Selective corticocortical connectivity suppression during propofol-induced anesthesia in healthy volunteers

Abstract

We comprehensively studied directional feedback and feedforward connectivity to explore potential connectivity changes that underlie propofol-induced deep sedation. We further investigated the corticocortical connectivity patterns within and between hemispheres. Sixty-channel electroencephalographic data were collected from 19 healthy volunteers in a resting wakefulness state and propofol-induced deep unconsciousness state defined by a bispectral index value of 40. A source analysis was employed to locate cortical activity. The Desikan-Killiany atlas was used to partition cortices, and directional functional connectivity was assessed by normalized symbolic transfer entropy between higher-order (prefrontal and frontal) and lower-order (auditory, sensorimotor and visual) cortices and between hot-spot frontal and parietal cortices. We found that propofol significantly suppressed feedforward connectivity from the left parietal to right frontal cortex and bidirectional connectivity between the left frontal and left parietal cortex, between the frontal and auditory cortex, and between the frontal and sensorimotor cortex. However, there were no significant changes in either feedforward or feedback connectivity between the prefrontal and all the lower-order cortices and between the frontal and visual cortices or in feedback connectivity from the frontal to parietal cortex. Propofol anesthetic selectively decreased the unidirectional interaction between higher-order frontoparietal cortices and bidirectional interactions between the higher-order frontal cortex and lower-order auditory and sensorimotor cortices, which indicated that both feedback and feedforward connectivity were suppressed under propofol-induced deep sedation. Our findings provide critical insights into the connectivity changes underlying the top-down mechanism of propofol anesthesia at deep sedation.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11571-021-09775-x.

Introduction

Anesthetics are commonly used during surgery to suppress the consciousness of patients undergoing painful surgery; however, the neural mechanism underlying their ability to suppress consciousness is still poorly understood (Brown et al. 2010). At the molecular level, various anesthetics achieve consciousness suppression by binding to different receptors, such as γ-aminobutyric acid type A (GABA_A) receptors, N-methyl-D-aspartate (NMDA) receptors and ${\alpha }_2$ adrenergic receptors (Alkire et al. 2008; Brown et al. 2010). Since the molecular targets bound by anesthetics are widely distributed in the neural system (Brown et al. 2011), it is likely that the consciousness suppression caused by anesthetics involves multiple brain regions and their interactions (Koch et al. 2016; Mashour and Hudetz 2018). Therefore, brain imaging combined with brain network analysis can provide a unique view of the action of anesthetics on large-scale neural networks, which makes it suitable for studying the neural mechanism of anesthetic-induced unconsciousness.

Earlier works on electroencephalography (EEG) have shown that frontoparietal communication is unidirectionally suppressed during anesthetic-induced unconsciousness (Lee et al. 2013b). In particular, the feedback connectivity from the frontal cortex to other cortices (Lee et al. 2013b) that provides top-down modulation was reduced during anesthesia, while feedforward communication in the posterior-to-anterior direction was left intact (Lee et al. 2013b). This unidirectional communication breakdown was demonstrated on various anesthetics with distinct molecular targets, including propofol, sevoflurane (Lee et al. 2009, 2013b; Ku et al. 2011; Boly et al. 2012; Jordan et al. 2013; Hudson and Pryor 2016; Ranft et al. 2016; Sleigh 2016; Palanca et al. 2017) and ketamine (Lee et al. 2013b; Schroeder et al. 2016; Bonhomme et al. 2016; Mashour 2016). Further studies also validated feedback communication breakdown in both humans and animals using functional magnetic resonance imaging (fMRI) (Boveroux et al. 2010; Bonhomme et al. 2012; Jordan et al. 2013; Ranft et al. 2016). Therefore, frontal-to-parietal communication breakdown has been advocated as the common functional pathway that underlies anesthetic-induced unconsciousness (Lee et al. 2013b; Hudetz and Mashour 2016; Mashour and Hudetz 2017).

However, divergence as evidence of feedback communication breakdowns has emerged in recent studies. First, an increase in frontoparietal connectivity strength was found in both patients and healthy volunteers using EEG after propofol-induced unconsciousness (Barrett et al. 2012; Nicolaou et al. 2012). A significant increase in theta frontoparietal communication was found in patients during anesthetic maintenance compared to the conscious state (Vlisides et al. 2019). Furthermore, a 27-fold increase in corticocortical phase-amplitude coupling between the frontal eye field and lateral intraparietal area was observed in macaque monkeys during propofol-induced anesthesia (Ma et al. 2019). Considering the directionality of functional connectivity, a study found that propofol-induced loss of consciousness was associated with impaired hierarchical feedforward connectivity rather than feedback connectivity (Sanders et al. 2018). Hence, whether the unidirectional suppression of feedback communication, in particular frontoparietal connectivity, is the common functional pathway for anesthetic-induced unconsciousness needs further investigation.

Functional connectivity has often been studied on EEG sensor space with coarse-grained cortex partitions. In earlier works, the calculation of frontoparietal connectivity was performed by averaging the connectivity strength obtained from all pairs of channels among the frontal and parietal cortex and across both hemispheres (Lee et al. 2009, 2013b; Ku et al. 2011; Jordan et al. 2013). However, the human brain possesses asymmetrical properties in both its structure and function (Vallortigara and Rogers 2005). Of relevance, it was found that the cortex is thicker in the left hemisphere in certain regions, including the frontal cortex, superior parietal cortex, and medial temporal cortex than in its right hemisphere counterparts (Kong et al. 2020). Such asymmetrical structural differences may also have functional implications. In the resting state, the frontoparietal network is preferentially coupled to the default mode network and language-related regions in the left hemisphere but to attention networks in the right hemisphere (Wang et al. 2014). Hence, an earlier approach involving averaging all connectivity measurements across both hemispheres can only reveal the quantitative differences in terms of overall connectivity strength. The hemispherical preference of propofol’s effect remains elusive. In this study, we aimed to study anesthetic-induced connectivity changes among cortical regions based on an analysis of the detailed cortical segmentation, including higher-order frontoparietal cortices and lower-order cortices, such as sensorimotor, auditory and visual cortices. We also investigated the interhemispherical connectivity changes during anesthesia.

