Breakdown of Cortical Effective Connectivity During Sleep

See allHide authors and affiliations

Science  30 Sep 2005:
Vol. 309, Issue 5744, pp. 2228-2232
DOI: 10.1126/science.1117256


When we fall asleep, consciousness fades yet the brain remains active. Why is this so? To investigate whether changes in cortical information transmission play a role, we used transcranial magnetic stimulation together with high-density electroencephalography and asked how the activation of one cortical area (the premotor area) is transmitted to the rest of the brain. During quiet wakefulness, an initial response (∼15 milliseconds) at the stimulation site was followed by a sequence of waves that moved to connected cortical areas several centimeters away. During non–rapid eye movement sleep, the initial response was stronger but was rapidly extinguished and did not propagate beyond the stimulation site. Thus, the fading of consciousness during certain stages of sleep may be related to a breakdown in cortical effective connectivity.

When awakened early in the night from non–rapid eye movement (NREM) sleep, people often report little or no conscious experience (1). It was first thought that this fading of consciousness was due to the brain shutting down. However, although brain metabolism is reduced, the thalamocortical system remains active, with mean firing rates close to those that occur during quiet wakefulness (2). Moreover, coherent or synchronized activity continues to be detected among distant cortical areas (35), and sensory signals still reach the cerebral cortex (6). Why, then, does consciousness fade?

Recently we have proposed that consciousness depends critically not so much on firing rates, synchronization at specific frequency bands, or sensory input per se, but rather on the brain's ability to integrate information, which is contingent on the effective connectivity among functionally specialized regions of the thalamocortical system (7). Effective connectivity refers to the ability of a set of neuronal groups to causally affect the firing of other neuronal groups within a system (8). The fading of consciousness during NREM sleep episodes early in the night, evidenced by short or blank reports of cognitive activity upon awakening (1), would then be associated with an impairment of cortical effective connectivity.

To test this prediction, we used a combination of navigated transcranial magnetic stimulation (TMS) and high-density electroencephalography (HD-EEG) to measure the brain response to the direct perturbation of a chosen cortical region noninvasively and with good spatiotemporal resolution (9, 10). Using TMS/EEG to investigate critical differences in the functioning of the waking and sleeping brain offers several advantages. Unlike sensory stimulation, direct cortical stimulation does not activate the reticular formation and bypasses the thalamic gate. Thus, it directly probes the ability of cortical areas to interact, unconfounded by peripheral effects. Also, since study subjects reported that they were not aware of the TMS pulse, neural responses are not contaminated by reactions that may result from becoming aware of the stimulation. Most important, the combination of TMS and HD-EEG dissociates effective connectivity (causal interactions) from functional connectivity [temporal correlations (8)].

Using a 60-channel TMS-compatible EEG amplifier, we recorded TMS-evoked brain responses while six subjects, lying with eyes closed on a reclining chair, progressed from wakefulness to NREM sleep. By means of magnetic resonance image (MRI)–guided estimation of the electric field induced on the surface of the brain (Fig. 1A), we targeted TMS to the rostral portion of the right premotor cortex. This is an area with extensive corticocortical connections that can be conveniently stimulated without eliciting muscle artifacts. Stimuli were delivered at random intervals (between 2 and 2.3 s) with intensity below the motor threshold (90%), resulting in a maximum electric field at the cortical target of between 75 and 84 V/m. We took special care to reduce the amount of auditory and somatosensory stimulation associated with each TMS pulse (11).

Fig. 1.

Navigated brain stimulation and EEG recordings during TMS. (A) The estimated electric field induced by TMS on the cortical surface in one subject is color-coded. The red area indicates the location of the maximal electric field strength (in this case, 81 V/m) and corresponds to the coordinates of the rostral premotor cortex, as identified on the three orthogonal projections of the subject's MRI. The brown pins represent the digitized electrodes. (B) Multichannel EEG recorded during wakefulness and NREM sleep while TMS (red) was delivered.

