Research ArticleConsciousness

Slow-Wave Activity Saturation and Thalamocortical Isolation During Propofol Anesthesia in Humans

See allHide authors and affiliations

Science Translational Medicine  23 Oct 2013:
Vol. 5, Issue 208, pp. 208ra148
DOI: 10.1126/scitranslmed.3006007

Abstract

The altered state of consciousness produced by general anesthetics is associated with a variety of changes in the brain’s electrical activity. Under hyperpolarizing influences such as anesthetic drugs, cortical neurons oscillate at ~1 Hz, which is measurable as slow waves in the electroencephalogram (EEG). We have administered propofol anesthesia to 16 subjects and found that, after they had lost behavioral responsiveness (response to standard sensory stimuli), each individual’s EEG slow-wave activity (SWA) rose to saturation and then remained constant despite increasing drug concentrations. We then simultaneously collected functional magnetic resonance imaging and EEG data in 12 of these subjects during propofol administration and sensory stimulation. During the transition to SWA saturation, the thalamocortical system became isolated from sensory stimuli, whereas internal thalamocortical exchange persisted. Rather, an alternative and more fundamental cortical network (which includes the precuneus) responded to all sensory stimulation. We conclude that SWA saturation is a potential individualized indicator of perception loss that could prove useful for monitoring depth of anesthesia and studying altered states of consciousness.

INTRODUCTION

Pharmacologically imposed unconsciousness allows controlled study of mechanisms of perception and conditions under which perception fails, such as anesthesia (or sleep). Despite the hundreds of thousands of anesthetics administered daily to patients, there is no robust individualized biomarker of perceptual awareness or its failure. The incidence of intraoperative awareness is consistently reported at 0.1 to 1.0% (1). Nevertheless, inferences about the awareness of an individual during administration of an anesthetic can be only made on the basis of pharmacodynamics (2) or scalp electroencephalogram (EEG) (3) from population data. The risk of this probabilistic approach is that perceptual awareness in an individual cannot be categorically ruled out. There is a need for an individualized neurophysiological marker that is linked to perception failure.

The widely supported “information integration” theory of consciousness holds that construction of a percept depends on synchronous activity in separate distributed nodes, each processing the discrete components of that percept (46). Breakdown of communication between these network nodes is thought to underlie loss of consciousness (79). Several studies exploring altered states of consciousness have identified a specific reduction in thalamic activity during the transition to loss of consciousness. This has led to the concept of a “thalamocortical switch,” whereby thalamic suppression acts as a barrier to further processing within cortical regions (1012). However, this attractive and intuitive construct has been questioned (13). The reduction during impaired consciousness in thalamic metabolism and blood flow could be secondary to functional thalamic deafferentation as a result of reduced inputs from brainstem or cortex. Alternatively, failure of information integration could be caused by a disruption of cortico-cortical communication (14) and reduction in thalamic connectivity (9, 15). Thus, thalamic deactivation may be a consequence of impaired consciousness rather than the cause.

The cerebral cortex itself provides the dominant afferent input to the thalamus, with corticothalamic neurons outnumbering thalamocortical neurons by an order of magnitude (16). These cortical neurons are subject to a number of hyperpolarizing and depolarizing influences that govern their capacity to respond to salient neural signals. The resting membrane potential in neocortical neurons is driven by physiological, circadian, pathological, or pharmacological changes in the output of sleep and wake-promoting centers. When the balance of these influences favors hyperpolarization, neocortical neurons become bistable oscillating at ~1 Hz between depolarized “UP” states of wake-like firing and hyperpolarized “DOWN” states of neuronal silence. These oscillations synchronize into traveling waves across the cortical surface and are observed on the scalp as EEG slow waves (17). This slow oscillation is a cortically generated rhythm, surviving callosotomy and thalamectomy and appearing even in isolated cortical tissue preparations (18). The slow oscillation is not an idling rhythm but is fundamental to sleep homeostasis (19). In later stages of non–rapid eye movement (non-REM) sleep, which are defined by the presence of slow waves, thalamic and cortical responses to stimulation become less consistent (20).