In this work, we aimed to comprehensively explore corticocortical connectivity under propofol anesthesia. We collected a 60-channel EEG signal at the resting wakefulness state (Rest) and propofol-induced unconsciousness state (Anes) defined using a bispectral index (BIS) of 40. A source analysis was performed to mitigate volume conduction and locate cortical activation on a finer scale compared to the EEG sensor space. Then, the source activity of each cortical region was averaged according to the Desikan-Killiany atlas (Desikan et al. 2006) to reflect the activity of the higher-order (frontal, prefrontal and parietal) and lower-order (sensorimotor, auditory and visual) cortices. Finally, the directional connectivity among cortical regions was estimated using normalized symbolic transfer entropy (NSTE) due to its superiority in nonlinear and model-free estimations of directed functional connectivity. The main contributions of this work are summarized as follows.

1. In addition to hotspot frontal-parietal connectivity, we provided a comprehensive analysis of corticocortical connectivity changes under propofol-induced deep sedation indicated by a BIS of 40, which is the lower bound of the recommended range of BIS for general anesthesia. On a fine scale in source space, our results showed that propofol suppressed frontoparietal connectivity in both the feedback and feedforward directions in the left hemisphere. Propofol also suppressed feedforward connectivity from the left parietal to right frontal cortex under deep sedation.

2. Within and between hemispheres, propofol significantly suppressed bidirectional connectivity between the frontal and auditory cortex and between the frontal and sensorimotor cortex.

3. Significant changes were not observed in either feedforward or feedback connectivity between the prefrontal and all the lower-order cortices or between the frontal and visual cortices. Our results showed that propofol anesthetic selectively suppressed corticocortical connectivity in both the feedback and feedforward directions under propofol-induced deep sedation.

Materials and methods

Participants

This study was approved by the ethics review board of the First Affiliated Hospital of Xi’an Jiaotong University. The study design complied with the guidelines of the Declaration of Helsinki. Written informed consent was obtained from all participants prior to the experiment. In total, 19 healthy male subjects ($33.3\pm 6.2\mathrm{yr},74.1\pm 10.8\mathrm{kg}$) participated in this study. All subjects were drug-free and not allergic to propofol. Exclusion criteria included a history of neurological or psychiatric conditions. The participants fasted for approximately 8 h before propofol infusion.

Propofol anesthesia

Propofol (Fresenius Kabi, Graz, Austria) was used as the sole anesthetic in this experiment. A target-controlled infusion (TCI) device (Injectomat TIVA Agilia, Fresenius Kabi Gmbh, Graz, Austria) based on the Marsh pharmacokinetic model (Marsh et al. 1991) was employed for propofol infusion. The anesthesia depth was continuously monitored using a bispectral index monitor (BIS, Covidien, Mansfield, MA, USA). Vital signs, including heart rate, blood pressure and oxygen saturation level, were monitored using intraoperative monitors (Philips MP50, Boeblingen, Germany). The anesthetic infusion was performed by at least two senior anesthesiologists to ensure the safety of the participant.

Throughout the experiment, the subjects were comfortably laid on bed. Before drug administration, the subjects were asked to close their eyes, remain relaxed and stay awake for 3 min to collect preanesthesia resting EEG data. Subsequently, the propofol infusion started with an initial plasma concentration set at 2.0 µg/ml. During propofol infusion, the BIS value was closely monitored. The anesthesia endpoint was set at BIS of 40 $\pm$ 5. The plasma concentration of propofol was increased with a step size of 0.2 µg/ml until the targeted anesthesia depth was achieved. A laryngeal mask was placed after subjects lost consciousness, and it was connected to a ventilator. Once a stable BIS reading of approximately 40 was achieved, the level of anesthesia was maintained for at least 3 min to collect EEG data under deep sedation. Then, the propofol infusion was stopped by setting the targeted plasma concentration to 0 μg/ml, and the subjects were transported to the recovery room to regain consciousness. In this experiment, the anesthesia endpoint was set at a BIS of 40 $\pm$ 5, which is equivalent to the lower bound of the recommended range of BIS for general anesthesia to perform surgery (Struys et al. 1998). The time when BIS reached 40 was presumed to be the time of loss of consciousness in our work. In addition, a subsequent behavioral test was carried out to verify the loss of behavioral responsiveness that is usually used as a surrogate of loss of consciousness (loudly named spoken and shoulder tapping by one of two senior anesthesiologists). All 19 subjects had no response to the behavioral test, which is equal to the modified observer’s assessment of alertness/sedation (MOAA/S) score of 0 (Pambianco et al. 2008).

EEG data acquisition and preprocessing

EEG data were acquired with a 60-channel SynAmps RT system (NeuroScan, Singen, Gemmary), referenced at left and right mastoids. The sampling frequency was 1000 Hz. The electrode impedance was kept below 5 kΩ throughout data acquisition. For each subject in both the resting state and anesthesia state, a 3-min long EEG recording was used in the analysis. After removing artifacts by visual inspection, a 2-min long EEG was downsampled to 256 Hz and then bandpass filtered into a 0.1–35 Hz frequency band with a zero-phase bandpass filter (filtfilt.m, MATLAB Signal Processing Toolbox, MathWorks, Natick, MA). The preprocessed EEG data were segmented into nonoverlapping 5-s epochs for source activity estimation. All EEG signal preprocessing was performed using MATLAB R2016a software (MathWorks, Natick, MA) with customized scripts based on the EEGLAB 13.5.4b toolbox (Swartz Center for Computational Neurosciences, La Jolla, CA; http://www.sccn.ucsd.edu/eeglab).

The source imaging of the EEG signal was obtained using Brainstorm (Tadel et al. 2011). The Montreal Neurological Institute (MNI) model (Colin27) was used as the standard head model, and the standard Neuroscan Quik-Cap64 registration channel file was used for electrode registration (Collins et al. 1998). The forward model was calculated using the OpenMEEG boundary element method (BEM) (Gramfort et al. 2010). The weighted minimum norm estimation (wMNE) method was selected to estimate the activation of 15,002 dipoles distributed on the cortex (Baillet et al. 2001). The wMNE method can produce a depth-weighted linear L2-minimum norm estimation of the current density distribution to compensate for the tendency to preferentially place source activity in superficial regions of the cortex in the classical MNE method. Moreover, the wMNE method shows advantages of less sensitivity to the inaccuracy of forward models, which is applicable in situations without individual structural magnetic resonance imaging (MRI) anatomy and precise digitized electrode positions (He et al. 2018; Hassan and Wendling 2018).

