 24 hours following sleep deprivation with no significant change (see Table
4). Similarly, there were no differences for the group in
NREM sleep time or REM sleep time for either the first
dark or light period. C3HeB/FeJ mice that did not show an
increase in sleep times did have an increase in delta density for NREM sleep. Thus, delta energy in the recovery
period was increased in all C3HeB/FeJ mice for the first
dark period. Afterwards, delta density declined without a
coinciding increase in NREM sleep time, such that delta
energy for the first light period was not different from baseline.
The impact of 12-hour sleep deprivation on REM sleep
latency was evaluated by comparing REM sleep latencies
for each mouse using average baseline values for the dark
period and values obtained in the first dark period following sleep loss. The REM sleep latencies for the 129/SvJ
mice were shortened from 8.95±1.0 minutes to 3.95±1.1
minutes (t=3.9 and p<0.03). In contrast, the C3HeB/FeJ
and C57BL/6J mice showed no differences in REM sleep
after sleep loss.
Accuracy and Validation of the Program
The overall agreement between computerized scoring of
sleep states by the ACQ 3.4 program and by human scorers
was greater than 95% for sleep staging in mice. This degree
of accuracy is comparable to the accuracy reported using
the original Benington and Heller sleep analysis algorithm
in rats,38 and is similar to the only other sleep analysis system validated in mice.40 The accuracy achieved in our
study was enhanced by excluding data from mice in which
behavioral sleep states were not identifiable by human
scorers. However, rejection of electrographic data must be
done with caution, particularly in the analysis of mutant
lines of mice. Aberrant EEG and EMG signals may result
from atypical physiological signals, which would be important to identify in mutant mice. It is necessary, therefore, to
review a sample of raw EEG and EMG signals in all mice.
That sample should include all behavioral states, and it is
equally essential to examine the power spectral analysis for
each mouse. A two-hour sample of the EEG and EMG data
can be reviewed within five minutes to assess for abnormal
waveforms, and spectral analysis can be performed in 10
minutes. Both of these measures should allow detection of
aberrant electrographic waveforms or spikes that may be
very important and otherwise missed if unscorable data
were simply discarded.
We found no difference in the accuracy of the initial
scored section and the validated section in either the light
or dark periods. Therefore, determining the accuracy of
computer scoring for any one-hour mid-light sample with
all sleep states is adequate to measure accuracy and adjust
the algorithm threshold to optimize computerized scoring
in each mouse. In contrast, there was a slight decline in the
accuracy of the program in detecting REM sleep after sleep
deprivation by gentle handling. We believe that in studies
involving sleep deprivation or extensive handling of the
mice, a validation after the handling of mice is warranted to
ensure continued accuracy. Accuracy in determining REM
sleep after sleep deprivation can be improved typically by
adjusting the EMG threshold for REM sleep.
The accuracy of the Benington and Heller algorithm in
determining REM sleep was greater in the C3HeB/FeJ
mice than in other strains. This particular strain demonstrated higher amplitude theta and more theta waveform
activity in REM sleep. Further, this strain makes a rapid
DISCUSSION
We tested the utility of an available sleep scoring program for developing a robust phenotypic assay for sleep
structure in mice. Although the algorithm for scoring
behavioral state was not changed, the characterization of
SLEEP, Vol. 23, No. 8, 2000
12
Sleep Analysis in Mice—Veasey et al
transition from NREM sleep into REM sleep, while the
other strains of mice have a longer NREM to REM sleep
transition (20—40 seconds of slow waves interspersed with
theta). The increased theta synchronization in this strain
may provide a mouse model for studying the hypothesized
REM sleep and theta activity mechanisms involved in hippocampal function.47
changes in both NREM sleep and REM sleep times. There
are two concerns with the 12-hour duration of sleep deprivation. First, after six hours of sleep deprivation in mice, it
is very difficult to maintain wakefulness, and later in the
12- hour period, stimulation is nearly constant and must be
more vigorous. Second, in the present study, we observed
large intra-strain differences in the response to sleep loss
ending at lights-off. This large degree of variance within a
genetically homogeneous group of mice is indeed interesting, but may result from environmental variance.
Alternatively, this large variance may reflect larger magnitudes in the ultradian rhythms of behavioral states seen at
this circadian time.46 Six hours of total sleep loss terminating within the lights-on period in mice results in much
less intra-strain variance.24 Therefore, for genetic studies
of sleep homeostasis, a protocol using six-hours of sleep
loss for the first half of the lights-on period is preferred.
Ideally, the homeostatic response should be characterized
as a dose-response curve, and this will require further
study.
