#### What We Already Know about This Topic

#### What This Article Tells Us That Is New

^{1},

^{2}the neurophysiological mechanisms of general anesthesia-induced loss of consciousness have not been fully clarified.

^{2}Over the last decade, many theories have been proposed to explain the mechanisms of consciousness, such as the cognitive binding paradigm

^{3}and the cortical information integration and information capacity theory.

^{2},

^{4–7}Common among these theories is the idea that consciousness is supported by an optimal balance between segregation of functionally diverse regions, and global coupling. The implication is that loss of consciousness during general anesthesia is caused by a breakdown of cortical connectivity, resulting in a reduction in coupling between widely distributed regions, but, paradoxically, with increased local synchronization.

^{2},

^{8}

^{9}Isoflurane impairs anteroposterior phase synchronization in 5–25 and 25–50 Hz frequency bands,

^{10}and disrupts frontal-to-posterior information transfer at high γ band frequencies in rats stimulated with light flashes.

^{11}The mean information integration capacity is reduced in several frequency bands after induction with propofol

^{12}; specifically, frontal-to-parietal transfer is inhibited, while communication in the other direction—from the parietal forward to the frontal region—persists.

^{13},

^{14}These studies focused on long-range neural communication associated with anesthesia-induced unconsciousness across multiple brain regions.

^{15–17}Because of the local cytoarchitectonic packing, cortical activity is inclined to synchronize into a common resonance mode with anesthetics.

^{18}Recently, Kreuzer

^{17}recorded local field potentials from the somatosensory cortex of three rats and found that the cortical signals were

*more*synchronous during anesthesia. Increase in local synchrony also accounts for the increase in amplitude of the electroencephalogram signal during anesthesia. In the aforementioned study, the neural synchrony between signals from each channel pair was measured in terms of cross-approximate entropy (AE), which is a nonlinear statistical parameter that quantifies the dissimilarity of patterns in a series of pairs.

^{19}We quantified the synchrony of signals from each channel pair by the coherence at different frequency bands, then an S-estimator was used to the bi-channel coherence matrix, which is a novel method to estimate the synchronization in multichannel electroencephalographic series

^{20},

^{21}, and finally, the index derived from the S-estimator (SI) was used to track the overall synchronization changes during anesthesia.

#### Materials and Methods

##### Data Recordings

^{19}After approval by the Animal Ethics Committee (University of Adelaide, Adelaide, South Australia), the effects of sevoflurane, desflurane, isoflurane, and enflurane were studied in eight adult merino sheep, each weighing approximately 50 kg. Under halothane anesthesia, a midline craniotomy was performed through the frontal bone, 1 cm anterior to the base of the cornual process. A standard, premanufactured linear array of eight stainless-steel electrodes (2 mm spacing–total length 14 mm) was placed over the para-sagittal frontoparietal cortex, penetrating 1–2 mm into the outer layers of the cortical gray matter. The electrocorticogram was obtained using two electroencephalogram monitors (A1000, Aspect Medical Systems, Natick, MA), and the signal was digitally sampled at 256 Hz. After at least 24 h for the animal to recover, the sheep were given a series of short inhalational anesthetics in oxygen. Sevoflurane, desflurane, isoflurane, and enflurane were administered in random order and increasing concentration until burst suppression was noted on the electrocorticogram, and the sheep was then allowed to recover fully (at least 2 h) before the administration of the next agent. More details can be found in the references.

^{19},

^{22}

^{23}(sample rate of 128 Hz, embedding of

*m*= 2, and noise threshold r = 0.2 × SD). Due to the incorporation of a lower threshold for noise, the AE could correctly classify the occurrence of burst suppression with increasing anesthetic drug effect.

^{24}

*N*= 8 for sevoflurane, and

*N*= 7 for desflurane, isoflurane, and enflurane anesthesia. The average of all the available electrocorticographic recordings was subtracted from each channel to remove the interference of baseline drift. Raw data segments of 1 min at different anesthetic states were extracted from each sheep to calculate the coherence-based SI indices. Each segment was visually inspected, and any segment with significant artifacts was excluded from the analysis.

