dissociated wake-like and sleep-like electro-cortical activity during sleep
TRANSCRIPT
NeuroImage 58 (2011) 612–619
Contents lists available at ScienceDirect
NeuroImage
j ourna l homepage: www.e lsev ie r.com/ locate /yn img
Dissociated wake-like and sleep-like electro-cortical activity during sleep
Lino Nobili a,b,c,⁎, Michele Ferrara d, Fabio Moroni e,f, Luigi De Gennaro e, Giorgio Lo Russo a,Claudio Campus c, Francesco Cardinale a, Fabrizio De Carli c
a Centre of Epilepsy Surgery “C. Munari”, Niguarda Hospital, Milan, Italyb Center of Sleep Medicine, Niguarda Hospital, Milan, Italyc Institute of Bioimaging and Molecular Physiology, Genoa Unit, National Research Council, Genoa, Italyd Department of Health Sciences, University of L'Aquila, Italye Department of Psychology, University of Rome “Sapienza”, Roma, Italyf Department of Psychology, University of Bologna, Italy
⁎ Corresponding author at: Centre of Epilepsy SurgerMedicine, Niguarda Hospital, Piazza Ospedale Maggiore
E-mail address: [email protected] (L. N
1053-8119/$ – see front matter © 2011 Elsevier Inc. Aldoi:10.1016/j.neuroimage.2011.06.032
a b s t r a c t
a r t i c l e i n f oArticle history:Received 5 April 2011Revised 11 June 2011Accepted 13 June 2011Available online 21 June 2011
Keywords:Motor cortexLocal sleepPrefrontal cortexActivationsArousalSleep regulation
Sleep is traditionally considered a global process involving the whole brain. However, recent studies haveshown that sleep depth is not evenly distributed within the brain. Sleep disorders, such as sleepwalking, alsosuggest that EEG features of sleep and wakefulness might be simultaneously present in different cerebralregions. In order to probe the coexistence of dissociated (wake-like and sleep-like) electrophysiologicalbehaviors within the sleeping brain, we analyzed intracerebral electroencephalographic activity drawn fromsleep recordings of five patients with pharmacoresistant focal epilepsy without sleep disturbances, whounderwent pre-surgical intracerebral electroencephalographic investigation. We applied spectral andwavelet transform analysis techniques to electroencephalographic data recorded from scalp and intracerebralelectrodes localized within the Motor cortex (Mc) and the dorso-lateral Prefrontal cortex (dlPFc). The Mcshowed frequent Local Activations (lasting from 5 to more than 60 s) characterized by an abrupt interruptionof the sleep electroencephalographic slow waves pattern and by the appearance of a wake-likeelectroencephalographic high frequency pattern (alpha and/or beta rhythm). Local activations in the Mcwere paralleled by a deepening of sleep in other regions, as expressed by the concomitant increase of slowwaves in the dlPFc and scalp electroencephalographic recordings. These results suggest that human sleep canbe characterized by the coexistence of wake-like and sleep-like electroencephalographic patterns in differentcortical areas, supporting the hypothesis that unusual phenomena, such as NREM parasomnias, could resultfrom an imbalance of these two states.
y “C. Munari”, Center of Sleep, 3, 20162, Milan, Italy.obili).
l rights reserved.
© 2011 Elsevier Inc. All rights reserved.
Introduction
Sleep has been traditionally defined in terms of whole-animalbehavioral state on the basis of the concept that it is imposed to thewhole brain by specialized sleep networks. However, this dominant“top-down” paradigm has been recently challenged by a “bottom-up”approach according to which sleep is a fundamental property of localneuronal networks in different brain structures (Krueger et al., 2008).In this view, sleep is orchestrated, but not fundamentally driven, bycentral mechanisms (Rector et al., 2005). Global (behaviorally andelectroencephalographically defined) sleep emerges when a largenumber of neuronal groups are in the altered input-output state thatcharacterizes the sleep-like state at the local network level (Kruegeret al., 2008).
Several evidences indeed support the idea that sleep intensity isnot a spatially global and uniform state. Topographical EEG studies innormal subjects shows that during sleep there are large regionalfrequency-specific EEG differences (Finelli et al., 2001; Ferrara et al.,2002; Marzano et al., 2010). These differences are stable, as the frontalpredominance of slow wave activity (SWA), the main indicator ofsleep depth.
From a temporal point of view, there is also evidence that the sleepprocess is not necessarily present simultaneously in the entire brain.The coexistence of wake-like and sleep-like EEG patterns has beenlong recognized in birds and aquatic mammals, such as in dolphins(Mukhametov, 1984; Lyamin et al., 2008). However, strong support tothe temporal acceptation of the local sleep theory comes also fromquantitative EEG studies in humans. Indeed, in normal subjects it hasbeen shown that different brain areas can fall asleep with a differenttiming (De Gennaro et al., 2001). Clinical evidence on patients withparasomnias (status dissociatus), e.g. sleepwalking, suggests thatthese individuals are awake (evidenced by their ability to negotiatearound objects), and asleep (indicated by their lack of awareness of
613L. Nobili et al. / NeuroImage 58 (2011) 612–619
their actions) simultaneously (Mahowald and Schenck, 2005).Therefore, according to these observations, vigilance states seem notnecessarily temporally discrete states.
