Source modeling sleep slow wavesMichael Murphya,b,1, Brady A. Riednera,b,c,1, Reto Hubera, Marcello Massiminia, Fabio Ferrarellia, and Giulio Tononia,2

aDepartment of Psychiatry, University of Wisconsin, Madison, WI 53719; and bNeuroscience Training Program and cClinical Neuroengineering TrainingProgram, University of Wisconsin, Madison, WI 53706

Edited by Marcus E. Raichle, Washington University School of Medicine, St. Louis, MO, and approved December 12, 2008 (received for reviewAugust 11, 2008)

Slow waves are the most prominent electroencephalographic (EEG)feature of sleep. These waves arise from the synchronization of slowoscillations in the membrane potentials of millions of neurons. Scalp-level studies have indicated that slow waves are not instantaneousevents, but rather they travel across the brain. Previous studies of EEGslow waves were limited by the poor spatial resolution of EEGs andby the difficulty of relating scalp potentials to the activity of theunderlying cortex. Here we use high-density EEG (hd-EEG) sourcemodeling to show that individual spontaneous slow waves havedistinct cortical origins, propagate uniquely across the cortex, andinvolve unique subsets of cortical structures. However, when thewaves are examined en masse, we find that there are diffuse hotspots of slow wave origins centered on the lateral sulci. Furthermore,slow wave propagation along the anterior�posterior axis of thebrain is largely mediated by a cingulate highway. As a group, slowwaves are associated with large currents in the medial frontal gyrus,the middle frontal gyrus, the inferior frontal gyrus, the anteriorcingulate, the precuneus, and the posterior cingulate. These areasoverlap with the major connectional backbone of the cortex and withmany parts of the default network.

brain mapping � default network � electroencephalography �slow oscillation � traveling wave

The transition from waking to sleep in humans is accompaniedby dramatic changes in the electroencephalograph (EEG). The

waking EEG is dominated by low-voltage fast activity, while duringnon-rapid eye movement (NREM) sleep low-frequency compo-nents appear in the EEG. During sleep, cortical neurons exhibit acharacteristic bistability, oscillating about once every second froma hyperpolarized downstate to a depolarized upstate (1). During thedownstate these neurons are almost entirely silent whereas duringthe upstate neuronal firing rates can be as high as in quietwakefulness (2). Intracranial recordings of anesthetized cats indi-cate that the origins of the slow oscillation are cortical and thatcortico–cortical connections are necessary for its synchronization(3). Large groups of slowly oscillating neurons synchronize duringsleep and the currents produced concomitantly with these changesin membrane potential sum to generate large slow waves in the scalpEEG (4). Increasing evidence suggests that slow waves may mediatesome of the functional benefits of sleep, at both the cellular and thesystems levels (5–7).

Topographical analyses of slow waves in humans have shown thatthey have a nonuniform scalp distribution, suggesting that areas ofthe cortex are differentially involved in slow waves (8, 9). Further-more, recent analyses have suggested that individual slow waves,rather than being simultaneous events, likely travel throughout thecortex (9). In addition, a growing body of evidence supports the ideathat sleep slow waves can be locally regulated (10–12). Therefore,questions remain about the role of specific cortical structures inindividual slow waves. Where do slow waves originate? Do theytravel through specific pathways or globally invade the cortex? Doall cortical areas participate equally in slow waves?

Although the slow oscillation has been investigated using severaldifferent techniques, source modeling of high-density EEG (hd-EEG) sleep slow waves offers a novel way to answer these questions.Scalp-level analyses of EEG sleep slow waves are made difficult bythe fact that local potentials can be generated by sources that are

distant from the recording electrodes (13). Intracranial recordingsare limited in scope and sensitivity; it is impossible to fully samplethe large cortical areas thought to be involved in slow waves.Imaging modalities like functional magnetic resonance imaging(fMRI) and positron emission tomography (PET) possess superiorspatial resolution to EEG source modeling, but their temporalresolution is insufficient to capture the changes in cortical activitythat occur during an individual slow wave (14, 15). Source-levelanalyses of hd-EEG sleep slow waves, however, offer a considerableadvantage in temporal resolution while achieving much higherspatial resolution than traditional scalp recordings, closer to that ofother functional imaging methods. Therefore, in this study, we usedhd-EEG source modeling to investigate the origin, propagation, andcortical involvement of individual sleep slow waves.

ResultsSource Modeling Individual Slow Waves Induced by TranscranialMagnetic Stimulation (TMS) Reliably Produces Maximal Sources Con-sistent with the Induction Site. In a previous study, we demonstratedthat individual pulses of TMS during sleep elicit EEG slow wavesthat closely resemble spontaneous sleep slow waves (16). For thecurrent study, we capitalized on this phenomenon by first sourcemodeling these evoked slow waves. This allowed us to ascertainwhether our source localization method produced consistent resultsin the case where the locus of the induction of the slow wave waspredetermined. Briefly, we recorded EEGs from 15 subjects witha TMS compatible system (Nexstim). After the subjects had tran-sitioned into stage 2 NREM sleep, we applied TMS pulses targetedto midline sensorimotor cortex (Fig. 1A). We selected a subject witha minimal TMS artifact and analyzed the TMS-evoked slow waveswith several different source modeling methods, including L2minimum norm with the local autoregressive average (LAURA)(17) constraint, low-resolution electromagnetic tomography(LORETA) (18, 19), standardized low-resolution electromagnetictomography (sLORETA) (20, 21), and the Bayesian minimumnorm (21) [see Methods and SI Text]. To normalize the currents, weselected 5 s of preceding EEG without slow waves, spindles, orartifacts as a baseline (see Methods). Regardless of the algorithm,TMS-triggered slow waves spread through source space in a ste-reotyped pattern (Fig. 1 B and C). About 120 ms after the TMSpulse, relative current begins to increase in the postcentral gyrus,precuneus, and cingulate gyrus, directly below the site of TMSstimulation on the scalp. Near the negative peak of the potentialwave, these sources increase in strength and are joined by relativecurrents in the anterior and posterior cingulate, lateral parietal, andsome temporal areas. As the wave weakens, most activation retreats