Although anesthesia and sleep are fundamentally different states, they do share some common mechanisms (21). Loss of consciousness from propofol-induced anesthesia has been linked to the appearance of slow waves that resemble the slow waves in non-REM sleep and have been found to share similar patterns of origin and propagation (22, 23).

Here, we hypothesize that the point at which the amplitude of slow-wave activity (SWA) reaches its maximum in an individual undergoing anesthesia marks an unresponsive yet active (unconscious) brain. To test this hypothesis, we analyzed simultaneously recorded EEG and functional magnetic resonance imaging (fMRI) data from subjects undergoing an ultraslow diminution of consciousness with the anesthetic agent propofol.

RESULTS

SWA saturates with deepening unconsciousness

While recording the EEGs of 16 subjects undergoing slow induction to loss of consciousness, we noted frequency-specific changes in EEG activity at both a group and individual level (Figs. 1 and 2). After 10 min of resting data acquisition, rising propofol concentrations in the brain were accompanied by changes in β (15 to 30 Hz), α (8 to 14 Hz), θ (4 to 8 Hz), and δ (0 to 4 Hz) activity and particularly by pronounced changes in slow-wave (0.5 to 1.5 Hz) activity. These activity changes reversed after the infusion was stopped.

Fig. 1. Temporal changes in EEG spectral power during propofol administration.

In experiment 1, 16 volunteers were assessed for EEG spectral power (measured in decibels, shown color-coded) for frequencies from 0.5 to 30 Hz (left-hand axis) as propofol ESC increased and decreased (dashed black line). Noxious laser stimuli, tones, and an auditory word task were delivered to the subjects during the phases containing stimulation (ii and iv). The data (expressed as grand mean change) have four phases: (i) resting awake (no stimuli), (ii) induction (with stimulation), (iii) peak dose (no stimuli), and (iv) emergence (with stimulation). Frequency bands of interest are as follows: SWA, 0.5 to 1.5 Hz; θ, 4 to 8 Hz; α, 8 to 14 Hz; β, 15 to 30 Hz.

Fig. 2. Relationship between slow-wave power, responsiveness, and propofol concentration in individual subjects.

The graph for each subject from experiment 1 presents the percentage of SWA power (0.5 to 1.5 Hz/0.5 to 30 Hz, red line) and the times at which loss (LOBR) (first blue line) and recovery (ROBR) (second blue line) of behavioral responses occurred (blue lines). The propofol ESC is shown as a black line in each plot.

The central nervous system propofol effect site concentration (ESC) at which each subject ceased to respond to the word task [loss of behavioral response (LOBR)] and the concentration at which he or she first responded again after the infusion was discontinued [recovery of behavioral response (ROBR)] showed significant interindividual variation [LOBR range: 0.7 to 2.75 μg/ml, mean (SD): 1.57 (0.6) μg/ml; ROBR range: 0.7 to 2.1 μg/ml, mean (SD): 1.23 (0.45) μg/ml]. Each individual’s EEG between these defined time points (LOBR and ROBR) was used for subsequent group analyses.

When the propofol concentration was further increased after the LOBR occurred, an intriguing phenomenon was revealed in the EEG data. In each subject, SWA increased sharply after the LOBR until reaching saturation, even though propofol concentrations continued to rise (Fig. 2). After discontinuation of the infusion, as drug levels dropped, the slow-wave power began to decrease and returned to baseline levels before recovery of the behavioral response. Hence, we were able to detect and define a second transition in addition to LOBR, namely, the onset of saturation in SWA for individual subjects (Fig. 2).

Group analysis performed by anchoring each individual’s EEG analysis to the behavioral transitions (LOBR and ROBR) allowed a clear identification of this additional neurophysiological transition of SWA saturation (Fig. 3A). The SWA was topographically distributed predominantly within frontal brain regions (Fig. 3B) during the unresponsive period, consistent with localization of SWA in natural sleep (17).

Fig. 3. SWA saturation after the LOBR during propofol administration.

