Decoupling of the brain’s default mode network during deep sleep Silvina G. Horovitza,b,1, Allen R. Braunc, Walter S. Carrd, Dante Picchionie, Thomas J. Balkine, Masaki Fukunagab, and Jeff H. Duynb aHuman Motor Control Section, Medical Neurology Branch, National Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, MD 20892; bAdvanced MRI, Laboratory of Functional and Molecular Imaging, National Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, MD 20892; cLanguage Section, Voice, Speech and Language Branch, National Institute on Deafness and Other Communication Disorders, National Institutes of Health, Bethesda, MD 20892; dNaval Medical Research Center, Silver Spring, MD 20910; and eDepartment of Behavioral Biology, Walter Reed Army Institute of Research, Silver Spring, MD 20910 Edited by Leslie G. Ungerleider, National Institute of Health, Bethesda, MD, and approved May 7, 2009 (received for review February 9, 2009) The recent discovery of a circuit of brain regions that is highly activity in humans during wakefulness and deep sleep, defined by active in the absence of overt behavior has led to a quest for standard spectral and polysomnographic criteria. Deep sleep is revealing the possible function of this so-called default-mode a phase of sleep during which the arousal threshold to external network (DMN). A very recent study, finding similarities in awake stimuli is maximal (18) and during which the level of conscious- humans and anesthetized primates, has suggested that DMN ness awareness is presumed minimal (16, 19). Thus, the deep- activity might not simply reflect ongoing conscious mentation but sleep condition allowed strong, ecologically valid modulation of rather a more general form of network dynamics typical of com- the level of consciousness without resorting to the less-natural plex systems. Here, by performing functional MRI in humans, it is conditions of absence seizure, coma, or anesthesia (20). shown that a natural, sleep-induced reduction of consciousness is Similar to previous studies, activity levels were inferred from reflected in altered correlation between DMN network compo- low-frequency (Ͻ0.08 Hz) fMRI signal fluctuation levels, and nents, most notably a reduced involvement of frontal cortex. This both coherence and amplitude of activity in the various regional suggests that DMN may play an important role in the sustenance components of the DMN were studied. Sleep level was assessed of conscious awareness. from concurrently acquired EEG data. EEG ͉ fMRI ͉ resting state ͉ connectivity ͉ consciousness Results ll except the most simple reflexive behaviors are mediated Data Quality. Good quality EEG–fMRI data were collected from Aby and reflected in the chemical and electrical signaling 12 of the 18 subjects studied (supporting information Table S1). within the brain’s circuitry. The ability to monitor the spatial Of these, 7 experienced at least 15 min of continuous deep sleep distribution of such activity with modern neuroimaging tech- required for the analysis. An example of the EEG during the 3-h niques such as PET (1) and functional MRI (fMRI) (2) has scan session for a subject is shown in Fig. S1. Sleep structure for greatly expanded knowledge of the brain’s architecture and the 7 subjects included in the deep-sleep analysis is described in functioning, identifying the brain circuitry that underlies a Table S2. variety of processes and behaviors. fMRI studies have demonstrated that a substantial level of DMN During Wake: Resting-State Network. In a subgroup of volun- brain activity also occurs in the absence of overt or task-related teers (n ϭ 5), we collected resting fMRI, during day hours and behavior (3–8). This spontaneous activity has spatial and tem- without sleep deprivation, using the same imaging parameters. poral characteristics that resemble those of intracortical electri- DMN by correlating each voxel’s time course with the average time cal and optical recordings (9–11). In particular, a network of course of the posterior cingulate cortex (PCC) defined according to brain regions has been identified in which activity is actually an anatomic template (21). During wake, the distribution of co- increased in absence of cognitively demanding tasks (12, 13). herent DMN activity was similar to that described in previous This network, termed the default mode network (DMN) (14), was reports (7, 8, 22). Network activity included the seed region and its initially thought to be associated with conscious, internally surroundings (PCC/precuneus [P]), bilateral inferior parietal mediated cognitive processes—for example, self-referential be- cortices (IPC) and angular gyri (AG), and frontal regions havior, moral reasoning, recollection, and imagining the future including medial prefrontal cortex (MPFC) and anterior cingu- (14)—cognitive