Author contributions: G.T. designed research; M. Murphy, B.A.R., R.H., M. Massimini, andF.F. performed research; M. Murphy and B.A.R. analyzed data; and M. Murphy, B.A.R., andG.T. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

1M. Murphy and B.A.R. contributed equally to this work.

2To whom correspondence should be addressed. E-mail: gtononi@wisc.edu.

This article contains supporting information online at www.pnas.org/cgi/content/full/0807933106/DCSupplemental.

© 2009 by The National Academy of Sciences of the USA

1608–1613 � PNAS � February 3, 2009 � vol. 106 � no. 5 www.pnas.org�cgi�doi�10.1073�pnas.0807933106

Dow

nloa

ded

by g

uest

on

Oct

ober

25,

202

0

back to the area around the induction site. The response is so robustthat the source model of any individual TMS-triggered slow waveclosely approximates the source model of the average of all of theresponses recorded during a single session. The variance withineach voxel across different waves is always less than the varianceacross the voxels within each single wave. We found that the origin,propagation, and involvement (see Methods) for TMS-evoked slowwaves were highly consistent across waves.

Source Modeling Spontaneous Slow Waves Reveals a Unique CorticalCurrent Distribution for Each Wave. Having confirmed that ouranalysis correctly localized the current sources of individual TMS-evoked slow waves, we next examined the source modeled behaviorof spontaneous sleep slow waves. Hd-EEG was collected from 6subjects during waking and sleep. After appropriate filtering of thedata (see Methods), we used a detection algorithm allowing for anunbiased detection of slow waves in terms of size and location, toidentify waves during slow wave sleep (NREM stages 3 and 4) ofthe first sleep cycle (see SI Text). The number of waves selected forthis analysis for each subject ranged from 60 to 237.

We found that, unlike the TMS-evoked slow waves, each spon-taneous slow wave showed a unique pattern of propagation acrossthe cortex. For example, two representative waves are shown in Fig.2. These waves originate from different cortical areas and involvedistinct but overlapping portions of the cortical surface. One wavespreads caudally and bilaterally from the anterior cingulate, while

the other wave starts in the left hemisphere near the insula, invadesthe cingulate gyrus, and then migrates through the right temporallobe. The results were qualitatively similar for other source mod-eling algorithms. Our data indicate that spontaneous sleep slowwaves are quite heterogeneous, and therefore the mathematical‘‘average track’’ is not a clear demonstration of how these waves act.Qualitatively, the ‘‘average’’ sleep slow wave probably beginsaround the insula, propagates posteriorly along midline structures,and includes large currents in the frontal gyri, anterior cingulate,precuneus, and posterior cingulate. However, because each spon-taneous wave is an idiosyncratic event, we systematically charac-terized individual waves by determining their origin, pattern ofpropagation, and cortical areas involved.

Spontaneous Slow Waves Originate More Frequently in the Insula andCingulate Gyrus. For every slow wave, we calculated which voxelswere likely to be the origin of that wave (see Methods and SI Text).Examination of the origins of spontaneous waves revealed a con-sistent pattern across subjects (Fig. 3 A and B and Fig. S1). Althoughslow waves could originate in either hemisphere, 17 of the top 20voxels with the most origins were in the left hemisphere. In termsof regional topography, we found a diffuse hot spot of originscentered on the lateral sulcus (including regions of insular, tempo-ral, frontal, and parietal cortices). Approximately 46% of sponta-neous slow waves across all subjects included at least one insularvoxel in the origin. There was a secondary disjoint hot spot for slowwave origins (�12%) around the medial cingulate gyrus (Fig. 3A).Like the lateral hot spot, the medial cingulate gyrus hot spot alsoincluded adjacent portions of the neighboring cortical structures. Thissecondary origin hot spot was not evident in all subjects (Fig. 3B).

Previous analysis of surface EEG slow waves suggested that manywaves originated from the anteriormost part of the scalp, oftenlaterally, while a second group of waves originated centromedially(9). To compare our results with those measured from the scalp, weseparated waves into 2 groups on the basis of whether they had atleast 1 probabilistic origin voxel in the left insula but not thecingulate gyrus (or vice versa). We then calculated the origins of thesurface EEG waves (see Methods) in each group (Fig. 3C). Wefound that the relative current waves that originated in the leftinsula most often manifested on the scalp as potential waves thatbegan at the left anterior portion of the scalp. Relative currentwaves that originated in the cingulate gyrus typically manifested onthe scalp as potential waves with centromedial origins.