Limitations of This System
The Benington and Heller program was designed to
analyze sleep off-line. Therefore, analysis with this system
requires time to download EEG signal files, perform a
Fourier transform, set thresholds, check accuracy and then
perform the data analysis through an Excel® program.
These steps can be tedious when dealing with records collected over several days for many mice. For this reason,
data analyzed in the present study were typically collected
for an entire week and then retrieved and analyzed. Downloading data requires cessation of recordings for at least
one hour. Additional programming would be required to
allow recording either directly onto a compact disc recorder
or other data archival systems. The on-line signals were
observed during collection to ensure adequate signal quality, prior to off-line analysis. Advantages of an off-line analysis system are that the experimenter may image raw signals on-line at any time. The accuracy may be determined
independently of the algorithm at any time after the data are
down-loaded.
A second limitation of the present program is that the
code is somewhat dated, in that it does not reflect changes
in computer architecture, or in programming languages. We
are currently attempting an update, revising the file handling routines and providing a Windows® interface. This
should reduce the time required for both the retrieval and
analysis steps, permit on-line analysis of ongoing recordings, and make the programs easier to use.
Phenotypic Characterization of Sleep in Genetic Studies of
Sleep in Mice
Over the past three decades, numerous parameters have
been used to characterize sleep in mice. However, in
rodents, these parameters have not been standardized. To
provide an overall description of sleep, we have compiled
a list of sleep parameters, including parameters commonly
used to describe human sleep, such as the intensity of each
behavioral state, the consolidation and the fragmentation of
sleep, and the timing of NREM and REM sleep. We have
provided detailed definitions and descriptions of the measurement for each sleep parameter in the Methods section.
It is hoped that this list will generate discussion and ultimately some consensus concerning the phenotypic characterization of sleep in mice.
In humans, NREM sleep is scored as stages 1, 2 and 3/4,
depending upon spindle and K complex frequency and the
density of delta waves. We were not able to discern discrete spindle complexes and K complexes in mice. We,
therefore, characterized NREM sleep with delta density,
sigma density, bout length, and the arousal index, while
REM sleep is characterized with theta density and bout
length.
Establishing a Recording Protocol
In the present study, we measured baseline sleep over
five to seven days, following a seven to nine day recovery
from surgery and acclimation to the recording cable and
chamber. Stable day-to-day NREM sleep and REM sleep
activity was present by day eight post-operatively in all
mice that exhibited steady-state activity for three or more
days. Steady-state sleep/wake behaviors were not observed
in all mice tested. Therefore, we recommend obtaining a
five-day baseline beginning eight days after surgical
implantation of electrodes.
Measuring the response to sleep deprivation provides
important information on the homeostatic control of sleep.
In this study, we examined the response to 12 hours of total
sleep loss throughout the lights-on period, terminating at
lights-off. This protocol in rats provides greater percentage
SLEEP, Vol. 23, No. 8, 2000
Characterization of Sleep in Mice Using this Sleep Analysis
System
The overall hourly distribution patterns of behavioral
states across the 24-hour light-dark cycle are remarkably
similar to those published.8,10,22-24,26,27,39,40 The C57BL/6J
mice in the present study demonstrated a second sleep-predominant period within the later portion of the dark period,
as previously reported.24 However, there were differences
13
Sleep Analysis in Mice—Veasey et al
between our observations of sleep state percentages for 24
hours and the light period, and those previously reported.
8,10,14,22,39,40 While the percentage of time spent in waking
for the 24-hour cycle and the percentage of time spent in
NREM sleep were within the range of previously reported
values (waking: 35% to 54% and NREM sleep: 35 to 57%
)8,10,22-24, 39,40 the average percentage of time spent in REM
sleep in our study (3.8%) was less than reported values
4.8%,24 5.4%,44 5.6%,8 5.8%,14,40 and 7.6%.10 We, therefore, instrumented and analyzed an additional four mice
and re-validated all C57BL/6J mice in the present study.
Accuracy for REM sleep scoring was >80% and with balanced false positives and negatives. Furthermore, the additional mice showed REM sleep percentages (4.2%) similar
to those obtained in our initial group of mice. We believe
the discrepancy between our percentages for REM sleep
and those from other studies may be attributed to the differences in defining the onset of REM sleep. We required
that theta waves comprise >75% of an epoch of REM sleep,
which may be a more conservative definition. Clearly, a
consensus on how to define REM sleep is needed among
researchers studying sleep in mice.
