Single Neuron Burst Firing in the Human
Hippocampus During Sleep
Richard J. Staba,1Charles L. Wilson,2,4*
Itzhak Fried,3,4and Jerome Engel, Jr.1,2,4
1Department of Neurobiology, UCLA School of Medicine,
Los Angeles, California
2Department of Neurology, UCLA School of Medicine,
Los Angeles, California
3Department of Neurosurgery, UCLA School of Medicine,
Los Angeles, California
4Brain Research Institute, UCLA School of Medicine,
Los Angeles, California
single neuron correlates of sleep-related hippocampal EEG patterns, very
limited hippocampal neuronal data are available for correlation with
human sleep. We recorded human hippocampal single neuron activity in
subjects implanted with depth electrodes required for medical diagnosis
and quantitatively evaluated discharge activity from each neuron during
episodes of wakefulness (Aw), combined stage 3 and 4 slow-wave sleep
(SWS), and rapid eye movement (REM) sleep. The mean firing rate of the
population of single neurons was significantly higher during SWS and Aw
compared with REM sleep (p ? 0.002; p < 0.0001). In addition, burst
firing was significantly greater during SWS compared with Aw (p ? 0.001)
and REM sleep (p < 0.0001). The synchronized state of SWS and associ-
ated high-frequency burst discharge found in human hippocampus may
subserve functions similar to those reported in non-primate hippocampus
that require burst firing to induce synaptic modifications in hippocampal
circuitry and in hippocampal projections to neocortical targets that par-
ticipate in memory consolidation. Hippocampus 2002;12:724–734.
© 2002 Wiley-Liss, Inc.
Although there are numerous non-primate studies of the
slow-wave sleep; REM sleep; firing rate; high-frequency
The marked changes in central neuronal activities that accompany mam-
malian sleep have generated numerous theories on the function of sleep (for
review, see Rechtschaffen, 1998). For example, “dream” sleep has been
proposed as a period during which memory traces are refined by weakening
disruptive synaptic connections among neuronal net-
works (Crick and Mitchison, 1983). The importance of
the hippocampal formation for memory is well estab-
lished on the basis of neuroanatomical and electrophysi-
ological studies (Squire and Zola, 1996; Eichenbaum,
1999). The idea that sleep subserves memory consolida-
tion derives from results showing that hippocampal
“place” cells exhibit experience-dependent replay of fir-
ing patterns during episodes of slow-wave sleep (SWS)
(Wilson and McNaughton, 1994; Kudrimoti et al.,
1999) and rapid eye movement sleep (REM) (Poe et al.,
2000; Louie and Wilson, 2001).
During SWS, the non-primate hippocampal enceph-
alogram (EEG) is dominated by irregular large-ampli-
tude activity with the intermittent appearance of sharp
waves (SPW) and related high-frequency (100–200-Hz)
namic interaction between hippocampal pyramidal cells
by a pronounced rhythmic slow activity (RSA) that re-
flects a shift from irregular synchronized firing to rhyth-
et al., 1995). On the basis of these neurophysiological
high-frequency bursts of hippocampal neuronal activity
neocortical networks (Buzsaki, 1998). REM sleep has
been described as a period during which weak associa-
tions among neocortical networks are strengthened and
representations relayed back to the hippocampus (Stick-
neuronal activity across behavioral states, it is important
to determine what state-dependent changes occur in hu-
Grant sponsor: National Institutes of Health; Grant number: NS-02808 and
*Correspondence to: Charles L. Wilson, UCLA School of Medicine, 2155
Reed Neurological Research Center, 710 Westwood Plaza, Los Angeles,
CA 90095. E-mail: email@example.com
Accepted for publication 27 September 2001
Published online 00 Month 2002 in Wiley InterScience (www.interscience.
HIPPOCAMPUS 12:724–734 (2002)
© 2002 WILEY-LISS, INC.
non-primate hippocampus. For example, does human hippocam-
pus show neuronal burst firing associated with the large amplitude
irregular activity that characterizes non-primate hippocampal ac-
tivity during SWS sleep? Also, does human hippocampal activity
show the prominent theta frequency bursting observed in non-
primates during REM sleep? Perhaps of greater importance is
whether there are patterns of discharge during sleep states that are
suitable for the transfer of hippocampal input or output to and
opportunity to record spontaneous human hippocampal single
neuron activity in subjects with chronically implanted hippocam-
pal depth electrodes required for medical diagnosis. During a pe-
riod of overnight recording, single neuron firing properties were
correlated with states of wakefulness, SWS, and REM sleep. After
these recordings, we quantitatively evaluated hippocampal activity
for high-frequency burst discharge, and on the basis of rates, pat-
terns, and variability of firing.
MATERIALS AND METHODS
Wide-band, high-frequency recordings were obtained from
eight patients with medically intractable complex partial seizures.
Before depth electrode implantation to investigate and localize
areas of seizure onset, patients gave their informed consent for
participation in these studies under the approval of the UCLA
8–14 flexible polyurethane depth electrodes stereotactically tar-
geted to clinically relevant brain areas, which were monitored on a
24-h basis to find those in which spontaneous seizure activity be-
gan first (Fried et al., 1999). Patients in whom a seizure onset area
could be localized became candidates for surgical removal of epi-
leptic sites if resection of the area would not produce an unaccept-
able neurological deficit. Localization was based on the recording
of 3–10 seizures during the average 2 weeks that patients spent in
when applicable, follow-up reports from resective surgery were
used to identify the epileptic area (Engel, 1996).