Spontaneous Slow Waves Stereotypically Propagate Along CertainPaths. For each slow wave, we used the timing of the relative currentmaxima in each voxel to map how the wave propagated across thecortex (see Methods and SI Text). We found that each slow wavetravels uniquely across the cortex (see individual example in Fig.4A). The propagation speed in source space ranged from 0.4 to 6.3m/s. The average speed was �2.2 m/s, was remarkably consistentacross subjects (ranging from 2.0 to 2.3 m/s), and was roughly inagreement with previous estimations from EEG and computersimulations (9, 23). Visual inspection of the streamlines for thespontaneous slow waves for each subject suggested the existence ofa mesial slow wave highway oriented along the anterior–posterioraxis (Fig. 4B and Fig. S2): regardless of where they originated, wavesthat traveled along the anterior–posterior axis usually did so alongthe mesial aspect of the cortex.

To determine what structures were included in this mesial slowwave highway, we calculated the number of streamlines that passedthrough each voxel (Fig. 4 C and D and Fig. S2). As expected giventhe origin results, there were high densities of streamlines in theinsula. However, the highest density of streamlines was in theanterior portions of the cingulate. There were also high densities ofstreamlines in the posterior cingulate and precuneus. These mesialstreamlines are not merely the result of the waves originating in thecingulate gyrus because more slow waves propagate through the

Fig. 1. TMS-evoked slow waves travel across the cortex in a stereotypedmanner. (A) Anatomical MRI showing the site of TMS stimulation (green cross-hairs). (B) Representative butterfly plot of all 60-channel mastoid referenced EEGtraces showing the response to an individual TMS pulse (green bar). (C) Corticalflat maps illustrating the relative current density during the slow wave at timesafter TMS indicated by the red lines in B. PCun, precuneus; CG, cingulate gyrus;AC, anterior cingulate; PosC, postcentral gyrus; INS, insula; PCC, posteriorcingulate.

Murphy et al. PNAS � February 3, 2009 � vol. 106 � no. 5 � 1609

NEU

ROSC

IEN

CE

Dow

nloa

ded

by g

uest

on

Oct

ober

25,

202

0

cingulate gyrus than originate there (paired t test, P � 0.005).Therefore, many waves originate elsewhere and then propagate toor through the cingulate cortex, confirming the existence of a mesial

highway. To a lesser extent this highway also incorporates dorsalmidline structures such as the paracentral lobule.

Propagation along the highway could occur in either direction,but most often traveling occurred from front to back. For eachvoxel, we also examined the average direction of all streamlines thatpassed through and calculated in what dimension the averagedirection vectors were anisotropic. In the highway, the largest groupof voxels was anisotropic along the anterior–posterior axis (43.4%,standard deviation 7.0%) with the average direction vector headingposteriorly for each subject. There was comparatively less move-ment along the anterior–posterior axis in the lateral aspect of thecortex. The percentage of anisotropic voxels in the highway wassignificantly larger than in the rest of the brain (26.1%, standarddeviation 1.8%, paired t test, P � 0.005). Furthermore, waves travelabout the same distance along the highway whether they are movinganteriorly (4.0 dipoles/wave) or posteriorly (3.6 dipoles/wave, pairedt test, P � 0.3). Waves that moved anteriorly on the highway tendedto have more posterior origins than waves that moved posteriorly,but this difference was not statistically significant.

Spontaneous Slow Waves Preferentially Involve a Subset of BrainAreas and Avoid Others. We next examined the amount of currentproduced by different cortical areas during a slow wave. For eachvoxel, we defined the involvement in that voxel for each individualslow wave as the average relative current during a 100 ms windowcentered on the negative peak of the EEG slow wave (see SI Text).This analysis revealed that certain brain structures are more in-volved than others by sleep slow waves. Relative current hot spotsincluded the precuneus, the cingulate gyrus, the posterior cingulate,the anterior cingulate, and left and to a lesser extent right frontalstructures including the middle frontal gyri, the medial frontal gyri,the inferior frontal gyri, and the insula (Fig. 5 and Fig. S1). Thisresult could not be inferred from the voltage topography alone (Fig.S3). While there was some variation between subjects (Fig. 5B), onthe whole, the left middle, medial, and inferior frontal gyri were

Fig. 2. Individual spontaneous slow waves are unique cortical events. (A and C) The butterfly plots of all 256-channel mastoid referenced EEG traces for 2 spontaneousslow waves are shown. (B) Cortical flat maps showing relative current density at the time points indicated by the red lines for wave A. (D) Cortical flat maps showingrelative current density at the time points indicated by the red lines for wave C. Note the difference between the first wave (A and B) where recruitment occurs froma relatively stationary hot spot and the second wave (C and D), where there is distinct traveling from one hemisphere to the other. AC, anterior cingulate; MedF, medialfrontal gyrus; INS, Insula; CG, cingulate gyrus; ITG, inferior temporal gyrus.

Fig. 3. Spontaneous slow waves can originate from anywhere in the cortex, butpreferentially originate in the insula or the cingulate gyrus. (A) Average corticalflat map showing the percentage of slow waves that included that voxel as anorigin.Therearetwodisjointoriginhotspots,onecenteredaroundtheinsulaandone centered around the cingulate gyrus. (B) Cortical flat maps for each subject.The insula origin hot spot is strongly present in every subject, and the cingulategyrus hot spot is at least weakly apparent in every subject. (C) The scalp origins ofwaves that originate in either the left insula or the cingulate gyrus. (Upper) Weselected waves with source origins that contained voxels from the left insula hotspot but not the cingulate gyrus hot spot. The topoplot indicates the percentageof waves that originate near each electrode, calculated using the scalp EEGmethod (9, 58). Waves with insula cortical origins often appeared on the surfaceas if they originated from the left anterior edge of the scalp. (Lower) Waves withorigins that contained voxels from the cingulate gyrus hot spot but not the leftinsula hot spot typically appear on the scalp as EEG waves originating centrome-dially. INS, Insula; CG, cingulate gyrus.

