THE neural correlates of anesthetic- and sleep-induced unconsciousness have yet to be fully clarified, but growing evidence suggests that fragmentation of cortical networks consistently occurs during pharmacologic, physiologic, and pathologic states of unconsciousness. Anesthetics with diverse molecular profiles and neurophysiologic effects have been shown to fragment functional networks in the cortex,^1–5  and both sleep and anesthesia reduce surrogates of cortical information transfer.^4–12  Several recent studies encourage further investigation of the role of cortex in anesthetic-induced unconsciousness. First, in rats, cross-modal cortical interactions are more sensitive to anesthetic effects than first-order thalamocortical circuits.¹³  Second, in humans, disrupted corticocortical connectivity has been reported to better distinguish propofol-induced unconsciousness from wakefulness and sedation compared with changes in thalamocortical connectivity.^14,15  Although correlative, these findings are mechanistically relevant given the apparent requirement for integration of cortical networks during normal conscious experience.^16,17  Furthermore, the relationship of anesthetic- and sleep-induced unconsciousness is a subject of active exploration,^18–25  but systematic comparisons of brain connectivity changes during anesthetic-induced (intravenous and inhaled) and sleep-induced unconsciousness have been lacking. Finally, despite the focus on functional, directed, and effective connectivity changes during general anesthesia,²⁶  there is little understanding of the underlying neurochemical control of connectivity patterns, which might provide an explanatory link to the molecular actions of general anesthetics.

In this study, we investigated the relationship of cortical acetylcholine and measures of brain connectivity—coherence and symbolic transfer entropy—derived from the electroencephalogram under multiple conditions: (1) before, during, and after unconsciousness induced by propofol or sevoflurane, (2) during spontaneous wakefulness, (3) during slow wave sleep (SWS), and (4) during rapid eye movement (REM) sleep. We report that electroencephalographic coherence and frontal-parietal directed connectivity in high γ (85 to 155 Hz) bandwidth are present during wakefulness and disrupted during physiologic (SWS and REM sleep) and pharmacologic (propofol and sevoflurane) states of unconsciousness. By contrast, coherence and bidirectional frontal–parietal connectivity in θ (4 to 10 Hz) bandwidth are present during states of cortical activation (low-amplitude fast-wave electroencephalogram) with (wakefulness) or without (REM sleep) behavioral arousal and correlate with cortical acetylcholine levels. We conclude that coherence and frontal-parietal directed connectivity in high γ band are neural correlates of wakefulness that are not mediated by cholinergic mechanisms.

All experiments were conducted on adult male Sprague–Dawley rats (n = 31; 300 to 350 g; Charles River Laboratories Inc., USA). The rats were housed in a temperature-controlled facility with 12-h light:12-h dark cycle (lights on at 6:00 am) and had ad libitum access to food and water. The experimental procedures were approved by the Institutional Animal Care and Use Committee at the University of Michigan (Ann Arbor, Michigan) and were in compliance with the Guide for the Care and Use of Laboratory Animals (eighth edition; The National Academies Press, Washington, D.C.) and Animals in Research: Reporting In Vivo Experiments (ARRIVE) guidelines.

Rats were anesthetized with 3 to 4% isoflurane in 100% oxygen and positioned in a stereotaxic frame (model 963; David Kopf Instruments, USA) using blunt ear bars. Isoflurane (1 to 2%) was delivered through a rat anesthesia mask (Kopf model 906) and was titrated to effect during surgery. General anesthesia was assessed by the absence of pedal and palpebral reflex. An anesthetic agent analyzer (Datex Medical Instrumentation, Inc., USA) was used for continuous monitoring of delivered isoflurane concentration. Core body temperature was monitored using a rectal probe (model 7001H; Physitemp Instruments, Inc., USA). A small animal far-infrared heating pad (Kent Scientific Co., USA) was used for the maintenance of body temperature at 37.0° ± 1.0°C. All rats (n = 31) were implanted with stainless steel screw electrodes to record electroencephalogram from the (1) frontal cortex: anterior–posterior = +3.0 mm and medial–lateral = 2.5 mm, (2) parietal cortex: anterior–posterior = −4.0 mm and medial–lateral = ±2.5 mm, and (3) occipital cortex: anterior–posterior = −8.0 mm and medial–lateral = ±2.5 mm, and a screw electrode over the nasal sinus served as the reference electrode; all coordinates were with respect to bregma. In a subgroup of these rats (n = 7), multistranded insulated (except at the tip) wires (AS636; Cooner Wires, Inc., USA) were implanted in the dorsal neck muscles for recording electromyogram. These rats were used for recording sleep-wake states. All other rats (n = 24) were implanted with a CMA/11 microdialysis guide cannula (CMA Microdialysis; Harvard Apparatus, USA) aimed at 1.0 mm above the prefrontal cortex (PFC) (anterior–posterior = +3.0 mm, medial–lateral = 0.5 mm, and ventral = 4.0 mm)²⁷  for microdialysis measurement of acetylcholine levels. The electrodes were interfaced with six-pin pedestal(s) (MS363; Plastics One, USA), and the entire assembly along with a microdialysis guide cannula was secured with dental cement (Cat No. 51459; Stoelting Co, USA). A subgroup of rats (n = 11) with the electrodes for electroencephalographic recording and a microdialysis guide tube was implanted with an indwelling catheter (Micro-Renathane tubing, MRE-040; Braintree Scientific, USA) in the internal jugular vein to provide access for propofol infusion. The catheter was tunneled under the skin and mated with a port (313-000BM-10; Plastics One) sutured on the back muscles between the scapulae. The remaining rats (n = 13) with the electrodes for electroencephalographic recording and a microdialysis guide tube but without an intravenous catheter were used for sevoflurane experiments. Buprenorphine hydrochloride (Buprenex^®; Reckitt Benckiser Pharmaceuticals Inc., USA) was used for presurgical (0.01 mg/kg body weight, subcutaneous) and postsurgical (0.03 mg/kg body weight, subcutaneous, every 8 to 12 h for 24 h) analgesia, and all rats received a single presurgical dose (20 mg/kg body weight, subcutaneous) of antibiotic cefazolin (West-Ward-Pharmaceutical Corp., USA). The rats implanted with catheters received an additional dose of antibiotic gentamicin (5.0 mg/kg body weight, intravenous) during the surgery. Animals were kept under observation until ambulatory and then returned to their home cage for postsurgical recovery. All rats were provided at least 7 to 10 days of postsurgical recovery during which they were conditioned to the recording chamber and cables. The jugular venous catheter was flushed with 0.3 ml heparinized saline (1 U/ml; Sagent Pharmaceuticals, USA) every other day.