(A) Group changes in normalized SWA power during an ultraslow induction of and emergence from propofol sedation. EEG power in the 0.5- to 1.5-Hz band for each subject (as shown in Fig. 2) was normalized to each individual’s maximum power across the 0.5- to 30-Hz frequency band and divided into t-ROIs on the basis of observed behavior during experiment 1. This behavioral banding removes the effect of interindividual differences in the dose-response relationship. At the start of each t-ROI, all subjects (n = 16) contributed to the data displayed. However, at the end of the t-ROIs “Responding during induction only” and “Unresponsive during recovery,” only four subjects are contributing. (B) Topographical brain maps detailing SWA in each of the t-ROIS are shown. EEG data are presented with respect to an average reference channel.

A voxel-based morphometric analysis was performed by using each individual’s structural MRI scan to calculate the volume of the gray matter in each individual’s prefrontal cortex. We observed a positive correlation between the peak SWA and total gray matter volume in the frontal cortex, suggesting that the maximum allowable SWA is dictated by the quantity of potentially recruitable neurons (fig. S2).

The UP state of the slow oscillation is characterized by spindle activity, reverberations of thalamocortical neurons at α band frequencies (7 to 14 Hz) (24). We observed a frequency-dependent change in spindle activity with increasing propofol concentrations: lower-frequency (10 to 12 Hz) spindles located in the prefrontal cortex rose in prevalence after LOBR (fig. S3).

SWA saturation reflects the transition to thalamocortical isolation

In experiment 2, 12 of the original 16 subjects completed an identical protocol in the fMRI scanner while EEG and fMRI data were collected simultaneously. The EEG data confirmed our findings from experiment 1 that SWA saturation occurred after the LOBR while propofol concentrations were still rising (Fig. 4B). The fMRI data were then divided for analysis into three temporal regions of interest (t-ROIs): (i) before LOBR, (ii) between the LOBR and the onset of SWA saturation, and (iii) during SWA saturation. These three t-ROIs were based on each individual’s simultaneously acquired behavioral and EEG data (Fig. 4A).

Fig. 4. Brain network responses to stimuli during propofol administration (experiment 2).

In 12 of the 16 subjects used for experiment 1, fMRI-BOLD data (n = 12) and simultaneous, stimulus-evoked behavioral responses and EEG SWA were collected. (A) Behavioral responsiveness during propofol administration in experiment 2 showing the presence or absence of motor responses to the auditory word task. (B) SWA during propofol administration in experiment 2. Slow-wave power is plotted as a percentage of the total EEG power in the SWA band (0.5 to 1.5 Hz/0.5 to 30 Hz) as residual cardioballistographic signal in the 0.5- to 1.5-Hz band remains after artifact removal. A residual high-frequency scanner artifact also remains and contributes to the total EEG power (0.5 to 30 Hz). SWP, slow-wave peak (onset of SWA saturation); t-ROIs are defined as in experiment 1 and Fig. 3. (C) fMRI-BOLD activation maps (group mean values), showing areas of significant stimulus-evoked BOLD responses to noxious stimuli (red) and auditory word task (blue) averaged over each t-ROI. BOLD responses before LOBR (left), between LOBR and SWP (middle), and during SWA saturation (right) are shown. Mixed-effects analysis, cluster threshold–corrected (Z > 2.3, P < 0.05).

Although the behavioral transition (LOBR) was associated with a significant reduction in activation in several cortical areas relevant to auditory and nociceptive inputs (for example, secondary somatosensory cortex, insula, and cingulate cortex), significant activity persisted in the thalamus and primary cortical processing regions of the now unresponsive subjects (Fig. 4C, left and middle). Moreover, the EEG transition to SWA saturation was specifically associated with loss of thalamic and primary cortical activation (Fig. 4C, right). However, beyond this thalamocortical isolation, the brain was neither inactive nor unresponsive. Both auditory and noxious stimuli were associated with evoked BOLD (blood oxygen level–dependent) signal changes. This activation was not within the sensory thalamocortical system but involved a network of regions: the precuneus and the posterior parietal and prefrontal cortices.