We chose the Desikan-Killiany atlas for cortical partitions in source space based on the following two criteria. First, the selected atlas should include all the brain cortical areas shown to relate to sustaining consciousness. Cortical areas, including the frontal cortex, prefrontal cortex, posterior parietal cortex, anterior cingulate cortex, and posterior cingulate/precuneus cortex, have been shown to play an important role in sustaining consciousness according to research in the field of sleep, anesthesia, and disorders of consciousness using multimodal imaging methods (Hobson and Pace-Schott 2002; Ferrarelli et al. 2010; Hudetz 2012; Lee et al. 2013a, b; Barttfeld et al. 2015; Uhrig et al. 2016, 2018; Hudetz and Mashour 2016; Pal et al. 2018; Demertzi et al. 2019; Huang et al. 2020; Mashour et al. 2020). The Desikan-Killiany atlas includes all the above consciousness-related cortical areas, as shown in Table 1. Second, the selected atlas should provide a high degree of homogeneity of the partitioned areas. Using an automated labeling system with a dataset of 40 MRI scans, the Desikan-Killiany atlas possesses high validity and reliability, with an average intraclass correlation coefficient (ICC) of 0.835 across all 68 areas compared to the manually labeled dataset (Desikan et al. 2006). Moreover, the Desikan-Killiany atlas has been adopted in many studies investigating source space functional connectivity (Rizkallah et al. 2019; Zhang et al. 2019a; Duclos et al. 2021). In this study, the Desikan-Killiany atlas was employed to provide cortical partitions. This partition produces 68 cortical areas with a high degree of homogeneity, where each cortical area belongs to one of seven cortices: frontal cortex, prefrontal cortex, central cortex, temporal cortex, parietal cortex, occipital cortex, and limbic cortex (Desikan et al. 2006). We retained all cortical areas belonging to the higher-order cortices, including the frontal, prefrontal, and parietal cortices, as detailed in Table 1. The cortical areas belonging to the lower-level cortices, including the auditory cortex, sensorimotor cortex, and visual cortex, were selected based on their functions, as detailed in Table S8 in the supplementary materials. We excluded deep cortical regions (anterior and posterior cingulate cortices) from our connectivity analysis to avoid possible estimation bias when conducting source imaging analysis (Lin et al. 2006; Grech et al. 2008). Cortical activations of the region were extracted by averaging the activity of all dipoles within the given region according to the Desikan-Killiany atlas (Desikan et al. 2006). The overall EEG signal preprocessing and analysis pipeline is shown in Fig. 1.

Table 1.

Detailed anatomic regions of brain cortices

Cortices	Regions
Frontal cortex	Caudal middle frontal gyrus
	Pars opercularis
	Pars triangularis
	Rostral middle frontal gyrus
	Superior frontal gyrus
Prefrontal cortex	Frontal pole
	Lateral orbitofrontal gyrus
	Medial orbitofrontal gyrus
	Pars orbitalis
Auditory cortex	Superior temporal gyrus
	Transverse temporal gyrus
Sensorimotor cortex	Paracentral gyrus
	Postcentral gyrus
	Precentral gyrus
Visual cortex	Lateral occipital gyrus
	Pericalcarine cortex
Parietal cortex	Inferior parietal gyrus
	Precuneus gyrus
	Superior parietal gyrus
	Supramarginal gyrus

Each brain region contains two parts, which are located in the left and right hemispheres.

Fig. 1.

Schematic diagram of EEG data processing and analysis. (FF: feedforward connectivity, FB: feedback connectivity, preF: prefrontal cortex, F: frontal cortex, P: parietal cortex, A: auditory cortex, S: sensorimotor cortex, V: visual cortex, NSTE: normalized symbolic transfer entropy)

The inverse method has been shown to influence EEG source localization and induce variability in functional connectivity estimates (Mahjoory et al. 2017). A supplementary analysis was carried out to verify our findings using another inverse estimation method, standardized low-resolution electromagnetic tomography (sLORETA), in the supplementary materials (Buemi et al. 2002).

Theories of consciousness postulate that brain functional integration modules play a key role in supporting normal consciousness (Park and Friston 2013; Oizumi et al. 2014). Hence, anesthetics may operate by suppressing the functional integration capacity to achieve their therapeutic effect. In this work, feedback connectivity from the frontal and prefrontal cortex to other cortices, including the parietal, sensorimotor, auditory and visual cortices, and feedforward connectivity from other cortices to the frontal and prefrontal cortex were investigated using normalized symbolic transfer entropy (NSTE).

Estimation of directional connectivity

Functional connectivity was estimated using normalized symbolic transfer entropy (NSTE) (Staniek and Lehnertz 2008; Lee et al. 2013b, 2015). To calculate NSTE, the symbolic transfer entropy (STE) between a pair of EEG time series is first estimated. NSTE is a directional connectivity measure; hence, to remove possible false connections, the obtained STE estimate is subtracted from the STE value obtained from a random shuffled version of the source time series. The unbiased STE estimate is further normalized with respect to the entropy of the target time series. This procedure was adopted in previous studies to analyze frontoparietal connectivity during anesthesia (Lee et al. 2013b).