Microdialysis samples were collected every 12.5 min (25 µl) with CMA/11 microdialysis probes (1-mm cuprophane membrane, 0.24-mm diameter, 6 kDa), which were perfused continuously with Ringer’s solution (147 mM NaCl, 2.4 mM CaCl₂, 4.0 mM KCl, 10 µM neostigmine; pH 6.0 ± 0.2) at 2.0 µl/min using a CMA/400 syringe pump (CMA Microdialysis; Harvard Apparatus). From each microdialysis sample, 22 µl was injected into a high-performance liquid chromatography paired with an electrochemical detector (Bioanalytical Systems, USA) for acetylcholine quantification. An ion-exchange (mobile phase: 50 mM Na₂HPO₄, pH 8.5) analytical column (MF-6150; Bioanalytical Systems) separated acetylcholine and choline in the dialysis samples, which were proportionately catalyzed into H₂O₂ by an immobilized enzyme reactor column (MF-6151; Bioanalytical Systems). H₂O₂ was detected by oxidation (applied potential at 500 mV, Ag⁺/AgCl reference electrode) at a platinum working electrode. The chromatograms were digitized to quantify acetylcholine levels using ChromGraph software (Bioanalytical Systems) and a seven-point standard curve (0.05 to 1.0 pmol).

The electrophysiologic signals were amplified with a Grass Model 15 LT bipolar portable physiodata amplifier system (15A54 Quad Amplifier; Natus Neurology Inc., USA). An MP150 data acquisition unit along with AcqKnowledge software (version 4.1.1; Biopac Systems, Inc., USA) was used for digitizing and storing the data. The electrode implanted over nasal sinus served as a reference for recording monopolar electroencephalogram (0.1 to 300 Hz, 1-kHz sampling rate) from the frontal, parietal, and occipital cortices, which were used for connectivity analyses. Bipolar electroencephalogram (frontal–parietal and parietal–parietal, 0.1 to 100 Hz, 250-Hz sampling rate) and electromyogram (1 to 100 Hz, 250-Hz sampling rate) were recorded for quantification of sleep-wake states. Electroencephalogram signals were amplified 5,000 times, while electromyogram signals were amplified 10,000 times.

The data were first downsampled to 500 Hz to reduce computation time, and an IIR notch filter was applied to remove 60 Hz line noise. Magnitude squared coherence was calculated at individual frequencies in 0.5-Hz intervals from 0.5 to 155 Hz between all electrode pairs in 2-s moving windows using the “mscohere.m” function in the MATLAB Signal Processing Toolbox (MathWorks Inc., USA). Using the approach demonstrated in previous publications from our^28,29  and other^30,31  laboratories, the coherence values were averaged over the following frequency bands: δ (0.5 to 4 Hz), θ (4 to 10 Hz), α (10 to 15 Hz), β (15 to 25 Hz), low γ (γ1: 25 to 55 Hz), medium γ (γ2: 85 to 125 Hz), and high γ (γ3: 125 to 155 Hz). The mean global coherence was obtained by averaging the coherence for individual channel pairs for each animal. To control for the possibility of spurious coherence affecting the analysis, the empirical data were statistically compared with a surrogate data set in which the phase relationships between the electroencephalogram signals were disrupted without altering the spectral content of each signal.³²  Paired Student’s t test showed a significant difference (P < 0.000001) in the coherence between the empirical and surrogate data sets, confirming that the measured coherence was not accounted for merely by spectral shifts of the signal across states.