As shown in Fig. 1B, TMS did not interfere conspicuously with ongoing wake or sleep EEG patterns nor did it cause visible artifacts. However, TMS elicited a time-locked response that was visible on a single-trial basis and that changed markedly from wakefulness to sleep. Figure 2A displays the single-trial responses recorded from one electrode located under the stimulator during a transition from wakefulness through stage 1 to NREM (stages 2 and 3) sleep (in the same subject as in Fig. 1). Figure 2B shows the averages calculated from the single trials collected in these three vigilance states. During wakefulness, TMS induced a sustained response made of recurrent waves of activity. Specifically, a sequence of time-locked high-frequency (20 to 35 Hz) oscillations occurred in the first 100 ms and was followed by a few slower (8 to 12 Hz) components that persisted until 300 ms. As soon as the subjects transitioned into stage 1 sleep, the TMS-evoked response grew stronger at early latencies but became shorter in duration: The amplitude of the initial components increased by 50 to 85% between 0 and 40 ms, whereas the subsequent waves were markedly dampened and fell below prestimulus noise levels (3 SD from the prestimulus baseline mean) within the first 150 to 200 ms. With the onset of NREM sleep, the brain response to TMS changed markedly. The initial wave doubled in amplitude and lasted longer. After this large wave, no further TMS-locked activity could be detected, except for a slight negative rebound between 80 and 140 ms. Specifically, fast waves, still visible during stage 1, were completely obliterated, and all TMS-evoked activity had ceased by 150 ms.

Fig. 2.

Changes in the TMS-evoked response during shifts in the state of vigilance. (A) Single trials recorded from one channel located under the stimulator while the subject (the same as in Fig. 1) transitioned from wakefulness through stage 1 to NREM sleep. Single-trial EEG data (filtered from 4 to 100 Hz) are color-coded for voltage. (B) Averaged TMS-evoked responses (filtered from 1 to 100 Hz) obtained during the three states of vigilance. The horizontal pink bands indicate the significance level (3 SD from the mean prestimulus voltage).

To better characterize the underlying neural events, we calculated the spatiotemporal dynamics of the currents induced by TMS in the cerebral cortex. We digitized and coregistered electrode positions to each subject's MRI, and we constructed a realistic head model. We then estimated current density on the cortical surface by using the weighted minimum norm least-squares method (11). Figure 3 shows the average responses recorded from all channels during wakefulness and NREM sleep in the same subject shown in Figs. 1 and 2. At early latencies, during both wakefulness and NREM sleep, TMS induced a clear dipolar voltage configuration that was centered under the coil and corresponded to maximum cortical activation in ipsilateral area 6. During wakefulness, this initial response was followed for about 300 ms by multiple waves of activity associated with rapidly changing configurations of scalp potentials. Current maxima shifted over time from the stimulation target to contralateral area 6, bilateral area 9, contralateral area 8, and ipsilateral area 7. The rostral premotor cortex has extensive transcallosal connections (12) and is linked to prefrontal areas (13). Thus, during wakefulness, the perturbation of the rostral premotor cortex was followed by spatially and temporally differentiated patterns of activation that appeared to propagate along its anatomical connections. In striking contrast, during NREM sleep the location of maximum current density remained confined to the stimulated area.

Fig. 3.

Spatiotemporal dynamics of scalp voltages and cortical currents evoked by TMS during wakefulness and sleep. (A and A′) Averaged TMS-evoked potentials recorded at all electrodes, superimposed in a butterfly diagram (black traces; the horizontal red line indicates the average reference), for the same subject as in Figs. 1 and 2. The time of TMS is marked by a vertical red bar. The red portions of the traces indicate the times at which TMS induced a significant response (see supporting online material for calculation details). Source modeling was performed at the local maxima of field power within periods of significant activity. (B and B′) Three-dimensional contour voltage maps (red, positive; blue, negative; step = 0.6 μV for wakefulness and 1 μV for NREM sleep). (C and C′) Corresponding current density distributions plotted on the cortical surface. At each time point, the results of the L2 Norm (see methods) were auto-scaled and thresholded at 80% to highlight maximum current sources (CDR, current density reconstruction).