^{25}Thus, in the current study, each segment was subdivided into 10-s epochs, with 75% overlap, then the SI values were computed for each epoch, and finally, a reliable estimate was obtained by averaging the SI values of all epochs.

##### Coherence Analysis and S-estimator

^{26}Given two electrocorticographic series

*x*

_{i}and

*x*

_{j},

*i*,

*i*= 1,2,……,

*M*, and

*M*is the number of available channels, the average synchronization at a frequency band [

*f*

_{L},

*f*

_{H}] can be calculated and denoted as

*d*

_{ij}. Detailed information is shown in the appendix. Then for the

*M*-channel electrocorticographic series, the synchrony at frequency band [

*f*

_{L},

*f*

_{H}] can be described by the coherence matrix,

*D*, with each element of it,

*d*

_{ij}, denoting the correlation between the electrocorticographic signals from the

*i*

^{th}channel and

*j*

^{th}channel.

_{i}

^{raw},

*i*=1,2,......,

*M*are eigenvalues, and

*v*

_{i}are the corresponding eigenvectors. As

*D*is a real symmetric matrix, all eigenvalues are real numbers, and they can provide information about the synchronization among individual elements of the matrix.

^{27}If the multichannel electrocorticographic signals

*x*

_{i},

*i*= 1, 2,……,

*M*are statistically independent, all the eigenvalues tend to be equal to 1. In contrast, if the signals are well-synchronized, only a few numbers of eigenvalues will remain prominent, and the others will be close to 0.

^{28}

^{21}In the current study, the estimate of coherence mainly depends on the phase relation between neural activities, thus we randomize the phase relationship using a typical nonlinear resampling method—the iterative amplitude-adjusted Fourier transform algorithm,

^{29}which can retain both the amplitude distribution and the power spectrum to a high degree of precision.

^{29}Details can be found in the references.

^{29},

^{30}For the two signals

*x*

_{i}and

*x*

_{j},

*i*,

*j*= 1, 2, ……,

*M*, first resampling one signal through the iterative amplitude-adjusted Fourier transform method, a surrogate coherence matrix is obtained, and surrogate eigenvalues can be derived by eigenvalue decomposition. This procedure is run

*N*

_{sum}(

*N*

_{sum}> 19) times to obtain

*N*

_{sum}groups of eigenvalues. The averaged surrogate eigenvalues are calculated and denoted as λ

_{i}

^{surro},

*i*=1,2,......,

*M*. Thus, the improved eigenvalues of statistical significance can be derived by the raw eigenvalues divided by the averaged surrogate eigenvalues,

*i.e.*,

^{20},

^{21}

*i*=1,2,......,

*M*are all equal to 1/

*M*, so SI = 0. On the contrary, if all the series are perfectly correlated, the variation in the coherence can be described by a single normalized eigenvalue

^{28}

##### Statistical Analysis

*post hoc*tests were used to explore the existence of significant differences in the synchronization at different study periods. If the assumption of homogeneity of covariance was violated by Mauchly’s test of sphericity, a Greenhouse–Geisser correction method was used. For each sheep, the association between the SI measure and the underlying anesthetic depth was assessed by the prediction probability (P

_{K}).

^{31}P

_{K}was calculated for each sheep, with the averaged AE index (as an indicator of the depth of anesthesia) as an independent variable, and the SI indices at the studied frequency bands as dependent variables. A P

_{K}value of 1 means that the SI index is perfectly concordant with the underlying anesthetic depth. A value of 0.5 means that the SI is not superior to that obtained by chance. The resultant P

_{K}value is replaced by 1−P

_{K}when there is a negative correlation between the AE and the SI indices. A further comparison of these P

_{K}values at different frequency bands was performed by using one-way repeated measures ANOVA with Bonferroni

*post hoc*tests. For all tests,

*P*< 0.05 were considered significant. Data are presented as mean ± SD, unless specially stated.

