Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2023 May 7.
Published in final edited form as: Neuroimage. 2022 Oct 6;263:119657. doi: 10.1016/j.neuroimage.2022.119657

Dynamic causal modelling of auditory surprise during disconnected consciousness: The role of feedback connectivity

Cameron P Casey a,*, Sean Tanabe a, Zahra Farahbakhsh a, Margaret Parker a, Amber Bo a, Marissa White a, Tyler Ballweg a, Andrew Mcintosh a, William Filbey a, Matthew I Banks a, Yuri B Saalmann b, Robert A Pearce a, Robert D Sanders c,d,e,*
PMCID: PMC10076444  NIHMSID: NIHMS1879500  PMID: 36209793

Abstract

The neural mechanisms through which individuals lose sensory awareness of their environment during anesthesia remains poorly understood despite being of vital importance to the field. Prior research has not distinguished between sensory awareness of the environment (connectedness) and consciousness itself. In the current study, we investigated the neural correlates of sensory awareness by contrasting neural responses to an auditory roving oddball paradigm during consciousness with sensory awareness (connected consciousness) and consciousness without sensory awareness (disconnected consciousness). These states were captured using a serial awakening paradigm with the sedative alpha2 adrenergic agonist dexmedetomidine, chosen based on our published hypothesis that suppression of noradrenaline signaling is key to induce a state of sensory disconnection. High-density electroencephalography was recorded from 18 human subjects before and after administration of dexmedetomidine. By investigating event-related potentials and taking advantage of advances in Dynamic Causal Modeling (DCM), we assessed alterations in effective connectivity between nodes of a previously established auditory processing network. We found that during disconnected consciousness, the scalp-level response to standard tones produced a P3 response that was absent during connected consciousness. This P3 response resembled the response to oddball tones seen in connected consciousness. DCM showed that disconnection produced increases in standard tone feedback signaling throughout the auditory network. Simulation analyses showed that these changes in connectivity, most notably the increase in feedback from right superior temporal gyrus to right A1, can explain the new P3 response. Together these findings show that during disconnected consciousness there is a disruption of normal predictive coding processes, so that all incoming auditory stimuli become similarly surprising.

1. Introduction

An estimated 313 million surgical procedures are performed globally each year, with this figure projected to increase dramatically with growing populations and spread of medical coverage (Weiser et al., 2015; Meara et al., 2015). General anesthesia has become an essential component of surgery, allowing physicians to perform operations by preventing patient movement and protecting patients from awareness of surgery. However, this is under ideal circumstances. While intraoperative awareness is not generally evident in clinical practice, focused investigations have shown it remains a pernicious problem, with estimated occurrence rates of roughly 5% (Sanders et al., 2017) in the general population and 11% among young adults (Lennertz et al., 2022). This implies that at least 15.6 million cases of intraoperative awareness occur each year. In order to better address this problem, it is essential that we develop a better understanding of how the brain becomes disconnected from the environment under general anesthesia.

A great deal of prior research has been conducted on the effects of anesthesia on the brain, especially in relation to loss of consciousness (Ku et al., 2011; Noreika et al., 2011; Lee et al., 2013; Blain-Moraes et al., 2015; Purdon et al., 2013). However, a common theme in this field of work is to assume that loss of consciousness has occurred when there is a loss of motor responsiveness to sensory input. This approach is problematic for our understanding of these processes because it conflates awareness of the environment, conscious experience itself, and motor responsiveness. Without experimentally distinguishing these processes, we face a fundamental limitation in our ability to understand the effects of anesthesia on the brain. Specifically, to understand sensory disconnection from the environment, it is necessary to directly compare states of ‘consciousness with sensory awareness’ versus ‘consciousness without sensory awareness’. We refer to the former as connected consciousness and the later as disconnected consciousness (Sanders et al., 2012; Casey et al., 2022).