processes thought to be unique to humans. late cortex (ACC) (Fig. 1A and Table 1). Recent studies have demonstrated that DMN connectivity per- sists during light sleep (8, 15); however, this persistence could be DMN During Deep Sleep. During deep sleep, only partial network expected because self-reflective thoughts do not abruptly cease involvement was observed, with apparent decoupling of frontal but rather decrease gradually as a person falls asleep, to the point areas from the DMN (Fig. 1B and Table 1). Whereas PCC and of being virtually absent during the deepest stages of sleep (16). Adding to these intriguing findings is a recent study of the IPC/AG correlations seem to strengthen, the correlations be- DMN in anesthetized nonhuman primates (17). Although the tween PCC and MPFC/ACC became nonsignificant. general extent of DMN activity was found to be similar to that in awake human, some notable differences can be seen from this Author contributions: S.G.H., A.R.B., W.S.C., D.P., T.J.B., M.F., and J.H.D. designed research; study’s data [see Fig. 4 in Vincent et al. (17)], including a S.G.H., A.R.B., W.S.C., and D.P. performed research; S.G.H. and M.F. contributed new somewhat reduced spatial involvement of the medial prefrontal reagents/analytic tools; S.G.H., D.P., and M.F. analyzed data; and S.G.H., A.R.B., W.S.C., D.P., cortex. The basis of such differences is not clear and might reflect T.J.B., M.F., and J.H.D. wrote the paper. (i) species-specific features/differences in cognitive abilities or The authors declare no conflict of interest. (ii) alterations in consciousness level due to the effects of the This article is a PNAS Direct Submission. anesthesia. 1To whom correspondence should be addressed. E-mail: [email protected]. To clarify these issues, the present study aimed at investigating This article contains supporting information online at www.pnas.org/cgi/content/full/ the relationship between level of consciousness and DMN 0901435106/DCSupplemental. 11376–11381 ͉ PNAS ͉ July 7, 2009 ͉ vol. 106 ͉ no. 27 www.pnas.org͞cgi͞doi͞10.1073͞pnas.0901435106 Downloaded by guest on September 26, 2021 Fig. 1. The default mode network during wake and sleep. Composite maps showing correlations with PCC during (A) wake and (B) deep sleep, and their significant difference as determined from statistical t test (C). A significant reduction of involvement of frontal regions is seen during deep sleep, whereas the posterior cingulate–inferior parietal correlations are preserved. The Z maps in A and B are both thresholded at Z ϭϮ5.0; the t map in C is thresholded at t ϭ Ϯ3.5. Both positive (yellow–red) and negative (blue) correlations are shown. Z values, t values, and Talairach coordinates of all significant clusters are reported in Tables 1 and 2. Comparison of the DMN in Wake and Deep Sleep. Statistical com- due to reduced frontal activity, given that the amplitude of the parisons of DMN connectivity during wake and deep sleep show fluctuations within that region remained unchanged (Table 4). a number of differences, most notably in the MPFC/ACC area In fact, robust local coherence of spontaneous activity within all (Fig. 1C and Table 2). Whereas wakefulness was characterized of the component regions of the DMN persisted during deep by significant positive correlation between PCC and MPFC/ACC sleep. regions, during deep sleep that correlation was absent. Quanti- Thus the differences we report are not due to changes in the tative revealed a significant reduction in the interconnections frontal regions per se. Nor are they due to seed selection: when between the MPFC and other regions during deep sleep. the seed was placed in the frontal area (Fig. S2) the results were Whereas significant correlations were seen between MPFC and similar to those obtained with the seed in PCC. As with the seed PCC (r ϭ 0.36, P Ͻ 0.001) and all MPFC and right IPC/AG (r ϭ in PCC, seed placement in the frontal region indicated a fully 0.38, P Ͻ 0.04) during the wake state, both became nonsignif- connected DMN in the wake condition (Fig. S2a), whereas it led icant during deep sleep (Fig. 2 and Table 3). The decoupling of to a dissociation between frontal and parietal areas during deep frontal and posterior regions during deep sleep was not simply sleep (Fig. S2b). At the same time the correlations within the Table 1. Center of mass for voxel-by-voxel correlations with PCC seed in wake and deep sleep Talairach coordinates Cluster size Peak Z Region xyz (mm3) value Wake (n ϭ 5) Posterior cingulate Ϫ2 Ϫ47 23 43,984 31.29 R. angular gyrus Ϫ36 68 33 7,409 9.76 L. angular gyrus 50 56 29 6,763 7.86 R. inferior frontal gyrus Ϫ46 Ϫ7 23 4,342 Ϫ7.11 R. post-central gyrus Ϫ55 29 38 3,944 Ϫ7.21 R. parahippocampal/fusiform Ϫ22 44 Ϫ11 3,629 Ϫ7.33 Medial frontal gyrus Ϫ6 Ϫ61 12 2,920 7.56 L.