Three parameters, namely, the embedding dimension ($\mathit{m}$), time delay ($\mathit{\tau }$) and prediction time ($\mathit{\delta }$), are needed for the NSTE calculation. To find the STE of a pair of EEG time series, the original time series of EEG was first symbolized. For example, a sequence of EEG signals of $\mathit{N}$ samples is given as $\mathit{X}=\{{\mathit{x}}_1,{\mathit{x}}_2,\cdots ,{\mathit{x}}_{\mathit{N}}\}$, and the symbolized version of $X$ is $\widehat{\mathit{X}}=\{{\widehat{\mathit{x}}}_1,{\widehat{\mathit{x}}}_2,\cdots ,{\widehat{\mathit{x}}}_{\mathit{N}-\left(\mathit{m},-,1\right)\mathit{\tau }}\}$, where ${\widehat{\mathit{x}}}_{\mathit{i}}(\mathit{i}\in \{1,2,\cdots ,\mathit{N}-\left(\mathit{m},-,1\right)\mathit{\tau }\}$ is the symbolized vector of $\{{\mathit{x}}_{\mathit{i}},{\mathit{x}}_{\mathit{i}+\mathit{\tau }},\cdots ,{\mathit{x}}_{\mathit{i}+\left(\mathit{m},-,1\right)\mathit{\tau }}\}$. For a given but otherwise arbitrary $i$, first, $m$ amplitude values of $\{{\mathit{x}}_{\mathit{i}},{\mathit{x}}_{\mathit{i}+\mathit{\tau }},\cdots ,{\mathit{x}}_{\mathit{i}+\left(\mathit{m},-,1\right)\mathit{\tau }}\}$ are arranged in ascending order $\{{\mathit{x}}_{\mathit{i}+\left({\mathit{k}}_{\mathit{i}1},-,1\right)\mathit{\tau }}\le {\mathit{x}}_{\mathit{i}+\left({\mathit{k}}_{\mathit{i}2},-,1\right)\mathit{\tau }}\le \cdots \le {\mathit{x}}_{\mathit{i}+\left({\mathit{k}}_{\mathit{i}\mathit{m}},-,1\right)\mathit{\tau }}\}$. Then, the symbolized sequence of ${\widehat{x}}_i$ can be given as ${\widehat{\mathit{x}}}_{\mathit{i}}\equiv ({\mathit{k}}_{\mathit{i}1},{\mathit{k}}_{\mathit{i}2},\cdots ,{\mathit{k}}_{\mathit{i}\mathit{m}})$. Given two symbolized EEG signal sequences, for instance, source sequence $\widehat{\mathit{X}}$ and target sequence $\widehat{\mathit{Y}}$, the STE from $\widehat{\mathit{X}}$ to $\widehat{\mathit{Y}}$ can be estimated using the following equation,

$${\mathit{STE}}_{\mathit{X}\to \mathit{Y}}=\sum \mathit{p}\left({\widehat{\mathit{y}}}_{\mathit{i}+\mathit{\delta },}{\widehat{\mathit{y}}}_{\mathit{i}},{\widehat{\mathit{x}}}_{\mathit{i}}\right){\mathit{log}}_2\frac{\mathit{p}({\widehat{\mathit{y}}}_{\mathit{i}+\mathit{\delta },}{\widehat{\mathit{y}}}_{\mathit{i}}|{\widehat{\mathit{x}}}_{\mathit{i}})}{\mathit{p}({\widehat{\mathit{y}}}_{\mathit{i}+\mathit{\delta },}|{\widehat{\mathit{y}}}_{\mathit{i}})}$$ 1

where the sum runs over all symbols and $p$ denotes the transition probability density. ${\mathit{STE}}_{\mathit{X}\to \mathit{Y}}$ is given in bits because the base of the log is 2.

Second, for estimating STE for a pair of EEG signals, the shuffled data method was used to remove the possible bias. The shuffled data are produced by exchanging the first half and the second half of the EEG signal, and this shuffling process is applied only to the source signal $\left(\mathit{X}\right)$, leaving the target signal $\left(\mathit{Y}\right)$ intact. Finally, the unbiased STE was normalized as followsfollowing,

$${\mathit{NSTE}}_{\mathit{X}\to \mathit{Y}}=\frac{{\mathit{STE}}_{\mathit{X}\to \mathit{Y}}-{\mathit{STE}}_{\mathit{X}\to \mathit{Y}}^{\mathit{Shuffled}}}{\sum \mathit{p}\left({\widehat{\mathit{y}}}_{\mathit{i}+\mathit{\delta },}{\widehat{\mathit{y}}}_{\mathit{i}}\right){\mathit{log}}_2\mathit{p}({\widehat{\mathit{y}}}_{\mathit{i}+\mathit{\delta }}|{\widehat{\mathit{y}}}_{\mathit{i}})}$$ 2

where,

$${\mathit{STE}}_{\mathit{X}\to \mathit{Y}}^{\mathit{Shuffled}}=\sum \mathit{p}\left({\widehat{\mathit{y}}}_{\mathit{i}+\mathit{\delta },}{\widehat{\mathit{y}}}_{\mathit{i}},{\widehat{\mathit{x}}}_{\mathit{i}}^{\mathit{shuffled}}\right){\mathit{log}}_2\frac{\mathit{p}({\widehat{\mathit{y}}}_{\mathit{i}+\mathit{\delta },}{\widehat{\mathit{y}}}_{\mathit{i}}|{\widehat{\mathit{x}}}_{\mathit{i}}^{\mathit{shuffled}})}{\mathit{p}({\widehat{\mathit{y}}}_{\mathit{i}+\mathit{\delta },}|{\widehat{\mathit{y}}}_{\mathit{i}})}$$ 3

The average NSTEs between the $\mathit{C}1$ and $\mathit{C}2$ cortices, denoted as $\overline{{\mathit{N}\mathit{S}\mathit{T}\mathit{E}}_{\mathit{C}1\to \mathit{C}2}}$ and $\overline{{\mathit{N}\mathit{S}\mathit{T}\mathit{E}}_{\mathit{C}2\to \mathit{C}1}}$, were calculated over all the pairs of cortical regions for each subject,

$$\overline{{\mathit{NSTE}}_{\mathit{C}1\to \mathit{C}2}}=\frac{1}{{\mathit{n}}_{\mathit{C}1}·{\mathit{n}}_{\mathit{C}2}}\sum _{\left(\mathit{i},\mathit{j}\right)}^{{\mathit{n}}_{\mathit{C}1},{\mathit{n}}_{\mathit{C}2}}{\mathit{NSTE}}_{\mathit{i}\to \mathit{j}}$$ 4

where $\mathit{i}$ is the number of all ${\mathit{n}}_{\mathit{C}1}$ regions belonging to the $\mathit{C}1$ cortex and $\mathit{j}$ is the number of all ${\mathit{n}}_{\mathit{C}2}$ regions belonging to the $\mathit{C}2$ cortex.

During NSTE calculation, we concatenated two successive epochs to form a 10 s window, which is similar to the study of Lee et al. (Lee et al. 2013b). To fix the three parameters, we adopted a maximization principle similar to that used in Lee et al. (Lee et al. 2013b). Briefly, the selected parameters should provide the maximum NSTE value, which indicates the maximum information transfer from the source signal to the target signal. First, we set the embedding dimension ($m$) to 3. Then, the prediction time ($\delta$) was found among the range between 1 and 100 with the maximum cross-correlation. Since the embedding dimension ($m$) and prediction time ($\delta$) were determined, we finally found the time delay ($\tau$) (from 1 to 30) producing maximum NSTE. For the detailed calculation procedure, please refer to Lee et al. (Lee et al. 2013b).