Electroencephalography (EEG) offers a convenient method for real time assessment of neurophysiology in a clinical setting. However, routine EEG signal metrics, such as spectral power or functional connectivity measures, are limited in what they can reveal about causal mechanisms within the brain. Recent advances in generative modeling methods now allow for more sophisticated inferences to be drawn from EEG data, permitting researchers to test mechanistic hypotheses about directed information flow within the brain. One such approach is dynamic causal modeling (DCM), which provides a framework for modeling functional relationships within an anatomically defined network, and importantly, the ability to test the effects of external stimuli on that network (Friston et al., 2003; Kiebel et al., 2008; Stephan et al., 2010).

In the current study, we employ an auditory roving oddball paradigm with high-density EEG recording and DCM modeling to probe sensory processing during connected and disconnected consciousness. To experimentally bias our participants towards a disconnected state, we administered dexmedetomidine, a commonly used sedative and anesthetic. We then confirmed sensory disconnection using serial awakenings with structured interviews. We hypothesized that disconnected consciousness would be associated with a loss of feedforward connectivity in response to auditory stimulation, indicating a reduction in environmental information in-flow to higher-order cerebral cortex (Sanders et al., 2021). To our knowledge, the current work represents the first attempt to study the changes in sensory processing of auditory cues associated with loss of sensory awareness but maintenance of consciousness.

2. Methods

2.1. Subjects and drug administration

The subjects presented here were enrolled in the UNderstanding Consciousness Connectedness and Intra-Operative Unresponsiveness Study (UN-ConsCIOUS, NCT03284307). The resting-state data from this study has been previously published (Casey et al., 2022), however, the data presented here have not. In brief, 20 healthy volunteers, ages 18–32 years old, without prior contraindications to anesthetics, were recruited for dexmedetomidine administration. Due to an early revision in the study protocol, the first two subjects were presented with a different auditory paradigm than the latter 18 and were thus excluded from analyses. Drug administration occurred under the supervision of an anesthetist, to achieve a series of stable drug plateaus throughout the visit (Fig. 1 A). A rapid infusion of 3.0 μg kg−1 h−1 was initially given over a 10-min period followed by a 0.5 μg kg−1 h −1 maintenance infusion to achieve the first drug step. The second drug step was similarly achieved by a 10 min infusion of 3.0 μg kg−1 h −1 followed by a 1.5 μg kg−1 h−1 maintenance infusion. The total duration of drug exposure was limited to 4 h for each subject.

Fig. 1.

Fig. 1.

Open in a new tab

Experimental Overview. A) Hypothetical drug dosing diagram illustrating the format of each subject’s study visit. Roving oddball auditory stimulation was administered throughout the visit and paired with subjective report information to assess the subject’s state of sensory awareness and consciousness during the auditory paradigm. B) A graphical illustration of the roving oddball paradigm. Tones were played in blocks for which the 1st tone (red) is considered and oddball and the 6th tone (blue) is considered a standard. C) An illustration of the hierarchical PEB modeling approach (PEB of PEBs) that was applied to the DCM results. At the first level, a PEB model is generated for each subject to contrast their baseline and disconnected data. At the second level, a PEB of PEBs models the group level effect of sensory disconnection. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article).

2.2. Roving oddball paradigm

We implemented an auditory roving oddball paradigm (Fig. 1 B) based on the work of Garrido et al. (2007, 2008). Participants were presented with a series of single frequency tones with a pseudo-random number of repetitions, between 2 and 11 repeats, constituting a stim- ulus block. Each stimulus block was followed by a block of a different auditory tone, for a duration of 7 minutes. The first tone in a new block was classified as an auditory oddball while the 6th repetition of a tone was considered an auditory standard. The auditory stimuli were presented through a set of headphones using E-Prime (Psychology Software Tools, Inc., Pittsburgh, PA, USA). The volume of tones was calibrated to each participant’s comfort level during the visit baseline. The paradigm was run a variable number of times for each subject through the sedation visit, interspersed between other study paradigms as time allowed. However, it was run at least once at baseline and per each drug infusion concentration.

2.3. EEG data acquisition, processing, and ERP extraction

High-density EEG data were collected using a NA300 EGI system with 256-channel gel-caps. Electrodes were manually prepared with application of electrolyte electrode gel to achieve electrode impedances below 50 k Ω. Data are recorded using EGI’s Net Station Acquisition 5.4 software.