To date, in humans the observation of the coexistence of dis-sociated sleep-like and wake-like EEG patterns in well-defined andrestricted cortical areas has been provided only in patients withparasomnia (Terzaghi et al., 2009). Therefore, in order to investigatewith a higher spatial resolution local (cortical) EEG dynamics duringsleep, here we analyzed intracerebral stereo-EEG data drawn fromovernight sleep recordings of five epileptic patients without sleepdisturbances, who underwent intracerebral EEG acquisition duringpreoperative investigation. Stereo-EEG recordings allow in part toovercome the low spatial resolution of the human sleep studies withsurface EEG which, even with high-density arrays, necessarily recordthe activity of relatively large groups of neurons. Our analysis focusedon NREM sleep, since we were interested in dissociated statesshowing EEG activation on one region accompanied by clear sleepEEG patterns (e.g., delta waves) on the other derivations.
We found that during NREM sleep some parts of the cortex can beelectrophysiologically fully activated while others are stronglydeactivated. Indeed, we observed local activations (LAs) occurringin the Motor cortex (Mc) that were accompanied by a deepening ofsleep in other cortical regions, as expressed by the simultaneousincrease of delta waves measured in the dorso-lateral Prefrontalcortex (dlPFc) and at the surface of the scalp.
Materials and methods
Patients and data recording
Intracerebral electroencephalographic (EEG) data were recordedfrom five patients (3 males, 2 females; mean age 26.8±5.9) withpharmacoresistant focal epilepsy. None reported any clinicallyevaluated sleep disorder.
Patients underwent an intracerebral EEG investigation (Stereo-EEG,SEEG) in order to accurately localize the epileptogenic zone beforesurgical treatment. EEG activity was recorded from platinum–iridium,semiflexible intracerebral electrodes with a diameter of 0.8 mm, 5–18contacts 2 mm in length and 1.5 mm intercontact distance. Position andnumber of electrodes varied according to individual condition: patientsincluded in this study shared the presence of contacts pairs unequiv-ocally localized within the Mc and the dlPFc, as confirmed by post-implantation magnetic resonance imaging (see Fig. 1 and Table 1), andintracerebral electrical stimulation (Cossu et al., 2005). Bipolar EEG fromboth regions showed no EEG abnormalities. Scalp EEG activity wasrecorded from two platinum needle electrodes placed during surgery at10–20 positions Fz and Cz. Electroocular activity was registered at theouter canthi of both eyes, and submental electromiographic activitywasacquired with electrodes attached to the chin. All signals were recordedusing a polygraphic recording system (XLTEK, Trex™) with a samplingrate of 512 Hz. Before stereotactic electrode implantation, patients gavewritten informed consent for participation in research studies and forpublication of data. Sleep was recorded during the fifth night afterelectrode implantation and no seizures were observed during the night.The EEG channels here analyzed were free of inter-ictal epilepticdischarges and were not involved in the ictal discharge during seizures.During the SEEG recording period, patients continued taking theirstandard doses of anticonvulsant medications (for details, see Table 1).
Acquisition files were transformed to EDF (European Data Format)to be handled by means of a MatLab software (MatLab 7.0, TheMatworks Inc., Natick, MA, USA).We used a bipolar montage betweenadjacent electrode contacts and between Fz–Cz scalp electrodes, EOGand EMG derivations. EEG and SEEG channels were 0.1–40 Hz bandpass filtered, EOG channel was 0.16–15 Hz band pass filtered and EMGchannel was 5–150 Hz band pass filtered. The same custom softwareallowed us to score the sleep stages according to the standard criteria
(Rechtschaffen and Kales, 1968). EEG artifacts were visually selectedand removed from the analysis. The overall rejection rate was 9.1%(SE: 2.5%).
Local activation scoring
We defined the transient electroencephalographic activationsappearing only on one of the three considered derivations as LocalActivations (LAs).
For LAs scoring we considered the standard definition of EEGarousals (American Sleep Disorders Association, 1992) as transientphenomena disrupting sleep, characterized by an abrupt increase inEEG frequency, which may include theta, alpha and/or beta rhythm,andwemanually scored them independently in each derivation. Sincewe were interested in dissociated states during which SEEG and scalpEEG channels could show opposite (wake-like and sleep-like)electrophysiological behaviors, no maximum duration criterion wasconsidered. A local Activation Index (AI) was then calculated as thenumber of events per hour of NREM sleep.
Finally, for each cortical LA we visually scored the presence andduration of concomitant EMG activations. An activation was consid-ered as an increase of muscle tone of at least twice the EMG level inthe 20 s epoch preceding the LA.
Spectral analysis
The time course of SWA across the night was assessed by spectralanalysis of EEG signal: the Fast Fourier Transform (Welch method)was applied to 2-second overlapping epochs preprocessed by a Tukeywindow. A mean spectrumwas estimated for 1-minute intervals afterartifact rejection. SWA was evaluated for each minute as EEG powerwithin the delta band (0.5–4 Hz). SWA power was normalized withineach recording to the overnight geometric mean of the total power.The first three NREM–REM cycles of each sleep recording were thendivided into an equal number of intervals (20 steps for each NREMphase and 5 steps for each REM) (Moroni et al., 2007): this enabledbetween subject averaging in order to estimate a mean SWA profilefor each sleep cycle and brain region. In order to compare SWAbetween regions and cycles, the normalized delta power wasaveraged for each recording, region and NREM cycle.
Time–frequency analysis
In order to evaluate EEG patterns characterizing the different deriva-tions in association with Mc LAs, we analyzed the time–frequencydistribution of EEG signals within a 40 s time window centered on theMc LA onset. Time–frequency analysis of bipolar EEG signals wasperformed by means of the discrete wavelet transform (Jobert et al.,1994; De Carli et al., 1999), particularly suitable for the analysis of non-stationary signals in the time–frequency domain.