Disruption of these thalamocortical networks at SWA saturation was further confirmed by a mixed-effects mean group subtraction across the time of the transition to SWA saturation (Fig. 5). With this method, we identified the brain regions that showed a statistically significant reduction in activity to sensory stimulation when comparing the mean activation in the behaviorally unresponsive pre–SWA saturation and the SWA saturation periods. We found modality-specific reductions in the thalamus and in the sensory-specific regions of the primary somatosensory cortex for noxious laser stimulation and in Heschl’s gyrus for auditory stimuli.

Fig. 5. Reduction in sensory-specific activity in thalamocortical areas at SWA saturation.

(A and B) Evoked BOLD responses to (A) noxious laser stimuli to the lower leg (red/yellow) and (B) auditory word task (blue) in unresponsive subjects before and after SWA (activation in the behaviorally unresponsive pre–SWA saturation period > SWA saturation period). Mixed-effects analysis, cluster threshold–corrected. Paired t tests (Z > 2.3, P < 0.05).

DISCUSSION

Slow waves are emergent oscillations that appear when cortical neurons become hyperpolarized under the balance of sleep-wake drivers (25). This rhythm dominates the EEG when an anesthetic drug, such as propofol, pharmacologically imposes and sustains this hyperpolarization (21). Recent evidence in both humans and animals from sleep and anesthesia studies suggests that the characteristics of the slow wave may influence the central processing of stimuli in altered states of consciousness (20, 26, 27). Using EEG recordings in healthy volunteers during an ultraslow induction of propofol-induced loss of consciousness, we have defined an additional electrophysiological transition during the unresponsive period, before burst suppression, based on the relative (percentage) power in the SWA band. We propose that the increase in SWA to apparent saturation reflects an increasing number of cortical neurons oscillating synchronously with deepening sedation, until neuronal involvement is maximal for that individual. Our data support this conclusion in that the frontal lobe is a dominant generator of SWA at saturation and that there is a positive correlation between the peak of the SWA and frontal gray matter volume, obtained by voxel-based morphometry (VBM).

Our second experiment, performed with simultaneous EEG and fMRI, allowed us to test whether there is functional deafferentation and thalamocortical isolation from external events at SWA saturation. The use of individualized behavioral and EEG-defined transitions to inform the fMRI analysis is a key component of this design and revealed changes in brain function specific to SWA saturation, an otherwise invisible phenomenon (see Fig. 4C). Our results show that when the thalamocortical network has been pharmacologically rendered refractory to external inputs, activation in an alternative and more fundamental network persists that includes the precuneus, posterior parietal, and prefrontal cortices.

In functional imaging studies of patients with altered states of consciousness, activity within this network is the first to show an increase in metabolism and blood flow in parallel with recovery (28). The precuneus has rich reciprocal connections to posterior parietal, retrosplenial, and prefrontal cortical areas but does not project directly to primary somatosensory cortices, brainstem nuclei, or the sensory thalamic nuclei (29). It does, however, have connections with the midline and intralaminar thalamic nuclei (30, 31), which are key to the regulation of consciousness (32). This may represent the most refractory level at which the central nervous system can respond to its environment. It is noteworthy that the precuneus is a key component of the default mode network (29), a core network associated with the resting human brain. Our finding of positive activity within the precuneus by external stimulation may be a consequence of thalamocortical isolation and also reflect a basic mode of brain function.

We have shown that before SWA saturation occurs, the unresponsive subjects process external stimulation with the same sensory-specific thalamocortical neural circuitry as that used for “normal” conscious perception during the responsive period. The persistence of α activity (Fig. 1) and the increased spindle prevalence at SWAS (fig. S3) show that failure of conscious perception, as indicated by the fMRI data, is unlikely to be explained by a thalamocortical switch, because clearly, thalamocortical exchange exists. However, it does not mean that the connection made between the thalamus and cortical regions ensures the transfer of meaningful information regarding the external world. Therefore, we believe that these thalamocortical networks are part of a disconnected consciousness (33) that is isolated from external events.