Electrodes and Localization
High-frequency EEG was recorded from bundles of nine plati-
num–iridium microwires, which were inserted through the lumen
of seven-contact clinical depth electrodes, so that they extended
3–5 ?m beyond the tip of the clinical electrode. Microwires were
tips were localized using the combined information from co-regis-
1.5-T magnetic resonance imaging (MRI) scans, and skull radio-
graphs. The imaging software used (Brain Navigator, Telefactor,
Philadelphia, PA) allowed for visualization and highlighting of
to the MRI scan. Anatomical boundaries were based on references
of hippocampal anatomy by Duvernoy (1998) and Amaral and
Insausti (1990). Only microwires verified to be located in the
hippocampus were used in analyses (Fig. 1).
Overnight Polysomnographic Sleep Studies
Sleep studies were conducted on the hospital ward within each
subject’s room. Studies were conducted 48–72 h after surgery and
typically began between the hours of 10 PM and midnight and
ended at 6 AM the following morning. Patients continued taking
their standard doses of anticonvulsant medications during this pe-
riod. Sleep staging was carried out according to the criteria of
electro-oculogram leads recording eye movements, two electro-
myogram (EMG) leads placed on the chin to record submental
to record cortical EEG activity (Fig. 2). Sleep–wake stages were
categorized as waking, drowsy, stage 1–4 sleep, and REM sleep.
Single neurons recorded during the stages defined as waking (Aw),
stages 3 and 4, hereafter referred to as slow-wave sleep (SWS), and
REM sleep were analyzed for firing rate and bursting activity.
States of drowsiness, stages 1and 2 were not included for analyses
in order to better contrast hippocampal activity during states of
wakefulness and non-REM sleep.
lustrating position of depth electrode. A: Electrode entering from the
lateral aspect of the temporal lobe and the tip of the electrode lying
the contralateral hippocampus. B: Enlarged view from A. Dotted
circles outline the hippocampus. Electrode position is indicated on
scan as signal dropout. The area of signal dropout is much larger com-
to be positioned within the hippocampus were included in the analyses.
Coronal magnetic resonance imaging (MRI) scan il-
HUMAN HIPPOCAMPAL SINGLE NEURON BURST FIRING DURING SLEEP
Continuous high-frequency EEG was recorded wide-band
(0.1–5 kHz) and sampled at 10 kHz with 12-bit precision using
RC Electronics software (Santa Barbara, CA). Data files were cop-
ied onto 1-GB Jaz disks for off-line analysis, and later archived on
CD. Data channels of interest were high-pass filtered at 300 Hz
Extracellularly recorded action potentials were triggered and
discriminated using DataWave Technologies CP Analysis soft-
ware (Longmont, CO). High-frequency, bi- or triphasic action
potential waveforms (“spikes”) with amplitude ?3:1 signal-to-
width ?2.0 ms) discriminators (Fig. 3); 3 ms surrounding the
main negative peak of the triggered spikes were saved in detection
acteristics that could be used in spike sorting. Up to eight param-
eters were extracted from each waveform and used in a spike-
sorting method called “cluster cutting.” Waveform parameters
included amplitude and width of the maximal negative peak
(width measured from peak negativity to the following peak posi-
tivity) and amplitude and duration of peaks or valleys surrounding
parameters were graphically displayed in x,y-point plots such that
spikes with similar waveform parameters would form clusters of
points. Boundaries separating point clusters were visually set in
a putative single neuron. Maximum boundary thresholds on the
eight extracted waveform parameters were set at ?2.5 SD of the
mean value of the point cluster. After replaying the clustered spike
waveforms to visually confirm the accuracy of the cluster bound-
aries, autocorrelograms with a time base of 1,000 ms and a bin-
width of 1 ms were constructed for each single neuron. A spike
in the 0–2-ms bins (refractory period) was considered a single
neuron. A spike train with counts greater than mean firing fre-
quency during the refractory period was considered multiple neu-
rons, and re-clustered. Upon re-clustering, a spike train with an
absence of a clear refractory period was termed multiple neuron
activity and omitted from the analysis. Cross-correlograms with
1-ms bin-widths were constructed for all simultaneously recorded
neurons. Pairs of neurons recorded on different microwires in the
same bundle that demonstrated 0-ms coincident interactions ex-
ceeding 99% confidence (Abeles, 1982) were considered the same
neuron, and one neuron was eliminated.
Single Neuron Analysis
To segregate hippocampal neurons on the basis of pathology,
two major criteria were used: hippocampal atrophy and electro-
graphic seizure onsets. A single neuroradiologist at UCLA evalu-
ated every subject’s MRI scans for the presence or absence of hip-
pocampal atrophy and its location as part of the clinical workup.
Electrographic seizure onsets were recorded during the patient’s
depth electrode telemetry monitoring, and attending neurologists
in the UCLA Seizure Disorders Center determined locations of
seizure onset. Neurons from atrophic hippocampi or from hip-
pocampi where seizure onsets were recorded were omitted. Neu-
rons that were successfully recorded during Aw, SWS, and REM
sleep for ?600 s in each state were included in the analysis.
Discharge activity was characterized for hippocampal neurons
by measuring mean firing rate, median interspike interval (ISI),
total spikes, burst duration, and mean number of spikes in bursts.
Interspike interval variability was measured by calculating the co-
efficient of variation (standard deviation/mean) between adjacent
ISI within each spike train (known as CV2) (Holt et al. 1996).