The discrete wavelet transform was used in this study by means ofthe recursive application of a pair of half-bandmirror filters, generatingwavelet-coefficient time series for each band in a multi-resolutionscheme. High frequencieswere estimatedwith high time resolution andlarge bandwidth, while time resolution decreased and frequencyresolution increased by halving the bandwidth at each step towardlower frequencies (Rioul and Vetterli, 1991). This process produced asequence of frequency bands with constant relative bandwidth(bandwidth/central frequency) well suited for EEG analysis. The 8–16 Hz bandwas further split into the conventional alpha (8–12 Hz) andsigma (12–16 Hz) EEG bands. Based on this decomposition, signalpowerwas computed and arranged in a grid as function of time (0.125 sresolution) for the following frequencybands: 0.5–1 Hz, 1–2 Hz, 2–4 Hz,4–8 Hz, 8–12 Hz, 12–16 Hz, 16–32 Hz. In order to get a better frequencydiscrimination, the wavelet filters were drawn from the application ofthe Remez exchange algorithm for orthonormal wavelets proposed by
Fig. 1. Example of a Magnetic Resonance Imaging (MRI) scan showing intracerebral electrodes implanted into the motor cortex (a) and the dorso-lateral prefrontal cortex (b) in asingle subject. From the left: axial, coronal and sagittal views. White circles indicate the location of the two electrode contacts on which sleep EEG analysis was performed.Superimposition of the ten bipolar derivations (yellow spheres) of the 5 subjects on the Montreal Neurological Institute (MNI) brain template (c: mesial projection; d: lateralprojection).
614 L. Nobili et al. / NeuroImage 58 (2011) 612–619
Rioul and Duhamel (1994) for the optimization of frequency selectivity(32 coefficients). Grids of power data were then averaged for eachrecording and normalized to the geometric mean of the signal powerwithin the 40-second time window.
Statistics
In order to evaluate the dynamics of SWA across the night, Analysisof Variance (ANOVA) with the factors Brain Region (Mc, dlPFc, Scalp)and Cycle (1st, 2nd, 3rd) was carried out on normalized EEG power inthe delta band (0.5–4.0 Hz).
Local Activation Index (AI) has been compared across Brain Region(Mc, dlPFc, Scalp) by means of a one-way ANOVA.
To analyze the time course of Mc LA across subsequent sleepperiods, each sleep cycle was divided into four time intervals. Thevalues of Mc LA index (events/hours) were then submitted to anANOVA with the factors Cycle (1st, 2nd, 3rd) and Segment (1st, 2nd,3rd, 4th).
Finally, a repeated measure ANOVA with the factors Brain Region(Mc, dlPFc, Scalp), Frequency Band (0.5–1 Hz, 1–2 Hz, 2–4 Hz, 4–8 Hz,8–12 Hz, 12–16 Hz, 16–32 Hz) and Time Interval (40 consecutive 1-sec intervals within the Mc LA-associated time window) was carriedout on log-transformed signal power. In order to limit the number offactor levels, time resolution for this analysis was reduced to 1 s. TheHuynh and Feldt adjustment was applied to the estimation ofsignificance levels in order to take into account time-dependentdeviation from sphericity assumption.
Table 1Demographic, MRI findings and clinical information for each patient.
Patient Gender Age(years)
Medications(mg/day)
MRI findings SEEG
Side Sample lobes dlPFca Mca Epileptogenic zoneb
1 F 33 Lamotrigine 400 Unrevealing L FC F3 Paracentral lobule Anterior cingulate gyrus2 M 21 Phenytoin 300 Unrevealing R FCT F3 Paracentral lobule Anterior cingulate gyrus3 F 32 Carbamazepine 800 Parietal–temporal
ischemic lesionL FCTP F1–F2 sulcus Precentral gyrus Superior temporal gyrus
Levetiracetam 20004 M 19 Valproic acid 1200 Unrevealing R FCT F3 Paracentral lobule Orbito-basal region
Topiramate 2005 M 36 Lamotrigine 400 Unrevealing R FC F3 Paracentral lobule Superior frontal gyrus
Topiramate 400
C=central; F=frontal; P=parietal; T=temporal. F1: superior frontal gyrus; F2: middle frontal gyrus; F3: inferior frontal gyrus. dlPFc=dorso-lateral prefrontal cortex; Mc=Motorcortex.
a Indicates the position of the bipolar SEEG derivations submitted to sleep EEG analysis.b Indicate the site of origin of the seizure.
Fig. 2. Overnight distribution of slow wave activity (SWA) and motor cortex LocalActivations (LAs). The vertical gray bars indicate REM sleep periods. Upper graph: SWA,marker of NREM sleep depth, has been calculated as the power in the 0.5–4.0 Hz band ofthe EEG signal recorded from the motor cortex (Mc) the dorso-lateral prefrontal cortex(dlPFc) and the scalp (Fz-Cz). Signal power has been normalized within each recordingto the geometric mean overnight value and averaged between recordings. In order toobtain an average SWA profile of each derivation, the first three NREM–REM cycles ofeach recording were divided into an equal number of intervals (20 steps for each NREMphase and 5 steps for each REM) and then averaged between patients. The light grayarea represents the confidence intervals for the three curves, calculated by assuming achi-square distribution of the band. The intervals overlap indicating no significantdifference at single time-step level. All three curves exhibit the typical cyclic NREM–
REM pattern and a progressive decrease from cycle to cycle. Lower graph: Distributionof the mean motor cortex LA index (# of events/hour of NREM sleep). LAs weredetected from motor cortex EEG as an abrupt increase of signal frequency interruptingsleep EEG pattern. LA index increases from cycle to cycle and at the end of eachcycle. Error bars show the confidence intervals associated to each column (assuming achi-square distribution of the event count) based on False Discovery Rate probabilitythreshold.