REM sleep percentage in C3HeB/FeJ mice is similar to
previously reported values,28 while NREM sleep percentages in our animals was lower (34% in our study and 51%
in the previous study). In studies validating sleep state
interpretation in C57BL/6J mice and C3HeB/FeJ mice, the
accuracy in scoring NREM sleep epochs are quite high
(>95%), and the variance is quite low.40 There are many
reasons why NREM sleep percentages may differ, including not only definition of behavioral states, but ambient
temperature and lighting, sex and age of the mice, food and
water access, the duration of the recovery period from electrode implantation, and access to a running wheel or mobility allowed with the implant and cable. These confounding
variables emphasize the need for very specific reference
standards for scoring sleep in mice.
tude of electrical signals after sleep deprivation, is unclear.
Of interest, the differences in baseline sleep parameters
across strains did not parallel the differences in recovery
sleep parameters across strains. This suggests that genetic
influences on homeostatic aspects of sleep control may differ significantly from the baseline genetic influences on
sleep control in laboratory mice.
Strain Differences in Sleep Structure
There were few significant differences in baseline sleep
times and behavioral state percentages between the
C3HeB/FeJ and C57BL/6J substrains and their progeny,
with the exception of an increase in REM sleep time for
C3HeB/FeJ mice in the light period. This suggests that that
C3HeB/FeJ and C57BL/6J may be excellent background
strains for studies of single gene mutations that have to be
analyzed on two backgrounds for the purpose of mapping.
In contrast, the 129X1/SvJ mice demonstrated significantly higher amounts of NREM sleep and REM sleep in the
dark period, thereby increasing sleep time and sleep efficiency, without affecting the NREM sleep:REM sleep ratio.
Large differences in sleep parameters compared to other
strains would pose a serious problem in selecting a background strain for breeding or a strain for comparison purposes. These data suggest that this particular substrain of
129/Sv mice is not well-suited for determining the impact
of an altered gene on sleep.
CONCLUSIONS
In summary, we have evaluated an automated sleep
scoring program for use in mice. Adequate agreement
between the computer and human scoring can be obtained,
provided care is taken in the implantation procedure to
acquire and maintain high quality recording signals. We
have modified the analysis to include parameters of sleep
consolidation, fragmentation, bout lengths and REM sleep
latencies, and these modifications will be made available to
sleep researchers. The automated sleep scoring and comprehensive analysis program and the recording protocol
have been designed to facilitate studies of the genetics of
sleep in mice. This recording and analysis programs, with
our modifications and source code, are available upon
request. We have used this system to describe baseline
sleep patterns in three commonly used inbred strains of
mice. Standardization of electrophysiological parameters
and measurement techniques, on a common set of inbred or
mutant lines, will facilitate the comparison of sleep anomalies identified and/or reported by different laboratories.
Response Patterns to 12 hour Sleep Deprivation
Sleep deprivation revealed patterns of responses unique
to each strain. All C3HeB/FeJ mice showed an increase in
delta energy for the dark period (or the first 12 hours of
recovery). In most C3HeB/FeJ mice, this was attributable
to an increase in NREM sleep time. In two of these mice,
however, this increase in delta energy was due to a large
increase in delta density in NREM sleep. The C57BL/6J
mice all showed increased total sleep, NREM sleep and
REM sleep times, each similar in magnitude to the amount
of increased sleep reported in a recent study of C57BL/6J
mice.24 In the present study, C57BL/6J mice did not show
an overall increase in delta energy. Whether the absence of
a delta energy change in our study occurred because of a
difference in the circadian time of sleep deprivation, the
duration of deprivation, or because of changes in the ampliSLEEP, Vol. 23, No. 8, 2000
ACKNOWLEDGMENTS
We thank Michael Padron, Dr. Michael Brennick and
Todd Michener for invaluable assistance in initiating the
14
Sleep Analysis in Mice—Veasey et al
Shiromani PJ. Effects of sleep on wake-induced c-fos expression. J of
Neurosci 1997;17:9746-50.
20. Chemelli RM, Willie JT, Sinton CM, Elmquist JK, Scammel T, Lee
C, Richardson JA, Williams SC, Xiong Y, Kisanuki Y, Fitch TE,
Nakazato M, Hammer RE, Saper CB, Yanagisawa M. Narcolepsy in
orexin knockout mice: molecular genetics of sleep regulation. Cell
1999;98:437-451.
21. Valatx JL, Bugat R, Jouvet M. Genetic studies of sleep in mice.
Nature 1972;238:226-7.
22. Friedmann JK. A diallel analysis of the genetic underpinnings of
mouse sleep. Physiol & Behav 1974;12:169-75.
23. Franken P, Malafosse A. Tafti M. Genetic variation in EEG activity during sleep in inbred mice. Am J Physiol 1998;275:R1127-37.