Thus, for a spike train of n spikes, there are n ? 2 CV2values. The
CV2measure effectively reduces the variation in the neuron firing
that occurs on a time scale longer than the mean ISI. For compar-
(EEG) during Aw, slow-wave sleep (SWS), and rapid eye movement
(REM) sleep. These three 10-s samples were taken from a sleep record
that was acquired during one patient’s overnight polysomnographic
sleep study. Sleep staging was carried out using a sleep record consist-
Example of hippocampal electroencephalogram
ing of recorded eye movements (EOG), muscle tone (electromyogram
[EMG], chin), and scalp EEG (EcoG, C3–A2 derivation). Note the
irregular large amplitude, slow-wave activity in hippocampus (Hip)
and EcoG during SWS, in contrast to the low-amplitude, mixed fre-
quency activity in Hip and EcoG during Aw and REM sleep.
STABA ET AL.
ison across states, we calculated the mean CV2for each spike train.
Burst detection involved serially scanning each spike train to iden-
defined by a series of three or more spikes with an ISI ?20 ms.
rhythmic discharge was established as the appearance of a peak(s)
ing the 99% confidence interval of the mean firing rate.
Discharge variables were analyzed using an analysis of variance
(ANOVA) repeated measures design. Consistent with the require-
ment for normality, variables were transformed with a logarithmic
function such that X? ? log (X). Significant (? ? 0.05) “main ef-
fects,” i.e., state (Aw vs SWS vs REM sleep), were further analyzed
using a Bonferroni post hoc analysis. Comparison of median ISI val-
ues and the number of spikes per burst were made using a one-way
Kruskal–Wallis and post hoc analysis with the Wilcoxon signed-rank
The primary goal of this investigation was to characterize the
firing patterns of hippocampal neurons during Aw, SWS, and
REM sleep. We found significant reductions in mean firing rate
among single hippocampal neurons during REM sleep. During
low-frequency waves, which characterize SWS, we observed an
increase in hippocampal high-frequency burst discharge that was
significantly greater than the bursting activity recorded during ep-
isodes of Aw and REM sleep.
Properties of the Single Neuron Population
were localized in hippocampus using post-implant CT and MRI
(Figs. 1, 3). Nineteen neurons in four patients were excluded on
the basis of localization in atrophic and/or epileptic (seizure-gen-
erating) hippocampi during each state. An additional 34 neurons
recorded in five patients did not meet the minimum sampling
criterion of 600 s recorded during each polysomnographically de-
fined state (Fig. 2). For the remaining 23 single neurons, a total of
the states of Aw, SWS, and REM sleep. For localization of record-
ing electrode tips using the techniques described in the Materials
and Methods section, hippocampi were visualized with 1.0-mm
MRI coronal slices. Starting at the anterior most point and pro-
ceeding posteriorly along the longitudinal axis of the hippocam-
pus, 11 of the 23 neurons were located within the anterior hip-
pocampal region (also known as uncal hippocampus or pes
the hippocampus (at anteroposterior levels identified by the pres-
ence of the lateral geniculate nuclei on the coronal slices), and the
remaining nine neurons were located in the posterior hippocam-
analyzed, six were recorded in hemispheres contralateral to epilep-
tic sites localized to the frontal lobe, 13 neurons were in hemi-
spheres contralateral to epileptic mesial temporal lobe sites, and
four neurons were 3–5 cm distant from an ipsilateral epileptic site
localized in the lateral temporal lobe. MRI-defined hippocampal
atrophy was present in two of the eight patients, and in both
patients, all seven neurons were contralateral to both the atrophic
hippocampus and the electrophysiologically defined epileptic site.
were lying supine in bed, eyes open, and participating in quiet
conversation with one of the investigators. The average epoch an-
alyzed for each neuron during each state was 699 ? 27 s (mean
? SE). The mean number of spikes comprising each spike train
Continuous wide-band electroencephalogram (EEG), digitized at 10
kHz, showing multi-neuron activity (top trace) was high-pass filtered
(middle trace) in order to set window discriminators to trigger neu-
ronal activity. Action potential waveforms (“spikes,” bottom) with
amplitudes >3:1 signal-to-noise (S:N) were triggered and separated
based upon extracted waveform characteristics using a graphical clus-
ter cutting method. In this example, the action potentials within the
two “boxed” sections of the filtered trace would be clustered to rep-
resent the activity from two different neurons. Spike amplitudes were
typically >50 ?V. However, for bursting neurons with spike ampli-
tude attenuation, it was occasionally observed that spikes toward the
end of the burst (i.e., 3rd spike in burst of neuron 2) fell below the
minimum 3:1 SNR threshold and would not be detected. Given that
not all bursts were comprised of spikes with attenuating amplitude,
given state, the loss of these “subthreshold” spikes would be negligible.
Detection and separation of single neuron activity.
