Abstract
Anaesthesia combined with functional neuroimaging provides a powerful approach for understanding the brain mechanisms that change as consciousness fades. Although propofol is used ubiquitously in clinical interventions that reversibly suppress consciousness, its effect varies substantially between individuals, and the brain bases of this variability remain poorly understood. We asked whether three networks that are primary sites of propofol-induced sedation and key to conscious cognition — the dorsal attention (DAN), executive control (ECN), and default mode (DMN) network — underlie responsiveness variability under anaesthesia. Healthy participants (N=17) underwent propofol sedation inside the fMRI scanner at dosages of ‘moderate’ anaesthesia, and behavioural responsiveness was measured with a target detection task. To assess information processing, participants were scanned during an active engagement condition comprised of a suspenseful auditory narrative, in addition to the resting state. A behavioural investigation in a second group of non-anesthetized participants (N=25) qualified the attention demands of narrative understanding, which we then related to the brain activity of participants who underwent sedation. 30% of participants showed no delay in reaction times relative to wakefulness, whereas the others, showed significantly delayed and fragmented responses, or full omission of responses. These responsiveness differences did not relate to information processing differences. Rather, only the functional connectivity within the ECN during wakefulness differentiated the participants’ responsiveness level, with significantly stronger connectivity in the fast relative to slow responders. Consistent with this finding, fast responders had significantly higher grey matter volume in the frontal cortex aspect of the ECN. For the first time, these results show that responsiveness variability during propofol anaesthesia relates to inherent differences in brain function and structure within the executive control network, which can be predicted prior to sedation. These results shed light on the brain bases of responsiveness differences and highlight novel markers that may help to improve the accuracy of awareness monitoring during clinical anaesthesia.
Introduction
Understanding how the human brain gives rise to consciousness remains a grand challenge for modern neuroscience. A first step to understanding consciousness is to define what it is not. To this end, anaesthesia combined with functional neuroimaging provides a powerful approach for studying the brain mechanisms that change as consciousness fades (Vatansever et al, 2020; Pal et al., 2020; Varley et al., 2020; Luppi et al., 2019; Demertzi et al., 2019; Mashour and Hudetz 2018; Naci et al., 2018; MacDonald et al., 2015; Sarasso et al., 2015; Stamatakis et al., 2010), and, conversely, that are necessary for realizing human consciousness. Anaesthesia has been used for over 150 years to reversibly abolish consciousness in clinical medicine, but its effect can vary substantially between individuals. At moderate dosages, the intended suppression of behavioural responsiveness is highly variable (Chennu et al., 2016; Bola et al., 2019), and at deep anaesthesia dosages, in rare cases (0.1-0.2%; Mashour and Avidan 2015; Sandin et al., 2000), individuals retain conscious awareness, also known as ‘unintended intraoperative awareness’ (Pandit et al., 2017; Sanders et al., 2017; Sandin et a., 2000; Mashour and Avidan 2015). A much higher percentage of patients presumed to be unconscious during general anaesthesia (22%; Leslie et al., 2007) may have subjective experiences, such as dreaming. The brain-bases of this considerable inter-individual effect variability remain poorly understood.
A key and mostly overlooked question is what these individual differences (Searle and Hopkins 2009; Palanca et al., 2009) can reveal about the unravelling of conscious cognition as consciousness falters during anaesthesia. Propofol is the most common anaesthetic agent in clinical interventions that require the reversible suppression of consciousness. Two recent electroencephalography (EEG) studies that measured variable behavioural responsiveness during mild and moderate propofol anaesthesia reported that the participants’ varying levels of responsiveness were differentiated by alpha band connectivity during wakefulness (Chennu et al., 2016), and changing patterns of EEG signal diversity from wakefulness to moderate sedation (Bola et al., 2019). However, the brain regional-or network-bases for this substantial individual variability (Searle and Hopkins 2009; Palanca et al., 2009), and its potential link to cognitive function, remain unknown. Propofol-induced sedation produces selective metabolic impairment and reduces brain activity bilaterally in frontal and parietal associative regions (Witon et al., 2020; Boly et al., 2008; Plourde et al., 2006; Alkire 2005; Baars et al., 2003; Fiset et al., 1999; for a review see MacDonald et al., 2015; but see Pal et al., 2020). Therefore, three brain networks with frontal and parietal lobe distribution, the dorsal attention network (DAN), executive control network (ECN), and default mode network (DMN) are primary candidate sources of individual variability under propofol sedation.
The DAN and ECN, distributed laterally across frontal and parietal lobes, are key to orchestrating stimulus-driven and goal-directed cognition. Their roles include, orienting and modulating attention to the saliency of incoming stimuli, monitoring and analysing information from the environment and integrating it with internally generated goals, as well as planning and adapting new behavioural schemas to take account of this information (Duncan 2010; Elliott 2003; Corbetta and Schulman 2002; Kroger et al., 2002; Shallice 1988). The DMN extends partially in lateral and medial frontal and parietal lobes and is thought to mediate internally-oriented cognition, e.g., autobiographical memory, imagination, and thinking about the self (Andrews-Hanna et al. 2010; Schneider et al., 2008; Buckner et al., 2008; Beer 2007; D’Argembeau et al., 2005; Wicker et al., 2003; Gusnard et al., 2001). Recent studies show that it is also involved in external environment monitoring (Spreng et al. 2014; Buckner et al. 2008; Hahn et al., 2007), including encoding of scenes, episodes or contexts (Vatansever et al., 2017), or shifts in contextually relevant information (Smith et al., 2018). Recent studies that investigate state-related manipulations (conscious vs unconscious) show that by, contrast to other brain networks, the DAN, ECN and DMN are selectively impaired during loss of consciousness, not only under anaesthesia but also during severe brain injury (Luppi et al., 2019; Naci et al., 2018). Thus, they provide strong evidence that these three networks are part of a common brain mechanism disrupted during loss of consciousness across vastly different conditions.
