Disruption of Frontal–Parietal Communication by Ketamine, Propofol, and Sevoflurane

KETAMINE is a phencyclidine analog that belongs to a class of anesthetics (including nitrous oxide) that do not act via the potentiation of γ-aminobutyric acid (GABA).^1–3  Studies of anesthetic-induced unconsciousness have consistently failed to identify common mechanisms of both GABAergic and non-GABAergic anesthetics, which are distinct at both the molecular and neurophysiologic levels. Identifying a common neural correlate of anesthetic-induced unconsciousness would be an important advance for the mechanistic study of consciousness and general anesthesia, and could facilitate the development of more sophisticated brain monitors for surgical patients.

Information feedback from the frontal cortex to other cortical regions has been referred to as “feedback” or recurrent processing and is thought to mediate conscious experience.^4,5  In contrast, feedforward information flow in the posterior-to-anterior direction is thought to mediate sensory processing, which can occur outside of consciousness.^6,7  There is emerging evidence for disrupted feedback and preserved feedforward processing in patients with persistent vegetative states,⁸  as well as anesthetized rats.^9,10  We previously demonstrated that frontal-to-parietal feedback connectivity significantly exceeds feedforward connectivity in conscious humans, and that this feedback dominance is attenuated by the GABAergic anesthetics propofol and sevoflurane.^11,12  In the current study, we used electroencephalography and normalized symbolic transfer entropy (NSTE) to assess directional connectivity across the frontal, parietal, and temporal regions of human surgical patients. On the basis of this analysis we demonstrate that drugs from three major classes of anesthetics selectively disrupt feedback communication. This finding supports the hypothesis that feedback connectivity and frontoparietal networks play an important role in consciousness and provide evidence for a common neurobiology of anesthetic-induced unconsciousness.

In brief, 30 surgical patients were anesthetized with ketamine (2 mg/kg intravenous infusion) while recording eight-channel electroencephalography of frontal, parietal, and temporal regions. NSTE is a model-free, nonlinear measure of the causal relationship between two signals, and can be used to infer directed functional connectivity.^13–17  NSTE between frontal and parietal or frontal and temporal regions was calculated for patients in the conscious state and after induction of anesthesia with ketamine. We also reanalyzed electroencephalographic data from surgical patients anesthetized with the intravenous anesthetic propofol (n = 9) or the inhaled anesthetic sevoflurane (n = 9),¹²  in order to compare directly the spectral changes and NSTE with that of ketamine. For the purpose of statistical analysis, the electroencephalogram time series during baseline consciousness and after anesthetic exposure was divided into three equal epochs for each state, resulting in six substates (baseline—B1, B2, B3; anesthesia—A1, A2, A3).

The study was approved by the Institutional Review Board of Asan Medical Center (Seoul, South Korea), and written informed consent was obtained in all cases; electroencephalographic data were gathered at Asan Medical Center and analyzed at the University of Michigan Medical School (Ann Arbor, MI). Investigators from both sites participated in the design of the study. Patients scheduled for elective stomach, colorectal, thyroid, or breast surgery (n = 30, male/female = 15 /15, American Society Anesthesiologists Physical Status I or II, aged 22–64 yr) were enrolled in this study. Exclusion criteria included previous cardiovascular disease (including hypertension), a previous brain surgery, a history of drug or alcohol dependence, known neurological or psychiatric disorders, or current use of psychotropic medications.

Patients received no sedatives or other medications before induction of anesthesia, and were monitored with electrocardiography, pulse oximetry, end-tidal carbon dioxide concentration, and noninvasive blood pressure measurement. Ketamine (2mg/kg diluted in 10ml of 0.9% normal saline) was infused over 2min (Baxter infusion pump AS40A; Baxter Healthcare Corporation, Deerfield, IL). We assessed noninvasive blood pressure every 30 s during ketamine infusion, and administered 5–10mg labetalol if systolic blood pressure was increased 30% higher than baseline blood pressure. Time to loss of consciousness was determined by checking every 10 s for the loss of response to verbal command (“squeeze your right hand twice”). Electroencephalographic and electromyographic data were acquired until 5 min after loss of consciousness. At the end of the study, patients received an effect-site target concentration (2.5 μg/ml) of propofol in combination with an effect-site target concentration (5ng/ml) of remifentanil.