It is possible that in the intermediate state, between the LOBR and SWA saturation, individuals are aware of stimuli but disinclined to respond (that is, implicitly rather than explicitly aware) and, because of the additional amnesic effects of anesthesia, have no memory of these events. We cannot draw this conclusion on the basis of our current data alone, but in the intermediate state, the brain appears to be processing stimuli in a way comparable to the behaviorally responsive state before LOBR, albeit with reduced efficiency. Only at SWA saturation do we see a marked shift in the way that the brain processes this sensory information, which we interpret as functional deafferentation and a disconnection from the external world.

This neurophysiologically defined transition at SWA saturation occurs at a lower anesthetic dose than does burst suppression and may provide a much sought-after individualized index of lack of perceptual awareness to sensory events in altered states of consciousness. In particular, translation of SWA saturation to the operating room could allow the optimum dose of anesthesia to be determined, preventing intraoperative awareness and over-anesthesia. The direct and specific nature of SWA saturation means that problems such as low-voltage EEGs that occur, for example, in older patients can be overcome.

Nevertheless, our results have limitations. As with commercially available electrophysiological depth-of-anesthesia monitors, the variability of EEG measures among patients with different ages and comorbidities will need to be addressed. Further work will be required to demonstrate that SWA saturation occurs for all anesthetic agents, in the presence of other co-induction agents (for example, analgesia and muscle relaxants), and with the increased nociceptive drive associated with surgery. Other necessary steps will include the development of algorithms to define SWA saturation in real time, taking into account the artifacts that are introduced into EEG recordings by the operating room environment. These issues have already been partly addressed in other commercially available depth-of-anesthesia monitors. We therefore believe that translation of SWA saturation as an index of perceptual awareness in the clinic is feasible.

MATERIALS AND METHODS

Experimental design

Two experiments were performed during induction of and emergence from deep sedation, using an ultraslow target-controlled intravenous infusion of propofol, a γ-aminobutyric acid type A (GABAA) agonist, in healthy volunteers while presenting a consistent intensity of auditory and noxious stimuli that required motor responses. In experiment 1, EEG data were collected in a clinical neurophysiology laboratory, and in experiment 2, simultaneous EEG and functional MRI (EEG-fMRI) BOLD data were obtained. The same paradigm was used for both experiments and consisted of data acquisition in four phases. The stimulation paradigm and propofol infusion regime used for both experiments are described further in the Supplementary Materials and fig. S1.

The first phase was a resting period of 10 min in which data were recorded while subjects were asked to remain still with their eyes closed. During the second phase, all subjects experienced an ultraslow propofol-induced loss of consciousness over the course of 48 min to an estimated maximum ESC of 4 μg/ml while a paradigm of laser stimuli, 1-kHz tones, and cognitive word tasks were presented. The peak dose was then maintained for 10 min without stimulation (phase 3). During the fourth and final phase, the infusion was turned off, and subjects were allowed to emerge from unconsciousness naturally while experiencing the same paradigm of sensory stimulation delivered during the induction phase.

Both experiments were approved by the Local Research Ethics Committee and performed in the same subjects. Data are presented from 16 subjects [8 male, 8 female; age, 28.6 ± 7.0 (SD) years; range, 19 to 43 years] for the bench EEG data from experiment 1. EEG-fMRI data from a subset of 12 subjects [5 male, 7 female; age, 26.3 ± 5.1 (SD) years; range, 19 to 34 years] are presented for the induction scan (that is, the second of the four functional scans) in experiment 2. Further details of participant recruitment, screening, and subject withdrawal are provided in the Supplementary Materials.

Stimulation paradigm

Each stimulation block was 16 min long and was a sequence of interleaved laser stimuli, computer-generated tones, and cognitive word tasks. Three stimulation blocks were delivered during both the induction and recovery phases. Stimuli were presented in a pseudorandomized order with a computer-generated tone, a word task, and two noxious laser stimuli delivered per minute (Presentation, Neurobehavioral Systems Inc.).