HUMAN HIPPOCAMPAL SINGLE NEURON BURST FIRING DURING SLEEP
within one spike train was 34, while the maximum was 6,674
The mean firing rate during Aw, SWS, and REM sleep showed
significant state-related changes in hippocampal single neuron ac-
tivity (F2,66? 11.94, P ? 0.0001). We observed a significant
reduction in hippocampal activity during REM sleep compared
0.0001) (Fig. 4A). Twenty of the 23 hippocampal neurons de-
creased discharge during REM sleep, in comparison with Aw and
SWS rates, while three showed an increase. No significant differ-
ence in mean firing rate was observed between the two electro-
graphically distinct states of Aw and SWS. However, the distribu-
total neurons and, during REM sleep,12 of the 23, had firing rates
?1 spike/s, while during SWS only two neurons had firing rates
Because neuronal firing rates may be influenced by the period
from when it occurred during the sleep process, we divided each
subject’s sleep recording into three time periods: beginning (22:
00–01:00), middle (1:00–4:00), and end (4:00–7:00), and com-
pared the neuronal activity recorded during Aw, SWS, and REM
episodes that occurred during the beginning period of the sleep
recording (n ? 5) was significantly higher compared with firing
rate during Aw episodes that occurred during the middle period of
1.49 vs 1.17 ? 0.19 spikes/s). None of the Aw episodes occurred
during the end period of the sleep recording. Six of the eight
subjects had SWS episodes during the middle of the sleep record-
ing (n ? 21), negating any meaningful comparison between SWS
episodes that occurred at the beginning (n ? 1) or end (n ? 1).
There was no difference in mean firing rate during REM sleep
episodes that occurred during the middle period of the sleep re-
cording (n ? 11) compared with those that occurred during the
end period of the recording (n ? 11; P ? 0.3). One subject had a
ing (n ? 1).
Variability of Firing
Having observed differences in firing rate associated with
sleep–wake states, we sought to characterize the variability of
neuron firing by quantifying the ISI distribution. Visual inspec-
tion of ISI histograms, like those in Figure 5A, typically re-
vealed a positively skewed distribution, thus the median was
used instead of the mean to estimate central tendency. Fluctu-
ations in hippocampal discharge were reflected in ISI histo-
grams as a broad range of ISI values, from a minimum ?2 ms to
ISI values exceeding 1,000 ms. The state-dependent differences
observed in firing rate (Fig. 4A) inversely correlated with the
median ISI values for each state (H2,66? 10.27, P ? 0.005).
Significantly lower firing rates during REM sleep were associ-
ated with significantly longer median ISI values (389 ms) com-
pared with the median ISI values during Aw (199 ms; P ?
0.005) and SWS (146 ms; P ? 0.0001).
Because a variable burst pattern of firing has been implicated in
many studies correlating hippocampal activity and behavior, we
employed joint ISI plots to obtain further information about dif-
ferences in the patterns of firing that were observed across states.
All the neurons we analyzed demonstrated the capacity to dis-
Mean firing rate for all 23 hippocampal neurons during states of Aw,
slow-wave sleep (SWS), and REM sleep (REMS). REM sleep was
associated with a significant reduction in firing rate compared with
Aw and SWS. No difference in firing rate was observed between Aw
and SWS. B: Frequency distribution of firing rates during each state.
Note the number of neurons firing >1 spike/s during SWS compared
REMS. *Aw vs REMS, P ? 0.002; **SWS vs REMS, P < 0.0001.
Firing rate of hippocampal neurons across states. A:
STABA ET AL.
or on the joint ISI plots (Fig. 6A) as the clustering of points near
values (ISIivs ISIi?1) to identify patterns that may exist in neuron
SWS. Such major changes in spike discharge indicate fluctuations
with such long time-based rate changes, the CV2(S.D./mean be-
tween all adjacent ISI values) was calculated and plotted for all 23
neurons combined during each state (Fig. 6B). A uniform distri-
bution of values of 0–2 would be expected if spike discharge fol-
lowed a Poisson process. As can be seen in the plots, the distribu-
the dark “cloud” of points at the top of the plot represents the
increased variability of neuron firing corresponding to short–long
or long–short interval–interval changes in neuron firing. Comput-
ing the mean CV2during each state, results revealed state-specific
differences in the spiking pattern (F2,66? 3.52, P ? 0.03). Dis-
charge variability was highest during SWS with a CV2of 1.14 ?
0.03, compared with Aw, which demonstrated the lowest CV2
value of 1.07 ? 0.02 (P ? 0.01). REM sleep was intermediate to
both Aw and SWS at 1.09 ? 0.03.
Burst Patterns and Spikes Per Burst
The ISI histograms (e.g., Fig. 5A) showed that the largest dif-
the initial 20 ms of the histogram. To better differentiate state-
dependent differences in high-frequency discharge, we set our
burst criteria as a series of three or more consecutive spikes sepa-
events observed in our recordings were often characterized by dec-
rementing amplitude during the successive discharge of spikes
(Fig. 3). The mean duration of a hippocampal burst event was
28.0 ? 0.0003 ms (n ? 2321). No statistical difference was noted
in burst duration across states. However, the mean number of
gram of two different single neurons during awake (Aw), slow-wave
sleep (SWS), and rapid eye movement (REM) sleep. A: ISI histograms
for this single neuron illustrate it was during the initial 20 ms that the
differences in ISI values across state were most pronounced, showing
a tendency for burst firing during SWS and Aw, but no bursting
Interspike interval (ISI) histogram and autocorrelo-
during REMS. B: Autocorrelogram for a different single neuron re-
veals rhythmic discharge with an approximate frequency of 1.5–2 Hz
during SWS. Note that rhythmic firing of higher frequencies at 4–7
REM sleep. ISI histograms and autocorrelograms were constructed
with 1-ms bin width.
HUMAN HIPPOCAMPAL SINGLE NEURON BURST FIRING DURING SLEEP
spikes discharged during a burst event did vary across state. Over-
all, there were fewer spikes discharged per burst during REM sleep
than there were during Aw and SWS (H2,2321? 9.18, P ? 0.01).