These data were originally gathered for a previous study of the frontal–parietal system, although NSTE was not applied for the original analysis. Eighteen patients scheduled for elective abdominal or breast surgery (n = 18, male/female = 8/10, American Society Anesthesiologists Physical Status I and II, aged 29–66 yr) were enrolled in the study. Propofol (n = 9) or sevoflurane (n = 9) was administered to induce general anesthesia while eight-channel electroencephalogram was recorded. Propofol (Diprivan; AstraZeneca, London, United Kingdom), was initially administered with a target-controlled infusion of 2.0 μg/ml and was increased at a rate 1.0 μg/ml per 20 s until loss of consciousness. Sevoflurane (Sevorane, Abbott, IL), was initially administered as 2 vol% and increased at a rate of 2 vol% per 20 s (up to 8%) until loss of consciousness; see study by Ku et al.¹²  for experimental details.

In all experiments, the electroencephalogram was recorded at eight monopolar channels in the frontal, parietal, and temporal regions (Fp1, Fp2, F3, F4, T3, T4, P3, and P4 referenced by A2, which followed the international 10–20 system for electrode placement) by a WEEG-32 (LXE3232-RF; Laxtha Inc., Daejeon, Korea) with a sampling frequency of 256 Hz. The electrode impedance was kept below 5 KΩ. Band-pass filtering with the fifth-order Butterworth filter was applied to electroencephalographic data (forward and backward), correcting the phase shifting after band-pass filtering (“butterworth.m” and “filtfilt.m” in Matlab Signal Processing Toolbox; MathWorks, Natick, MA). Electromyography was concurrently recorded at four bipolar channels (bilateral frontalis and temporalis muscles) by a QEMG-4 (Laxtha Inc.) with a sampling frequency of 1,024 Hz.

Transfer entropy (TE) offers a nonlinear and model-free estimation of directed functional connectivity based on information theory, quantifying the degree of dependence of Y on X or vice versa.^16,17  The TE can be defined as the amount of mutual information between the past of X (X^P) and the future of Y (Y^F) when the past of Y (Y^P) is already known, i.e.,

where H (Y^F/Y^P) is the entropy of the process Y^F conditional on its past.

The distributions of X^P, Y^P, and Y^F can be written explicitly as

Equation (3) shows that TE represents the amount of information provided by the additional knowledge of the past of X with respect to the future of Y.

One disadvantage of TE is the subjective decision for the bin size in the probability calculation in Equation (2). To avoid this problem, symbolic TE (STE) can be used. In STE, each vector for Y^F, X^P, and Y^P in Equation (2) is a symbolized vector point. For instance, a vector Y_t consists of the ranks of its components , where is replaced with the rank in ascending order, for j = 1, 2,…,m. Here m is the embedding dimension and τ is the time delay. STE is defined in the same way as Equation (2), but replacing the embedded vector points with the symbolized vector points. In comparison with TE, STE is a more robust and computationally efficient method.^16,18  A schematic illustrating the principles of TE is presented in figure 1.

To remove the bias of STE for a given electroencephalographic data set, the shuffled-data method was used.¹⁹  The shuffled data retain the same signal characteristics as the original signal does, but the causal relation is completely eliminated. This shuffling process was applied only to the source signal (X), leaving the target signal (Y) intact. The STE with the shuffled source signal , estimates the bias caused by the signal characteristics of the source signal (X). The unbiased STE was normalized as following,

NSTE is normalized STE (dimensionless), in which the bias of STE is subtracted from the original STE and then divided by the entropy within the target signal, H (Y^F/Y^P). Intuitively, NSTE represents the fraction of information in the target signal Y, which is not explained by its own past and explained by the past of the source signal X.¹⁹ 

Finally, the asymmetry between NSTE_{x → y} and NSTE_{y → x} was defined as following,

Therefore, if DF_{x → y} has a positive value, the connectivity from X to Y is dominant, and vice versa for a negative value. The feedback and feedforward connections in the frontoparietal network were evaluated with NSTE_{f → p} and NSTE_{p → f} over the 48 subjects (30 ketamine, 9 propofol, 9 sevoflurane). The average and were calculated over the eight pairs of electroencephalogram channels between the frontal and parietal regions for each subject; , where n_f = 4 and n_p = 2. The asymmetry of information flow between the two brain regions was defined as (Equation 5) for each subject.

