Network Properties in Transitions of Consciousness during Propofol-induced Sedation

Reliable electroencephalography (EEG) signatures of transitions between consciousness and unconsciousness under anaesthesia have not yet been identified. Herein we examined network changes using graph theoretical analysis of high-density EEG during patient-titrated propofol-induced sedation. Responsiveness was used as a surrogate for consciousness. We divided the data into five states: baseline, transition into unresponsiveness, unresponsiveness, transition into responsiveness, and recovery. Power spectral analysis showed that delta power increased from responsiveness to unresponsiveness. In unresponsiveness, delta waves propagated from frontal to parietal regions as a traveling wave. Local increases in delta connectivity were evident in parietal but not frontal regions. Graph theory analysis showed that increased local efficiency could differentiate the levels of responsiveness. Interestingly, during transitions of responsive states, increased beta connectivity was noted relative to consciousness and unconsciousness, again with increased local efficiency. Abrupt network changes are evident in the transitions in responsiveness, with increased beta band power/connectivity marking transitions between responsive states, while the delta power/connectivity changes were consistent with the fading of consciousness using its surrogate responsiveness. These results provide novel insights into the neural correlates of these behavioural transitions and EEG signatures for monitoring the levels of consciousness under sedation.


Individual results of wPLI during the levels of responsiveness
The individual changes of wPLI were investigated for all frequencies during the levels of responsiveness ( Supplementary Fig. S2). We focused on the delta and beta bands. In the beta band, the individual change of wPLI was noticeably increased only at the transition point between consciousness and unconsciousness. This was remarkably consistent across subjects.
Although the delta connectivity statistically increased in the fronto-parietal interaction and parietal region during unresponsiveness. Most subjects showed clear changes of wPLI across five states; one did not. Comparing the unconscious state, including the transition into unresponsiveness or responsiveness, with the conscious state (baseline and recovery), the delta wPLI certainly increased on average during unconsciousness ( Supplementary Fig. S1).
Nonetheless, there was the outlier marked with a red plus sign on the changes of parietal wPLI in the Supplementary Figure S1c. The Sub03 did not respond purposefully to auditory stimuli during propofol-induced unresponsiveness but tried to wake up or moved severely as if there was sleepwalking. In Sub03, although the beta wPLI during the transition points into unresponsiveness or responsiveness rapidly increased, the delta wPLI at baseline continued to decrease until propofol infusion (propofol-induced unresponsiveness) and recovery of responsiveness. This feature could be thought that Sub03 was caused by not reaching the stable deep sedation state that does loss of spontaneous movement. Future experiments with a large number of subjects are expected to have smaller SD and even more clear changes of delta wPLI during the five states of propofol-induced unresponsiveness.

Traveling wave analysis
The negative peak is the criteria point to calculate the brain dynamics of the delta wave (slow oscillations) because this measure is pointed and simply distinguishable, and its timing is independent on the EEG baseline 1 . Specifically, the delta wave was detected on the midline channels placed along the anterior-posterior axis. The timing of negative peak was calculated at each electrode. In the case of the responsive state being the traveling wave, we also measured the velocity of the wave using correlation and linear regression. The distance was calculated on the anterior-posterior axis based on the Fpz electrode.

Graph theory
We calculated network properties based on the weighted phase lag index. Global efficiency, local efficiency, and small-worldness are appropriate for investigating integrated information, local segregation, and information sharing in the functional network.
Network integration indicates the availability to combine information of various brain areas and transmit information in the network 2 . Global efficiency is defined as the average inverse shortest (weighted) path length from one node to all other nodes in the functional network of global information integration.
where is the number of nodes composing the graph, and is the shortest path length between node and node . A higher value means shorter lengths and easily communicable in the network. Thus, an efficient global network means that the nodes are remarkably integrated and the path length between nodes is constantly shortened 3 .
where is the sub-graph of the first neighbours of node . indicates the effectiveness of information delivery to the first neighbour of a certain node when this node is eliminated from the network. It can be interpreted as efficiency of information exchange in the local communities for measuring fault tolerance network between connected neighbours 4 .
Altogether, small-worldness is defined as the ratio of clustering coefficient to characteristic where and indicate the clustering coefficients and and indicate the characteristic path lengths in the tested network and a random network, respectively. A smallworld network is characterised by high clustering coefficients and low characteristic path lengths. Therefore, this measure can reveal the balance between global integration and local segregation to examine self-organized critical dynamics in segregated and integrated processing within the brain 5 .