615L. Nobili et al. / NeuroImage 58 (2011) 612–619
For all the ANOVAs carried out in the present study, the normalityassumption was checked by Lilliefors test, a 2-sided goodness-of-fittest, using the Kolgomorov–Smirnov statistic and suitable fornormality testing in small samples (Lilliefors, 1967). The level ofsignificance was set at pb0.01. Post-hoc comparisons were conductedon confidence intervals (CI) based on ANOVA result and evaluated at95% level with the application of the Dunn-Sidák adjustment formultiple comparisons. For the post-hoc analysis of time–frequencydistribution within the Mc LA-associated time window, the followingprocedure was adopted: the first 10 s of the time window wereconsidered as background; for each frequency band, the sample ofvalues relevant to each point in time (one value for each recording,0.125 s resolution) was compared by a two sample t-test with thebackground values and a probability level was associated to eachpoint in the time–frequency plane; a probability threshold was thenset according to the False Discovery Rate (FDR) method (Benjaminiand Hochberg, 1995); all points in the time–frequency plane, with aprobability level lower than selected threshold were consideredsignificant.
ANOVAs, the associated multiple comparisons and Lilliefors test tocontrol for normality of residuals were performed by the StatisticsToolbox of Matlab software (Mathworks Inc, Natick, MA, USA).
Results
Dynamics of slow wave activity and distribution of local activations
The dynamics of SWA across the night was analyzed bymeans of EEGspectral analysis applied to the first three NREM-REM cycles. Thepresence of SWA during NREM sleep and its progressive, physiologicaldecay across sleep cycles, typically present in the scalp recordings, wasalso clearly visible both in the dlPFc and in the Mc (see Fig. 2). The BrainRegion x Cycle ANOVA showed a significant main effect for Brain Region(F(2,16)=8.58, p=0.003). Post-hoc comparisons indicated that Mchave lower SWA levels compared to both dlPFc and scalp derivations,that do not differ between them (means±CI: Mc=0.78±0.11, dlPFc=1.14±0.11, scalp=1.05±0.11). The main effect for Cycle (F(2,16)=13.58, p=0.0004) indicated that the first two sleep cycles show higherSWA compared to the third cycle (Means±CI: 1st=1.21±0.11,2nd=1.01±0.11, 3rd=0.75±0.11). The Brain Region×Cycle interac-tion was not significant (F(4,16)=1.63, p=0.21).
In spite of the similar global dynamics of SWA recognized in bothcortical structures, the Mc showed frequent local activations character-ized by an abrupt interruption of the sleep EEG slowwaves pattern andby the appearance of awake-like EEGhigh frequency pattern (includingalpha and/or beta rhythm). These LAs, differently from those occurringsimultaneously on the threederivations,were paralleledby adeepeningof sleep in other cortical regions, as expressed by the concomitantincrease of slowwaves in the dlPFc and scalp EEG recordings (see Fig. 3).
In order to quantify the intranight-distribution of LAs, we visuallyscored all the activations during NREM sleep independently in eachderivation, as specified in theMethods.Mean local Activation Index (AI),measured as events/hour of NREM sleep, was 18.7 for Mc, 4.1 for dlPFcand 6.7 for scalp. One-way ANOVA on local AI showed a significantdifference for Brain Region (F(2,8)=77.0, pb0.0001). Post-hoc com-parisons showed that mean AI for dlPFc and scalp do not significantlydiffer, but they are both significantly lower compared toMcmean value(see Fig. 4a).
Eighty-eight percent of Mc activations were strictly local in nature(being discernible only on this derivation), while LAs account for only6% of dlPFc activations and 10% of scalp EEG activations. Specifically,
Fig. 3. Sample patterns of intracerebral EEG. The first three traces show EEG recordings from the motor cortex (Mc), dorso-lateral prefrontal cortex (dlPFc) and scalp (Fz–Cz), and arefollowed by the electrooculogram (EOG) and electromiogram (EMG, chin). a) Quiet wakefulness (eyes closed): fast rhythms prevail in all the three EEG tracings. b) NREM sleep: slowwaves prevail in all the three EEG tracings. c) A local activation (LA, grey shadowed area), characterized by fast EEG activity, appears in the motor cortex derivation and continue fortens of seconds while sleep EEGwith slowwaves prevails in the other tracings. d)When a short LA appears in themotor cortex, a burst of slowwaves characterizes the other two EEGderivations. e) A diffuse short burst of delta waves followed by a complete awakening with fast rhythms in all the three EEG recordings.
616 L. Nobili et al. / NeuroImage 58 (2011) 612–619
we observed that 89.2% of all Mc LAs occurred during Stage 2, 6%during Stage 3 and 4.8% during Stage 4. Mc LAs generally lasted lessthan 30 s (mean: 15.63, standard error, SE: 1.9); however, suchcontrasting behavior (fast EEG activities in the Mc and SWA in thedlPFc and scalp recordings) sometimes continued for a substantialperiod of time (30–120 s; Fig. 4b). The rate of these long lasting(N30 s) Mc LAs was 2.6 events/h of NREM sleep (SE: 0.3) with acumulative duration of 132.8 s/h of NREM sleep (SE: 41.2 s). DlPFcand scalp EEG did not show long lasting LAs.