During ketamine anesthesia, we found that NSTE_f→_p and NSTE_p→_f have multiscale properties, showing distinct information transfer between frontal-parietal regions on short- and long-term scales. This may be associated with simultaneous increases of gamma and delta powers (relatively short- and long-term dynamics). Therefore, information transmission of a single time scale would not be able to represent the multiscale connectivity of ketamine anesthesia. In this study, we found that the maximum information transfer between frontal and parietal regions provides a consistent connectivity feature of ketamine, propofol, and sevoflurane. Three embedding parameters, embedding dimension (d_E), time delay (τ), and prediction time (δ), are needed for NSTE. The parameter set that provides the maximum information transfer (NSTE) from the source signal to the target signal was selected as the primary connectivity for a given electroencephalographic data set, instead of applying a conventional embedding method. By investigating the NSTE in the broad parameter space of d_E (from 2 to 10) and τ (from 1 to 30), we fixed the embedding dimension (d_E) at three, which is the smallest dimension providing a similar NSTE, and found the time delay (τ) producing maximum NSTE. In this parameter space, a vector point could cover from 11.7ms (with τ = 1 and d_E = 3) to 351ms maximally (with τ = 30 and d_E = 3). If a parameter set for maximum information transfer was determined in one direction, we used the same parameters for the opposite direction. Taking the maximum NSTE as the primary connectivity for a given electroencephalographic data set, it can be concluded that all other processes are nonparametric without subjective decisions for embedding parameters. The prediction time was determined with the time lag (1–100, 3.9–390ms) resulting in maximum cross-correlation, assuming the time lag as the interaction delay between the source and target signals.

For the purpose of statistical evaluation, we segmented 10-min long electroencephalographic data into six substates: three substates for baseline (B1, B2, and B3) and three substates for the period of anesthetic exposure (A1, A2, and A3), with each substate being a 100-s long electroencephalographic epoch. The feedback or feedforward connectivity of each substate was defined as the mean value of connectivity over 10 small windows (10-s long). For each small window, the average NSTE, , and , between four frontal and two parietal electroencephalogram channels was calculated. The small window size of 10 s may satisfy pseudostationary conditions for NSTE calculation, and the mean value over 10 small windows may reflect the connectivity of a substate. Therefore, the data format of the ketamine group is 30 subjects and 6 substates for each connection. The feedback and feedforward connectivity, and , were compared across the six substates; the significance was assessed by a repeated measure ANOVA and a post hoc analysis using Tukey multicomparison test. The mean ± SD for each connection is presented in Supplemental Digital Content 1 (https://links.lww.com/ALN/A923) and the results of the post hoc test are presented in the Results. The D’Agostino-Pearson omnibus normality test and Freedman nonparametric test were conducted before the ANOVA test. A P value less than 0.01 was considered significant. GraphPad Prism Version 5.01 (GraphPad Software Inc., San Diego, CA) was used for the tests.

Despite efforts to remove bias from NSTE measurements, there is still a potential for false-positive causality. We therefore applied the permutation and time-shift tests to evaluate, respectively, the significance and the linear mixing effect on the connectivity. At each substate (B1, B2, B3, A1, A2, and A3), 10 trial data sets were generated by continuously dividing 100-s long electroencephalographic epochs into 10-s long epochs. The permutation and time-shift tests were then applied to the 10 trial data sets, derived from eight-channel electroencephalogram. The test statistics () were calculated for eight pairs of electroencephalographic channels.