We used normalized symbolic transfer entropy (NSTE) to assess directed connectivity. NSTE is an information theoretic measure and serves as a surrogate for directed cortical communication. Our previous studies have validated the use of NSTE to measure connectivity changes in humans^7,8,11  and rats,²⁸  and other laboratories have supported our findings with NSTE using different approaches: functional connectivity and dynamic causal modeling.^9,14  Transfer entropy, the basis of NSTE, was also used in a previous study of evoked potentials in low γ range (less than or equal to 50 Hz) in rats³³  and head-restrained ferrets.³⁴  We selected θ and γ bands for NSTE analysis because these have been linked to electroencephalographic activation, changes in cortical acetylcholine, and changes in states of behavioral arousal and consciousness.^29,35  NSTE was analyzed between frontal and parietal areas because studies from human volunteers, patients, and animals suggest the importance of frontal–parietal networks in consciousness of the environment.^{7–9,11,14,26,33,36}  NSTE was calculated between left frontal and left parietal areas in θ (4 to 10 Hz), low γ (γ1: 25 to 55 Hz), medium γ (γ2: 85 to 125 Hz), and high γ (γ3: 125 to 155 Hz) bands as has been described previously.^11,28  In brief, the sampling rate was kept at 1 kHz to improve the stability of the analysis. The continuous raw electroencephalographic data were first filtered for the frequency band of interest (θ, low γ, medium γ, and high γ) with a zero-phase FIR filter of order 400 designed using firls2.m (MATLAB Signal Processing Toolbox; MathWorks Inc.) and implemented using filtfilt.m (MATLAB Signal Processing Toolbox; MathWorks Inc.). Data were then extracted from time periods of interest and analyzed in 10-s bins. Note that all electroencephalographic analyses were performed on the same data epochs. NSTE requires three parameters: embedding dimension (d_E), time delay (τ), and prediction time (δ). These parameters were selected such that the transfer entropy from the source signal to the target signal was maximized for a given data window, as described previously.^11,28  In brief, d_E was kept fixed at 3, δ was set by selecting the lag that maximized the cross-correlation between the two signals, and τ was varied from 1 to 30. The potential bias of symbolic transfer entropy was removed using a shuffled data set, after which the unbiased data set was normalized.^11,28 

The rats were connected to the electroencephalogram recording system at least 30 min before the start of an experimental session (9:30 am to 10:00 am), and a microdialysis probe being infused with Ringer’s solution at 2.0 µl/min was lowered into the PFC. The electroencephalogram was recorded continuously while the microdialysis samples were collected every 12.5 min throughout the experiment. Previous reports from our²⁹  and other³⁷  laboratories have shown that it takes up to 37.5 min for the in vivo baseline acetylcholine levels to stabilize after probe insertion. Therefore, the first three microdialysis samples were excluded and six preanesthesia baseline microdialysis samples were collected. In order to hold the behavioral state constant during the baseline condition, the rats were kept awake using gentle tapping on the recording chamber. At the completion of six wake microdialysis samples, the rats were either connected to a microsyringe pump (SP 101I; WPI Inc., USA) through the venous catheter for continuous propofol infusion (800 µg ⋅ kg⁻¹ ⋅ min⁻¹) or exposed to sevoflurane anesthesia (2.0 to 2.2%), and six microdialysis samples were collected during the administration of the anesthetic. Sevoflurane experiments were conducted in a custom-made clear round recording chamber, which could be rotated in a direction opposite to that of the rat’s movement. This apparatus allows the rat to stay in the center of the chamber, thereby preventing any kinks in the microdialysis tubing. The recording chamber can be sealed during anesthesia and was fitted with a door to allow access to the animal during anesthetic exposure. The chamber also had ports for inlet and outlet of anesthetic vapors, which were monitored using two anesthetic agent analyzers (Datex Medical Instrumentation, Inc.). The anesthetic concentration was based on preliminary experiments conducted in our laboratory and was titrated to produce loss of righting reflex along with slow wave (high-amplitude δ wave) electroencephalogram, both of which are used as a surrogate for loss of consciousness.^38,39  Propofol and sevoflurane produced loss of righting reflex within about 20 min, after which the rats were maintained in a recumbent position for the rest of the anesthetic administration. The core body temperature was monitored and maintained at 37.0° ± 1.0°C using a far-infrared heating pad. At the completion of six microdialysis samples during the administration of anesthetic, the propofol infusion or sevoflurane exposure was stopped. Thereafter, six microdialysis samples were collected in the postsevoflurane recovery period. Initial experiments with propofol (n = 3) showed that acetylcholine levels during the postpropofol recovery period (six microdialysis samples) remained below baseline wake levels. The data from these initial propofol experiments, which only had six postpropofol recovery samples, were excluded from the analysis. In addition, it took much longer for the rats to recover righting reflex after propofol anesthesia (approximately 29 min) as compared to sevoflurane (less than 10 min), which likely reflects pharmacokinetics. Therefore, in order to ensure a complete recovery from the effects of propofol anesthesia, microdialysis samples were collected for an extended recovery period (12 microdialysis samples) in the remaining eight rats. In the propofol group, we could not collect microdialysis samples in one of the eight rats with extended recovery period because of a microdialysis probe malfunction. In the sevoflurane group, two rats did not have good electroencephalogram signals and the microdialysis probe failed in one of the rats. The last epoch during wake, anesthetic-induced unconsciousness, and postanesthetic recovery period was taken as the most stable and representative of the corresponding states, and therefore the electroencephalogram and acetylcholine analysis was restricted to these 12.5-min epochs.