As far as the evaluation of the coincidence of Mc activations andperipheral muscle activity changes is concerned, we observed that45±4.3% of all Mc LAs were accompanied by an EMG activation.Considering only these coupled activations, the percentage of timeof chin muscle tone increase during the Mc LA was 33±3.2% of theentire LAs duration.
The overnight distribution of Mc LAs is depicted in Fig. 2 (lowerpanel). The Cycle x Segment ANOVA showed a main effect for Cycle(F(2,24)=8.83, p=0.0013). The mean LA index (LAs/h) increasedacross NREM sleep cycles (Means±CI: 10.5±4.42, 18.3±4.42 and27.2±4.42). Post-hoc comparisons showed that each cycle wassignificantly different from each other (see Fig. 4c). The increasefrom the first to the third cycle was significantly fitted by a lineartrend (r2=0.34, pb0.0001).
The main effect for Segment was also significant (F(3,24)=6.40,pb0.0024). Mc LAs increased in the last part of each NREM period(Means±CI: 13.3±5.65, 15.3±5.65, 15.1±5.65 and 30.9±5.65).Post-hoc comparisons indicated that the last segment differed sig-nificantly from the others (Fig. 4d). The Cycle×Segment interactionwas not significant (F(6,24)=0.53, p=0.78).
Time–frequency analysis of local cortical activations
For each artifact-free Mc LA, the time–frequency distribution wasevaluated by discrete wavelet transform. The mean time–frequencydistribution of signal power (after square root transformation) isreported in Fig. 5 for the three derivations.
Repeated measure ANOVA did not show significant differencesamong mean log-transformed power values as function of Region(F(1,13)=2.68, p=0.13), Frequency (F(6,78)=2.00, p=0.16) andTime (F(39,507)=2.47, p=0.08). Significant first order interactionswere found for Frequency×Region (F(6,78)=18.36, pb0.0001) andFrequency×Time (F(234,3042)=13.3, pb0.0001), while the Time×Region interaction was close to significance at the selected threshold(F(39,507)=3.88, p=0.018).
Moreover, the crucial Frequency×Time×Region interaction wasalso significant (F(234,3042)=9.20, pb0.0001). Therefore, post-hoc
Fig. 4. Panel a: Means and confidence intervals (CI) of the Activation index (AI) values as a function of brain region. The AI forMotor cortex (Mc) was significantly higher compared tothe dorso-lateral prefrontal cortex (dlPFc) and scalp values (Fz–Cz). Panel b: Distribution of Mc LA Index as a function of the activation duration. Panel c: Means and confidenceintervals (CI) of the Mc LA index (events/hour) as a function of the sleep cycle. The LA index increased from the first to the third cycle: the first and last cycle weresignificantly different (their confidence intervals did not overlap) while the second one had an intermediate value. The global increase was fitted by a significant linear trend(pb0.0001). Panel d: Means and confidence intervals (CI) of the Mc LA index as function of within-cycle segments. The CI for the first 3 segments largely overlapped, indicating thatmean values were not significantly different. The last segment of the cycle showed a significantly higher mean LA index.
Fig. 5. Time frequency distribution of EEG patterns recorded in Motor cortex (Mc, upper part), dorso-lateral prefrontal cortex (dlPFc, middle part) and scalp (Fz–Cz, bottom part) inassociationwithmotor cortex Local Activations (LAs). A 40-sec timewindowwas set around the start (0 time)of eachMc LAand signal powerwas calculatedbywavelet transform for eachtime unit within a time frequency grid (0.125 s time resolution and frequency-dependent bandwidth). Time–frequency distributionwas averagedwithin each recording and then amongrecordings afterwithin-subject normalization. Thedistribution of averaged signal amplitude (square root of the power) is presented for eachbrain regionby a color scale. The points, in thetime–frequency plane,which didn't significantly differ from thebackground, aremarked by the black symbol ‘X’ (ANOVA followed byFDR-corrected post-hoc comparison—see the text fordetails). The abrupt shift towards higher frequencies (4–32 Hz) in themotor cortex is accompaniedbyan increaseof EEG activity in the low frequencies (0.5–2.0 Hz) in theother regions, inassociation with the onset of the Mc local activation.
617L. Nobili et al. / NeuroImage 58 (2011) 612–619
618 L. Nobili et al. / NeuroImage 58 (2011) 612–619
analysis focused on band power changes as a function of time in eachregion. In Fig. 5, the points in the time–frequency plane which did notsignificantly differ from backgroundweremarked by the black symbol‘X’. A significant shift of power from low to higher frequenciesassociated to LA onset is evident for Mc (top diagram), characterizedby a significant reduction of SWA (0.5–4.0 Hz) and by an increase inthe upper frequency bands (4.0–32.0 Hz) particularly evident for thealpha band (8.0–12.0 Hz). On the other hand, dlPFc (central diagram)and scalp (bottom diagram) were characterized by a transient andsignificant increase of SWA (0.5–4.0 Hz) around the Mc LA, actuallybeginning just before the Mc LA. An increase of low-frequencyactivities was also detected in Mc slightly preceding the increase offast activities.
Discussion
Here we showed, by means of intracerebral stereo-EEG recordings,that human sleep can be characterized by the coexistence of wake-likeand sleep-likeEEGpatterns indifferent cortical areas, as indicated by thehigh number of local activations within the motor cortex that wereaccompanied by deep-sleep EEG patterns in the prefrontal cortex andscalp.