It is well established that the DAN/ECN and the DMN display an antagonistic relationship (Huang et al., 2020;Fox et al., 2005; Fransson et al., 2005) during spontaneous thought and goal-directed cognition. During conditions that engage external-attention, the reduction of functional activity in the DMN (Greicius et al., 2003, Raichle et al., 2001, Shulman et al. 1997), is concomitant with an increase of activity in the DAN and ECN (Seeley et al. 2007; Dosenbach et al. 2007; Fox et al., 2005; Sridharan et al. 2008). The anticorrelation between these is related to individual differences in performance variability (Kelly 2008), and directly supports sustained attention (Kucyi et al., 2020). More broadly, it has been suggested to facilitate ongoing switches between internally- and externally biased modes of attention (Honey et al., 2017; Buckner et al., 2013) that contribute to consciousness (Carhart-Harris and Friston 2010). Conversely, this antagonistic relationship breaks down in state-related manipulations of consciousness, such as during sleep (Tagliazucchi et al., 2013), anaesthesia (Bonhomme at el., 2012; Boveroux et al., 2010), and severe brain injury (Haugg et al., 2018; Boly et al., 2009). Despite this accumulating evidence for the key and interrelated roles of the DAN, ECN, and DMN in supporting conscious cognition (Huang et al., 2020; Demertzi et al., 2013; Vanhaudenhuyse et al., 2011), and their primacy as target sites of propofol-induced sedation, their roles in individual differences under anaesthesia have not been previously investigated.
To address this gap, in two studies we tested whether variability or impairments in functional connectivity within and between the DAN, ECN, and DMN underlie individual differences in responsiveness during propofol anaesthesia. To directly investigate their individual and joint roles, in a first fMRI study we scanned healthy volunteers (N=17) during wakefulness and during the administration of propofol at dosages of ‘moderate anaesthesia,’ expressly aimed at engendering individual differences. Prior to the wakefulness and moderate anaesthesia scans, participants were asked to perform a verbal recall memory test that assessed basic memory function. Variability in behavioural responsiveness in each state was assessed with an auditory target detection task prior to scanning. To test whether any individual differences in responsiveness were related to differences in information processing that were invisible to the clinical sedation scale (Ramsay et al., 1974), participants were scanned during an active information processing condition comprised of a brief (5 minutes) engaging auditory narrative, and during a control condition of spontaneous thought, i.e., the resting state.
Listening to plot-driven narratives is naturally engaging, requires minimal behavioural collaboration from volunteers, and therefore, is highly suitable for testing information processing independently of behavioural output or eye opening (Naci et al., 2018; 2017), which are impaired in moderate anaesthesia. In a second behavioural study, we assessed the high-level attention demands of narrative understanding in an independent behavioural group (N=25), and related them to the brain activity of participants who underwent scanning.
Finally, to test whether individual differences under anaesthesia related to inherent brain features independent from the propofol sedation, we investigated the functional connectivity between and within the three networks during wakefulness, as well as grey matter volume differences across participants.
Methods
Participants
Study 1
Healthy participants for the fMRI scanning anaesthesia study (N=17; 18-40 years; 13 males) were tested at the Robarts Research Institute, Western University, in Canada. Ethical approval was obtained from the Health Sciences Research Ethics Board and Psychology Research Ethics Board of Western University. Data from the wakefulness condition was previously reported in Naci et al., 2018.
Study 2
The independent group of healthy participants (N=25; 18-40 years; 7 males) for the behavioural suspense rating study were tested at the Global Brain health Institute at Trinity College Dublin, in Ireland. Ethical approval was obtained from the School of Psychology Research Ethics Board, Trinity College Dublin.
All healthy participants were right-handed, native English speakers, and had no history of neurological disorders. Both studies were performed in accordance with the relevant guidelines and regulations set out by the research ethics boards. Informed consent was obtained for each participant prior to the experiment.
Stimuli and Design
Study 1
Behavioural testing
Memory recall task
The brief recall from the Mini Mental State Exam (Folstein et al., 1975) was adopted to test memory function under sedation. The researcher named three unrelated objects clearly and slowly and asked the volunteer to name each of them. The volunteer was instructed to remember the words in order to be able to repeat them in a short while. After 10 minutes, the volunteer was asked the repeat the words. Two different lists (a. Ball-Flag-Tree; b. Flower-Egg-Rope) were used to avoid familiarity effects between the wakeful and moderate anaesthesia states. Their presentation was counterbalanced across participants.