Monopolar (frontal, parietal, and occipital cortices) and bipolar electroencephalogram (frontal–parietal and parietal–parietal) along with electromyogram were recorded for 12 h between 6:00 am and 6:00 pm during the light phase. Bipolar electroencephalogram and electromyogram in 10-s bins were used for the characterization and quantification of wake (low-amplitude fast-wave electroencephalogram with high muscle tone), SWS (high-amplitude slow wave electroencephalogram with low muscle tone), and REM sleep (low-amplitude fast-wave electroencephalogram along with muscle atonia). Only pure sleep or wake epochs were selected for the analysis, i.e., wakefulness epoch was characterized by the presence of low-amplitude fast-wave electroencephalogram and high muscle tone for 100% of the time in the 10-s epoch. Similarly, for REM sleep, the selected epochs were characterized by low-amplitude fast-wave electroencephalogram and muscle atonia for 100% of the time in 10-s epoch. Selection of these pure epochs ensured that there was no potential confound in the connectivity analysis arising from mixed states within a single epoch. Of a total of 4,320 10-s epochs, we selected a subset of 99 pure epochs for connectivity analysis. Note that the state of wakefulness in these experiments was spontaneous, while the state of baseline wakefulness in the study of propofol- and sevoflurane-induced unconsciousness was maintained by the experimenters.

After 3 to 7 days of microdialysis and electroencephalogram data collection, rats were deeply anesthetized (ketamine and xylazine, 80 and 10 mg/kg body weight, intraperitoneal) and transcardially perfused with 100 ml of phosphate buffered saline (0.1 M, pH 7.2; 1219SK; EM Sciences, USA) followed by 250 ml fixative (4% paraformaldehyde, 4% sucrose, 0.1 M phosphate buffer, pH 7.2; 1224SK; EM Sciences) solution. The brains were extracted and stored in a fixative solution for at least 24 h and then allowed to equilibrate in 30% sucrose in phosphate buffer. Forty-micron coronal sections were cut through the PFC on a cryostat (Leica Microsystems Nussloch GmbH, Germany). The sections were mounted on slides and stained with cresyl violet to confirm the site of microdialysis.

Statistical analyses were conducted in consultation with the Center for Statistical Consultation and Research at the University of Michigan. A priori power analysis (nQuery Advisor + nTerim; Statistical Solutions Ltd, USA) using the acetylcholine pilot data indicated that a minimum sample size of 3 would have 80% power (effect size = 4.0) to detect a difference in means of 0.8 pmole (SD of difference = 0.2) with Student’s paired t test with one-sided α = 0.05/3 (Bonferroni correction for three pairwise tests). All statistical comparisons were conducted in a within-group design, and P < 0.05 was considered statistically significant. Each group of rats in this study had different surgical implants: jugular venous catheter and microdialysis probe in the propofol group, microdialysis probe without jugular venous catheter in the sevoflurane group, and neither catheter nor microdialysis probe in the sleep-wake group. Therefore, random allocation of rats to different experimental conditions was not possible. Additionally, the experimenters could not be blinded to the experiments because the experimental interventions (propofol infusion, sevoflurane exposure, and sleep-wake recordings) were clearly different and could not be masked. Repeated-measures ANOVA (RMANOVA) with Tukey multiple comparisons test was used for the comparison of acetylcholine levels and the electroencephalographic coherence in each frequency band between the last epochs from the periods of (1) wakefulness, (2) anesthetic-induced unconsciousness (propofol and sevoflurane), and (3) postanesthetic recovery wakefulness. These epochs were also used for comparison of directed connectivity between the frontal and parietal areas for θ and γ (low, medium, and high) bands. Similarly, RMANOVA with Tukey post hoc test was used for the comparison of coherence and directed connectivity during sleep-wake states. The data are reported as mean ± SEM along with 95% CI. To enhance readability, only P values are provided in the Results section, while mean ± SEM, 95% CI, and F statistic are provided in a tabular format (tables 1 to 5, Supplemental Digital Content 1, http://links.lww.com/ALN/B316, which show descriptive and inferential statistics). Statistical comparisons were performed with Graph Pad Prism 6.05 (Graph Pad Software, Inc., USA).