As shown in Fig. 4, this breakdown in effective connectivity during sleep was evident and reproducible in all six subjects. We estimated current density whenever the global power of the evoked field was higher (>6 SD) than mean prestimulus levels and plotted the location of the strongest TMS-evoked activation on each subject's cortical surface, color-coded according to its latency (11). During wakefulness, the site of maximum activation moved back and forth among premotor and prefrontal areas in both hemispheres and, in some subjects, it also involved the motor and posterior parietal cortex. During NREM sleep, by contrast, the activity evoked by TMS did not propagate in space and time in any of the subjects. In two subjects, we were also able to stimulate the parietal cortex (area 5), and we found a similar impairment of intracortical information transmission during NREM sleep (fig. S2). Thus, although TMS during sleep elicits an initial response that is even stronger than during wakefulness, this response remains localized, does not propagate to connected brain regions, dissipates rapidly, lacks high-frequency components, and is stereotypical regardless of stimulation site.

Fig. 4.

Spatiotemporal cortical current maps during wakefulness and NREM sleep in all six subjects. Black traces represent the global mean field powers, and the horizontal yellow lines indicate significance levels. For each significant time sample, maximum current sources were plotted and color-coded according to their latency of activation (light blue, 0 ms; red, 300 ms). The yellow cross marks the TMS target on the cortical surface.

Various mechanisms could account for the enhancement of early TMS-EEG responses in sleep, including a stronger driving force in hyperpolarized postsynaptic neurons (14), an increased discharge synchrony of cortical populations (15), a reduction in synaptic depression (16, 17), and thalamic bursting triggered by the TMS-induced corticothalamic volley (18). These mechanisms may also produce the enhancement of cortical components of visual, auditory, and somatosensory evoked potentials that has been reported during NREM sleep (6).

What causes the dramatic breakdown in cortical effective connectivity during sleep? During NREM sleep, cortical neurons are depolarized and fire tonically just as in quiet wakefulness, but these depolarized up-states are interrupted by short hyperpolarized down-states when neurons remain silent (19). The transition from up- to down-states appears to be due to depolarization-dependent potassium currents that increase with the amount of prior activation (19). Perhaps because of this bi-stability of cortical networks during NREM sleep (16, 17), any local activation, whether occurring spontaneously or induced by TMS, will eventually trigger a local down-state that prevents further propagation of activity. Alternatively, the block may occur in the thalamus, whose neurons, when hyperpolarized, fire a single burst in response to corticothalamic volleys and then enter a prolonged inhibitory rebound (20). Finally, there may be sleep-related changes in the balance between excitation and inhibition (21), as suggested by paired-pulse TMS studies (22).

Whatever the precise mechanisms, they are most likely engaged by the progressive reduction of the firing of diffuse neuromodulatory systems that occurs when we fall asleep (23). Indeed, the blockade of intracortical signaling did not begin suddenly, and the spatiotemporal pattern of cortical activation during stage 1 sleep was intermediate between those of wakefulness and NREM sleep (fig. S3). Specifically, during stage 1 sleep, the TMS-evoked response propagated from the right premotor cortex to the homotopic contralateral site within the first few tens of milliseconds; however, this initial activation was not sustained nor did it reach prefrontal or parietal areas.

By using a combination of TMS and HD-EEG, we have found evidence for a breakdown of transcallosal and long-range effective connectivity during NREM sleep. This breakdown in the ability of cortical areas to interact effectively contrasts with the persistence or increase in interhemispheric and interareal broadband coherence that can be observed in EEG studies of sleep (3, 24). Thus, an impairment in the ability to integrate information among specialized thalamocortical modules—a proposed theoretical requirement for consciousness (7)—may underlie the fading of consciousness in NREM sleep early in the night. It will be important to see whether cortical effective connectivity recovers in part during late-night sleep, especially during REM sleep, a time at which conscious reports become long and vivid (1). More generally, probing the brain's effective connectivity directly may prove useful in pharmacologically induced unconsciousness and in several psychiatric and neurological conditions in which consciousness is affected and neural interactivity may be compromised above and beyond neural activity and neural synchrony (25).

Supporting Online Material

Materials and Methods

Figs. S1 to S3

References and Notes

References and Notes

View Abstract

Stay Connected to Science

Navigate This Article