During REM sleep, the average number of spikes per burst was
3.6 ? 0.08 spikes, which was significantly fewer in comparison to
the 4.1 ? 0.08 spikes/burst during Aw (P ? 0.005) and the 3.9 ?
0.04 spikes/burst during SWS (P ? 0.003). Higher-order bursts
(?4 spikes/burst) (Suzuki and Smith, 1985b) represented 11.8%
and 11.6% of total bursts detected during Aw and SWS, respec-
tively, and 7.1% of the total bursts during REM sleep. One hy-
pothesis explaining the state differences in spikes per burst is that
faster firing cells discharge more spikes per burst than do slower
firing cells. We addressed this hypothesis by testing whether there
was a correlation between neuron firing rate and mean number of
spikes in bursts during each state. Using Spearman’s rank correla-
tion, we did not detect any significant correlation between a neu-
ron’s firing rate and the number of spikes per burst (r ? ?0.003,
P ? 0.9). Despite differences in the number of spikes per burst
across states, the mean ISI between spikes discharged during a
burst event was similar across states. The mean rate of discharge
within bursts was 162 ? 1.5 spikes/s, ranging from 75 spikes/s to
Bursts were detected in 22 of the 23 hippocampal neurons.
Changes in state were associated with significant changes in burst
rate (number of bursts per minute; F2,66? 18.89, P ? 0.0001).
During SWS, the mean burst rate of hippocampal neurons was
5.1 ? 1.2 (Table 1). During Aw, the burst rate was significantly
lower than SWS (P ? 0.001), while during REM sleep, it was
significantly less during Aw or SWS, respectively (P ? 0.01 and
P ? 0.0001).
Mean burst rate of neurons recorded during Aw episodes occur-
ring during the beginning period of the sleep recording were sig-
nificantly higher compared with the burst rate during Aw episodes
18; t ? 2.19, df ? 18, P ? 0.04; 9.98 ? 3.80 vs 1.45 ? 0.55
bursts/min). There was no difference in mean burst rate during
REM sleep episodes that occurred during the middle period of the
strated rhythmic burst discharge at frequencies of 2–7 Hz. The
autocorrelograms from a single neuron in Figure 5B illustrates
neuron firing during wakefulness, slow-wave sleep (SWS), and rapid
eye movement (REM) sleep. A: The spike train from a single hip-
ISI values as a point along the x- and y-time axes. Pairs of ISI values
plotted for awake (Aw), n ? 2,354; slow-wave sleep, n ? 2,834;
REMS, n ? 910. Note that some of the ISI values, which exceeded
Interval-by-interval fluctuations in hippocampal
1,000 ms, are not illustrated in the plot. B: Variability in neuronal
firing was plotted as the mean of adjacent ISI values versus CV2(refer
to Methods for CV2calculation). Neuron activities from all 23 hip-
pocampal neurons during each state are represented (Aw, n ? 32790;
SWS, n ? 37,498; REMS, n ? 19386). The mean of adjacent ISI
values was plotted on a logarithmic scale.
STABA ET AL.
rhythmic discharge only during SWS. Rhythmic bursting was ob-
served during SWS in two of the four neurons, during Aw in one
neuron, and during both Aw and REM in one neuron.
As measured above, burst propensity is based on burst rate, i.e.,
the number of bursts over time. Although burst rate is the most
common measure reported, to a certain extent, increases in firing
rate might inflate burst rate (Fig. 4). To remove differences found
in burst firing due to changes in firing rate, we calculated a second
on the number of bursts per 500 spikes. Although similar state-
dependent modulation for the burst ratio was observed as seen for
burst rate, this measure did unmask bursting during REM sleep
that was not detected with the first measure. Seventeen of the 22
ing SWS than during Aw and REM (Fig. 7A), and five neurons
demonstrated variable bursting patterns (Fig. 7B). The burst ratio
for all 23 neurons was 18.2 ? 2.7 during SWS (Table 1). This
rate, there was no difference in the burst ratio between Aw and
REM sleep (P ? 0.1).
A third technique by which we quantified bursting of hip-
pocampal neurons was calculation of the percentage of spikes
found within bursts out of the total number of spikes in the spike
train, i.e., the extent to which the burst firing contributes to the
total amount of neuron firing. State significantly affected the pro-
portion of spikes found within bursts (F2,66? 12.78, P ? 0.0001;
Table 1). An average of 14.1 ? 2.3% of spikes discharged by
hippocampal neurons were found within bursts during SWS. This
was a significantly higher proportion than that calculated for Aw
(P ? 0.001) and REM sleep (P ? 0.0001). Statistically, Aw and
REM sleep failed to demonstrate any difference in the ratio of
spikes within bursts to total spikes discharged (P ? 0.2). No dif-
of spikes found within bursts across sleep–wake episodes that oc-
curred during any of the three periods of sleep recording.
The results of this study provide evidence for several significant
alterations in firing properties of human hippocampal single neu-
rons associated with different states of consciousness. Overall, the
mean firing rate was lowest during REM sleep compared with Aw
and SWS, but no significant difference was observed between hip-
pocampal neurons during Aw and SWS. More specific analysis of
firing patterns within the spike trains revealed significantly higher
ISI variability during SWS compared with Aw; it also showed that
the state of SWS was associated with a significant increase in high-
frequency burst firing compared with Aw and REM sleep.