Statistical analysis

This paper presents an exploratory study investigating the changes in directional connectivity between brain regions of interest during Anes. We selected a sample size of 19 subjects according to earlier works investigating directional connectivity during Anes (Ku et al. 2011; Lee et al. 2013b; Jordan et al. 2013). Our sample size is similar to or larger than earlier works, such as in (Ku et al. 2011; Lee et al. 2013b; Jordan et al. 2013). As a post hoc analysis, we calculated the sample size according to a previous related study that investigated frontoparietal connectivity under Anes (Lee et al. 2013b). Based on the mean and standard deviation of connectivity strength reported in (Lee et al. 2013b), the effective size (Cohen's d) of the connectivity strength between Anes and Rest was 1.8750 (Cohen 2013; Zhang and Yuan 2018). With a significance level of 0.05 and statistical power of 0.9, we found that 5.257 subjects per group were required. We collected data from 19 subjects for the experiment, which fulfilled the requirement for drawing a statistically significant conclusion.

The feedback connectivity and feedforward connectivity, denoted as $\overline{{\mathit{N}\mathit{S}\mathit{T}\mathit{E}}_{\mathit{C}1\to \mathit{C}2}}$ and $\overline{{\mathbf{N}\mathbf{S}\mathbf{T}\mathbf{E}}_{\mathbf{C}2\to \mathbf{C}1}}$, were compared across Rest and Anes. After the normality test with the Lilliefors corrected Kolmogorov–Smirnov test (Lilliefors 1967), we found that some connectivity strength distributions violated the normality assumption (p < 0.05). Hence, we adopted a paired sample permutation test to test the median difference in estimated functional connectivity between Rest and Anes (Kaiser 2007). The number of permutations is set to 10,000 (van den Heuvel et al. 2017). The false discovery rate method was used to correct multiple comparisons with a significance level of the adjusted p < 0.05 (Benjamini and Hochberg 1995). For the overall connectivity between the frontal/prefrontal cortex and the parietal, auditory, sensorimotor, and visual cortex, the number of multiple comparisons was 8 (feedforward and feedback connectivity for each pair). For intra- and interhemispheric connectivity, the number of multiple comparisons was 16 (4x2x2 = 16, 4 cortices, 2 hemispheres, 2 directions).

Results

Decreased overall directional connectivity under propofol anesthesia

We first investigated the overall connectivity changes between higher-order cortices (frontal and prefrontal cortex) and lower-order cortices (auditory, sensorimotor and visual). In addition, we also investigated changes in frontoparietal connectivity during propofol anesthesia. Compared to Rest, we found that there were no significant changes in the feedback connectivity from the prefrontal cortex to the parietal cortex ([median, minimum, maximum], [0.0042, 0.0033, 0.0089] for Rest, [0.0046, 0.0038, 0.0128] for Anes, p = 0.3382) or all the lower-order cortices (Rest: [0.0050, 0.0038, 0.0141], [0.0042, 0.0032, 0.0143], [0.0037, 0.0029, 0.0116], Anes: [0.0051, 0.0041, 0.0107], [0.0040, 0.0034, 0.0120], [0.0045, 0.0038, 0.0112], for auditory, sensorimotor and visual cortices, respectively, all FDR-adjusted p > 0.05), as shown in Fig. 2a and Table S2 in the supplementary materials. The feedforward connectivity from parietal ([0.0045, 0.0033, 0.0089] for Rest, [0.0043, 0.0038, 0.0128] for Anes, p = 0.6708) and all the lower-order cortices (Rest: [0.0054, 0.0038, 0.0141], [0.0046, 0.0032, 0.0143], [0.0042, 0.0029, 0.0116], Anes: [0.0052, 0.0041, 0.0107], [0.0043, 0.0034, 0.0120], [0.0043, 0.0038, 0.0122], for auditory, sensorimotor and visual, respectively, all FDR-adjusted p > 0.05) to the prefrontal cortex also showed no significant changes during propofol anesthesia, as shown in Fig. 2b and Table S2 in the supplementary materials. Hence, propofol might exert little hypnotic effect through the connectivity between the prefrontal and parietal cortex and between the prefrontal cortex and the lower-order cortices.

Fig. 2.

Feedforward and feedback connectivity between higher- and lower-order cortices in the resting wakefulness state (red and blue) and propofol-induced unconsciousness state (magenta and black). Boxplot shows the median and 25th/75th percentiles. The red cross represents outliers. (NSTE: normalized symbolic transfer entropy, FB: feedback connectivity, FF: feedforward connectivity, A: auditory cortex, S: sensorimotor cortex, P: parietal cortex, V: visual cortex, ***: p < 0.001, **: p < 0.01, *: p < 0.05 with FDR correction.)

On the other hand, significant bidirectional connectivity suppressions were observed during propofol anesthesia between frontal and auditory cortex (Rest: [0.0062, 0.0042, 0.0129], Anes: [0.0050, 0.0041, 0.0086], p < 0.0001 for feedback connectivity; Rest: [0.0063, 0.0042, 0.0118], Anes: [0.0049, 0.0039, 0.0067], p = 0.002 for feedforward connectivity), and between frontal and sensorimotor cortex (Rest: [0.0057, 0.0038, 0.0137], Anes: [0.0042, 0.0037, 0.0108], p = 0.0002 for feedback connectivity; Rest: [0.0056, 0.0036, 0.0122], Anes: [0.0042, 0.0038, 0.0089], p = 0.006 for feedforward connectivity), as shown in Fig. 2c, d and Table S1 in the supplementary materials. There was also a decrease in feedforward connectivity from the parietal to frontal cortex without existing FDR correction for multiple comparisons (Rest: [0.0052, 0.0035, 0.0118], Anes: [0.0046, 0.0035, 0.0132], p = 0.0402), as shown in Fig. 2d and Table S1 in the supplementary materials. Significant changes were not observed in either feedforward or feedback connectivity between the frontal and visual cortex (Rest: [0.0041, 0.0032, 0.0097], Anes: [0.0043, 0.0037, 0.0103], p = 0.4774 for feedback connectivity; Rest: [0.0041, 0.0031, 0.0090], Anes: [0.0043, 0.0033, 0.0117], p = 0.7228 for feedforward connectivity) were found, as shown in Fig. 2c, d and Table S1 in the supplementary materials.