In particular, we observed that: i) The physiological andprogressive decay of SWA across NREM sleep cycles is comparablebetween the derivations (dlPFc, Mc, scalp) investigated; ii) In spite ofthis similarity, Mc shows a higher rate of local activation during NREMsleep compared to dlPFc and scalp, that increases across sleep cyclesand within each cycle towards the end of each NREM period, beingpossibly associated with the decrease of SWA and the approach ofREM sleep; iii) Mc LAs generally have a duration of 5–10 s, but theycould sometimes last up to 120 s; iv)More than 50% of all Mc LAswerenot accompanied by an EMG activation and only one third of the McLA cumulative duration is accompanied by an increase of chin muscletone; v) The time–frequency analysis of the EEG around each localactivation indicates that Mc LAs were characterized by an increase ofhigh frequency EEG activity of Mc at the onset of the LA, paralleled bythe increase of low-frequency EEG activities (peaking at 0.5–2 Hz) atdlPFc and scalp sites.
A number of scalp EEG studies (Finelli et al., 2001; Ferrara et al.,2002; Marzano et al., 2010) have already highlighted the presence ofregional differences in the topographic distribution of slow waves;our results go beyond these findings, indicating for the first time inhumans, by means of intracerebral recordings, that the sleep processis not necessarily present simultaneously in the entire brain.
The appearance of localized simultaneous sleep-like andwake-likeEEG activity during NREM sleep can explain poorly understood sleepevents, such as sleepwalking and confusional arousal. In particular,during such clinical phenomena, considered from the first pioneeringdescription as arousal disorders (Broughton, 1968), the coexistence ofactivated and deactivated brain regions have been shown by SinglePhoton Emission Computed Tomography (Bassetti et al., 2000) andintracerebral electrophysiological studies (Terzaghi et al., 2009). Ourresults demonstrate that the occurrence of local dissociated statesis actually an intrinsic feature of physiological NREM sleep, thussupporting the hypothesis that sleep and wakefulness could be notmutually exclusive and that unusual phenomena might result froman imbalance of these two states (Mahowald and Schenck, 2005).Genetic factors or external triggers such as sleep deprivation (Zadraet al., 2008), inducing a modification of the arousal threshold, mayfavor the persistence of local dissociated activity, and this may causethe appearance of motor phenomena such as those observed in NREMparasomnias. Moreover, if local awakenings can appear during sleep,we cannot exclude that such a dissociation may also occur duringwakefulness (in this case with cortical areas showing local sleepfeatures) and this could explain other phenomena such as sleepinertia (Ferrara et al., 2006; Marzano et al., 2011), that subjective
feeling of grogginess accompanied by decreased levels of performancewhich typically follows awakening.
The occurrence of LAs is consistent with experimental studiesshowing that sleep and wakefulness can be restricted to small groupsof neurons (Pigarev et al., 1997), individual cortical columns (Rectoret al., 2005) or to larger brain regions, as in some birds and marinemammals that in order to continue flying, swimming or scanning thesurrounding environment can simultaneously exhibit sleep in onecerebral hemisphere and wakefulness in the other one (Mukhametov,1984; Lyamin et al., 2008). Moreover, our data are in accordance withrecent studies which, applying experimental manipulations ofsensory and learning systems before sleep, provided evidence of alocal regulation of sleep (Vyazovskiy et al., 2000; Cantero et al., 2002;Huber et al., 2004; Huber et al., 2006; Nelini et al., 2010). Moregenerally, the coexistence of sleep-like and wake-like patterns is inagreement with the predictions of the neuronal group theory ofsleep function (Krueger et al., 1999, 2008), which posits that sleep islocal in nature, being a fundamental property of small neuronalgroups.
In an evolutionary perspective, we can speculate that a lowerarousal threshold and a higher level of activation in the motor cortexduring the least activated physiological state (NREM sleep) may havebeen selected, because they increase the probability of survival,facilitating motor behaviors in case of sudden awakenings. In thesame vein, the significant enhancement of slow frequencies in thedlPFc immediately before and during the Mc LAs could be interpretedas a behavior that allows the global sleep process to easily proceed,even when a local activation appears. This finding also suggests thatsleep (and sleep intensity) is not a spatially global state. In fact, a formof local homeostasis seems to occur, characterized by the enhance-ment of delta activity in one cortical area to compensate for a decreaseof slow activity in another area.
Our results may also shed light on the interpretation of certainsleep EEG features, such as K complexes and transient bursts of deltawaves, which have alternatively been considered a hallmark of deepsleep (Amzica and Steriade, 1997; De Gennaro et al., 2000) or asarousal reactions (Terzano et al., 1990; Wauquier et al., 1995; De Carliet al., 2004; Halász et al., 2004). In particular, since transient Kcomplexes and bursts of delta waves can be elicited by external orinternal stimuli (Terzano et al., 1990; Nobili et al., 2006; Terzaghiet al., 2008), they have been interpreted as a kind of ‘anti-arousal’response (Wauquier et al., 1995; Halász et al., 2004). However, thetransient increase of delta band in the dlPFc observed in our studyseems not to represent an anti-arousal response to the Mc LA. In fact,the increase of delta power is present and precedes the motor cortexactivation (see Fig. 5). Furthermore, cross-correlations betweenprefrontal delta activity and alpha activity over Mc (data notshown) indicate that the phasic increase of prefrontal delta activityis significantly related to the increase in motor cortex activation(alpha activity, 8.0–12.0 Hz), and precedes this activationwith ameanlag of −1.5 s. Nevertheless, these phenomena could be interpreted asindicating that, during NREM sleep, different cortical areas respond toactivation (due to internal or external stimuli) with differentelectrophysiological features and timing. Further studies includingthe analysis of more cortical regions and the assessment of vegetativefunctions, which have been shown to be linked to reactive slowwavesover the frontal regions (Church et al., 1978; Ferini-Strambi et al.,2000) might be helpful to evaluate possible hierarchical regulations oflocal activations.