Physiological Identification of Hippocampal Cell
Since the recordings made by Ranck (1973) describing com-
plex-spike cells and theta cells, three criteria have commonly been
used to discriminate putative pyramidal cells from interneurons:
spike duration, firing rate, and burst propensity (Skaggs and Mc-
firing rates ?10 Hz and high-frequency spike bursts with attenu-
ating amplitude have been associated with pyramidal cells. The
hippocampal neurons we recorded discharged an average of 2
spikes/s and 96% demonstrated decrementing amplitude, com-
recordings, the sampling rate was insufficient to reduce variable
spike waveform alignments, limiting our ability to measure aver-
rons were reliably isolated single neurons using 40-?M-diameter
microwires suggests that the microwire tips were recording from
measured by burst ratio. A: Seventeen of the 22 bursting neurons
demonstrated enhanced burst firing during the period of slow-wave
sleep (SWS) compared with awake (Aw) and rapid eye movement
(REMS). B: SWS episodes were associated with an activation of burst
firing for most hippocampal neurons; five neurons had variable burst
patterns across sleep–waking states.
State-related burst patterns of individual neurons as
Measures of Hippocampal Bursting*
3.3 ? 1.1**
12.4 ? 2.8
9.8 ? 2.2%
5.1 ? 1.2†,‡
18.2 ? 2.7†,‡
14.1 ? 2.3%†,‡
1.0 ? 0.5
8.4 ? 2.8
6.3 ? 2.1%
Aw, awake; SWS, slow-wave sleep; REMS, rapid eye movement sleep.
*All three burst measures show significantly higher burst firing during
SWS compared with both Aw and REM sleep. Burst rate is the number
of burst per minute. Burst ratio is the number of bursts per 500 spikes.
Values are mean ? S.E.
**Aw vs. REMS, P ? 0.005
†SWS vs REMS, P ? 0.0001.
‡SWS vs Aw, P ? 0.01.
HUMAN HIPPOCAMPAL SINGLE NEURON BURST FIRING DURING SLEEP
relatively sparsely packed large neurons rather than the densely
packed small neurons of the granule layer of the fascia dentata.
These electrophysiological data combined with the characteristics
of our neurons were pyramidal cells.
Studies in the rodent have found an increased probability of
episodes of SWS and waking immobility (Buzsaki, 1986). How-
ever, pyramidal cell discharges decreased during the periods when
rodents show robust RSA associated with REM sleep and ambula-
tory behavior (Suzuki and Smith, 1987; Chrobak and Buzsaki,
1996). Similar sleep-related changes in firing rate were found in
the other in human recordings of multi-unit hippocampal activity
(Ravagnati et al., 1979). In contrast, human functional neuroim-
aging studies during sleep have revealed a deactivation of the hip-
pocampal formation during SWS (Maquet et al., 1997) and acti-
vation during REM sleep (Nofzinger et al., 1997). While imaging
results provide evidence for a net increase in hippocampal activity
during REM sleep, the electrophysiological evidence in non-pri-
mate point to specific cell types, and human studies point to hip-
pocampal locations, that preferentially decrease discharge activity
during REM sleep.
consistent with some of the same studies cited above regarding firing
hippocampal single neuron burst activity (Colder et al., 1996). Burst
structure analysis revealed similarities in burst duration and mean ISI
between spikes in bursts, which suggest that mechanisms involved in
burst generation were stable across all behavioral states. However, we
higher-order bursts might enhance the transfer of hippocampal out-
put (Huerta and Lisman, 1995).
with those that occurred during the middle period of the recording.
The reduction in firing rate and burst rate from Aw episodes that
occurred during middle period of the sleep recording may reflect the
level of monoaminergic and cholinergic activation from preceding
sleep episodes imposed on the waking state. No difference was ob-
episodes that occurred during the middle period compared with the
SWS because six of eight subjects experienced SWS episodes during
the middle period of the sleep process.
Propensity for burst firing recorded from our hippocampal sin-
allow us to quantify the hippocampal EEG for the presence of
SPWs, the increased propensity for hippocampal neuron burst
firing during SWS reflects a synchronized state (Fig. 2) in which
the probability of hippocampal SPW occurrence is greatest. Con-
versely, the reduction in burst firing associated with Aw and REM
sleep reflects states in which the hippocampal EEG is low ampli-
tude, mixed frequency and the probability for SPW occurrence is
reduced. High-frequency burst discharge has been shown to in-
crease the probability of signal transmission between neurons, as
well as induce long-term synaptic modifications (for review, see
Lisman 1997). In accordance with Buzsaki’s theory (1998), the
SWS would increase the probability of hippocampal output influ-
encing neocortical networks.
response to repetitive stimuli, as well as in spontaneous activity
from hippocampus, thalamus, and midbrain nuclei during sleep
(Yamamoto et al, 1986; Kodama et al., 1989; Britten et al. 1993;
revealed a state-dependent increase in the variability of spike dis-
increased variability is probably an increase in burst firing. In spite
of similarities in mean firing rate between Aw and SWS, CV2plots
(Fig. 6B) identified high variability at mean ISI values of approxi-
mately 1,000 ms in duration. This variance reflects long intervals
of neuronal silence after periods of high-frequency burst firing.
the significantly longer median ISI values compared with Aw and
SWS, contrary to a previous finding of short ISI hippocampal
al., 1969). Discharge variability during REM sleep in human hip-
upon a reduced background of neuronal activity. This pattern of
firing is in contrast to the highly rhythmic bursting that accompa-
nies the theta activity dominating REM sleep in non-primate hip-
during SWS (one neuron illustrated in Fig. 5B) and one neuron
our sample, may not fully represent the rhythmicity of neuronal
activity in the human hippocampus. Other recordings from hu-
man hippocampus and entorhinal cortex in this laboratory (un-
published data) have shown ample evidence of rhythmic bursting
during all three states.