The permutation test is a nonparametric statistical significance measure that is used to assess whether the test statistics of two groups are interchangeable.^13,14  The null hypothesis is that the means of the test statistics (NSTE) of original and randomized electroencephalographic data are interchangeable. The null hypothesis was considered to be rejected with P value less than 0.01. The significance test for connectivity was applied to each pair of electroencephalographic channels with 10 trials and the corresponding 10 shuffled data sets.

With the same data format, the time-shift test was applied to evaluate possible false-positive connectivity due to an instantaneous linear mixing effect.^13,14  In the test, because the future of the source signal has no causal relationship with the present of the target signal, if NSTE (t) < NSTE (t'), t'= t + δ, where t' is the future time and δ is an interaction time delay between two signals, then the two signals have an instantaneous linear mixing with time delay δ. This also holds for the shifted signal with the past of source signal (t'= t − δ). As with the permutation test, the time-shift test was applied to the original and time-shifted (i.e., shuffled) electroencephalographic data. The significance of the mean difference between the two groups was evaluated with the permutation test (nonparametric method, P < 0.01).

The relative power of the electroencephalogram spectrum is the percentage of power contained in a frequency band relative to the total spectrum (0.1–35 Hz). We removed frequency bands above 35 Hz to avoid muscle artifact. The relative powers of δ, θ, and γ bands increased after injection of ketamine (across 3 substates of baseline consciousness and 3 substates of ketamine anesthesia: F(5, 179) = 3.54, P = 0.0047 for δ; F(5, 179) = 11.28, P < 0.0001 for θ; and F(5, 179) = 68.86; P < 0.0001 for γ), whereas the relative powers of α and β bands decreased (across 6 substates: F(5, 179) = 52.25, P < 0.0001 for α; F(5, 179) = 14.38, P < 0.0001 for β). The simultaneous increase of the relative powers for both slow waves (δ and θ) and fast waves (γ) was unique to the ketamine power spectrum (fig. 2, first column). Propofol and sevoflurane (fig. 2, second and third columns) induced a spectral pattern distinct from that of ketamine anesthesia: there was an increase of δ, θ, and α powers (F(5, 53) > 3.3, P < 0.05 for propofol; F(5, 53) > 4.4, P < 0.001 for sevoflurane) and decrease of β and γ powers (F(5, 53) > 3.5, P < 0.01 for propofol; F(5, 53) > 11, P < 0.001 for sevoflurane). The difference among the three anesthetics was most salient in α (reduced by ketamine, increased by propofol and sevoflurane) and γ (increased by ketamine, reduced by propofol and sevoflurane). Thus, the effects of GABAergic and non-GABAergic anesthetics were associated with distinct electroencephalographic spectra. In terms of the regional difference of relative powers for the five frequency bands, the F3, F4 and P3, P4 channels showed a significant power difference. See table 1 in Supplemental Digital Content 1 (https://links.lww.com/ALN/A923), which shows regional differences in relative power after administration of ketamine.

To examine whether the increased γ power after administration of ketamine was due to electromyographic artifact, we investigated Pearson correlation coefficients between electromyography and electroencephalography in the γ frequency bands (25–35 Hz). The γ waves for the electromyogram and electroencephalogram are presented in figure 3, A and B; a scatter plot of both signals during anesthesia is presented in figure 3C. The mean and SD of Pearson correlation coefficient was −0.0106±0.05, which was calculated only with the pairs of channels with P value less than 0.01. No correlation was found between the twosignals.

Distinctive patterns of connectivity were demonstrated across different brain regions (fig. 4, A–F). Figure 4A demonstrates feedback dominance between frontal–parietal regions during the baseline conscious state, which is consistent with our previous findings.^11,12  During administration of ketamine, feedback connectivity was gradually reduced and significantly inhibited after loss of consciousness (F(5, 179) = 7.785, P < 0.0001 for feedback connectivity; P < 0.001 for A2 and B1, B2, B3). Mean time to loss of consciousness was 89.4 ± 18.4 s. All but two patients were unresponsive at the end of the 2-min infusion and all were within 30 s thereafter. The reduced feedback connectivity began a notable but statistically insignificant rebound (P > 0.01, not significant for B2, B3, and A3), which may reflect the waning effect of the single drug bolus. By contrast, the feedforward connectivity was preserved for the entire experimental period (F(5, 179) = 2.07, P = 0.072; P > 0.01, not significant for any substates).

