Reconfiguration of network hub structure after propofol-induced unconsciousness

Abstract

Introduction: General anesthesia induces unconsciousness along with functional changes in brain networks. Considering the essential role of hub structures for efficient information transmission, the authors hypothesized that anesthetics have an effect on the hub structure of functional brain networks.

Methods: Graph theoretical network analysis was carried out to study the network properties of 21-channel electroencephalogram data from 10 human volunteers anesthetized on two occasions. The functional brain network was defined by Phase Lag Index, a coherence measure, for three states: wakefulness, loss of consciousness induced by the anesthetic propofol, and recovery of consciousness. The hub nodes were determined by the largest centralities. The correlation between the altered hub organization and the phase relationship between electroencephalographic channels was investigated.

Results: Topology rather than connection strength of functional networks correlated with states of consciousness. The average path length, clustering coefficient, and modularity significantly increased after administration of propofol, which disrupted long-range connections. In particular, the strength of hub nodes significantly decreased. The primary hub location shifted from the parietal to frontal region, in association with propofol-induced unconsciousness. The phase lead of frontal to parietal regions in the α frequency band (8-13 Hz) observed during wakefulness reversed direction after propofol and returned during recovery.

Conclusions: Propofol reconfigures network hub structure in the brain and reverses the phase relationship between frontal and parietal regions. Changes in network topology are more closely associated with states of consciousness than connectivity and may be the primary mechanism for the observed loss of frontal to parietal feedback during general anesthesia.

Fig. 1

Schematic representation of analysis. For each original and surrogate data set, directed Phase Lag Index (dPLI) matrices were generated (only five signals are shown). The fully connected dPLI matrix becomes sparse via the Wilcoxon signed-rank test. The diagonal and disconnected components are denoted in black whereas others are denoted in gray. This dPLI matrix is transformed to Phase Lag Index (PLI) matrix, resulting in an undirected weighted network. Some network properties (average path length, clustering coefficient) are measured with normalization of a random network whereas others (modularity and centrality) are measured only with the PLI network.

Fig. 2

Different frequency bands show different functional connectivity during general anesthesia. The Phase Lag Index (PLI) averaged over all windows and subjects for 0.5–4 Hz increases whereas the PLIs of other frequency ranges do not show significant changes after loss of consciousness (LOC). The changed PLI after anesthesia does not return to its original values after recovery of consciousness (ROC). The error bar indicates SD (**P < 0.01 and ***P < 0.001; adjusted P values after Dunn multicomparison tests).

Fig. 3

Functional brain networks become segregated after loss of consciousness (LOC) and reintegrated after recovery of consciousness (ROC). (A) Average path length is compared after normalization by average path length of random network. The normalized average path length (Lw/Lr) increases after LOC in all frequency bands and decreases after ROC. The error bar indicates SD (*P < 0.05, **P < 0.01, and ***P < 0.001; adjusted P values after Tukey multicomparison tests). (B) Clustering coefficient is compared after normalization by clustering coefficient of a random network. In 8–13 Hz, the normalized clustering coefficient (Cw/Cr) increases after LOC. Recovery of Cw/Cr with ROC is shown in 4–8 Hz. (C) Functional brain networks become modularized after anesthesia. In 13–25 Hz, modularity (Q) increases during the LOC. (D) Small worldness (σ) is larger than 1 in all states and is increased after the ROC point in 13–25 Hz.

Fig. 4

Propofol affects hub nodes with the highest centrality in the brain network. The nodes are ranked in descending order of average betweenness centrality for the four frequency bands in wakefulness (W), loss of consciousness (LOC), and recovery consciousness (ROC), respectively. The highest-ranked node (defined as a hub) is mainly affected by propofol and returns with the ROC. This feature is observed in all four frequency bands.

Fig. 5

The betweenness centrality of the hubs is reduced after loss of consciousness (LOC) and returns with the recovery of consciousness (ROC). (A) The betweenness centrality of the hub nodes is more salient in wakefulness and after ROC compared with LOC. The betweenness centrality of the hubs correlates well with the states. The error bar indicates SD (*P < 0.05, **P < 0.01, and ***P < 0.001; adjusted P values after Tukey multicomparison tests). (B) The regional proportion for the hubs of different brain regions (from dark to light color: frontal, central, temporal, and parietal cortices). Dominant parietal hubs (in white) are observed during wakefulness (W) and ROC, especially in 8–13 and 13–25 Hz. These dominant parietal hubs are diminished after LOC. Instead, the frontal region (in black) emerges as the dominant hub.

Fig. 6

The relative degree centrality of the hubs is reduced after loss of consciousness (LOC) and returns after the recovery of consciousness (ROC). (A) The absolute degree centrality for the hubs is variable depending on the frequency bands. The strengths were similar with those of connectivity measured by Phase Lag Index (fig. 2). The error bar is SD (*P < 0.05, **P < 0.01, and ***P < 0.001; adjusted P values after Tukey multicomparison tests). (B) The relative degree centrality for the hubs is reduced after LOC over all frequency bands. During the ROC, return of relative degree centrality is shown in most frequency bands. The relative degree centrality correlates with the state change. (C) The regional proportion for the hubs defined by highest-ranked nodes in relative degree centrality of different brain regions (from dark to light color: frontal, central, temporal, and parietal cortices). The parietal hubs (in white) are dominant during wakefulness (W), diminished after LOC, and return to dominance after ROC. The frontal region (in black) emerges as the dominant hub during LOC, especially over the 8–13 and 13–25 Hz bands.

Fig. 7

Feedback-dominant connectivity is reduced in the α frequency band after loss of consciousness (LOC) and returns with recovery of consciousness (ROC). (A) The directed Phase Lag Index (dPLI) changes of frontoparietal connections across the three states are distinctive over the three frequency bands. The α frequency band (8–13 Hz) showed the most prominent changes according to the state transition. The error bar indicates SD (** P < 0.01 and *** P < 0.001; adjusted P values after Dunn multicomparison tests). (B) Time courses of the dPLI for three frequency bands. The time course of the α frequency band (8–13 Hz) shows the most significant change during anesthesia. (C) The changes of betweenness centrality of frontal and parietal hubs for the three frequency bands. The nodes with the largest betweenness centrality are selected as the hub nodes in the frontal and parietal regions. The α frequency band (8–13 Hz) manifests a significant reversal in terms of frontal and parietal hub structures. The parietal hub is dominant during wakefulness, whereas the frontal hub is dominant after LOC. This shift correlates with shifts in reversal of directional connectivity. The error bar indicates standard error. (D) The phase relationships among electroencephalogram channels in the α frequency band. The phase lead and lag relationship measured by dPLI is presented with different colors. A channel lagging in phase against the other channels is denoted by black. Otherwise, a channel leading in phase against the other channels is denoted by gray. If a channel has no significant phase lead or lag relationship (Wilcoxon signed-rank test with median 0.5, P > 0.05, P values are not corrected) it is denoted by white. During wakefulness (W) three parietal channels are significantly lagged in phase, whereas five frontal channels lead in phase relationship. This pattern tends to be disrupted during LOC and returns with the ROC.