The application of our findings to the current understanding of
normal hippocampal activity during sleep was limited to the anal-
ysis of 23 hippocampal neurons, so that we could base our conclu-
sions only on single neurons showing clear isolation, and from
neurons recorded in sites free from any electrographic or anatom-
ical evidence of epileptic potential; i.e., no seizure onsets were
recorded from these electrodes, and no hippocampal atrophy was
detected in these recording sites on the basis of neuroradiologic
STABA ET AL.
analysis of MRI. While we cannot discount the presence of inter-
ictal epileptiform EEG activity in our recordings, studies have
shown interictal epileptiform discharges are widespread during
SWS (Sammaritano et al., 1991; Malow et al., 1998), and that
areas demonstrating epileptic potential positively correlate with
human studies have failed to demonstrate a clear relationship be-
tween interictal EEG spikes and neuronal burst discharge (Wyler
et al., 1981; Telfeian et al., 1999). Although there is evidence that
epilepsy can influence sleep architecture (Shouse and Sterman,
1983; Gigli and Gotman, 1992; Bazil et al., 2000), characteriza-
tion of neuronal activity was based on stable, 10-min episodes of
wakefulness, SWS, and REM sleep that were clearly defined by
standard polysomnographic criteria. Nevertheless, evaluation of
these data must take into consideration the possible influences of
epileptic activity and anticonvulsant medications to which these
hippocampal neurons were exposed.
Our results demonstrate state-related changes in the frequency
and pattern of hippocampal neuron firing. The timing of such
neuronal discharge in relation to the network EEG rhythm may
determine whether neuronal synaptic transmission is enhanced or
have demonstrated that cortical and thalamic neurons discharge
rhythmically in the low-frequency range (0.5–4 Hz) during states
characterized by high-amplitude, low-frequency EEG, i.e. SWS
(Steriade et al., 1996; Contreras and Steriade, 1997; Weyland et
al., 2001). The SWS-related interaction between cortical and
thalamic neurons is effective in increasing their spatiotemporal
coherence over widespread brain areas (Steriade, 1999). As a
firing during SWS may provide the critical increase in hippocam-
pal output that is required for potentiation of neocortical targets.
may correspond to SPW-like episodes optimal for hippocampal
output to influence neocortical targets, while the reduction in fir-
ing during REM sleep may signal a shift in hippocampal state
favorable to the reception of neocortical feedback. Further studies
neuronal burst firing are needed to distinguish the comparative
roles of SWS and REM sleep in memory processing.
The authors thank A. Bragin, J. Zeitzer, and F. Lopez-Rodri-
guez for their helpful discussions, and T. Fields and E. Behnke for
excellent technical assistance.
Abeles M. 1982. Quantification, smoothing, and confidence limits for
single-units’ histograms. J Neurosci Methods 5:317–325.
Amaral D, Insausti R. 1990. Hippocampal Formation. In: Paxinos G,
Bazil CW, Castro LH, Walczak TS. 2000. Reduction of rapid eye move-
Arch Neurol 57:363–368.
of neurons in macaque MT to stochastic motion signals. Vis Neurosci
Brain Res 398:242–252.
Buzsaki G. 1998. Memory consolidation during sleep: a neurophysiolog-
ical perspective. J Sleep Res 7:17–23.
Buzsaki G, Leung LWS, Vanderwolf CH. 1983. Cellular basis of hip-
pocampal EEG in the behaving rat. Brain Res Rev 6:139–171.
Cobb SR, Buhl EH, Halasy K, Paulsen O, Somogyi P. 1995. Synchroni-
zation of neuronal activity in hippocampus by individual GABAergic
interneurons. Nature 378:75–78.
Colder BW, Wilson CL, Frysinger RC, Harper RM, Engel J. 1996. In-
terspike intervals during interictal periods in human temporal lobe
epilepsy. Brain Res 719:96–103.
Contreras D, Steriade M. 1997. State-dependent fluctuations of low-
Crick F, Mitchison G. 1983. The function of dream sleep. Nature 304:
coupling of hippocampal pyramidal cells and interneurons in the be-
having rat. J Neurosci 19:274–287.
Debanne D, Gahwiler BH, Thompson SM. 1996. Cooperative interac-
tions in the induction of long-term potentiation and depression of
Proc Natl Acad Sci U S A 93:11225–11230.
Draguhn A, Traub RD, Bibbig A, Schmitz D. 2000. Ripple (?200-Hz)
oscillations in temporal structures. J Clin Neurophysiol 17:361–376.
Duvernoy HM. 1998. The human hippocampus. New York: Springer.
Eichenbaum H. 1999. The hippocampus and mechanisms of declarative
memory. Behav Brain Res 103:123–133.
Engel J Jr. 1996. Current concepts: surgery for seizures. N Engl J Med
Fried I, Wilson CL, Maidment NT, Engel J Jr, Behnke E, Fields TA,
MacDonald KA, Morrow JW, Ackerson L. 1999. Cerebral microdi-
alysis combined with single-neuron and electroencephalographic re-
cording in neurosurgical patients. Technical note. J Neurosurg 91:
Gigli GL, Gotman J. 1992. Effects of seizures, kindling, and carbamaz-
epine on sleep organization in cats. Epilepsia 33:14–22.