Changes in the two directions of connectivity gave rise to a significant change in the asymmetry of feedback and feedforward connectivity after anesthesia (F(5, 179) = 23.88, P < 0.0001). The asymmetry of information flow in the frontal–parietal network was largely reduced during ketamine injection, which induced a balanced information flow during the first 1 min after loss of consciousness (P < 0.001 between B1, B2, B3 and A1, A2, A3). This balanced bidirectional connectivity began to break down from the substate A3 (P < 0.01, for A2 and A3). On investigating the frontal and temporal regions (fig. 4B), we also found feedback dominance in the baseline conscious state and selective inhibition of feedback connectivity after ketamine injection (F(5, 179) = 6.59 and P < 0.0001 for feedback, but F(5, 179) = 2.42, P = 0.038 for feedforward; A2 is most significant). Figure 4E demonstrates that ketamine was associated with a reduction in the asymmetry, but the feedback connectivity still maintained a relatively greater information flow. Thus, the inhibition of feedback connectivity in the frontal–temporal network was not as robust as that of the frontal–parietal network. The tendency of rebound of feedback and feedforward connectivity in the frontal–temporal network in substate A3 was still observed.

Ketamine had no effect on horizontal connectivity across the hemispheres (F(5, 179) = 1.76, P = 0.124 for left to right; F(5, 179) = 2.25, P = 0.052 for right to left), despite the disruption of the rostral–caudal connectivity in the brain. Figure 4, C and F shows that the information flow between left and right hemispheres was balanced and no asymmetry was identified, even in the baseline waking state. The connectivity among eight electroencephalographic channels in six substates is presented in figures 1 and 2 of Supplemental Digital Content 1 (https://links.lww.com/ALN/A923). The variance of individual connectivity during ketamine anesthesia is presented in figure 3 of Supplemental Digital Content 1 (https://links.lww.com/ALN/A923).

The simultaneous increase of slow and fast electroencephalogram waves (δ and γ bands) were associated with multiscale information transmission structures during ketamine anesthesia (fig. 5, A–F). In particular, the rebound of feedback/feedforward connectivity was identified in short time scales (τ = 1, 2, and 3 with d_E=3; 11, 23, and 35ms in fig. 5, A and D) during ketamine anesthesia, but did not appear at longer time scales (τ = 10, 15, and 20 with d_E = 3; 117, 175, and 234ms in fig. 5, B and E). However, despite the reactivation of connectivity, the reduced asymmetry between the feedback and feedforward connectivity was maintained (F(5, 179) > 7, P < 0.001; B1, B2 and B3 > A1, A2, and A3 for the asymmetries of all τ in fig. 5, C and F). Distinct from the baseline conscious state, the rebound feedback connectivity was accompanied by increasing feedforward connectivity, which maintained the reduced asymmetry during ketamine anesthesia. Thus, although activity of feedback/feedforward connections was variable across time scales, the reduced asymmetry was consistently associated with the loss of consciousness. Furthermore, the feedback dominance in the baseline state and the selective inhibition of feedback connectivity during anesthesia were found irrespective of time scales.