In summary, we found that propofol most likely exerted its hypnotic effect through the suppression of connectivity linked to the frontal cortex.

Decreased intra- and interhemispheric directional connectivity under propofol anesthesia

In the previous section, we identified a reduction in overall connectivity between the frontal and parietal cortex and between the frontal cortex and other lower-order cortices in Anes compared to Rest. However, connectivity between cortices within and between hemispheres remains unknown. To address this problem, we further investigated intra- and interhemispheric NSTE between the frontoparietal cortices, frontal cortex and other lower-order cortices.

Compared with Rest, the suppressed frontoparietal connectivity may be attributed to the reduced feedforward connectivity from the left parietal to left frontal cortex (Rest: [0.0055, 0.0038, 0.0114], Anes: [0.0049, 0.0038, 0.0077], p = 0.0016), from the left parietal to right frontal cortex (Rest: [0.0047, 0.0030, 0.0079], Anes: [0.0043, 0.0034, 0.0053], p = 0.0088) and feedback connectivity from the left frontal to left parietal cortex (Rest: [0.0059, 0.0038, 0.0114], Anes: [0.0048, 0.0038, 0.0077], p = 0.0232), as shown in Figs. 3a, 4a and Table S6 in the supplementary materials. We found a significant reduction of bidirectional connectivity within and between hemispheres between frontal and auditory cortex: feedforward connectivity from the left auditory to left frontal cortex (Rest: [0.0068, 0.0051, 0.0132], Anes: [0.0051, 0.0038, 0.0088], p = 0.0004), from the left auditory to right frontal cortex (Rest: [0.0054, 0.0034, 0.0120], Anes: [0.0045, 0.0036, 0.0061], p = 0.0176), from the right auditory to left frontal cortex (Rest: [0.0057, 0.0035, 0.0188], Anes: [0.0045, 0.0039, 0.0093], p = 0.0008) and from the right auditory to right frontal cortex (Rest: [0.0068, 0.0045, 0.0118], Anes: [0.0057, 0.0038, 0.0078], p = 0.0010), feedback connectivity from the left frontal to left auditory cortex (Rest: [0.0075, 0.0051, 0.0132], Anes: [0.0061, 0.0038, 0.0088], p = 0.0004), from the left frontal to right auditory cortex (Rest: [0.0066, 0.0035, 0.0188], Anes: [0.0049, 0.0039, 0.0093], p = 0.0004), from the right frontal to left auditory cortex (Rest: [0.0050, 0.0032, 0.0089], Anes: [0.0044, 0.0036, 0.0078], p = 0.0100) and from the right frontal to right auditory cortex (Rest: [0.0066, 0.0043, 0.0136], Anes: [0.0053, 0.0040, 0.0087], p = 0.0024), as shown in Figs. 3b, 4b and Table S3 in the supplementary materials. We also found significantly reduced bidirectional connectivities both within and between hemispheres between the frontal and sensorimotor cortex, as shown in Figs. 3c, 4c and Table S4 in the supplementary materials (p values and the median, minimum, maximum values of NSTE). There were no significant changes in bidirectional connectivity both within and between hemispheres between the frontal and visual cortex, as shown in Figs. 3d, 4d and Supplementary Table S5 (p values and the median, minimum, maximum values of NSTE). A schematic diagram of the whole intra- and interhemispheric directional connectivity between the frontal cortex and the other four cortices is shown in Fig. 5.

Fig. 3.

Intra- and interhemispheric feedforward connectivity between the higher-order frontal cortex and lower-order cortices in the resting wakefulness state (Rest) and propofol-induced unconsciousness state (Anes). Boxplot shows the median and 25th/75th percentiles. The red cross represents outliers. (L: left hemisphere, R: right hemisphere, e.g., L–R represents the left frontal cortex and right parietal/auditory/sensorimotor/visual cortex, ***: p < 0.001, **: p < 0.01, *: p < 0.05 with FDR correction)

Fig. 4.

Intra- and interhemispheric feedback connectivity between the higher-order frontal cortex and lower-order cortices in the resting wakefulness state (Rest) and propofol-induced unconsciousness state (Anes). Boxplot shows the median and 25th/75th percentiles. The red cross represents outliers. (L: left hemisphere, R: right hemisphere; e.g., L–R represents left frontal cortex and right parietal/auditory/sensorimotor/visual cortex, ***: p < 0.001, **: p < 0.01, *: p < 0.05 with FDR correction)

Fig. 5.

Schematic diagram of the intra- and interhemispheric feedforward and feedback connectivity between the higher-order frontal cortex and other cortices in the propofol-induced unconsciousness state (Anes). The curve with red and blue arrows represents a significant reduction in feedback and feedforward connectivity, respectively, in Anes compared with the resting wakefulness state (Rest). The thickness of the curve is proportional to the degree of reduction. Feedforward connectivity from the left parietal to right frontal cortex, bidirectional connectivity between the left frontal and left parietal cortex, between the frontal and auditory cortex, and between the frontal and sensorimotor cortex were suppressed under Anes. However, there were no significant changes in either feedforward or feedback connectivity between the frontal and visual cortex in either the intra- or interhemispheres. The feedforward connectivity from the right parietal cortex to either the left or right frontal cortices also showed no significant changes in Anes. The above results showed selective suppression of corticocortical connectivity under propofol anesthesia at deep sedation. (LH: left hemisphere, RH: right hemisphere)

Discussion

In this study, a source analysis of EEG signals and NSTE estimation of directional functional connectivity showed that the anesthetic propofol suppressed consciousness by selectively decreasing corticocortical communication between higher-order frontoparietal cortices and between higher-order frontal and lower-order cortices in both feedforward and feedback directions. Our results demonstrated that the connectivity between the prefrontal cortex and lower-order cortices showed no significant changes in either feedforward or feedback connectivities; however, a unidirectional reduction in connectivity between the parietal and frontal cortex and bidirectional connectivity suppression from the frontal to other lower-order cortices except the visual cortex were observed during deep sedation with propofol anesthesia.