Auditory target detection task
Before the wakeful and moderate anaesthesia fMRI sessions, participants were asked to perform a computerized auditory target detection task (4 minutes), which aimed to assess individual differences during moderate anaesthesia. Participants were instructed via headphones to press a button with their index finger as soon as they heard an auditory beep and to keep their eyes on the fixation cross on the screen. The response box was tapped to the hand to ensure the participant did not lose contact during anaesthesia. The averaged reaction time (RT) for each participant was computed for further analyses.
MRI testing
Inside the MRI scanner, participants underwent two functional scans during wakefulness and moderate anaesthesia. A plot-driven audio narrative (5 minutes) was presented over MRI compatible noise cancelation headphones (Sensimetrics, S14; www.sens.com), and participants were asked to simply listen with eyes closed. The story comprised a highly engaging auditory excerpt from the movie ‘Taken’ depicting dramatic events, where a young girl travelling abroad without her family is kidnapped while speaking on the phone to her father. A similar eyes-closed, resting state condition (8 minutes) was also acquired in each state for comparison with the narrative condition.
Sedation Procedure
The level of sedation was measured with the Ramsay clinical sedation scale for each participant by 3 independent assessors (two anaesthesiologists and one anaesthesia nurse) in person inside the scanner room. Before entering the fMRI scanner, a 20G I.V. cannula was inserted into a vein on the dorsum of the non-dominant hand of the participants. The propofol infusion system was connected to the cannula prior to the first scanning session. No propofol was administered during the wakeful session. Participants were fully wakeful, alert and communicated appropriately (Ramsay 1) and wakefulness (eye-opening) was monitored with an infrared camera placed inside the scanner. At the commencement of the moderate anaesthesia session, intravenous propofol was administered with a Baxter AS 50 (Singapore). An effect-site/plasma steering algorithm was used combined with the computer-controlled infusion pump to achieve step-wise increments in the sedative effect of Propofol. This infusion pump was manually adjusted to achieve the desired levels of sedation, guided by targeted concentrations of Propofol, as predicted by the TIVA Trainer (the European Society for Intravenous Anaesthesia, eurosiva.eu) pharmacokinetic simulation program. The pharmacokinetic model provided target-controlled infusion by adjusting infusion rates of Propofol over time to achieve and maintain the target blood concentrations as specified by the Marsh 3 compartment algorithm for each participant, as incorporated in the TIVA Trainer software (Marsh et al., 1991). In accordance with the Canadian Anaesthesia Society (CAS) guidelines, non-invasive blood pressure (NIBP), heart rate, oxygen saturation (SpO2) and end-tidal carbon dioxide (ETCO2) were monitored continuously through the use of a dedicated MR compatible anaesthesia monitor. Complete resuscitation equipment was present throughout the testing.
Propofol infusion commenced with a target effect-site concentration of 0.6 μg/ml and oxygen was titrated to maintain SpO2 above 96%. Throughout sedation, participants remained capable of spontaneous cardiovascular function and ventilation. Supplemental oxygen was administered via nasal cannulae to ensure adequate levels of oxygen at all times. If the Ramsay level was lower than 3 for moderate sedation, the concentration was slowly increased by increments of 0.3 μg/ml with repeated assessments of responsiveness between increments to obtain a Ramsay score of 3. During administration of propofol, participants generally became calm and slowed in their response to verbal communication. Once participants stopped engaging in spontaneous conversation, and speech became sluggish, they were classified as being at Ramsey level 3. Nevertheless, when asked via loud verbal communication participants agreed to perform the auditory target detection and memory recall tasks, as would be expected at Ramsey level 3. During moderate anaesthesia (Ramsey 3), the mean estimated effect-site propofol concentration was 1.99 (1.59-2.39) μg/ml and the mean estimated plasma propofol concentration was 2.02 (1.56-2.48) μg/ml.
Study 2
To provide a subjective measure of attention during the story, participants rated how “suspenseful” it was every 2 seconds, from ‘least’ (1) to ‘most suspenseful’ (9). Participants heard the stimulus through over-ear headphones in sound-isolated rooms at the Global Brain Health Institute at Trinity College Dublin. Laboratory computers were used to present the stimulus, cue and collect responses. The audio excerpt was divided into 156 clips, each 2 seconds long to match the repetition time (TR, 2 sec) used in independent participant group who underwent MRI scanning. Participants had up to 3 seconds to make a response, at which point the next sequential clip began immediately. At the end of experiment, participants indicated via a feedback questionnaire that the interruptions did not disrupt the coherence of the story’s plot and their perception of suspense throughout.
Data Analyses
Study 1
MRI Data Acquisition
Functional images were obtained on a 3T Siemens Prisma system, with a 32-channel head coil. The high-resolution brain structural images were acquired using a T1-weighted 3D MPRAGE sequence with following parameters, voxel size: 1 × 1 × 1 mm, TA=5 minutes and 38 seconds, echo time (TE) = 4.25ms, matrix size=240×256×192, flip angle (FA) = 9 degrees. The functional echo-planar images (EPI) images were obtained with following parameters, 33 slices, voxel size: 3 × 3 × 3, inter-slice gap of 25%, repetition time (TR) = 2000ms, TE=30ms, matrix size=64×64, FA=75 degrees. The audio story and resting state had 155 and 256 scans, respectively. A volume level deemed comfortable by each individual was used for the duration of the scanning.