Subjects were allowed to rest with their eyes closed for up to 17 min (10 min of silence + 7 min of roving oddball paradigm) at a time without researcher intervention. Each rest period was concluded by a researcher calling the participant’s name and initiating a brief structured interview which has been discussed in detail previously (Casey et al., 2022). Briefly, participants were asked to recount the last thing going through their mind, whether they were awake or asleep, whether they were awake, dreaming, or unconscious, and whether they were aware of the external world around them. Participants who reported having an experience, being asleep, dreaming, and unaware of the world around them were considered to be in a state of disconnected consciousness. Participants who reported having an experience, being awake, and aware of the world around them were considered to be in a state of connected consciousness. Reports of no experience and unconsciousness as well as conflicting reports that could not be confidently labeled were excluded from analysis.

All data processing was performed by a member of the research team experienced in EEG analysis but blinded to the conscious state, using EEGLab (v14.1.2b) (Delorme and Makeig, 2004). Data were filtered between 1 and 30 Hz. Filtered data were visually inspected for noisy channels and noisy epochs, which were removed. Independent Components Analysis (ICA) were then computed using the InfoMax (Bell and Sejnowski, 1995) algorithm and components dominated by eye movements or muscle artifacts were rejected. After these cleaning steps, data were average referenced, linearly detrended, and baseline corrected by subtracting the average amplitude over the 50 ms preceding each trial.

2.4. Scalp ERP analysis

Grand-average ERPs were generated by averaging all trials for a particular tone-type (standard or oddball) and condition (baseline or disconnection). The baseline data, collected before dexmedetomidine administration, were used as the connected consciousness condition. This averaging procedure resulted in 4 ERPs: baseline-standard, baseline-oddball, disconnection-standard, and disconnection-oddball. Topographical views of each ERP over time were plotted using functions from FieldTrip toolbox (v20201229, https://www.fieldtriptoolbox.org/). An a prioi decision was made to make statistical comparisons between ERPs across time at electrode Cpz based on prior studies (Garrido et al., 2007; Boly, 2011). Statistical comparisons were made using within-subject permutations (nperm = 1000) of data labels (either tone-type or condition depending on the contrast) with significance defined based on a Welch t-statistic greater than 95% of t-statistics in the permutation-based null distribution. The null distribution was constructed by storing the most extreme t-statistic, across all time points, from each permutation. Because the null distribution is based on the entire time window, the results of this test are inherently corrected for multiple-comparisons over time- points (Groppe et al., 2011). Regions of significance, i.e. one or more significant sequential time point, are summarized in the results section using the maximum t-value within that range and its corresponding p-value. In cases where no significant differences are observed, a time range is reported and the maximum t-value/p-value are reported for that range. As an exploratory analysis, the same permutation testing was ap- plied across all 256 electrodes. The resulting p-values were corrected for multiple comparisons across electrodes using the Benjamini-Hochberg false discovery rate procedure (Benjamini and Hochberg, 1995).

2.5. Dynamic causal modeling with Bayesian model selection

The DCM for ERPs framework, as implemented in SPM12 (https://www.fil.ion.ucl.ac.uk/spm), was used to estimate source level effective connectivity between brain regions in response to auditory stimulation. The DCM framework, as applied to EEG, may be conceptualized as a source reconstruction procedure with additional Bayesian priors, based on experimentally derived physiological constants, that constrain the estimation of the inverse model (Kiebel et al., 2008). This model, based on biologically plausible neuronal architecture, allows for inferences to be made about the origins of the observed neural responses as well as the flow of information (effective connectivity) between those sources.