24. Franken P, Malafosse A, Tafti M. Genetic determinants of sleep
regulation in inbred mice. Sleep 1999;22:155-169.
25. Eleftheriou BE, Zolovick AJ, Elias MF. Electroencephalographic
changes with age in male mice. Gerontologia 1975;21:21-30.
26. Daszuta A, Gambarelli F. Early postnatal development of EEG and
sleep-waking cycle in two inbred mouse strains. Brain Res 1985;
354:39-47.
27. Roussel B, Turrillot P, Kitahama K. Effect of ambient temperature
on the sleep-waking cycle in two strains of mice. Brain Res 1984;294:
67-73.
28. Gourmelon P, Briet D, Clarencon D, Court L. Tsiang H. Sleep
alterations in experimental street rabies virus infection occur in the
absence of major EEG abnormalities. Brain Res 1991;554:159-65.
29. Toth LA, Rehg JE, Webster RG. Strain differences in sleep and
other pathophysiological sequelae of influenza virus infection in naive
and immunized mice. J of Neuroimmun 1995;58:89-99.
30. Gourmelon P, Briet D. Clarencon D, Court L, Tsiang H.
Electrophysiological and sleep alterations in experimental mouse rabies.
Brain Res 1986;398:128-140.
31. Toth LA. Strain differences in the somnogenic effects of interferon
inducers in mice. Journal of Interferon & Cytokine Research 1996;
16:1065-72.
32. Fang J, Wang Y, Krueger JM. Mice lacking the TNF 55 kDa receptor fail to sleep more after TNF-alpha treatment. J of Neurosci
1997;17:5949-55.
33. Branchey M, Brebbia DR, Kohn M, Litchfield D. Results of studies using an on-line automatic analyzer of sleep stages in the rat.
Electroenceph Clin Neurophysiol 1974; 37:501-506.
34. Johns TG, Piper DC, James GWL, Birtley RD, Fischer M.
Automated analysis of sleep in the rat. Electroenceph Clin Neurophysiol
1977;27:76-83.
35. Gottesmann C, Kirkham PA, lacoste G, Rodrigues L, Arnaud CH.
Automatic analysis of the sleep-waking cycle in the rat recorded by
miniature telemetry. Brain Res 1977;132:562-568.
36. Neuhaus HU, Borbèly A. Sleep telemetry in the rat. II. Automatic
identification and recording of vigilance states. Electroenceph Clin
Neurophysiol 1970;29:358-362.
37. Chouvet G, Odet P, Valatx JL, Pujol JF. An automatic sleep classifier for laboratory rodents. Waking and Sleeping 1980;4:9-31.
38. Benington JH, Kodali SK, Heller HC. Scoring transitions to REMsleep in rats based on the EEG phenomena of pre-REM-sleep: an
improved analysis of sleep structure. Sleep 1994;17:28-36.
39. Tobler I, Deboer T, Fischer M. Sleep and sleep regulation in normal and prion protein-deficient mice. J of Neurosci 1997;17:1869-79.
40. Van Gelder RN, Edgar DM, Dement WC. Real-time automated
sleep scoring: validation of a microcomputer-based system for mice.
Sleep 1991;14:48-55.
41. Sleep Disorders Atlas Task Force. EEG arousals: scoring rules and
examples. A preliminary report from the sleep disorders atlas task force
of the American Sleep Disorders Association. Sleep 1992; 15:174-184.
42. Hengemihle JM, Long JM, Betkey J, Jucker M, Ingram DK. Agerelated psychomotor and spatial learning deficits in 129/SvJ mice.
Neurobiology of Aging 1999; 20:9-18.
sleep recording system, Drs. Jeff Noebels, Wally
Mendelson, Gillian Leach, Ted Abel and Lisa Tarantino for
useful comments on the manuscript. This work was supported by research grant, NIH AR45325; NIH MH57855.
REFERENCES
1. Festing MFW. From character to gene: some strategies for identifying single genes controlling behavioral characters. In: Goldwitz,
Wahlsten, Wimer, eds. Techniques for the genetic analysis of brain and
behavior. Amsterdam: Elsevier, 1992:17-38.
2. Silva AJ, Giese KP. Gene targeting: A novel window into the biology of learning and memory. In: Martinez, Kesner, eds. Neurobiology
of learning and memory. New York: Academic Press, 1998:89-134.
3. “The inheritance of complex traits.” Principles of genetics, Second
Edition, D. P. Snustad and M. J. Simmons, eds. New York: J. Wiley and
Sons, 2000:91-112.