MRI preprocessing
1) Functional data. Standard preprocessing procedures and data analyses were performed with SPM8 (Wellcome Institute of Cognitive Neurology, http://www.fil.ion.ucl.ac.uk/spm/software/spm8/) and the AA pipeline software (Cusack et al., 2015). In the preprocessing pipeline, we performed slice timing correction, motion correction, registration to structural images, normalization to a template brain, and smoothing. The data were smoothed with a Gaussian smoothing kernel of 10mm FWHM (Peigneux et al. 2006). Spatial normalization was performed using SPM8’s segment-and-normalize procedure, whereby the T1 structural was segmented into grey and white matter and normalized to a segmented MNI-152 template. These normalization parameters were then applied to all EPIs. The time series in each voxel was high pass-filtered with a cutoff of 1/128 Hz to remove low-frequency noise, and scaled to a grand mean of 100 across voxels and scans in each session. The preprocessed data were analyzed in using the general linear model. Prior to analyses, the first five scans of each session were discarded to achieve T1 equilibrium and to allow participants to adjust to the noise of the scanner. To avoid the formation of artificial anti-correlations (Murphy et al. 2009; Anderson et al., 2011), we performed no global signal regression. 2) Structural data. The brain structural images were processed using FreeSurfer package (https://surfer.nmr.mgh.harvard.edu/), a well-documented automated program which is widely used to perform surface-based morphometric analysis (Dale et al. 1999). The processing steps include: 1) removing the non-brain tissue; 2) transforming the skull-stripping brain volume to Talairach-like space; 3) segmenting brain tissues to GM, WM, cerebrospinal fluid (CSF); 4) performing intensity normalization to remove the effect of bias field; 5) building a surface tessellation to generate a triangular cortical mesh consisting of about 300,000 vertices in the whole brain surface; 6) correcting topological deficits of cortical surface; and 7) deforming brain surface to generate optimized models of GM/WM and GM/CSF boundaries.
Analyses of fMRI data
To investigate whether the behavioural variability under anaesthesia related to perceptual or higher-order processing differences among participants, we performed two mixed data-driven and model-based analyses. First, we extracted the sound envelope of the auditory narrative via the Matlab MIRtoolbox (http://www.jyu.fi/hum/laitokset/musiikki/en/research/coe/materials/mirtoolbox) and built a generalised linear model (GLM) by using statistical parametric mapping to derive auditory characteristic-related brain activation for each individual. Subsequently, the z-scored average suspense ratings of the narrative obtained by the independent group of participants in study 2 were used as a regressor in the fMRI data of individuals who underwent propofol anaesthesia in study 1, to measure the neural correlates of information processing during the narrative condition. The regressors were generated by convolving boxcar functions with the canonical hemodynamic response function. Also included in the general linear model were nuisance variables, comprising the movement parameters in the three directions of motion and three degrees of rotation, as well as the mean of each session. Fixed-effect analyses were performed in each subject, corrected for temporal auto-correlation using an AR (1)+white noise model. Linear contrasts were used to obtain subject-specific estimates for each effect of interest. Linear contrast coefficients for each participant were entered into the second level random-effects analysis. Clusters or voxels that survived at p<0.05 threshold, corrected for multiple comparisons with the family wise error (FWE) were considered statistically significant.
Functional connectivity (FC) within and between the DAN, ECN, and DMN was assessed by computing the Pearson correlation of the fMRI time courses between 19 regions of interests (ROIs) (from Raichle 2011) constituting these different brain networks, as identified by resting state studies (Table 1). Permutation (1000 times) tests were used to explore FC difference between conditions and were FDR corrected for multiple comparisons. All of the analyses were conducted using Fisher z-transformed correlation (Pearson r). Pearson correlation cannot directly imply causal relations between neural regions, but it is a simple FC measure with minimal assumptions. It is adequate for our purpose because the time-course and spatial extent of fronto-parietal and default mode networks encompassed a vast swath of hierarchical processing cascade and, thus, many regions of cause-effect space were triggered by the stimulus. FC within and between these networks, indicated by Pearson correlation, reflected their interactions over several minutes and the resulting computations on the information content of the narrative condition, or of spontaneous thought during the resting state condition. Glass’s delta was used to compute the effect size of the comparison between wakeful and moderate anaesthesia states, because of the different standard deviations of the two states, while Hedges’ g was used to compute the effect size of the comparison between fast and slow responders because of the different sample sizes for the two groups.
Grey matter volume analyses
Grey Matter Volume (GMV) was computed as the volume of a truncated tetrahedron. The reconstructed surface was smoothed with a 10-mm full-width at half maximum (FWHM) Gaussian kernel to perform the statistics. Based on group-level comparison, we detected the clusters showing significant GWM differences at the group-level. Monte Carlo simulation cluster analysis and a cluster-wise threshold of p < 0.05 was adopted for multiple comparisons correction. In this way, we identified the regions with significant group differences in GMV. We then extracted averaged GMV of each significant cluster for individual participants and performed permutation tests (1000 times) to identify the group difference between fast and slow responders.