Spurious interpretation of correlation or causality can occur when measuring connectivity. As such, the time-shift and permutation tests^13,14  were applied to all pairs of electroencephalographic channels between frontal and parietal regions in order to test the significance and assess the false-positive connections generated by NSTE analysis. To apply these tests, 10 trials (10-s long electroencephalogram epochs) and 10 corresponding shuffled data sets were generated from each substate for each subject receiving ketamine. The rejection rates of the null hypothesis for the two statistical tests are summarized in table 1. If the null hypotheses of both tests are rejected, the strength of connectivity is deemed robust and reliably deviates from random connectivity generated from potential linear mixing. The percentage of connections passed for both tests was evaluated over all possible pairs of the subjects (2,880 pairs: 2 directions × 6 substates × 8 channel pairs for frontal and parietal regions × 30 subjects). The feedback connections in the baseline state had a higher null hypothesis rejection rate for both tests (30% among all 720 feedback connections for B1, B2, and B3), indicating that the feedback connections in the baseline state were relatively less susceptible to the potential risk of linear mixing. The feedforward connections in the baseline state had a rejection rate of 16%; the feedback and feedforward connections during ketamine anesthesia had rejection rates, respectively, of 23 and 17% on average. Comparing the directional connections, the feedforward connections had relatively lower null hypothesis rejection rates for both tests, which may be due to relatively smaller NSTE values in the feedforward direction. Regarding the dependence on state, the baseline conscious state had a higher rejection rate than the anesthetized state.

Analysis of only the connections that survived both the permutation and time-shift tests for each substate demonstrated the same patterns of the feedback/feedforward connectivity. The selectively inhibited feedback connection and preserved feedforward connection during ketamine anesthesia, as well as the significantly reduced asymmetries of feedback/feedforward connectivity, were confirmed (F(5,179) = 9.4, P < 0.0001, A2/A3 < A1 and baseline states for feedback connections; F(5, 179) = 0.54, P = 0.74 across six substates for feedforward connections; F(5, 179) = 8.13, P < 0.0001 for feedback/feedforward asymmetry).

Figure 6, A–F shows feedback/feedforward connectivity and feedback/feedforward asymmetry in frontal–parietal networks during baseline consciousness and unconsciousness induced by ketamine, propofol, and sevoflurane. Because propofol and sevoflurane had longer electroencephalogram epochs for the induction of anesthesia compared with ketamine (from start of drug administration to loss of consciousness: 4.09±1min for propofol; 3.8±0.73min for sevoflurane), we resampled the feedback and feedforward connections measured by NSTE with the same data length. For convenience of comparison, we rescaled the time axis. The epoch lengths for induction and anesthesia were set as 5 and 8min, respectively, and lengths were matched using Matlab function (“resample.m”). Thus, the x-axis in figure 6, B, C, E, and F does not represent the true timeline for each patient. The mean and standard error of NSTE for the baseline state and the mean and standard error for the “resampled NSTE” for induction and anesthesia over all subjects (9 subjects for propofol and 9 subjects for sevoflurane) are presented as continuous in figure 6, B, C, E, and F.

The dominant feedback connectivity in the baseline state and the selective inhibition of feedback connectivity after induction was clearly demonstrated across all three anesthetics (F(5, 179) = 9.18, P < 0.0001 for propofol and F(5, 179) = 4.11, P = 0.0042 for sevoflurane 6 six substates). In contrast, feedforward connectivity was preserved irrespective of state, a consistent finding across all three anesthetics (F(5, 179) = 0.179, P = 0.967 for propofol and F(5, 179) = 0.7, P = 0.625 for sevoflurane across 6 substates). As a result, the reduction of feedback dominance and feedback/feedforward asymmetry (F(5, 179) = 4.97, P = 0.0012 for propofol and F(5, 179) = 5.3, P = 0.0008 for sevoflurane across six substates) in the frontal–parietal network was a common neural correlate of anesthetic-induced unconsciousness across ketamine, propofol, and sevoflurane, which represent three molecularly distinct classes of anesthetics. The mean ± SD of feedback and feedforward connections for six substates are presented for ketamine, propofol, and sevoflurane in table 2 of Supplemental Digital Content 1 (https://links.lww.com/ALN/A923).

This is the first study to assess changes in directional connectivity during ketamine anesthesia in humans and also the first to provide evidence for a common correlate of both non-GABAergic (ketamine) and GABAergic (propofol, sevoflurane) anesthetics that is based in the neurobiology of consciousness. Despite the distinct spectral changes of ketamine compared with propofol and sevoflurane, all anesthetics inhibited feedback connectivity, preserved feedforward connectivity, and thus inhibited the feedback dominance observed during the baseline conscious state. These findings have implications for the measure and mechanism of general anesthesia, as well as the role and mechanism of feedback connectivity in consciousness. It is important to note that this study assessed only connected or external consciousness (i.e., consciousness of environmental stimuli), which is thought to be mediated by lateral frontoparietal networks, rather than disconnected or internal consciousness (e.g., dream states), which is thought to be mediated by more medial networks. Unresponsiveness to an unambiguous verbal command was used as a surrogate for anesthetic-induced loss of external consciousness.