#### Results

*post hoc*tests) were conducted for SI values from awake to burst suppression (states I–III), and from burst suppression to recovery (states III–V), respectively. As shown in table 1, all the comparisons revealed significant differences (

*P*< 0.05). Further,

*post hoc*tests were carried out and are shown in figure 2. In almost all the cases, the burst suppression (III) can always be distinguished from the anesthesia (II and IV) and awake/recovery states (I and V;

*P*< 0.05). The SI values at the α and β bands could also differentiate the anesthesia and awake/recovery states (

*P*< 0.05).

*P*< 0.01), isoflurane (

*P*< 0.05), and enflurane anesthesia (

*P*< 0.01); the SI values across the whole frequency spectrum and γ band also revealed significant differences under isoflurane (

*P*< 0.05) and enflurane anesthesia (

*P*< 0.01), and those at the δ band revealed significant differences under isoflurane anesthesia (

*P*< 0.01). These results demonstrate that the cortical electrocorticographic signals are more synchronous in the presence of volatile anesthetic agents.

_{K}) between the SI and averaged AE values. As shown in figure 3, overall there was a modest, but significant, correspondence between AE and SI for all the frequencies. With the Bonferroni correction for multiple comparisons, there was no significant difference (

*P*> 0.05) in the P

_{K}values between the different frequency bands. However, the strongest correlations with the AE seemed to be found for the SI applied to the α and β bands—with mean P

_{K}values between 0.85 and 0.9. These results probably reflect the well-described appearance of strong synchronous α activity during volatile-based anesthesia—the so-called anesthesia spindles.

#### Discussion

^{15–17}In the awake state, synaptic inputs into closely spaced neuronal populations represent diverse patterns, and these populations operate quite independently of each other. However, with volatile anesthesia, the diversity of these synaptic activities is transformed into a more uniform, synchronous pattern, reflecting a reduction in the independence of operation. Our results are in close agreement with those in a recent paper

^{32}by Lewis

*et al.*, who found that—during propofol anesthesia—the slow (<1 Hz) oscillations strongly entrained all the high-frequency activity (see fig. 7 in Lewis’ article), thus increasing small-scale coherence (<4 mm). Conversely, they found that increased propofol caused

*asynchrony*at large scales in the low-frequency band. Diverse anesthetic drugs have been shown to

*disrupt*long-range cortical coupling.

^{33}Thus it would appear that anesthetic drugs have the dual effect of reducing global cortical integration, while enhancing synchronization on a local scale. Indeed, we would speculate that the anesthetic-induced disruption of long-range cortical integration might actually be achieved by the anesthetic-induced increase in small-scale synchrony. By making small groups of neurons hyper-synchronous, they become insensitive to long-range input—the so-called cortical block phenomenon—that has been proposed as the final common pathway for unconsciousness.

^{15}This effect may arise from the enhancement of inhibitory GABAergic (GABA: γ-aminobutyric acid) synaptic transmission. Augmented GABAergic interneuron action is likely to influence the cortical activity by coercing independent neuronal populations into more synchronous, uniform, activity patterns.

^{17}If this shift in cortical dynamics reflects a final common explanation of anesthetic action, the mechanism by which non-GABAergic drugs (

*e.g.*, ketamine and nitrous oxide) achieve the same endpoint needs to be considered. A simplistic explanation may be that GABAergic activity

*per se*is not the critical element, but instead the relative balance of excitation

*versus*inhibition is important. Thus,

*n*-methyl-D-aspartate antagonism would shift the balance in favor of inhibitory domination in similar fashion to direct enhancement of GABAergic activity.

^{4}The diversity of synaptic inputs into neighboring neuronal populations during the awake state implies that the repertoire of discriminable patterns available to the corticothalamic system is extraordinarily informative. In contrast, during deep anesthesia, volatile anesthetics produce a stereotypic burst-suppression pattern in which low amplitude suppressions are interrupted every few seconds by brief, quasi-periodic bursts of global activation, corresponding to an extreme loss of information flux.

^{34}; the effect of sevoflurane on the amplitude is intermediate between enflurane and isoflurane

^{35}; and desflurane minimally decreases it, similar to isoflurane.

^{19}