Study 2
Analyses of suspense ratings
A group-averaged set of suspense ratings were calculated by averaging participants’ ratings at each time-point. To determine how similar suspense ratings were across the group, the inter-subject correlation of suspense ratings was computed as the average of the Pearson correlations of each participant’s data with the mean data from the rest of the group. To compute the average correlation value, each correlation coefficient was normalized using Fisher’s r-to-z transformation and the average of these values was computed and then back-transformed into a correlation coefficient.
Results
The effect of moderate anaesthesia on brain network connectivity
Initially, we investigated how moderate propofol anaesthesia perturbed the patterns of functional connectivity (FC) relative to wakefulness across all the participants (Figure 1). For the narrative condition, a 2×2 ANOVA that explored the main effects of state (wakefulness, moderate anaesthesia) and network connectivity (DMN, DAN, ECN, DMN–DAN, DMN– ECN, DAN–ECN) showed significant main effects of state (F (1, 16) = 5.3, p < 0.05), network connectivity (F (2.9, 46.7) = 28.1, p < 0.0001), and a significant interaction effect of state by network connectivity (F (5, 80) = 8.2, p < 0.0001). We investigated further this interaction effect by testing the effect of sedation state on each network connectivity. We found significantly higher DMN–DAN (permutation test, p < 0.01, FDR corrected), as well as DMN–ECN (permutation test, p < 0.05, FDR corrected) connectivity in moderate anaesthesia relative to wakefulness, during the narrative condition (Figure 1B).
For the resting state condition, a 2×2 ANOVA that explored the main effects of state (wakefulness, moderate anaesthesia) and network connectivity (DMN, DAN, ECN, DMN– DAN, DMN–ECN, DAN–ECN) showed a significant main effect of network connectivity (F (2, 32.3) = 18.9, p < 0.0001) and a significant interaction effect of state by network connectivity (F (2.4, 38.8) = 4.3, p < 0.05). However, we found no significant differences between wakefulness and moderate anaesthesia for each network connectivity (Figure 1B).
A direct comparison between the narrative and resting state conditions, with a 2×2 ANOVA [condition (narrative, resting state) x network connectivity (6 levels)] on the FC difference values (moderate anaesthesia - wakefulness), showed a significant main effect of network connectivity (F (5, 80) = 9.9, p < 0.0001), and no main effect of condition or interaction effect. Although the 2×2 ANOVA did not reveal any interactions, effect size analyses showed higher effect sizes of moderate anaesthesia in the narrative relative to the resting state condition for both DMN–DAN (Glass’ Delta = 1.20 vs 0.52) and DMN–ECN (Glass’ Delta = 1.02 vs 0.46). In summary, these results suggested that the antagonistic relationship between the DMN and the DAN/ECN and was reduced during moderate anaesthesia, with a stronger and significant result in the narrative condition relative to the resting state.
The effect of moderate anaesthesia on behavioural responsiveness
Despite the same Ramsay 3 clinical score, independently determined by three assessors, we observed significant heterogeneity in the reaction time of the auditory detection task (Figure 2). 5/17 participants were not delayed significantly relative to their non-sedated responses, 9/17 were significantly delayed and had fragmented responses (showing 2–40% missing trials), and 3/17 failed to make any responses (full omission of responses), despite agreeing to have understood the task instructions and to make responses via the button box affixed to their dominant hand. This large individual variability was at odds with the propofol infusion rates titrated for each participant based on the pharmacokinetic model adjusted for demographic variables (Marsh et al., 1991), to maintain stable target blood concentrations consistent with Ramsay 3 sedation level (Ramsay, 1974). Additionally, all participants successfully completed the brief delayed recall memory task.
To further investigate these individual differences in responsiveness, individuals were divided into two groups, fast responders (FRs; N=11), and slow responders (SRs; N=6) based on the mean split of reaction time (Figure 2). Individual responsiveness variability was not explained by demographic differences, as suggested by no differences in age (between-samples t-test, t = −0.42, p = 0.677) and gender (Fisher’s exact test, odds ratio = 0.47, p = 0.584) between FRs and SRs.
Rather, the brain-bases of this variability may be related to three other factors. First, the individual variability could be related to underlying fine-grained perceptual and/or higher-order information processing differences between participants, which may have been invisible to the behavioural examination during the Ramsay assessment. For example, participants who made no responses during moderate anaesthesia, may have appeared to understand the task instructions based on their limited verbal responses, despite genuine lack of understanding, which may have led to failure of button-press responses. Second, independently of the extent of information processing in each individual, responsiveness differences may be driven by connectivity alterations within and between the DAN, ECN and DMN networks, which would suggest impairments in other cognitive processes. Third, individual responsiveness differences may be related to structural brain differences, which may, in turn, manifest as differences in information processing and/or functional connectivity. In the following analyses, we explored each of these factors in turn.
Information processing and behavioural responsiveness during moderate anaesthesia
To measure information processing during the auditory narrative, we used a previously established method (Naci et al., 2018; 2017; 2014), where we demonstrated that the extent of stimulus-driven cross-subject correlation provides a measure of regional stimulus-driven information processing. During wakefulness, we observed widespread and significant (p<0.05; FWE cor) cross-subject correlation between healthy participants, bilaterally in sensory-driven (primary and association) auditory cortex, visual cortex, as well as higher-order frontal and parietal cortical regions (Figure 3A). By contrast, during moderate anaesthesia, significant (p<0.05; FWE cor) cross-subject correlation was observed only in primary and secondary auditory cortex and three small clusters (the left post central gyrus, left paracentral lobule and right precentral gyrus) of premotor and motor cortex that have been implicated in language comprehension (Hauk et al., 2004) (Figure 3B). As expected, and consistent with previous studies (Naci et al., 2018; Davis et al., 2007), these results suggested reduced information processing during moderate anaesthesia relative to wakefulness.