Although our findings derive from SEEG investigations in epilepticpatients, we are confident that they can be extended to the generalpopulation, as we investigated patients without any sleep complaint.Moreover, the EEG channels analyzed here were free of inter-ictalepileptic discharges and were not involved in the ictal dischargeduring seizures. Therefore, we can reasonably rule out a possibleepileptic origin of the described local activations.
619L. Nobili et al. / NeuroImage 58 (2011) 612–619
In conclusion, our data show that the boundaries between sleep andwakefulness are less clearly defined than expected and support thenew,local interpretation of the electrophysiology of sleep and vigilance.Although these states have been usually considered as whole-brainphenomena, our findings suggest that their electroencephalographicfeatures can coexist in different brain areas. In the future, it would beinteresting to specifically assess the relations between motor cortexactivations and muscular activity recorded from the muscular districtsdirectly corresponding to the cortical regions explored by theintracerebral electrodes. Moreover, it would be fascinating to evaluatewhether motor cortex EEG activations are part of normal dreaming inNREM sleep, associated with fictive movement mentation.
Acknowledgments
This work has been in part supported by grants from Compagnia diSan Paolo, ProgrammaNeuroscienze 2008/09 (3896 SD/sd, 2008.2130),the University of L'Aquila (Ricerche di Ateneo ex 60%) and MIUR, Italy,(PRIN: n. 2007BNRWLP-002) toMichele Ferrara, andby theESRS Sanofi-Aventis Research Grant 2008-10 to Fabio Moroni.
References
American Sleep Disorders Association (ASDA), 1992. EEG arousals: scoring rules andexamples: a preliminary report from the Sleep Disorders Atlas Task Force of theAmerican Sleep Disorders Association. Sleep 15, 173–184.
Amzica, F., Steriade, M., 1997. The K-complex: its slow (b1-Hz) rhythmicity andrelation to delta waves. Neurology 49, 952–959.
Bassetti, C., Vella, S., Donati, F., Wielepp, P., Weder, B., 2000. SPECT during sleepwalking.Lancet 356, 484–485.
Benjamini, Y., Hochberg, Y., 1995. Controlling the false discovery rate: a practical andpowerful approach to multiple testing. J. R. Stat. Soc. Ser. B 57, 289–300.
Broughton, R.J., 1968. Sleep disorders: disorders of arousal? Science 159, 1070–1078.Cantero, J.L., Atienza, M., Salas, R.M., Dominguez-Marin, E., 2002. Effects of prolonged
waking-auditory stimulation on electroencephalogram synchronization andcortical coherence during subsequent slow-wave sleep. J. Neurosci. 22, 4702–4708.
Church, M.W., Johnson, L.C., Seales, D.M., 1978. Evoked K-complexes and cardiovascularresponses to spindle-synchronous and spindle-asynchronous stimulus clicksduring NREM sleep. Electroencephalogr. Clin. Neurophysiol. 45, 443–453.
Cossu,M., Cardinale, F., Castana, L., Citterio, A., Francione, S., Tassi, L., Benabid,A.L., Lo Russo,G., 2005. Stereoelectroencephalography in thepresurgical evaluationof focal epilepsy:a retrospective analysis of 215 procedures. Neurosurgery 57, 706–718.
De Carli, F., Nobili, L., Gelcich, P., Ferrillo, F., 1999. A method for the automatic detectionof arousals during sleep. Sleep 22, 561–572.
De Carli, F., Nobili, L., Beelke, M., Watanabe, T., Smerieri, A., Parrino, L., Terzano, M.G.,Ferrillo, F., 2004. Quantitative analysis of sleep EEG microstructure in the time–frequency domain. Brain Res. Bull. 63, 399–405.
De Gennaro, L., Ferrara, M., Bertini, M., 2000. The spontaneous K-complex during stage2 sleep: is it the "forerunner" of delta waves? Neurosci. Lett. 291, 41–43.
De Gennaro, L., Ferrara, M., Curcio, G., Cristiani, R., 2001. Antero-posterior EEG changesduring the wakefulness-sleep transition. Clin. Neurophysiol. 112, 1901–1911.
Ferini-Strambi, L., Bianchi, A., Zucconi, M., Oldani, A., Castronovo, C., Smirne, S., 2000.The impact of cyclic alternating pattern on heart rate variability during sleep inhealthy young adults. Clin. Neurophysiol. 111, 99–101.
Ferrara, M., De Gennaro, L., Curcio, G., Cristiani, R., Corvasce, C., Bertini, M., 2002.Regional differences of the human sleep electroencephalogram in response toselective slow-wave sleep deprivation. Cereb. Cortex 12, 737–748.
Ferrara, M., Curcio, G., Fratello, F., Moroni, F., Marzano, C., Pellicciari, M.C., De Gennaro,L., 2006. The electroencephalographic substratum of the awakening. Behav. BrainRes. 167, 237–244.
Finelli, L.A., Borbely, A.A., Achermann, P., 2001. Functional topography of the humannonREM sleep electroencephalogram. Eur. J. Neurosci. 13, 2282–2290.