Holt GR, Softky WR, Koch C, Douglas RJ. 1996. Comparison of dis-
charge variability in vitro and in vivo in cat visual cortex neurons.
J Neurosci 75:1806–1814.
Huerta PT, Lisman JE. 1995. Bidirectional synaptic plasticity induced by
a single burst during cholinergic theta oscillation in CA1 in vitro.
Kodama T, Mushiake H, Shima K, Nakahama H, Yamamoto M. 1989.
neurons in cats. I. Relation to natural sleep and alert states. Brain Res
Kudrimoti HS, Barnes CA, McNaughton BL. 1999. Reactivation of hip-
pocampal cell assemblies: effects of behavioral state, experience, and
EEG dynamics. J Neurosci 19:4090–4101.
Quantitative analysis of depth spiking in relation to seizure foci in
HUMAN HIPPOCAMPAL SINGLE NEURON BURST FIRING DURING SLEEP
patients with temporal lobe epilepsy. Electroencephalogr Clin Neuro- Download full-text
Lisman JE. 1997. Bursts as a unit of neural information: making unreli-
able synapses reliable. Trends Neurosci 20:38–43.
Louie K, Wilson MA. 2001. Temporally structured replay of awake hip-
Malow BA, Lin X, Kushwaha R, Aldrich MS. 1998. Interictal spiking
increases with sleep depth in temporal lobe epilepsy. Epilepsia 39:
Maquet P, Degueldre C, Delfiore G, Aerts J, Peters JM, Luxen A, Franck
G. 1997. Functional neuroanatomy of human slow wave sleep. J Neu-
Noda H, Manohar S, Adey WR. 1969. Spontaneous activity of cat hip-
pocampal neurons in sleep and wakefulness. Exp Neurol 24:217–231.
Nofzinger EA, Mintum MA, Wiseman M, Kupfer DJ, Moore RY. 1997.
Forebrain activation in REM sleep: an FDG PET study. Brain Res
Otto T, Eichenbaum H, Wiener SI, Wible CG. 1991. Learning-related
patterns of CA1 spike trains parallel stimulation parameters optimal
for inducing hippocampal long-term potentiation. Hippocampus
Poe GR, Nitz DA, McNaughton BL, Barnes CA. 2000. Experience-de-
pendent phase-reversal of hippocampal neuron firing during REM
sleep. Brain Res 855:176–180.
Ranck JB. 1973. Studies on single neurons in dorsal hippocampus forma-
repertoires. Exp Neurol 41:462–531.
hippocampal formation and amygdala neurons during sleep. Sleep
Rechtschaffen A. 1998. Current perspectives on the function of sleep.
Perspect Biol Med 41:359–390.
Rechtschaffen A, Kales A. 1968. A manual of standardized terminology,
techniques and scoring system for sleep stages of human subjects.
Bethesda, MD: Neurological Information Network. 57 p.
Sammaritano M, Gigli GL, Gotman J. 1991. Interictal spiking during
wakefulness and sleep and the localization of foci in temporal lobe
epilepsy. Neurology 41:290–297.
pathology predicts kindled seizure susceptibility in cats. Brain Res
Skaggs WE, McNaughton BL. 1996. Replay of neuronal firing sequences
in rat hippocampus during sleep following spatial experience. Science
Softky WR, Koch C. 1993. The highly irregular firing of cortical cells is
inconsistent with temporal integration of random EPSPs. J Neurosci
Squire LR, Zola SM. 1996. Structure and function of declarative and
Steriade M. 1999. Coherent oscillations and short-term plasticity in cor-
ticothalamic networks. Trends Neurosci 22:337–345.
fast (30–40Hz) spontaneous oscillations in intrathalamic and
thalamocortical networks. J Neurosci 16:2788–2808.
Stickgold R. 1998. Sleep: off-line memory reprocessing. Trends Cogn Sci
Suzuki SS, Smith GK. 1985a. Single-cell activity and synchronous burst-
ing in the rat hippocampus during waking behavior and sleep. Exp
Suzuki SS, Smith GK. 1985b. Burst characteristics of hippocampal com-
plex spike cells in the awake rat. Exp Neurol 89:90–95.
Suzuki SS, Smith GK. 1987. Spontaneous EEG spikes in the normal
hippocampus. I. Behavioral correlates, laminar profiles and bilateral
synchrony. Electroencephalogr Clin Neurophysiol 67:348–359.
Telfeian AE, Spencer DD, Williamson A. 1999. Lack of correlation be-
tween neuronal hyperexcitability and electrocorticograhpic respon-
siveness in epileptogenic human neocortex. J Neurosurg 90:939–945.
Weyand T, Boudreaux M, Guido W. 2001. Burst and tonic response
modes in thalamic neurons during sleep and wakefulness. J Neuro-
Wilson MA, McNaughton BL. 1994. Reactivation of hippocampal en-
semble memories during sleep. Science 265:676–679.
Wyler AR, Ojemann GA, Ward AA. 1982. Neurons in human epileptic
cortex: correlation between unit and EEG activity. Ann Neurol 11:
Yamamoto M, Nakahama H, Shima K, Kodama T, Mushiake H. 1986.
Markov-dependency and spectral analyses on spike-count in mesence-
phalic reticular neurons during sleep and attentive states. Brain Res
Sharp wave-associated high-frequency oscillation (200Hz) in the in-
tact hippocampus: network and intracellular mechanisms. J Neurosci
STABA ET AL.