We delved deeper into these effects to investigate whether each individual’s sensory-driven (auditory) perceptual processes and high-order attention processes during the story condition, related to their response times in the target detection task. First, we used SPM to model the relationship between the story’s sound envelope and changes in brain activity over time. At the group level, during wakefulness, the sound envelope predicted brain activity in bilateral primary and secondary auditory cortex and right inferior frontal gyrus (Figure 3C). This activity was dramatically reduced in moderate anaesthesia to one small cluster in left primary auditory cortex (superior temporal gyrus) that did not reach statistical significance (p=0.05 FWE cor) (Figure 3D). At the individual level, the sound envelope predicted significant activations in the auditory regions in 14/17 participants (3/17 showing subthreshold activations) during wakefulness, and in 10/17 during moderate anaesthesia. We observed no correlation between the participants’ reaction times in either wakefulness or moderate anaesthesia and the extent of their individual activations in auditory regions (Figure 3 G–H).
Second, to investigate differential recruitment of high-order attention processes during the story, we used a previously established qualitative measure of the listeners’ sustained attention throughout the narrative — the perception of suspense on a moment-by-moment basis (Naci et al., 2014). Beyond basic physical properties captured in the sound envelope, such as the amplitude and tone of the musical soundtrack, ‘suspense,’ in dramatic narratives such as the one used in this study, arises through ongoing sustained attention, as the current features of the movie (e.g., a young girl is alone in a foreign country, far away from her family) are compared to stored knowledge of the world (e.g., human traffickers pray on vulnerable targets), what has happened previously in the story (e.g., the girl’s close friend has been kidnapped), and what may happen in future (e.g., she may be kidnapped).
Again, we used SPM to model the relationship between this qualitative measure of the story’s ongoing high-level attention demands and changes in brain activity over time, measured in the independent group of participants, who had listened to the story without a secondary task in the fMRI scanner, during the wakeful and moderate anaesthesia states. During wakefulness, suspense ratings significantly (p<0.05; FWE cor) predicted activity in regions of the auditory attention network (the left superior temporal gyrus, left and right inferior frontal cortex, the inferior precentral gyrus/sulcus) (Tobyne et al., 2018; Michalka et al., 2015; Naci et al., 2013) and the salience network (the left/right anterior cingulate, insula, inferior frontal, and the thalamus) (Seeley 2019; Menon 2015), with clips rated as ‘highly suspenseful’ predicting stronger activity in this networks (Table 2) (Figure 3E). As expected, during moderate anaesthesia, the activation predicted by suspense ratings was reduced to one cluster extending in right superior temporal and marginal gyri (Figure 3F).
In the behavioural group, the suspense ratings throughout the narrative showed a very robust (r=0.90; SE=0.07)) and significant inter-subject correlation (t(24)=20.56, p=9.6e-17), confirming the common understanding of the story across different individuals listening to this narrative. In previous work, we have shown that a common neural code that is reliably present in individual participants underlies common understanding of engaging audio-visual narratives (Naci et al., 2017; 2014). Therefore, the similarity of the story’s suspense ratings enabled model-based predictions of the underlying brain activity that could be applied to individual participants. The suspense ratings predicted significant activations in the auditory attention network and salience network regions in 17/17 of participants during wakefulness and 14/17 during moderate anaesthesia. We observed no correlation between the participants’ reaction time in wakefulness or moderate anaesthesia and the extent of their individual activation in the auditory attention and salience networks regions (Figure 3 I–J).
In summary, these results suggested that differences in behavioural responsiveness under anaesthesia, were neither driven by differences in auditory perceptual processes, nor in sustained attention processes and the ability to integrate complex and dynamic information that unveils over time.
Network connectivity and behavioural responsiveness during moderate anaesthesia
Rather, alterations in the connectivity within and between the DAN, ECN, and DMN during moderate anaesthesia may drive these individual differences. We observed that functional connectivity within the ECN was the only FC type that differentiated the two response groups, with significantly higher FC for FRs relative to SRs, in the narrative condition during wakefulness (permutation test, p < 0.005, FDR corrected) (Figure 4). No FC differences between FR and SR were observed within the DAN or DMN, or between any of the three network pairs.
To further investigate the within–ECN FC differences between FRs and SRs, we performed a 2×2×2 ANOVA that explored the main effects of condition (narrative, resting state), state (wakefulness, moderate anaesthesia) and response group (FRs, SRs), as well as any interaction among them. We found a main effect of response group (F (1,15) = 7.40, p < 0.05) that was driven by higher connectivity for FRs, and no effects on condition or state. Although the 2×2×2 ANOVA did not reveal any interactions, effect size analyses showed a higher difference between FRs and SRs in the narrative condition during wakefulness (Hedges’ g = 1.58) relative to moderate anaesthesia (Hedges’ g = 1.10), and relative to the resting state condition during wakefulness (Hedges’ g = 0.68), and moderate anaesthesia (Hedges’ g = 0.96). In summary, the within–ECN functional connectivity during the narrative condition in the wakeful state most strongly related to the individual differences in behavioural responsiveness observed during moderate anaesthesia.