General anesthetics can be roughly organized into three groups.²⁰  Group 1 includes primarily GABA_A receptor agonists such as propofol, etomidate, and barbiturates, which induce unconsciousness but do not effectively suppress movement. Group 2 includes ketamine, nitrous oxide, and xenon, which are not thought to act primarily via GABA_A receptors, but rather through N-methyl-d-aspartate or other receptor types. These drugs have a potent analgesic effect, but are relatively weak hypnotics and immobilizers. Finally, group 3 drugs include volatile anesthetics, such as sevoflurane, desflurane and isoflurane, which are potent hypnotics, immobilizers, and amnesics, and have diverse molecular effects such as GABA_A and glycine receptor agonism, two-pore potassium channel activation, and excitatory neurotransmitter receptor antagonism. As can be seen in table 2, ketamine (group 2) does not share the same molecular, neuroanatomic, or neurophysiologic properties of propofol (group 1) or sevoflurane (group 3). Ketamine is thought to act via glutamatergic N-methyl-d-aspartate¹  receptors or HCN1 channels³; unlike virtually all other inhaled and intravenous anesthetics, ketamine does not metabolically inhibit the thalamus.²¹  Furthermore, ketamine does not activate the sleep-associated ventrolateral preoptic nucleus as other Group 1 and Group 2 anesthetics do, but rather activates numerous brainstem and diencephalic arousal centers.²²  This arousal mechanism may contribute to the increase in high-frequency and desynchronized waveforms observed on the electroencephalogram.²³  Due to their unique neurophysiologic profile, the effects of ketamine and nitrous oxide are not accurately reflected in indices generated by commercially available brain monitors,²⁴  which generally have algorithms that rely on a slowing electroencephalogram frequency. Therefore, the effects of ketamine and other non-GABAergic drugs have eluded generalized frameworks of anesthetic mechanism (table 2) and are not reliably detected in the clinical setting by conventional monitoring technology.

The current study suggests that measurement of directional connectivity in the frontoparietal or frontotemporal region could reflect the effects of all major groups of anesthetics. Because the analytic measure of NSTE is rooted in information theory—and because the frontoparietal network is an important site of information convergence—our findings are consistent with the integrated information theory of consciousness. We predict that measures of information integration, such as the complex phi (Φ),²⁵ would also be reduced by diverse anesthetics. However, Φ as a real-time measure of integrated information may require high-density electroencephalography, a perturbational approach such as transcranial magnetic stimulation, and significant computational burden for accurate calculation, all of which are currently impractical for a routine clinical care setting. Our past and current studies demonstrate proof of principle that feedback and feedforward values—as well as the asymmetry between them—can be determined with relatively few channels in human surgical patients, although real-time analysis has not yet been attempted.

A recent study using high-density electroencephalography also found selective inhibition of feedback connectivity in frontoparietal networks after propofol-induced unconsciousness and, using dynamic causal modeling, suggested that this finding reflects a corticocortical interaction rather than the cortical signature of an underlying thalamocortical event.²⁶  Another recent study using positron-emission tomography suggested a unique importance of frontoparietal connectivity in the recovery of consciousness after anesthesia.²⁷  Thus, it is likely that the measurement of feedback and feedforward connectivity demonstrated in the current study reflects a corticocortical interaction and the important role of frontal–parietal communication in consciousness and anesthesia.^25,28  These findings support the hypothesis that the disruption of top–down communication from the frontal cortex is a common mediator of general anesthetic effects,⁵  and provide a framework for future mechanistic studies of anesthetic-induced unconsciousness that link distinct molecular and neurophysiologic actions to a common information-processing endpoint. For example, the thalamocortical hypersynchrony and increased power of α waves caused by propofol^29,30  likely block the flexible corticocortical communication required for consciousness. However, we demonstrate that ketamine—unlike propofol—decreases α power in association with anesthetic-induced unconsciousness (although synchrony was not measured directly). Thus, there is likely an alternative neurophysiologic pathway that nonetheless results in the inhibition of frontal–parietal communication.