4. Takahashi, JS, Pinto, LH, Vitaterna, MH. Forward and reverse
genetic approaches to behavior in the mouse. Science 1994; 264:172433.
5. Nolan PM, Houpt TA, Bucan M. Chemical mutagenesis and
screening for mouse mutations with an altered rest-activity pattern. In:
Lydic R, ed. Molecular regulation of arousal states. Boca Raton: CRC
Press, 1998:143-156.
6. Oliverio A. Sleep and activity rhythms in mice. Waking and
Sleeping 1980;4:155-166.
7. Daszuta A, Gambarelli F, Ternaux JP. Sleep variations in C57BL
and BALBC Mice from 3 weeks to 14 weeks. Develop Brain Res
1983;7: 87-96.
8. Welsh DK, Richardson GS, and Dement WC. Effect of age on the
circadian pattern of sleep and wakefulness in the mouse. J of Gerontol
1986;4:579-586.
9. Kitahama K, Valatx J-L. Instrumental and pharmacological paradoxical sleep deprivation in mice: strain differences. Neuropharm.
1980;19:529-535.
10. Richardson GS, Moore-Ede MC, Czeisler CA Dement WC.
Circadian rhythms of sleep and wakefulness in mice: analysis using
long-term automated recording of sleep. Am J Physiol 1985;R3:20-30.
11. Toth LA, Rehg JE. Effects of sleep deprivation and other stressors
on the immune and inflammatory responses of influenza-infected mice.
Life Sciences 1986;3:701-9.
12. Homanics GE, Quinlan JJ, Firestone LL. Pharmacologic and
behavioral responses of inbred C57BL/6J and strain 129X1/SvJ/SvJ
mouse lines. Pharm Biochem & Behev 1999;63:21-6.
13. Chen Z, Gardi J, Kushikata T, Fang J, Krueger JM. Nuclear factorkappa B-like activity increases in murine cerebral cortex after sleep
deprivation. Am J Physiol 1999;276:R1812-8.
14. Homanics GE, Harrison NL, Quinlan JJ, Krasowski MD, Rick CE.
de Blas Al, Mehta AK, Kist F, Mihalek RM, Aul JJ, Firestone LL.
Normal electrophysiological and behavioral responses to ethanol in mice
lacking the long splice variant of the gamma2 subunit of the gammaaminobutyrate type A receptor. Neuropharmacology 1999;38:253-65.
15. O’Donnell CP, Schaub CD, Haines AS, Berkowitz DE, Tankersley
CG, Schwartz AR, Smith PL. Leptin prevents respiratory depression in
obesity. Am J of Resp & Crit Care Med 1999;159:1477-84.
16. Boutrel B, Franc B, Hen R, Hamon M, Adrien J. Key role of 5HT1B receptors in the regulation of paradoxical sleep as evidenced in 5HT1B knock-out mice. J of Neurosci 1999;19:3204-12.
17. Krueger JM, Fang J, Taishi P, Chen Z, Kushikata T, Gardi J. Sleep.
A physiologic role for IL-1 beta and TNF-alpha. Annals of the New York
Academy of Sciences 1998;856:148-59.
18. Huntsman MM, Porcello DM. Homanics GE, DeLorey TM,
Huguenard JR. Reciprocal inhibitory connections and network synchrony in the mammalian thalamus. Science 1999;283:541-3.
19. Basheer R, Sherin JE, Saper CB, Morgan JI, McCarley RW,
SLEEP, Vol. 23, No. 8, 2000
15
Sleep Analysis in Mice—Veasey et al
43. Welsh DK, Richardson GS, Dement WC. Effect of running wheel
availability on circadian patterns of sleep and wakefulness in mice.
Physiol Behav 1988;43:771-77.
44. Oliverio A, Malorni W. Wheel running and sleep in two strains of
mice: plasticity and rigidity in the expression of circadian rhythmicity.
Brain Res 1979;163:121-133.
45. Gehrig AE, Warneke-Wittstock R, Sauer CG, Weber BH. Isolation
and characterization of the murine X-linked juvenile retinoschisis
(Rs1h) gene. Mammalian Genome 1999;10:303-7.
46. D’Olimpio F, Renzi P. Ultradian rhythms in young and adult mice:
further support for the basic rest-activity cycle. Physiol & Behav 1998;
64:697-701.
47. Poe GR, Nitz DA, McNaughton BL, Barnes CA, Experiencedependent phase-reversal of hippocampal neuron firing during REM
sleep. Brain Res 2000;855:176-80.
SLEEP, Vol. 23, No. 8, 2000
16
Sleep Analysis in Mice—Veasey et al