The temporal course of the changes in PFC acetylcholine levels before, during, and after anesthetic-induced unconsciousness is illustrated in figure 1A. Histologic analysis confirmed the probe placement in the PFC for all rats in the propofol and sevoflurane groups (fig. 1, B and C). The effect of propofol and sevoflurane on cortical acetylcholine is shown in figures 2 and 3, respectively. There was a statistically significant decrease in acetylcholine levels during propofol-induced (P = 0.005) and sevoflurane-induced (P = 0.0006) unconsciousness as compared to the levels observed in the waking state (figs. 2A and 3A; table 1, Supplemental Digital Content 1, http://links.lww.com/ALN/B316, which shows descriptive and inferential statistics). The decrease in acetylcholine was comparable between the propofol (approximately 90%) and sevoflurane (approximately 85%) groups. Compared to anesthetic-induced unconsciousness, the recovery wake epoch for both propofol (P = 0.004) and sevoflurane (P = 0.002) groups was characterized by a significant increase in acetylcholine levels that was not statistically different from preanesthesia wake acetylcholine levels (figs. 2A and 3A; table 1, Supplemental Digital Content 1, http://links.lww.com/ALN/B316, which shows descriptive and inferential statistics). Previous reports have demonstrated that cortical acetylcholine levels are high during wakefulness, low during SWS, and high again during REM sleep, approximating the levels observed during the waking state.^35,40  These changes in cortical acetylcholine across sleep-wake states have been well described and accepted. Therefore, we did not measure changes in cortical acetylcholine during sleep-wake states in the current study.

The decrease in acetylcholine levels in PFC during anesthetic-induced unconsciousness was associated with a reduction in long-range corticocortical coherence (figs. 2B and 3B; table 2, Supplemental Digital Content 1, http://links.lww.com/ALN/B316, which shows descriptive and inferential statistics). As compared to wakefulness, anesthetic-induced unconsciousness produced a significant decrease in electroencephalographic coherence in δ (propofol: P = 0.001 and sevoflurane: P = 0.004), θ (propofol: P = 0.0004 and sevoflurane: P = 0.003), α (propofol: P = 0.0004 and sevoflurane: P = 0.006), β (propofol: P = 0.0003 and sevoflurane: P = 0.0006), low γ (propofol: P < 0.0001 and sevoflurane: P < 0.0001), medium γ (propofol: P < 0.0001 and sevoflurane: P < 0.0001), and high γ (propofol: P = 0.0002 and sevoflurane: P < 0.0001) bands (figs. 2B and 3B; table 2, Supplemental Digital Content 1, http://links.lww.com/ALN/B316, which shows descriptive and inferential statistics). Compared to anesthetic-induced unconsciousness, coherence during the recovery wake epoch showed a significant increase and returned to wake levels in all bands except the δ band in the propofol and sevoflurane groups: δ (propofol: P = 0.08 and sevoflurane: P = 0.05), θ (propofol: P = 0.001 and sevoflurane: P = 0.03), α (propofol: P = 0.0002 and sevoflurane: P = 0.02), β (propofol: P = 0.0007 and sevoflurane: P = 0.004), low γ (propofol: P < 0.0001 and sevoflurane: P < 0.0001), medium γ (propofol: P = 0.0004 and sevoflurane: P = 0.0006), and high γ (propofol: P = 0.003 and sevoflurane: P = 0.0002); δ coherence during recovery from sevoflurane-induced unconsciousness remained significantly lower as compared to that observed during the wake state (P = 0.01; figs. 2B and 3B; table 2, Supplemental Digital Content 1, http://links.lww.com/ALN/B316, which shows descriptive and inferential statistics). The effect of general anesthetics on cortical coherence was driven by similar changes in the coherence in individual channel pairs across the cortex (figs. 1 and 2, Supplemental Digital Content 2, http://links.lww.com/ALN/B317).

Next, we analyzed NSTE between left frontal and left parietal cortices in θ and γ bands before, during, and after anesthetic-induced unconsciousness. As compared to wakefulness, propofol-induced unconsciousness was marked by a significant decrease in frontal-to-parietal connectivity in θ (P = 0.02) and γ (low: P = 0.003, medium: P < 0.0001, and high: P = 0.0001) bands (fig. 2C; table 3, Supplemental Digital Content 1, http://links.lww.com/ALN/B316, which shows descriptive and inferential statistics). The parietal-to-frontal connectivity also decreased in θ (P = 0.02), medium γ (P = 0.0004), and high γ (P = 0.0002) bands, while there was no statistically significant change in low γ band (P = 0.1; fig. 2D; table 3, Supplemental Digital Content 1, http://links.lww.com/ALN/B316, which shows descriptive and inferential statistics). As compared to propofol-induced unconsciousness, the connectivity during the postpropofol recovery epoch in both directions in θ (frontal-to-parietal: P = 0.0004 and parietal-to-frontal: P = 0.01), medium γ (frontal-to-parietal: P = 0.003 and parietal-to-frontal: P = 0.01), and high γ (frontal-to-parietal: P = 0.004 and parietal-to-frontal: P = 0.004) bands showed a significant increase and returned to the preanesthesia wake levels. The group-level statistical changes in medium and high γ frontal–parietal connectivity were supported by changes in connectivity in each individual rat; all rats showed a decrease in frontal–parietal connectivity in medium and high γ bands. The response was more variable in the low γ band. Sevoflurane-induced unconsciousness was also marked by a decrease in frontal-to-parietal connectivity in γ bands (low: P = 0.01, medium: P = 0.0009, and high: P < 0.0001; fig. 3C; table 3, Supplemental Digital Content 1, http://links.lww.com/ALN/B316, which shows descriptive and inferential statistics). There was no change in frontal-to-parietal θ connectivity (P = 1.0) during sevoflurane-induced unconsciousness as compared to wakefulness. The parietal-to-frontal connectivity decreased in θ (P = 0.003) and higher γ (medium: P = 0.0004 and high: P < 0.0001) bands, while there was no statistically significant change in low γ band (P = 0.4; fig. 3D; table 3, Supplemental Digital Content 1, http://links.lww.com/ALN/B316, which shows descriptive and inferential statistics). Compared to sevoflurane-induced unconsciousness, bidirectional connectivity increased in medium (frontal-to-parietal: P = 0.01 and parietal-to- frontal: P = 0.01) and high (frontal-to-parietal: P = 0.0003 and parietal-to-frontal: P = 0.0005) γ bands during the postsevoflurane recovery wake epoch and returned to the presevoflurane wake levels. Frontal-to-parietal θ connectivity was significantly higher during post-sevoflurane recovery wake epoch than during presevoflurane wake (P = 0.04) and the unconscious state (P = 0.02), while the parietal-to-frontal θ connectivity showed an increase (P < 0.0001) during postsevoflurane recovery wake epoch and returned to the presevoflurane wake levels.