Structural brain differences and behavioural responsiveness during moderate anaesthesia
Finally, we tested whether individual differences in behavioural responsiveness were related to structural brain differences, which may also underlie functional connectivity differences. To this end, we performed whole-brain vertex-based comparisons of grey matter volume between FRs and SRs. We found that FRs had uniformly significantly higher grey matter volume relative to the SRs, in five regions bilaterally distributed: the left and right superior and dorsolateral frontal cortex (L SFC, R SFC), including the presupplementary motor area (pre-SMA), left and right rostral middle frontal cortex (L rMFC, R rMFC) and the right precentral cortex (R Precentral) (Figure 5 and Table 3). These regions overlapped with the three functional ROIs that defined the frontal nodes of the executive control network (Table 1). The opposite effect —higher grey matter for SR than FR—was not observed anywhere in the brain. These results lend support to the functional connectivity results above, and together they strongly suggest that connectivity within the ECN, and especially the frontal aspect of this network, underlies individual differences in behavioural responsiveness under moderate anaesthesia.
This was further confirmed by a direct comparison of the FC differences between FRs and SRs in the ROIs covering the frontal (dorsal medial PFC, left and right anterior PFC) and parietal (left and right superior parietal cortex) aspects of the ECN separately. A 2×2×2 ANOVA with factors state (wakefulness, moderate anaesthesia), region (frontal ECN, parietal ECN), and response group (FRs, SRs) showed a main effect of region [F (1,15) = 11.791, p < 0.01], driven by higher connectivity in the parietal aspect of the ECN, and three-way interactions (state x region x response group) [F (1,15) = 8.358, p < 0.05). During wakefulness, connectivity within the frontal or the parietal aspects of the ECN, independently, could distinguish FRs and SRs (permutation test, p < 0.05 for frontal aspect; permutation test, p < 0.05 for parietal aspect, FDR corrected). By contrast, in the moderate anaesthesia, only FC in the frontal aspect could significantly distinguish the FRs and SRs (permutation test, p < 0.05, FDR corrected) (Figure 6).
Discussion
Although anaesthesia has been used for over 150 years to reversibly abolish consciousness in clinical medicine, the brain-bases of its considerable inter-individual effect variability remain poorly understood. To address this gap, we asked whether the connectivity within and between three networks that are primary sites of propofol-induced sedation and key to conscious cognition — the DAN, ECN and DMN — underlie individual responsiveness differences under anaesthesia. Consistent with previous studies (Chennu et al., 2016; Bola et al., 2019), we observed substantial individual differences in reaction times under moderate anaesthesia. These were not related to differences in perceptual or higher-order information processing. Rather, only the functional connectivity within the ECN during the wakeful narrative condition differentiated the participants’ responsiveness level, with significantly stronger ECN connectivity in the fast, relative to slow responders. These results were further supported by our finding that fast responders had significantly higher grey matter volume in the frontal cortex aspect of the executive control network. Taken together, the functional and structural imaging results, provided very strong evidence that individual differences in responsiveness under moderate anaesthesia related to inherent differences in brain function and structure within the executive control network, which can be predicted prior to sedation.
Consistent with previous studies, we found that the antagonistic relationship between the DAN/ECN and the DMN was reduced during moderate anaesthesia (Bonhomme et al., 2012; Boveroux et al., 2010), but the connectivity between the DAN/ECN and DMN, or within the DMN network showed no relationship to individual responsiveness variability. In addition to orchestrating internally-oriented cognition (Andrews-Hanna et al. 2010; Schneider et al., 2008; Buckner et al., 2008; Beer 2007; D’Argembeau et al., 2005; Wicker et al., 2003; Gusnard et al., 2001), the DMN was recently found to play a role in broad associative processes (Smith et al., 2018;Vatansever et al., 2017). The absence of a relationship between the DMN connectivity and responsiveness differences is fitting given that the target detection task did not elicit rich episodic information or any contextual shifts. Furthermore, it suggests that individual differences in responsiveness were not driven by differences in internally-oriented, or self-referential cognitive processes.
The differential impact of moderate anaesthesia on cognitive process can be inferred in a post hoc manner, based on the perceptual and cognitive processes recruited by the target detection task and the brain networks and regions that showed functional and/or structural differences between the participants. The task involved a hierarchy of perceptual and cognitive processes. These included (a) auditory perception, (b) understanding of linguistic instruction, short-term memory to remember instructions for the duration of the experiment, (d) attention to auditory stimuli, (e) attention to intention to make a response, (f) cognitive control over other, competing and distracting mental processes, and the (g) execution of a motor response, when prompted by the auditory stimuli. The Ramsay clinical assessment determined similar basic language communication (b), and we found short-term memory function (c) across participants during moderate anaesthesia. Furthermore, our results showed no relationship between behavioural responsiveness and sensory/auditory (a), or higher-order, including linguistic information-processing that relied on sustained attention (b, d), across participants. Therefore, differences in responsiveness were likely underpinned by differences in complex mental faculties, such as attention to intention, cognitive control and action execution.