Feedback or recurrent processing, both within sensory modalities and across association cortices, has been proposed to be critical for consciousness.^4–7  Although focused mainly on the frontal, parietal, and temporal regions using low-resolution scalp electroencephalography, our past studies^11,12  and the current study demonstrate, by two distinct analytical methods, that consciousness of the environment is associated with a dominant cortical feedback originating in the frontal region. These empirical findings are consistent with a recent neural mass model based on diffusion tensor imaging of structural connectivity, in which phase lag index measures revealed a dominant flow of information from the frontal cortex to an information-processing hub of the posterior parietal cortex.³¹  On the molecular level, it has been suggested that feedback processing is mediated by N-methyl-d-aspartate receptor activity whereas feedforward connectivity is mediated by α-amino-3-hydroxy-5-methyl-isoxazolepropionic acid receptors,⁴  which has recently been confirmed in a study of the macaque visual system.³²  The current study further supports this hypothesis because ketamine—which has glutamatergic N-methyl-d-aspartate-receptor antagonist effects—selectively inhibits feedback connectivity while preserving feedforward connectivity.

This investigation is limited by the spatial resolution of an eight-channel electroencephalogram and thus our findings cannot be extrapolated beyond the regions analyzed. However, obtaining such findings in surgical patients with relatively few channels in a real-world clinical setting enhances the translational potential of this study. As noted, a recent study of propofol using high-density electroencephalography is consistent with our finding and suggests that it reflects a direct corticocortical interaction; the stringent statistical tests applied in the current study also indicate that the effects of ketamine on feedback connectivity are robust. Nonetheless, further study with high-density electroencephalography and stepwise titration of multiple anesthetics will be important confirmation. Further confirmation of increased frontoparietal connectivity and communication during recovery from general anesthesia will also be required.^12,27  The current study did not address recovery because ketamine was used only for induction in these surgical cases.

It should be noted that two recent studies assessing directional connectivity with Granger causality did not find a decrease of connectivity after propofol-induced unconsciousness.^33,34  These results, however, are inconsistent with the known effects of general anesthetics. General anesthesia is associated with a metabolic deactivation of frontoparietal networks³⁵  and a decrease in frontoparietal functional connectivity²⁸ —it is therefore unlikely that feedback connectivity would be maintained or increased under these conditions. However, we must acknowledge that different analytic techniques and protocols may yield different findings. For instance, variations in electroencephalogram reference, commands, eye opening, and time scale of analytic techniques may influence connectivity results. Furthermore, like other connectivity measures, NSTE analysis has numerous limitations that weaken the assertion of a truly “causal” interaction between brain regions. NSTE does not work well in higher dimensions (more than 5) because the number of possible symbols increases by a factorial order with dimension. Thus, NSTE requires a data set long enough for proper estimation of conditional probability; the short stationary epoch in the electroencephalogram conflicts with this requirement. The NSTE measure itself is not sensitive to the effects of a latent third source, which can generate spurious causality. The high dimensionality, nonstationarity, and third-source problems limit many current analyses of connectivity; considerably more work must be done to establish definitively that a particular direction of connectivity or communication is critical to the mechanism of consciousness and anesthesia. However, the consistent results across multiple anesthetics, multiple analytic methods, and multiple studies encourage further research to test the hypothesis that impaired communication in the frontal–parietal network reflects a common final pathway of anesthetic mechanism.

In conclusion, we demonstrate that anesthetics from three major drug classes disrupt frontal–parietal communication in a characteristic and consistent way, as measured by NSTE in the feedback and feedforward directions. These data provide the foundation for a common metric and common neurobiology of anesthetic-induced unconsciousness.