Halász, P., Terzano, M., Parrino, L., Bódizs, R., 2004. The nature of arousal in sleep. J. SleepRes. 13, 1–23.
Huber, R., Ghilardi, M.F., Massimini, M., Tononi, G., 2004. Local sleep and learning.Nature 430, 78–81.
Huber, R., Ghilardi, M.F., Massimini, M., Ferrarelli, F., Riedner, B.A., Peterson, M.J.,Tononi, G., 2006. Arm immobilization causes cortical plastic changes and locallydecreases sleep slow wave activity. Nat. Neurosci. 9, 1169–1176.
Jobert, M., Tismer, C., Poiseau, E., Schulz, H., 1994. Wavelets—a new tool in sleepbiosignal analysis. J. Sleep Res. 3, 223–232.
Krueger, J.M., Obal, F., Fang, J., 1999. Why we sleep: a theoretical view of sleep function.Sleep Med. Rev. 3, 119–129.
Krueger, J.M., Rector,D.M., Roy, S., vanDongen,H.P., Belenky,G., Panksepp, J., 2008. Sleep asa fundamental property of neuronal assemblies. Nat. Rev. Neurosci. 9, 910–919.
Lilliefors, H.W., 1967. On the Kolmogorov–Smirnov test for normality with mean andvariance unknown. J. Am. Stat. Assoc. 62, 399–402.
Lyamin, O.I., Manger, P.R., Ridgway, S.H., Mukhametov, L.M., Siegel, J.M., 2008. Cetaceansleep: an unusual form of mammalian sleep. Neurosci. Biobehav. Rev. 32, 1451–1484.
Mahowald, M.W., Schenck, C.H., 2005. Insights from studying human sleep disorders.Nature 437, 1279–1285.
Marzano, C., Ferrara, M., Curcio, G., De Gennaro, L., 2010. The effects of sleep deprivationin humans: topographical electroencephalogram changes in non-rapid eyemovement (NREM) sleep versus REM sleep. J. Sleep Res. 19, 260–268.
Marzano, C., Ferrara, M., Moroni, F., De Gennaro, L., 2011. Electroencephalographicsleep inertia of the awakening brain. Neuroscience 10, 308–317.
Moroni, F., Nobili, L., Curcio, G., De Carli, F., Fratello, F., Marzano, C., De Gennaro, L.,Ferrillo, F., Cossu, M., Francione, S., Lo Russo, G., Bertini, M., Ferrara, M., 2007. Sleepin the human hippocampus: a stereo-EEG study. PLoS ONE 2, e867.
Mukhametov, L.M., 1984. Sleep in marine mammals. Exp. Brain Res. 8, 227–238.Nelini, C., Bobbo,D.,Mascetti,G.G., 2010. Local sleep:a spatial learning taskenhances sleep in
the right hemisphere of domestic chicks (Gallus gallus). Exp. Brain Res. 205, 195–204.Nobili, L., Sartori, I., Terzaghi, M., Stefano, F., Mai, R., Tassi, L., Parrino, L., Cossu, M., Lo,
Russo G., 2006. Relationship of epileptic discharges to arousal instability andperiodic leg movements in a case of nocturnal frontal lobe epilepsy: a stereo-EEGstudy. Sleep 29, 701–704.
Pigarev, I.N., Nothdurft, H.C., Kastner, S., 1997. Evidence for asynchronous developmentof sleep in cortical areas. NeuroReport 8, 2557–2560.
Rechtschaffen, A., Kales, A., 1968. A manual of standardized terminology. Techniquesand Scoring System for Sleep Stages of Human Subjects. Public Health Service,Washington DC. NIH Publication No. 204, U S Government Printing Office.
Rector, D.M., Topchiy, I.A., Carter, K.M., Rojas, M.J., 2005. Local functional statedifferences between rat cortical columns. Brain Res. 1047, 45–55.
Rioul, O., Duhamel, P., 1994. A Remez exchange algorithm for orthonormal wavelets.IEEE Trans. Circuits Syst. 41, 550–560.
Rioul, O., Vetterli, M., 1991. Wavelets and signal processing. IEEE Signal Process. 8, 14–38.Terzaghi, M., Sartori, I., Mai, R., Tassi, L., Francione, S., Cardinale, F., Castana, L., Cossu, M.,
LoRusso, G., Manni, R., Nobili, L., 2008. Coupling of minor motor events andepileptiform discharges with arousal fluctuations in NFLE. Epilepsia 49, 670–676.
Terzaghi, M., Sartori, I., Tassi, L., Didato, G., Rustioni, V., Lo Russo, G., Manni, R., Nobili, L.,2009. Evidence of dissociated arousal states during NREM parasomnia from anintracerebral neurophysiological study. Sleep 32, 409–412.
Terzano, M.G., Parrino, L., Fioriti, G., Orofiamma, B., Depoortere, H., 1990. Modificationsof sleep structure induced by increasing levels of acoustic perturbation in normalsubjects. Electroencephalogr. Clin. Neurophysiol. 76, 29–38.
Vyazovskiy, V., Borbély, A.A., Tobler, I., 2000. Unilateral vibrissae stimulation duringwaking induces interhemispheric EEG asymmetry during subsequent sleep in therat. J. Sleep Res. 9, 367–371.
Wauquier, A., Aloe, L., Declerck, A., 1995. K-complexes: are they signs of arousal or sleepprotective? J. Sleep Res. 4, 138–143.
Zadra, A., Pilon, M., Montplaisir, J., 2008. Polysomnographic diagnosis of sleepwalking:effects of sleep deprivation. Ann. Neurol. 63, 513–519.