Word task. A simplified version of the word task described in (34) was used in which a complex auditory and decision-making task was followed by the generation of a motor response. A list of 200 single-syllable words was derived (MRC Psycholinguistics Database, Machine Usable Dictionary v2.0) with a familiarity of 488 ± 99 (SD) and concreteness of 438 ± 120 (SD). The words were spoken by a male actor and recorded with Audacity (http://audacity.sourceforge.net). Words were selected from the list at random and presented in pairs (mean interstimulus interval between pairs of words, 56.8 s; range, 15.0 to 103.4 s) so that each 16-min stimulation block contained eight “same” and eight “different” pairs of words. Once a word had been presented, it was not used again. Subjects were asked to respond through a two-option button box whether the words were the same or different.

Noxious laser stimuli. Brief laser stimuli were applied to provide a noxious stimulus because a key behavioral feature of anesthesia is the loss of protective withdrawal to pain. Infrared lasers provide a highly reproducible nociceptive stimulus that selectively activates A delta and C nerve fibers (35). Laser stimuli (5-ms duration, 5-mm 1/e2 diameter, 1.34-μm wavelength) were generated by a Nd:YAP laser (STIMUL 1340, Electronic Engineering) and applied 2 cm above the lateral malleolus on the lower right leg at a rate of about two per minute (mean interstimulus interval, 30.4 s; range, 13.4 to 49.4 s). To avoid skin damage and nociceptor sensitization/habituation, the site was moved after each laser stimulus within a marked 6-cm2 area. For both experiments, the subject and the investigators wore protective goggles.

The applied laser stimuli produced a characteristic double sensation, an initial sharp pinprick, followed by a brief more diffuse after sensation. The laser energy level in joules required to achieve a subjective intensity of 5/10 (using a numerical rating scale with anchors of 0 = no sensation, 1 = just painful, and 10 = most intense sensation) was determined for each individual at the start of each experimental session before propofol administration, and this energy level was used throughout the session. During experiment 1 only, subjects were asked to rate the intensity 3 s after each laser stimulus on a computerized numerical rating scale by using the same button box until they became unresponsive. Pain ratings were not sought in experiment 2.

Tones. Additionally, tones of frequency of 1 kHz and duration of 60 ms were presented with interstimulus intervals ranging from 16 to 94 s with a mean of 60 s. Subjects were asked to listen passively to these tones, and no response was sought. All auditory stimuli were presented binaurally with MR-compatible electrostatic headphones (MRC Institute of Hearing Research). Auditory volume adequacy was checked before each experiment (and while the scanner was running for experiment 2) by playing sample stimuli. The tones were not analyzed as part of this experiment because the fMRI-BOLD responses to tones did not reach threshold, probably a result of subjects’ difficulty in discriminating tones over background scanner noise.

EEG analysis

Data preprocessing. Preprocessing was carried out with BrainVision Analyzer version 2.0 (BrainProducts GmbH), custom-written MATLAB code (MathWorks Inc.), and the EEGLAB v7.2.9 analysis toolbox (36). After cardioballistographic, fMRI gradient artifact and blink artifact removal (as detailed below), data sets were down-sampled to 500 Hz, re-referenced to a common average, and band pass–filtered from 0.5 to 30 Hz with a 48-dB/octave Butterworth zero-phase filter. Visual inspection of the data revealed that three subjects demonstrated brief periods of minor burst suppression in the later stages of the induction period for experiment 1 and none for experiment 2. Because these periods in experiment 1 were very short (that is, 2 to 6 s in duration) and did not meet the clinical criteria for burst suppression (37), all data were used for the spectral analysis.

Time-frequency analysis. Spectral analysis of activity band was carried out with Chronux (www.chronux.org) (38), which applies a multitaper spectral estimation and fast Fourier transform algorithm (window size, 3 s; step size, 4 s). The frequency power spectrum over time (and therefore with varying propofol dose) within the 0.5 to 30 Hz range was plotted for each subject and channel. Additionally, phasic changes in EEG absolute and relative power in the specific frequency bands of interest, that is, β (15 to 30 Hz), α (9 to 14 Hz), θ (4 to 8 Hz), and slow-wave (0.5 to 1.5 Hz) bands, were calculated for each volunteer. Each subject’s activity was averaged across channels, and then a grand mean average was generated for the group. Temporal smoothing was carried out with a median filter of order 20.