1610 � www.pnas.org�cgi�doi�10.1073�pnas.0807933106 Murphy et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

25,

202

0

more involved than the right middle, medial, and inferior frontalgyri (paired t test, P � 0.05). There was also a subset of areas thattypically showed weak involvement. This subset included somesensory areas (such as the postcentral gyrus, visual cortex, andassociation areas) and the inferior parietal lobule, the middletemporal gyrus, and regions that were electrotonically distant fromthe recording electrodes (such as the uncus, hippocampus, andfusiform gyrus) (Fig. 5).

Discussion Slow waves are the most prominent electrophysiologicalfeature of NREM sleep. In this paper, we first validated our sourcemodeling results by analyzing slow waves evoked by TMS. We found

that each of these waves has a stereotyped source topography withthe largest relative currents under the site of stimulation. We thensource modeled individual spontaneous slow waves and determinedthat each corresponds to a unique relative current distribution.While individual slow waves could originate from a wide variety ofcortical locations, as a group, there was some indication that someareas were more likely to be origins, namely the insula and, to alesser extent, the cingulate gyrus. Slow waves often travel along theanterior–posterior axis via the cingulate cortices. In addition, slowwaves preferentially involve several cortical areas including theinferior frontal gyrus, the medial frontal gyrus, the middle frontalgyrus, the insula, the anterior cingulate gyrus, the posterior cingu-late gyrus, and the precuneus. Slow waves for the most part avoidthe visual and sensory cortices and the fusiform gyrus.

Source Modeling hd-EEG Is Uniquely Suited to Studying IndividualSlow Waves. While other neuroimaging techniques like PET andfMRI offer superior spatial resolution, source modeling ofhd-EEG is uniquely suited to the analysis of sleep slow waves.First, the superior time resolution of EEG permits tracking theevolution of individual slow waves. Furthermore, althoughsource modeling typically restricts the space to cortical sources,evidence from several different modalities suggests that the sleepslow oscillations do indeed originate in, travel across, and aremaintained by the cortex. Isolated cortical slabs undergo slowoscillations, suggesting that the cortex is sufficient for thegeneration of slow waves (24, 25). Disrupting thalamo–corticalconnections leaves the slow oscillation intact while damagingcortico–cortical connections disrupts the synchronization of theslow oscillation (3). Additionally, TMS does not directly reachsubcortical structures (26), yet it can reliably elicit slow waves(15). Spreading upstates have been observed in multiple-electrode single-unit recordings of cortical neurons during slowwave oscillations in anesthesia (27). The possibility of a subcor-tical basis for EEG slow waves is also discounted by the fact thatcortical decreases in regional cerebral blood flow during NREMsleep are correlated with EEG delta power and are independentof thalamic activity (28). Furthermore, the variability betweenindividual spontaneous EEG slow waves suggests that a singlesubcortical substrate is unlikely (9).

While the above evidence suggests that source modeling isparticularly well suited to describing slow waves, there are severalcaveats. First, despite improvements in source modeling, there is the

Fig. 4. Spontaneous slow waves often propagate via the cingulate. For each wave, we calculated a streamline for each origin voxel. (A) The origin voxels (red circles)and streamlines (red tubes) for a single slow wave overlaid on an anatomical MRI. This slow wave originates primarily from right anterior cortex (including the inferiorfrontal and middle frontal gyri). It propagates medially and posteriorly. (B) Streamlines for an individual subject. Only the longest 10 streamlines for each wave areshown. A dense streamline highway can be seen running along the anterior–posterior axis in the medial portion of the cortex. (C) For each voxel, we calculated thenumber of waves with streamlines that passed through it. A flat cortical map of the propagation density is shown for each voxel, expressed as a percentage of the totalnumber of waves. Propagation hot spots are seen along the cingulate cortices. (D) Flat cortical maps of the propagation density for each subject. AC, anterior cingulate;CG, cingulate gyrus; INS, insula; PCC, posterior cingulate.

Fig. 5. Spontaneous slow waves preferentially involve certain cortical areas. Foreach voxel, we averaged the relative current over a 100-ms window centered onthe negative peak of the EEG wave. (A) Cortical flat map displaying for all wavesthe total involvement of each voxel, expressed as a percentage of the maximum.Involvement hot spots include the anterior cingulate, the cingulate, the posteriorcingulate, the precuneus, and the left inferior frontal gyrus and insula. Involve-ment cold spots include the inferior and middle temporal gyri, the middleoccipital gyrus, the postcentral gyrus, the inferior parietal lobule, the fusiformgyrus, and the parahippocampal gyrus. (B) Cortical flat maps for each subject.Although the relative values of the involvement hot spots vary between subjects,the same areas are highly involved in all subjects. IPL, inferior parietal lobule; IFG,inferior frontal gyrus; MOG, middle occipital gyrus; FG, fusiform gyrus; MTG,middle temporal gyrus; PHG, parahippocampal gyrus. AC, anterior cingulate; CG,cingulate gyrus; INS, insula; PosC, postcentral gyrus; PCC, posterior cingulate;PCun, precuneus; ITG, inferior temporal gyrus.