Previous works using sensor-level EEG signals found that suppression of feedback connectivity from the frontal to parietal cortex was accompanied by anesthesia induced by propofol, sevoflurane and ketamine (Lee et al. 2013b). Later, multiple studies confirmed that weakened frontal-to-parietal connectivity is associated with unconsciousness caused by anesthetics (Mashour 2014; Palanca et al. 2015; Schroeder et al. 2016; Bonhomme et al. 2016; Ranft et al. 2016; Mashour and Hudetz 2017). In the present study, however, a suppression of feedforward connectivity from the parietal to frontal cortex was initially found at the global level. A follow-up intra- and interhemispheric analysis identified suppressed feedback connectivity from the left frontal to left parietal cortex as well as suppressed feedforward connectivity from the left parietal to both the left and right frontal cortices. The discrepancy between our results and previous work could have occurred because the anesthetic protocol achieved a much deeper unconsciousness state in our study. In our experiment, the level of unconsciousness was indicated by a BIS of 40, equivalent to the lower bound of the recommended range of BIS for general anesthesia. Verified by the lack of responsiveness to behavioral tests, the unconsciousness level in our work was at least at the level measured by a MOAA/S score of 0. Previous studies on functional connectivity during anesthesia usually perform behavioral tests to define loss of consciousness, among which most are at the level of MOAA/S score of 1 (see the review listed in Table S7 in supplementary materials). Therefore, the level of unconsciousness in our study was deeper than that in most previous related studies. Our results suggested that the suppression of frontoparietal connectivity occurred in both feedback and feedforward directions at propofol-induced deep sedation and that the feedback connectivity within the left hemisphere was mostly affected by propofol anesthetic.

Frontoparietal connectivity has often been studied after the induction phase of anesthesia, with a relatively light anesthesia level compared to the anesthesia level required to perform surgery (Lee et al. 2013b; Vlisides et al. 2019). To probe the existence of consciousness, behavior assessments are usually conducted (Zhong et al. 2005). However, multiple studies have pointed out that unconsciousness is not an equivalent state to unresponsiveness (Sanders et al. 2012, 2017; Linassi et al. 2018; Wang et al. 2021). A study using simultaneous fMRI and EEG under propofol anesthesia with a peak effect site concentration of 4.0 µg ml⁻¹, combined with an auditory stimulus, found that the functional module related to primary auditory processing was still active after loss of responsiveness (Mhuircheartaigh et al. 2013). The activation diminished after slow-wave saturation occurred, which corresponded to a deep level of anesthesia where most neurons were depressed (Mhuircheartaigh et al. 2013). This finding was consistent with the current study, where both auditory and sensorimotor feedforward processes were suppressed under propofol-induced deep sedation. Therefore, our results might indicate that propofol disrupted the cognitive process by blocking the projection of sensory information to higher-order processing networks and thus preventing information integration at propofol-induced deep sedation.

A study using event-related fMRI found that during anesthesia induction, brain activity first declined in both hemispheres of the frontal cortex but was preserved in both hemispheres of the temporal cortex involved in auditory language processing (Heinke et al. 2004). These declining and preserved brain activities may result in decreased bidirectional connectivity both within the unilateral hemisphere and between the bilateral hemispheres of the frontal and auditory cortices. During maintenance of anesthesia, brain activity decreased in the bilateral hemispheres of both the frontal and auditory cortices, which might result in decreased bidirectional connectivity between the frontal and auditory cortices (Heinke et al. 2004). Liu et al. found not only the same deactivation but also decreased connectivity between the primary auditory and frontal cortices using seed-based correlation analysis under propofol anesthesia (Liu et al. 2012). It is notable that under deep anesthesia (plasma concentration of 0.75–1.0 μg/ml) in the study of Liu et al., there were also minor preserved connectivities mainly in the right hemisphere (Liu et al. 2012), which was likely associated with the activity of information retrieval (Habib et al. 2003). However, under depth of propofol-induced deep sedation in our present study, connectivity between frontal and auditory cortices was depressed both within unilateral and between bilateral hemispheres.

In this study, we found an asymmetric pattern of bidirectional frontoparietal connectivity suppression in the left hemisphere in propofol-induced deep sedation. It seems that this asymmetry is strange, and intuitively, this suppression is also expected to symmetrically occur in the right hemisphere. Although the molecular targets of anesthetics are widely distributed in the brain (Brown et al. 2011), the resulting macroscopic cortical communication breakdown seems to be diversified. As the depth of anesthesia deepened, the brain activity at primary cortices gradually declined and constrained until eventually deactivated (Mhuircheartaigh et al. 2013). This extensive activity suppression could result in the disruption of both the intra- and interhemispheric communication between frontal and primary cortices. However, the brain was still active under deep anesthesia in certain regions, including the precuneus and the posterior parietal and prefrontal cortices (Mhuircheartaigh et al. 2013). Notably, the extent of activation in the posterior parietal cortex was larger in the right hemisphere than in the left hemisphere (Mhuircheartaigh et al. 2013). The remaining activation in the posterior parietal cortex in the right hemisphere might be the reason for the observed selective suppression of frontoparietal connectivity under propofol-induced deep sedation in this work.

The asymmetry found in this work might be related to the role of lateralization of cerebral function, which is a fundamental aspect of nervous system organization (Geschwind and Miller 2001; Vallortigara and Rogers 2005; Frasnelli et al. 2012). The central nervous system of humans, which consists of regions distinct in anatomy and function, also differs in sensitivity to anesthetics (Heinke and Koelsch 2005). Earlier reports found that propofol-produced metabolic decrease varies among cortical region (Fiset et al. 1999). Such a difference may be attributed to the lateral differences in the GABA binding site, in which a higher level of GABA binding was found on the left side of most brain areas (Guarneri et al. 1988). Additionally, it was reported that in the thalamocortical model, connectivity from the thalamus to the left frontal lobe was significantly reduced compared to the right frontal lobe (Liu et al. 2013). Combined with our observation that a significant reduction existed in the bidirectional connectivity between the left parietal cortex and left frontal cortex, these results suggested that propofol’s anesthetic effect may be related to preferable degradation in the left hemisphere.