To identify the associated topographic distribution of the frequency-specific changes, the average EEG power at each electrode coordinate was plotted for each frequency band across the experimental t-ROI using spline interpolation.

Spindle oscillations. Spindles consist of a series of waxing and waning oscillations in the α frequency band from 8 to 14 Hz. The prevalence of spindles was calculated from 8 to 14 Hz in 1-Hz steps by using a modification of the pattern detection algorithm by MacKay and colleagues (39). This was done for each EEG channel on a single-subject basis. Temporal changes in the overall prevalence of spindles at each frequency were calculated by averaging the spindle count in 3-s epochs across channels. Group spindle activity was obtained before and after the subjects’ LOBR and averaging spindle count across subjects. As with the time-frequency changes across the broader frequency bands, the topographic distribution of spindle prevalence for the group data was obtained by plotting the spindle prevalence at each electrode coordinate.

Blink artifact removal. Eye blinks were identified with an automated algorithm (BrainVision Analyzer version 2.0) that parsed the VEOG channel and were confirmed by visual inspection. Independent component analysis was used to remove blink artifact from the remaining EEG channels by constraining the data domain to the time interval around the blinks. Data were again inspected visually for the adequacy of artifact removal.

fMRI gradient artifact removal. High-amplitude (~10 mV) fMRI artifacts are generated in the EEG data by the switching of gradient fields and by the radiofrequency pulse generation in the scanner bore. fMRI gradient artifacts are temporally stable and occur at a rate about 500 times faster than the spontaneous EEG, so they can be removed using artifact template subtraction (BrainVision Analyzer version 1.0). Data were first up-sampled to 50 kHz using spline interpolation. Gradient artifact averaging was then performed on the basis of the gradient onset markers recorded during acquisition. A sliding average of 16 volumes was applied to optimize the gradient artifact removal and account for a slight drift in relaxation time (TR) (~1 ms) over the course of the 48-min scans, which causes a slow change in the artifact structure. After gradient artifact removal, data were down-sampled to 0.5 kHz, and a low-pass filter of 30 Hz (Butterworth zero-phase filter, 48 dB/octave) was applied.

Cardioballistographic artifact removal. The cardioballistographic artifact (CBA) in the EEG data during fMRI acquisition arises from very small movements of the electrodes within the magnetic field because of pulsatile effects at the scalp. CBA removal was performed with R-wave markers of the QRS complex from the electrocardiogram channel and a template correction algorithm (BrainVision Analyzer version 2.0). The delay time between the R wave and the CBA was estimated by calculating the smoothed global field power (GFP) distribution for all EEG channels. The local minima to the left and right of the GFP maximum were determined, and the temporal midpoint between these two minima was defined as the delay time corresponding to the middle of the GFP power distribution. This delay time was then used to define an artifact template on the basis of the first 101 R waves.

MRI data analysis

Data were analyzed with the FMRIB’s Software Library (FSL) version 4.1.6 (40).

Preprocessing of functional MRI data. Preprocessing included removal of the initial four dummy volumes to exclude any non–steady-state magnetization effects, motion correction with MCFLIRT (Motion Correction FMRIB’s Linear Image Registration Tool), spatial smoothing using a Gaussian kernel of 5-mm full-width half-maximum, global intensity normalization, and temporal high-pass filtering with a cutoff of 50 s to remove low-frequency scanner drift. Automated removal of nonbrain tissue was initially performed before statistical analysis using BET (Brain Extraction Tool) with further manual correction in FSLView. The functional scans were co-registered to the individual T1 high-resolution structural image, and then onto a standard brain [Montreal Neurological Institute 152 (MNI152) brain], using FLIRT (FMRIB’s Linear Image Registration Tool).

Because these were dynamic scans where the cardiorespiratory physiology was expected to be correlated with drug level, each of the fMRI data sets was decomposed into spatially independent linearly mixed components with MELODIC (Multivariate Exploratory Linear Optimized Decomposition into Independent Components) during preprocessing. For each scan, components were inspected visually for spatial distributions or signal time courses that could be attributed to pulsations (cerebrospinal fluid and circumferential rings) or respiration (slice artifacts and/or time courses matching the recorded respiratory excursion). The identified physiological noise components were regressed out, and the data sets were reconstructed so that drug level–induced physiological changes did not confound further analyses.