Murphy et al. PNAS � February 3, 2009 � vol. 106 � no. 5 � 1611

NEU

ROSC

IEN

CE

Dow

nloa

ded

by g

uest

on

Oct

ober

25,

202

0

fundamental limitation that infinitely many cortical current distri-butions can give rise to a given scalp voltage topography (13). Inaddition, there are a variety of technical limitations specific to ouranalysis. We used the same MRI and set of channel locations foreach subject; as individual subjects and net placements vary fromthis ideal, the appropriateness of this approximation decreases. Ourforward model was relatively simple, consisting of 4 nested isotropicspheres. More sophisticated forward models incorporate complexgeometry and/or anisotropy (29). While each of the source mod-eling techniques used here revealed similar results, future refine-ments in forward or inverse modeling should further improve thelocalization of current sources through hd-EEG.

Current Sources Reflect the Cellular Physiology of the Slow Oscilla-tion. We found large relative current maxima around the negativepeak of the EEG slow wave. Prior results indicate that this negativepeak occurs at the same time as a positive peak in the local fieldpotential (LFP) in the underlying cortex (30, 31). The timing ofthese peaks reflects the beginning of the transition from downstateto upstate and the resumption of neural firing (9, 32, 33). Unitrecording studies also support this interpretation (34). While cor-tical currents are typically thought to arise from the summation ofinhibitory and excitatory postsynaptic potentials from the localneural population (13), synaptic input is greatly reduced during thedownstate (23, 35–37). This suggests that postsynaptic potentials arenot entirely responsible for the large currents recorded during a slowwave. Although the precise mechanism by which these cortical currentsare generated is unknown, contributing factors could be synchro-nous changes in cellular currents including the hyperpolarization-activated cation current and/or the potassium leak current (23).

Origins, Propagation, and Involvement Each Reveal Novel InformationAbout the Cortical Generators of Slow Waves. We used 3 parametersto characterize slow waves in source space: origin, propagation, andinvolvement. Slow waves often originate from areas adjacent to thelateral sulci (insula, superior temporal gyri), preferentially propa-gate via the cingulate cortices, and typically involve the anteriorcingulate gyrus, the middle, medial, and inferior frontal gyri, andthe posterior cingulate and precuneus. This means that, as is thecase with surface analyses (9), slow waves are not always largestwhere they originate; rather, they can grow in relative currentamplitude as they travel. We also noted that slow waves were morelikely to originate in the left insula than in the right insula and thatslow waves on average involved left frontal gyri more than theyinvolved right frontal gyri. Although topographic studies of humansleep EEG have failed to detect consistent interhemispheric dif-ferences in slow wave activity (SWA, EEG power 0.5–4.5 Hz), suchdifferences have been observed in other animals (38–40). Further-more, sleep deprivation elicits larger rebounds in SWA in the lefthemisphere (8, 41).

The probabilistic origin of the slow wave is an estimate of wherethe transition from the downstate to the upstate first occurs beforeadditional areas are recruited. Although every voxel was includedin the origin of at least one slow wave, there was a hot spot of originscentered on the lateral sulci and spreading into the adjacent insular,temporal, parietal, and frontal cortices. Our findings suggest thatthe transition to the upstate can begin anywhere in the cortex butpreferentially begins in the area around the insula and superiortemporal gyrus. The reason for this regional origin selectivity isunknown. It may be that upstates that originate in these areas arebetter able to propagate to other cortical areas and manifest on thescalp. This agrees with the observation that the area around thelateral sulcus is important for the initiation and propagation ofseizures in many cases of temporal lobe epilepsy (42, 43). Intrigu-ingly, patients with insular strokes report significantly more tired-ness than patients with similar strokes that do not involve the insula(44). It is possible that impaired slow waves contribute to thisclinical presentation.

We found that slow waves were associated with large relativecurrents in the middle frontal, medial frontal, and inferior frontalcortices, as well as the anterior cingulate, cingulate, insula, posteriorcingulate, and precuneus. By contrast, sensory areas, such as thepostcentral and occipital gyri, did not contribute much current toslow waves. This result is in agreement with previous PET studiessuggesting that decreases in regional metabolism during NREMsleep occur in the anterior cingulate, insula, medial frontal cortex,and precuneus (28, 45, 46). There was also considerable overlapbetween areas that were consistently not involved in slow waves andareas that showed increases in regional metabolism. It is unclearwhy different brain areas are differentially involved in slow waves.It may be that only certain areas of the brain possess the necessarydendritic geometry to produce such large scalp potentials (13).However, increasing evidence suggests that sleep plays an impor-tant role in brain plasticity and that it can be locally regulated (11,47, 48). Therefore, one intriguing possibility is that during wakingprimary sensory areas may undergo less plasticity than the higher-order cortical areas and therefore may be less involved in slowwaves.