Our functional connectivity results demonstrated that FC within the ECN differentiated the participants’ level of responsiveness, suggesting that moderate anaesthesia impacts processes of goal and action execution subserved by the ECN (Marek and Dosenbach 2018; Dosenbach et al., 2007; Ernst and Paulus 2005). Furthermore, our structural imaging results demonstrated that grey matter volume differences in the frontal cortex aspect of the ECN, including the dorsolateral prefrontal cortex (DLPFC) and pre-supplementary motor area (pre-SMA), underscore individual responsiveness differences. In addition to their role in action selection and execution (Lee et al., 1999; Boly et al., 2007), the DLPFC and pre-SMA represent higher-order attention, i.e., attention to intention. Lau et al (2004) investigated the brain regions that were preferentially involved in attending to the intention to move, relative to the actual movement, and found that brain activity in pre-SMA, as well as its functional connectivity with right DPFC represented the intention to move. The pre-SMA has also been implicated in sustained cognitive control (Nachev et al., 2005), another process that is key to successful task performance. In summary, a post-hoc interpretation of our findings would be that moderate anaesthesia impacts attention to intention, sustained cognitive control, and action execution, and that individuals with smaller grey matter volume in frontal regions and weaker functional connectivity within the ECN may have a vulnerability for stronger suppression of behavioural responsiveness than those with higher values for these features.
As the suppression of behavioural responsiveness is one of the key aims of anaesthesia, the converse finding was also highly relevant. Surprisingly, we found that 30% of participants showed no delay in reaction times under moderate anaesthesia relative to wakefulness, and critically, these exhibited inherent brain differences to participants who were significantly delayed. This result is highly relevant to the prediction of responsiveness variability under anaesthesia (Chennu et al., 2016; Palanca et al 2009) and monitoring depth-of-anaesthesia for the detection of unintended intraoperative awareness. Although rare (0.1-0.2%, Mashour and Avidan 2015; Sandin et al., 2000), unintended awareness can be very traumatic and lead to negative long-term health outcomes, such as post-traumatic stress disorder (up to 70%), as well as clinical depression or phobias (Pandit et al., 2017; 2014; Mashour and Avidan 2015). Given its rarity, unintended intraoperative awareness does not lend itself to direct investigation in the relatively small groups of typical research studies with vulnerable populations, including the present study, where individuals were deeply anaesthetized without intubation in a research MRI setting.
Our results suggest that individuals with larger grey matter volume in frontal regions and stronger functional connectivity within the ECN may require higher doses of propofol to become non-responsive, to the same extent as individuals with smaller grey matter volume and weaker connectivity in these regions. If replicated in a clinical context, these findings will provide novel markers that may help to improve the accuracy of awareness monitoring during clinical anaesthesia. It is worth noting that propofol was used here due to its wide prominence in clinical interventions, and future studies that employ the same paradigm across different agents will determine whether the brain-behaviour relationships we report here generalize to other anaesthetic agents.
Finally, we observed higher power for detecting individual responsiveness differences under anaesthesia from ECN functional connectivity in the wakeful narrative condition, than in the resting state. This may be due to higher signal to noise during the narrative condition. Previous studies have shown that the active attention engagement during well-crafted narratives leads to reduced movement (Centeno et al., 2016) and less sleep in scanner (Centeno et al., 2016, Vanderwal et al., 2015), resulting in higher signal-to-noise relative to resting state (rs) fMRI (Wang et al., 2017). Moreover, it may be due to higher power to detect functional roles of brain networks during naturalistic stimuli relative to rsfMRI. Rs fMRI is entirely unconstrained, and thus it is difficult to separate signal due to cognitive processes from extraneous sources, including motion, cardiac and respiratory physiological noise, arterial CO2 concentration, blood pressure, and vasomotion (Murphy et al., 2013; Birn et al., 2006). By contrast, as we have previously established, naturalistic narratives drive diverging responses across brain networks, highlighting their functionally distinct roles, and thus may be more sensitive for investigating how their differing cognitive roles are impacted by anaesthesia than rsfMRI (Naci et al., 2018; Haugg et al., 2018; Naci et al., 2014). Our results are consistent with a growing recognition of the importance of naturalistic stimuli for studying cognitive processes and understanding the neural basis of real-world functioning (Sonkusare et al., 2019).
Conflict of Interest
The authors declare no conflict of interest.
Author contributions
Study Conceptualization, L.N., N.T.; Data Acquisition, L.N. Methodology, R.C., F.D., L.N.; Formal Analyses, F.D., N.T.; Writing, F.D., L.N.; Funding and resource acquisition, A.M.O., L.N. Project administration, L.N.; Supervision, L.N.
Acknowledgements
F.D. was funded by the Provost PhD Award Scheme from Trinity College Dublin, to L.N. L.N. was also funded by an L’Oreal for Women In Science International Rising Talent Award, and the Welcome Trust Institutional Strategic Support grant. AMO is a CIFAR Fellow and is supported by the Canadian Institutes for Health Research (CIHR) and the Natural Sciences and Engineering Research Council of Canada (NSERC). Data acquisition (fMRI) was supported by a Canada Excellence Research Chairs award to AMO (Grant No. 215063)