As compared to wakefulness, SWS was characterized by a significant decrease in corticocortical coherence levels in δ (P < 0.0001), θ (P = 0.0001), α (P = 0.007), medium γ (P = 0.0002), and high γ (P = 0.0001) bands; there was no statistical difference in coherence in β (P = 0.07) and low γ (P = 0.2) bands (fig. 4A; table 4, Supplemental Digital Content 1, http://links.lww.com/ALN/B316, which shows descriptive and inferential statistics). Compared to SWS, corticocortical coherence decreased during REM sleep in β (P = 0.0004) and γ range (low: P = 0.001, medium: P = 0.0004, and high: P = 0.008), while it increased in θ (P < 0.0001) and α (P = 0.0001) bands (fig. 4A; table 4, Supplemental Digital Content 1, http://links.lww.com/ALN/B316, which shows descriptive and inferential statistics). There was no significant difference (P = 0.2) in δ coherence between SWS and REM sleep. Compared to wakefulness, REM sleep was characterized by reduced coherence in δ (P < 0.0001), β (P = 0.004), and the entire γ range (low: P = 0.0001, medium: P < 0.0001, and high: P < 0.0001); there was no significant difference in θ (P = 0.2) and α (P = 0.6) bands between wakefulness and REM sleep (fig. 4A; table 4, Supplemental Digital Content 1, http://links.lww.com/ALN/B316, which shows descriptive and inferential statistics). Global changes in cortical coherence during sleep-wake states were also evident in the changes in coherence observed in individual channel pairs across the cortical areas (fig. 3, Supplemental Digital Content 2, http://links.lww.com/ALN/B317).

Bidirectional frontal–parietal connectivity in θ band (fig. 4, B and C) followed a pattern that closely mirrored the cortical acetylcholine levels reported across sleep-wake states.^35,40  Compared to the wake state, frontal–parietal θ connectivity decreased during SWS (frontal-to- parietal: P = 0.0006 and parietal-to-frontal: P < 0.0001) and increased during REM sleep (frontal-to-parietal: P = 0.003 and parietal-to-frontal: P = 0.007; fig. 4, B and C; table 5, Supplemental Digital Content 1, http://links.lww.com/ALN/B316, which shows descriptive and inferential statistics). As opposed to frontal–parietal θ connectivity, frontal–parietal connectivity in the γ range showed a progressive reduction from wake to REM sleep and did not mirror the known changes in cortical cholinergic tone (fig. 4, B and C). The frontal-to-parietal connectivity in medium (P = 0.006) and high γ (P = 0.002) bands showed a significant decrease from wake to SWS, and all three γ bands were significantly reduced during REM sleep as compared to both wake (low: P = 0.009, medium: P = 0.0007, and high: P = 0.0002) and SWS (low: P = 0.01, medium: P = 0.003, and high: P = 0.04). Frontal-to-parietal connectivity in the low γ band was not significantly different between wake and SWS (P = 0.12; fig. 4B; table 5, Supplemental Digital Content 1, http://links.lww.com/ALN/B316, which shows descriptive and inferential statistics). Similarly, as compared to wakefulness, the parietal-to-frontal connectivity in γ bands decreased during SWS (low γ: P = 0.04, medium γ: P = 0.003, and high γ: P = 0.0005) and during REM sleep (low γ: P = 0.009, medium γ: P = 0.0004, and high γ: P = 0.0001). As compared to SWS, the parietal-to-frontal connectivity in medium γ (P = 0.002) and high γ (P = 0.009) bands decreased further during REM sleep; there was no statistical difference in low γ between SWS and REM sleep (P = 0.05; fig. 4C; table 5, Supplemental Digital Content 1, http://links.lww.com/ALN/B316, descriptive and inferential statistics). The group-level statistical changes in medium and high γ frontal–parietal connectivity were supported by changes in connectivity in each individual rat; all rats showed a decrease in frontal–parietal connectivity in medium and high γ bands. The response was more variable in the low γ band.