Sleep Slow Waves Propagate Across the Default Network. Slow wavespreferentially propagate along the anterior–posterior axis throughthe cingulate gyrus and neighboring structures. This slow wavehighway may act to functionally connect anterior high-involvementareas (the anterior cingulate, middle frontal gyri, and inferiorfrontal gyri) with posterior high-involvement areas (the precuneus,the posterior cingulate). Diffusion spectrum imaging (DSI) hasrevealed that an anatomical backbone of fibers runs along theanterior–posterior axis (49). The overlap between the highly inter-connected regions making up the backbone and the cortical areasthat are highly traveled by slow waves is striking. This suggests thata functional role of the connectional backbone revealed by DSI maybe to mediate the propagation of sleep slow waves between distinctcortical areas. Perhaps the repeated occurrence of downstates inthis connectivity backbone (49) may contribute to the disruption ofinformation transmission within the cerebral cortex that accompa-nies deep stages of NREM sleep (50). The areas showing maximalinvolvement in slow waves also show considerable overlap with thedefault network, consistent with recently published event-relatedfMRI data, triggered on the negative peak of spontaneous slowwaves (15). This functional network was first identified as a set ofbrain regions active during rest that undergo task-dependent de-creases in activity as measured by fMRI and PET (51, 52).Subsequent observations revealed that this network exhibits cor-relations in spontaneous BOLD signal (53, 54) that persist intosleep (55). The overlap between default network structures and theareas maximally involved by sleep slow waves is intriguing, consid-ering that this network has been implicated in monitoring theexternal environment, social cognition, and memory (51, 56) andthat it can be altered by sleep deprivation (57).

MethodsSubjects and Recordings. Six male subjects (ages 24–35, 3 right-handed) partic-ipatedinthespontaneoussleepportionofthestudy.Allparticipantsgavewritteninformed consent, and the experiment was approved by the University of Wis-consin Human Subjects Committee. Hd-EEG (Geodesic, 256 electrodes) was re-corded across an entire night of sleep. EEG recordings were sampled at 500 Hz.Noisy channels were replaced with spline interpolations using NetStation soft-ware (EGI). The EEG was visually scored for sleep stages (20-s epochs) on the basisof standard criteria (58). Artifacts were rejected on the basis of an automatedthreshold-crossing detection algorithm. Subjects who had TMS also had T1weighted MRIs (resolution 0.5 mm) acquired with a 3T GE Signa scanner.

Transcranial Magnetic Stimulation Data. The data for the TMS-evoked slowwaves were collected from a single male subject (age 21) as part of a previousstudy (16). In this prior study, we used a 60-channel TMS-compatible EEG net(Nexstim) to record the response to TMS in 15 sleeping males (ages 21–36).Stimulation of the sensorimotor cortex at 65% of maximal stimulator output

1612 � www.pnas.org�cgi�doi�10.1073�pnas.0807933106 Murphy et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

25,

202

0

reliably elicited high-amplitude (�80 �V) slow wave responses (negative zerocrossing to subsequent positive zero crossing duration of 0.25–1 s) that began�125 ms after delivery of a pulse. Targeting of the sensorimotor cortex wasachieved using a brain navigated stimulation of the individual subject’s MRI. Thesound of the TMS click was masked throughout the stimulation period of sleep byplaying a digitized continuous version of the click through ear phones.

Source Localization. Source localization was performed on filtered (0.5–6 Hz)EEG data, using a 4-shell head model derived from a magnetic resonance imageof an individual whose head closely approximates the Montreal NeurologicalInstitute head. A standard, coregistered set of electrode positions was used forthe construction of the forward model. The inverse matrix was calculated usingthe minimum norm least-squares (L2) method, subject to depth weighting,orientation weighting, truncated singular value decomposition regularization at10�3 to stabilize the solution, and the LAURA constraint (17). The source spacewas restricted to 2447 cortical voxels (7 mm3) that were each assigned to a gyruson the basis of the Montreal Neurological Institute probabilistic atlas. All inversemodeling was performed using GeoSource (EGI). Relative current was deter-mined by dividing the inverse model of each slow wave by the average inversemodel of several (average � 4.5) seconds of quiet waking data from the samesubject during the same recording. For TMS, the baseline period was stage 2 sleepdominated by low-voltage fast activity; there were no spontaneous slow waves,

K-complexes, or movement artifacts. Relative current calculations were per-formed using MATLAB (The Math Works).

Relative Current Maxima, Origins, and Propagation. For each slow wave, wedefined a window of 100 ms centered on the detection of the negative voltagepeak. We restricted our analysis to the behavior of relative current voxels withinthis window. We set a threshold equal to 25% of the maximum relative currentvalue achieved on this interval. Next, for each voxel, we determined the timing ofany local maxima that occurred during our time window of interest. We rejectedall maxima that did not exceed the threshold. For each voxel, we selected themaxima that occurred most closely to the voltage peak. If 2 maxima wereequidistant from the detection time, we selected the earliest maximum. Thisprocedure ensured that every voxel had at most 1 maximum. These maxima weresorted by the time they occurred. We defined the probabilistic origin as theearliest 10% of voxels to have a maximum for each wave (see SI Text). A3-dimensional gradient based upon the timing was used to calculate streamlinesfor the propagation of each slow wave (see SI Text).

ACKNOWLEDGMENTS. We thank David Balduzzi and Chiara Cirelli for theirhelpful comments. This work was supported by Grant MH077967 (to G.T.) fromthe National Institute of Mental Health and Grant NS05515 (to G.T.) from theNational Institute of Neurological Disorders and Stroke.