In our study, we found a decrease in bidirectional connectivity between the left frontal and left parietal cortex and the frontal cortex and lower-order cortices under propofol anesthesia. In the existing research, a large body of evidence indicated a decrease in feedback connectivity during anesthesia, whereas the feedforward connectivity was left intact (Lee et al. 2013b). However, recent studies reported a reduction in feedforward connectivity (Sanders et al. 2018). In this work, we found that propofol mostly acted on the connectivity linked to the frontal cortex, and it primarily deceased the bidirectional connectivity from the frontal cortex to other lower-order cortices. Our results suggested that the top-down process was suppressed in not only feedback but also feedforward direction under propofol-induced deep sedation, as indicated by a BIS value equal to 40.

In a recent animal study under propofol anesthesia, a significant increase in the connectivity strength of corticocortical interactions from the frontal eye field to the lateral parietal area in macaque monkeys was found using electrocorticogram (ECoG) data (Ma et al. 2019). It should be noted that the increase in connectivity strength was found on a specific neural circuit that has strong anatomic projections, as indicated by the high spatial resolution of ECoG compared to EEG. In this study, we used source imaging of EEG and demonstrated that corticocortical functional connectivity was selectively depressed on a macroscopic level. It was possible for a certain region to gain a high connectivity strength. However, our results suggested that disconnection rather than connectivity strengthening was more evident when anesthetic took effect.

The purpose of this work was to explore the changes in directional connectivity in Anes. We found a pattern of selective corticocortical connectivity suppression during propofol-induced deep sedation. Recent studies using unsupervised machine learning methods have shown that functional connectivity can be clustered into several limited connectivity patterns termed connectivity states (Vlisides et al. 2019; Zhang et al. 2019b; Li et al. 2019). Using whole-brain connectivity, several connectivity states with a large occurrence rate were found to dominate the anesthetic-induced unconsciousness state (Zhang et al. 2019b). While using connectivity between frontal and parietal cortices, connectivity states characterized by suppressed alpha frontal-parietal connectivity were shown to alternate over time during anesthetic maintenance (Vlisides et al. 2019; Li et al. 2019). Through a dynamic functional connectivity analysis, rich information related to temporal, spatial, and spectral properties can be revealed. In the static analysis, we primarily focused on the changes in functional connectivity strength. Moreover, to facilitate an unsupervised clustering algorithm to reveal the common connectivity patterns among consciousness states, a long signal recording is required for conducting dynamic connectivity analysis. Given the short recording time in the present study, it would be difficult to carry out dynamic functional connectivity analyses.

In this study, we explored changes in directional connectivity between brain regions of interest at Anes compared to Rest. We found selective corticocortical connectivity suppression in altered states of consciousness during propofol anesthesia. However, whether the relationship between consciousness state alterations and connectivity changes is causal cannot be asserted based on the experiment in this study. Since the changes in connectivity were accompanied by both the effect of propofol and alteration of the consciousness state, it cannot be concluded that an altered state of consciousness alone causes changes in connectivity. Determining the causal relationship between the changes in functional connectivity and the altered consciousness state requires a sophisticated experimental procedure that can reverse the anesthetic-induced unconsciousness state to a wakefulness state while keeping the anesthetic concentration constant, similar to that conducted in (Pal et al. 2018) with rats. However, such a procedure is difficult to carry out on human volunteers. An alternative awakening test procedure that attempts to arouse the subjects to regain responsiveness after the loss of consciousness while continuing to administrate anesthetic may be feasible for human subjects (Scheinin et al. 2018).

The current study has several limitations. First, in the directional connectivity analysis, we excluded deep cortical regions from our connectivity analysis to avoid possible estimation bias when conducting the source imaging analysis. Future works using simultaneous fMRI and EEG can potentially improve localization accuracy in deep cortical areas and explore their connectivity to other cortices under Anes. Second, in our experiment, we adopted a commonly used anesthetic scheme, that is, Rest preceded Anes. Theoretically, the order of the consciousness states (Rest vs. Anes) should be counterbalanced to control for possible order effects. Since Rest was chosen as the baseline condition to contrast with Anes, the influence on the baseline condition due to alternating the experimental order cannot be completely ruled out. Third, in this work, we selected propofol as the sole anesthetic agent to induce unconsciousness. Hence, our results could not be extended to other anesthetics that do not bind to γ-aminobutyric acid (GABA_A) receptors. Future work should be carried out on non-GABAergic anesthetics, including ketamine and sevoflurane. Fourth, only male subjects were recruited in this study. It has been shown that sex could also affect propofol action. Our results hence require further validation in the female subject group. Last, the current study is a single-center study with a limited number of subjects. Future work can be carried out to validate our findings with a large number of subjects in multiple centers.

In conclusion, propofol selectively decreases interactions between higher-order left frontoparietal cortices and between higher-order frontal cortex and lower-order auditory and sensorimotor cortices, which might indicate that both feedback and feedforward connectivity are depressed under propofol-induced deep sedation.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

HW: Formal analysis, Data curation, Methodology, Visualization, Software, Writing–original draft, Writing—review & editing. YZ: Investigation, Data curation, Writing—review & editing. HC: Visualization, Software. FY: Investigation, Data curation. DS: Investigation, Data curation. QW: Project administration, Conceptualization, Supervision. SC: Formal analysis, Methodology. YW: Formal analysis, Methodology, Writing—original draft, Writing—review & editing, Funding acquisition. LH: Project administration, Conceptualization, Supervision, Funding acquisition.

Funding

This work was supported by [the National Natural Science Foundation of China] (Grant numbers 81701787, 81671778, 31271063, 62101413, and U1401255), [the Natural Science Basic Research Plan in Shaanxi Province of China] (Grant number 2019JQ-138), and [the Fundamental Research Funds for the Central Universities, Xidian University] (Grant number No. XJS201213).

Data and material availability

The datasets generated and/or analyzed during the current study are available from the corresponding author (Dr. Liyu Huang) on reasonable request.

Code availability

The custom datasets generated and/or analyzed during the current study are available from the corresponding author (Dr. Liyu Huang) on reasonable request.

Declarations

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Ethics approval

This study was approved by the ethics review board of the First Affiliated Hospital of Xi’an Jiaotong University. The study design complied with the guidelines of the Declaration of Helsinki.

Consent to participate

Written informed consent was obtained from all participants prior to the experiment.

Consent for publication

All the authors agree to publish this manuscript.