Event-related BOLD. Each individual’s fMRI propofol induction scan was denoised (see the “Preprocessing of functional MRI data” section) and then split into three separate neurophysiological phases using transitions defined by behavior and the simultaneously acquired EEG (as shown in Fig. 4B). The transitions were defined as (i) LOBR and (ii) SWP. These transition-bounded epochs were classified as “responsive,” “unresponsive,” and “SWA saturation.” Data from the induction phase were only used for this analysis because the emergence data contained large motion artifacts at the point of ROBR.

fMRI data for each phase were analyzed with a three-level general linear model approach in FEAT version 5.98 (FMRIB’s Expert Analysis Tool). At the first level, areas of BOLD signal changes in response to the noxious stimuli, tones, and word task for each phase of the experiment and subject were identified. The three stimuli categories were included as separate regressors with each laser stimulus, tone, and word task of 1-s duration. Motion correction parameters from MCFLIRT and the subjects’ button press timings were also included as regressors of no interest. These regressors were each convolved with a gamma hemodynamic response function [mean lag, 6 ± 3 (SD) s]; temporal filtering and temporal derivative were also applied to account for intersubject differences in the hemodynamic response.

At the second level, the within-subject main effects of each stimulation type were generated for the neurophysiologically defined phases as well as for contrast maps to indicate the transition between phases. The main effects and contrast activation maps were then passed to a third-level mixed-effects group-level analysis with FLAME (FMRIB’s Local Analysis of Mixed Effects). Cluster-based thresholding (Z = 2.3, P < 0.05) was used to reveal significant group-level brain activation corrected for multiple comparisons across space.

Voxel-based morphometry. Structural data were analyzed with FSL’s VBM analysis tool. First, nonbrain tissue was removed with BET. Next, tissue-type segmentation was carried out with FAST4. The resulting gray matter partial volume images were then aligned to MNI152 standard space with the affine registration tool FLIRT, followed by nonlinear registration with FNIRT (FMRIB’s Nonlinear Image Registration Tool). The resulting images were averaged to create a study-specific template, to which the native gray matter images were nonlinearly re-registered. The registered partial volume images were modulated to correct for local expansion or contraction by dividing by the Jacobian of the warp field. Voxelwise estimates of gray matter volume were averaged across the whole prefrontal cortex with the Harvard-Oxford structural atlas, and these individual estimates of percentage of gray matter were used to calculate frontal gray matter volume for each subject.

Statistical analysis

The statistics used in our study are discussed individually within the descriptions of the analysis tools above.

SUPPLEMENTARY MATERIALS

www.sciencetranslationalmedicine.org/cgi/content/full/5/208/208ra148/DC1

Materials and Methods

Results

Fig. S1. Schematic of propofol infusion regime and stimulation paradigm.

Fig. S2. Individual peak slow-wave power correlates with prefrontal gray matter volume.

Fig. S3. Spindle activity provides evidence that thalamocortical exchange persists.

Table S1. Physiological parameters at baseline and peak propofol dose.

REFERENCES AND NOTES

  1. Acknowledgments: We thank G. Hadjipavlou for his help in preparing the figures and the radiographers S. Wilson and C. Young for their involvement in the study. Funding: Supported by the Medical Research Council of Great Britain and Northern Ireland (I.T., R.N.M., and FMRIB Centre), Wellcome Trust (I.T.), the National Institute for Academic Anaesthesia, and the International Anesthesia Research Society. Author contributions: All authors designed and R.N.M., C.W., and R.R. performed the experiments. I.T. and R.N.M. provided funding. All authors contributed to data analysis/interpretation and writing of the paper. Competing interests: Patent applications have been filed by Isis Innovation Ltd., the technology transfer company of the University of Oxford, on perception loss detection for anesthesia monitoring (US61/746975 and PCT/GB2013/051445). All authors are listed as inventors.
View Abstract

Navigate This Article