1. Steriade M, Nunez A, Amzica F (1993) A novel slow (� 1 Hz) oscillation of neocorticalneurons in vivo: depolarizing and hyperpolarizing components. J Neurosci 13(8):3252–3265.

2. Destexhe A, Hughes SW, Rudolph M, Crunelli V (2007) Are corticothalamic ‘up’ statesfragments of wakefulness? Trends Neurosci 30(7):334–342.

3. Amzica F, Steriade M (1995) Disconnection of intracortical synaptic linkages disruptssynchronization of a slow oscillation. J Neurosci 15(6):4658–4677.

4. Steriade M (2000) Corticothalamic resonance, states of vigilance and mentation. Neuro-science 101(2):243–276.

5. Born J, Rasch B, Gais S (2006) Sleep to remember. Neuroscientist 12(5):410–424.6. Tononi G, Cirelli C (2006) Sleep function and synaptic homeostasis. Sleep Med Rev

10(1):49–62.7. Maquet P (2001) The role of sleep in learning and memory. Science 294(5544):1048–1052.8. Finelli LA, Borbely AA, Achermann P (2001) Functional topography of the human nonREM

sleep electroencephalogram. Eur J Neurosci 13(12):2282–2290.9. Massimini M, Huber R, Ferrarelli F, Hill S, Tononi G (2004) The sleep slow oscillation as a

traveling wave. J Neurosci 24(31):6862–6870.10. Huber R, et al. (2006) Arm immobilization causes cortical plastic changes and locally

decreases sleep slow wave activity. Nat Neurosci 9(9):1169–1176.11. Huber R, Ghilardi MF, Massimini M, Tononi G (2004) Local sleep and learning. Nature

430(6995):78–81.12. Esser SK, et al. (2006) A direct demonstration of cortical LTP in humans: a combined

TMS/EEG study. Brain Res Bull 69(1):86–94.13. Nunez PL, Srinivasan R (2006) Electric Fields of the Brain: The Neurophysics of EEG (Oxford

Univ Press, Oxford).14. Volkow ND, Rosen B, Farde L (1997) Imaging the living human brain: magnetic resonance

imaging and positron emission tomography. Proc Natl Acad Sci USA 94(7):2787–2788.15. Dang-Vu TT, et al. (2008) Spontaneous neural activity during human slow wave sleep.

Proc Natl Acad Sci USA 105(39):15160–15165.16. Massimini M, et al. (2007) Triggering sleep slow waves by transcranial magnetic stimula-

tion. Proc Natl Acad Sci USA 104(20):8496–8501.17. Grave de Peralta Menendez R, Gonzalez Andino S, Lantz G, Michel CM, Landis T (2001)

Noninvasive localization of electromagnetic epileptic activity. I. Method descriptions andsimulations. Brain Topogr 14(2):131–137.

18. Pascual-Marqui RD, et al. (1999) Low resolution brain electromagnetic tomography(LORETA) functional imaging in acute, neuroleptic-naive, first-episode, productive schizo-phrenia. Psychiatry Res 90(3):169–179.

19. Pascual-Marqui RD, Michel CM, Lehmann D (1994) Low resolution electromagnetic to-mography: a new method for localizing electrical activity in the brain. Int J Psychophysiol18(1):49–65.

20. Fuchs M, Kastner J, Wagner M, Hawes S, Ebersole JS (2002) A standardized boundaryelement method volume conductor model. Clin Neurophysiol 113(5):702–712.

21. Pascual-Marqui RD (2002) Standardized low-resolution brain electromagnetic tomogra-phy (sLORETA): technical details. Methods Find Exp Clin Pharmacol 24(Suppl D):5–12.

22. Phillips C, Mattout J, Rugg MD, Maquet P, Friston KJ (2005) An empirical Bayesian solutionto the source reconstruction problem in EEG. Neuroimage 24(4):997–1011.

23. Hill S, Tononi G (2005) Modeling sleep and wakefulness in the thalamocortical system.J Neurophysiol 93(3):1671–1698.

24. Sanchez-Vives MV, McCormick DA (2000) Cellular and network mechanisms of rhythmicrecurrent activity in neocortex. Nat Neurosci 3(10):1027–1034.

25. Timofeev I, Grenier F, Bazhenov M, Sejnowski TJ, Steriade M (2000) Origin of slow corticaloscillations in deafferented cortical slabs. Cereb Cortex 10(12):1185–1199.

26. Ruohonen J, Ilmoniemi R (2002) Physical principles for transcranial magnetic stimulation.Handbook of Transcranial Magnetic Stimulation, eds Pascual-Leone A, Davey N, RothwellJ, Wasserman E, Puri B (Oxford Univ Press, New York), pp 18–29.

27. Volgushev M, Chauvette S, Mukovski M, Timofeev I (2006) Precise long-range synchroni-zation of activity and silence in neocortical neurons during slow-wave oscillations [cor-rected]. J Neurosci 26(21):5665–5672.

28. Dang-Vu TT, et al. (2005) Cerebral correlates of delta waves during non-REM sleeprevisited. Neuroimage 28(1):14–21.

29. Hallez H, et al. (2007) Review on solving the forward problem in EEG source analysis.J Neuroeng Rehabil 4:46.

30. Contreras D, Steriade M (1995) Cellular basis of EEG slow rhythms: a study of dynamiccorticothalamic relationships. J Neurosci 15(1 Pt 2):604–622.