In this study of multiple states of arousal, we demonstrate that corticocortical coherence and frontal-parietal directed connectivity in high γ range (85 to 155 Hz) is a consistent neural correlate of wakefulness. Reduction of high γ frontal-parietal directed connectivity was not simply found at a group level but also in each animal under the state of anesthetic- or sleep-induced unconsciousness. Frontal–parietal connectivity in low γ range (25 to 55 Hz) showed similar patterns, but the response was not consistent across all animals studied. In contrast to frontal–parietal high γ connectivity, frontal–parietal connectivity in θ bandwidth was correlated with cholinergic tone in the presence (wakefulness) or absence (REM sleep) of behavioral arousal. The relationship among different electroencephalographic indices, cortical acetylcholine, and the behavioral states is summarized in table 1.

The emergence of cortical γ oscillations has been posited to be a result of local interactions between inhibitory and excitatory processes—a reciprocal interaction between inhibitory fast-spiking parvalbumin-positive γ-aminobutyric acid neurons (interneuron–interneuron γ or the ING model) or the interaction of excitatory pyramidal neurons with these inhibitory neurons (pyramidal–interneuron γ or the PING model).^41–43  Human and animal studies have demonstrated both local and long-range γ synchronization across neuronal networks.^44–47  Previous studies of low γ oscillations (less than or equal to 50 Hz) showed that anterior–posterior corticocortical phase synchronization in rats,⁴⁸  and coherence in surgical patients,⁴⁹  decreased under anesthetic-induced unconsciousness. In a recent study conducted in surgical patients, Nicolaou and Georgiou³⁰  reported that global field synchrony in the γ band was significantly reduced during surgical anesthesia with propofol, sevoflurane, and desflurane. γ coherence was also found to be disrupted during REM sleep in animals.^50–52  In this study, we confirmed these previous findings, and in addition we demonstrated a decrease in high γ (125 to 155 Hz) coherence during propofol- and sevoflurane-induced anesthesia as well as during SWS and REM sleep.

The changes in coherence and similar phase-based measures of functional connectivity are indicative of statistical covariation in neural oscillations across the brain but do not provide any information on the influence of one brain region over another. Therefore, in order to gain insights into the information exchange between different cortical areas, we studied the effect of anesthetic- and sleep-induced unconsciousness on corticocortical directed connectivity. Past work in humans and animals supports a role for directed connectivity from frontal cortex to more posterior cortical areas in consciousness. In human volunteers and surgical patients, we have demonstrated that three distinct general anesthetics selectively suppress anterior-to-posterior directed connectivity,^7,8,11  and related studies in human volunteers show that diverse anesthetics reduce the complexity of cortical response to perturbation with transcranial magnetic stimulation.^2,4  However, these studies were either restricted to a lower γ range (because of the use of scalp electroencephalogram) or utilized a perturbational approach involving transcranial magnetic stimulation of cortex and measuring the spread of response as an index of brain connectivity. The current study in rats is the first study on spontaneous frontal–parietal connectivity across multiple states of arousal with an emphasis on high γ bandwidth. Our data confirm previous reports from human and animal studies^{7,8,11,14,33,34}  that show a preferential inhibition of frontal-to-parietal connectivity during anesthesia in the low γ range, but in addition demonstrate for the first time that high γ (85 to 155 Hz) frontal-parietal directed connectivity appears to be most consistently affected and closely correlated to loss of wakefulness during general anesthesia, SWS, and REM sleep. There are two major possibilities that account for this finding. The first is that there is a fragmentation of cortical networks, which confines information processing in the high γ bandwidth to a modular network that cannot effectively communicate. The second possibility—not mutually exclusive—is that information processing itself is diminished even within the local modules. Discriminating among these possibilities depends, in part, on the direction of the connectivity and the oscillation bandwidths. For example, the parietal-to-frontal connectivity is preserved in low γ bandwidth during anesthetic-induced unconsciousness and correlates with visual sensory processing, suggesting that feedforward information transfer in this frequency range is preserved.³³  Importantly, as opposed to the effective connectivity approach based on perturbation of cortical networks through transcranial magnetic stimulation,^2,4,6  the information theoretic approach used in the current study reliably correlated with the behavioral quiescence during REM sleep, which is often associated with disconnected consciousness, i.e., dreams.^53,54  The consistent finding of disrupted frontal–parietal connectivity in high γ bandwidth across sleep and propofol- or sevoflurane-induced unconsciousness provides further evidence that unconsciousness due to sleep and anesthesia might have shared cortical mechanisms.^55–57  Frontal–parietal high γ connectivity consistently correlated with behavioral arousal and may therefore be involved in what has been referred to as “connected consciousness,” i.e., consciousness of environmental stimuli during wakefulness. Based on our past work relating network hub structure to directionality in brain and nonbiologic networks,¹²  we propose that these changes in directed connectivity reflect shifts in network topology.