31. Steriade M, Contreras D, Amzica F (1994) Synchronized sleep oscillations and their parox-ysmal developments. Trends Neurosci 17(5):199–208.

32. Destexhe A, Contreras D, Steriade M (1999) Spatiotemporal analysis of local field poten-tials and unit discharges in cat cerebral cortex during natural wake and sleep states.J Neurosci 19(11):4595–4608.

33. Molle M, Marshall L, Gais S, Born J (2002) Grouping of spindle activity during slowoscillations in human non-rapid eye movement sleep. J Neurosci 22(24):10941–10947.

34. Vyazovskiy VV, et al. (2008) Cortical spike activity during NREM sleep up-states reflectssleep homeostasis. Soc Neurosci Abstr: 285.7 (abstr).

35. Compte A, Sanchez-Vives MV, McCormick DA, Wang XJ (2003) Cellular and networkmechanisms of slow oscillatory activity (�1 Hz) and wave propagations in a corticalnetwork model. J Neurophysiol 89(5):2707–2725.

36. Cossart R, Aronov D, Yuste R (2003) Attractor dynamics of network UP states in theneocortex. Nature 423(6937):283–288.

37. Haider B, Duque A, Hasenstaub AR, McCormick DA (2006) Neocortical network activity invivo is generated through a dynamic balance of excitation and inhibition. J Neurosci26(17):4535–4545.

38. Goldstein L, Stoltzfus NW, Gardocki JF (1972) Changes in interhemispheric amplituderelationships in the EEG during sleep. Physiol Behav 8(5):811–815.

39. Roth C, Achermann P, Borbely AA (1999) Frequency and state specific hemispheric asym-metries in the human sleep EEG. Neurosci Lett 271(3):139–142.

40. Vyazovskiy VV, Borbely AA, Tobler I (2002) Interhemispheric sleep EEG asymmetry in therat is enhanced by sleep deprivation. J Neurophysiol 88(5):2280–2286.

41. Ferrara M, De Gennaro L, Curcio G, Cristiani R, Bertini M (2002) Interhemispheric asym-metry of human sleep EEG in response to selective slow-wave sleep deprivation. BehavNeurosci 116(6):976–981.

42. Isnard J, Guenot M, Ostrowsky K, Sindou M, Mauguiere F (2000) The role of the insularcortex in temporal lobe epilepsy. Ann Neurol 48(4):614–623.

43. Ryvlin P, et al. (2006) Nocturnal hypermotor seizures, suggesting frontal lobe epilepsy, canoriginate in the insula. Epilepsia 47(4):755–765.

44. Manes F, Paradiso S, Robinson RG (1999) Neuropsychiatric effects of insular stroke. J NervMent Dis 187(12):707–712.

45. Hofle N, et al. (1997) Regional cerebral blood flow changes as a function of delta andspindle activity during slow wave sleep in humans. J Neurosci 17(12):4800–4808.

46. Braun AR, et al. (1997) Regional cerebral blood flow throughout the sleep-wake cycle. AnH2(15)O PET study. Brain 120(Pt 7):1173–1197.

47. Gais S, Born J (2004) Declarative memory consolidation: mechanisms acting during humansleep. Learn Mem 11(6):679–685.

48. Tononi G, Cirelli C (2003) Sleep and synaptic homeostasis: a hypothesis. Brain Res Bull62(2):143–150.

49. Hagmann P, et al. (2008) Mapping the structural core of human cerebral cortex. PLoS Biol6(7):e159.

50. Massimini M, et al. (2005) Breakdown of cortical effective connectivity during sleep.Science 309(5744):2228–2232.

51. Buckner RL, Andrews-Hanna JR, Schacter DL (2008) The brain’s default network: anatomy,function, and relevance to disease. Ann NY Acad Sci 1124:1–38.

52. Raichle ME, et al. (2001) A default mode of brain function. Proc Natl Acad Sci USA98(2):676–682.

53. Fox MD, et al. (2005) The human brain is intrinsically organized into dynamic, anticorre-lated functional networks. Proc Natl Acad Sci USA 102(27):9673–9678.

54. Greicius MD, Krasnow B, Reiss AL, Menon V (2003) Functional connectivity in the restingbrain: a network analysis of the default mode hypothesis. Proc Natl Acad Sci USA100(1):253–258.

55. Horovitz SG, et al. (2008) Low frequency BOLD fluctuations during resting wakefulnessand light sleep: a simultaneous EEG-fMRI study. Hum Brain Mapp 29(6):671–682.

56. Raichle ME, Snyder AZ (2007) A default mode of brain function: a brief history of anevolving idea. Neuroimage 37(4):1083–1090.

57. Gujar N, Hu P, Yoo S, Walker M (2008) The sleep-deprived brain at rest: a doubledissociation in abnormal ‘‘default-mode’’ activity. Sleep 31(Abstract Suppl):A110.

58. Rechtschafflen A, Kales A (1968) A Manual of Standardized Terminology, Techniques, andScoring System for Sleep Stages of Human Subjects (National Institutes of Health,Washington, DC).

59. Riedner BA, et al. (2007) Sleep homeostasis and cortical synchronization: III. A high-densityEEG study of sleep slow waves in humans. Sleep 30(12):1643–1657.

Murphy et al. PNAS � February 3, 2009 � vol. 106 � no. 5 � 1613

NEU

ROSC

IEN

CE

Dow

nloa

ded

by g

uest

on

Oct

ober

25,

202

0