Cholinergic neurons in the basal forebrain, laterodorsal tegmentum, or pedunculopontine tegmentum are part of the arousal-promoting circuitry.^35,58  Our data show a decrease in acetylcholine in PFC during propofol- and sevoflurane-induced unconsciousness, which recovered back to baseline levels during recovery from anesthesia. Similar findings have been reported by previous studies showing a decrease in acetylcholine across the frontal cortex and in hippocampus during isoflurane, sevoflurane, and propofol anesthesia.^59–61  Cortical acetylcholine levels are also known to be high during REM sleep, a state in which consciousness can occur in the absence of behavioral arousal.^35,40,53  As opposed to high γ coherence and frontal–parietal connectivity, which correlated with the state of behavioral arousal, our data reveal that coherence and frontal–parietal connectivity in the θ bandwidth parallel levels of cortical acetylcholine. Therefore, cortical acetylcholine and θ connectivity correlate with an active cortex during wakefulness and REM sleep, rather than being associated with behavioral arousal. The dissociation between cortical acetylcholine and behavioral arousal is also supported by previous reports that ablation of cholinergic neurons in basal forebrain either did not affect total wakefulness⁶²  or had a transient effect on wakefulness,⁶³  while systemic atropine produced a dissociated state with waking behavior in the presence of slow wave electroencephalogram.⁶⁴  Conversely, cortical acetylcholine is increased after ketamine administration, despite the behavioral phenotype of general anesthesia.²⁹  Interestingly, the observed stepwise decrease in high γ frontal–parietal directed connectivity across wakefulness, SWS, and REM sleep parallels progressive reduction in levels of cortical norepinephrine across the sleep-wake cycle.⁶⁵  Furthermore, infusion of norepinephrine into basal forebrain under desflurane anesthesia in rats has been shown to produce microarousals along with increased cross-approximate entropy and decrease in the frontal delta power,⁶⁶  while antagonism of norepinephrine in rat barrel cortex produced synaptic quiescence.⁶⁷  However, there has been no direct study of the role of cortical norepinephrine in frontal–parietal brain connectivity.

A limitation of the current study is that we cannot draw causal conclusions regarding the relationships of cortical acetylcholine, behavioral arousal, and changes in electroencephalographic indices of connectivity. Furthermore, anesthetic induction and emergence are inherently unstable states and are particularly difficult to characterize in rodents, which makes it difficult to reliably quantify the neurochemical and neurophysiologic changes during state transitions. Therefore, in order to identify correlates of clearly defined states of consciousness, we focused on the most stable and representative epochs for anesthetic-induced unconsciousness. We acknowledge that the epochs of wakefulness we evaluated were states of complete wakefulness, and thus the neural correlates identified might relate to cognition rather than to consciousness, per se. As such, this investigation does not provide direct information regarding the transition points for states of wakefulness. However, unlike the many gradations along the continuum of sedation and general anesthesia that could have occurred between our data points, sleep and wakefulness are posited to discretely alternate or “flip-flop” across states.⁶⁸  Thus, the finding of impaired frontal–parietal connectivity during sleep states is supportive of observations during wakefulness and general anesthesia, despite the fact that the transitions were not studied directly.

It is also important to note that, although our study assessed changes in corticocortical connectivity as correlates of wakefulness and anesthetic or sleep-induced unconsciousness, we cannot exclude important influences of the thalamus and/or basal forebrain through the ascending reticular activation system^21,55,58  on cortical activation and intracortical connectivity. Furthermore, in order to minimize the use of research animals and to avoid redundancy, we relied on the previously published literature^35,40  and did not measure cortical acetylcholine during sleep-wake states.

Bioelectric potentials originating in muscle tissue are a known source of artifacts in scalp electroencephalographic data. In the current study, electroencephalogram was recorded using stainless steel electrodes directly implanted into the cranium. The electrodes were neither in physical contact with muscle tissues nor in close proximity (approximately 3 to 4 mm away from nearest muscle tissue), and to our knowledge there is no preclinical rodent study showing interference from muscle tissues in the electroencephalographic data recorded using transcranial stainless steel electrodes. This issue was also addressed in a previous rat study (same electrodes, montage, and recording parameters) from our laboratory²⁸  and found no significant correlation of brain and muscle activity over a wide frequency range (0.1 to 250 Hz). However, a focused analysis to exclude completely the possibility of interference from muscle tissues in electroencephalographic data was not conducted in this study.

This is the first study to report suppression of electroencephalographic coherence and frontal–parietal directed connectivity in high γ bandwidth across multiple states of unconsciousness (propofol anesthesia, sevoflurane anesthesia, SWS, REM sleep) compared with spontaneously occurring wakefulness. These data suggest that coherence and frontal–parietal connectivity in high γ range are neural correlates of wakefulness and that the fragmentation of these oscillations might contribute to the loss of behavioral arousal.

The authors thank Ralph Lydic, Ph.D. (Robert H. Cole Professor of Neuroscience, Department of Anesthesiology, University of Tennessee, Knoxville, Tennessee), for training and assistance with acetylcholine quantification. The authors also thank Chris Andrews, Ph.D. (Statistical Consultant, Center for Statistical Consultation and Research, University of Michigan, Ann Arbor, Michigan), for help with statistical analysis and Stella Wisidagamage, M.S. (Laboratory Technician, Department of Anesthesiology, University of Michigan, Ann Arbor, Michigan), for help with data collection.

Supported by grant No. R01 GM098578 from the National Institutes of Health (Bethesda, Maryland) to Dr. Mashour. Additional funding support received from the Department of Anesthesiology, University of Michigan Medical School, Ann Arbor, Michigan.

The authors declare no competing interests.