DCMs were fit using the ERP mode within SPM12, which models each source with three interacting subpopulations of neurons: pyramidal cells, spiny stellate cells, and inhibitory interneurons (Kiebel et al., 2008; David et al., 2005). Within this framework, all extrinsic (between source) connections are modeled as excitatory, while the intrinsic (self-connections) are inhibitory. Models were fit to the ERP data between 0 and 400 ms after stimulus onset using 8 empirical modes. Modulatory ef- fects (B-matrix) were modeled using standards as the reference trials and oddballs as the change trials (i.e., standards = 0, oddballs = 1). The cor- tical surface (IMG) method was chosen for the electromagnetic model. The auditory stimulus was modeled as a Gaussian (M = 50, SD = 16) arriving at auditory cortex 50 ms after tone presentation based on SPM12 recommendations for auditory paradigms.

The model space investigated here (Fig. 3 C) was based upon previously published work applying DCM to the auditory roving oddball paradigm (Garrido et al., 2008; Boly, 2011; Garrido et al., 2007; Rosch et al., 2019). This space consisted of 30 models (M1-M30) varying in their anatomical sources and connectivity. The full model (M30) consisted of left A1 (L-A1), right A1 (R-A1), left superior temporal gyrus (L-STG), right superior temporal gyrus (R-STG), left inferior frontal gyrus (L-IFG), and right inferior frontal gyrus (R-IFG) (Fig. 3 A), with feedforward (FF), feedback (FB), lateral (Lat), and intrinsic (Self) connections. All connections included within a model were also allowed to be modulated differently by standard and oddball tones. M1-M29 are all subsets of M30.

Fig. 3.

Fig. 3.

Open in a new tab

Bayesian Model Selection. A) 3D brain models showing the spatial location of each region of interest included in the DCM model space. The MNI coordinates used as source priors, presented as ROI [X, Y, Z], were L-A1 [−42, −22, 7], R-A1 [46, −14, 8], L-STG [−61, −32, 8], R-STG [59, −25, 8], L-IFG [−46, 20, 8], and R-IFG [46, 20, 8]. B) Exceedance probabilities for each model specification indicating how much relative evidence there is for each model. BMS results are shown for the baseline data alone (top row), the disconnected data alone (middle row), and all the data pooled together (bottom row). In addition to the individual models (left column), we present family-level evidence based on the number of ROIs (middle column), and the types of connections included in the models (right column). C) Model specifications for all 30 models included in the BMS model space.

The 30 model specifications were compared using random-effects Bayesian model selection (BMS). Each subject’s baseline and disconnected consciousness data were modeled separately. Individual models and model families based on number of modeled regions (2, 4, 5, or 6) and connection types (None, FF, FF/FB, or FF/FB/Self) were compared. This procedure was applied to the baseline and disconnected data separately as well as pooled together.

2.6. Parametric Empirical Bayes (PEB)

Because M30 outperformed all other models regardless of how the data were subdivided, this model was selected for further analysis using PEB (Friston et al., 2016). A within-subjects hierarchical approach (Zeidman et al., 2019) was applied (Fig. 1 C); at the first level, each subject’s disconnected consciousness and baseline DCMs were contrasted using PEB; at the second level, each subject’s PEB model was used as input to a group level PEB (a PEB of PEBs) which tested for consistent group level effects. Both the ‘A’ and ‘B’ matrices were included in the PEB analysis, representing the effective connectivity observed in response to standards and the modulation of that connectivity by oddballs respectively. The contrast was specified such that positive values indicate parameters that were greater in the disconnected condition than in baseline, while negative values indicate the reverse. Parameters with a posterior probability of being non-zero of > 95% were considered to have sufficient evidence for interpretation.

2.7. DCM simulations

In order to test the relationship between the PEB and scalp ERP findings, we simulated scalp level ERPs of standard tone responses using DCMs with varying connectivity strengths. First, an average DCM (M30 architecture) was generated by averaging all data across subjects. PEB coefficients for the connection with > 95% confidence of differing between disconnected and connected consciousness (Fig. 4 A–B, yellow and purple connections) were then extracted for use in modifying the DCM connectivity strength. DCM connectivity strength was adjusted by adding multiples, from 1 to 4 by intervals of 0.1, of the PEB coefficients to the average parameter values and simulating scalp ERPs at electrode Cpz. More formally:

di=D+ai(P∘C)ai=1,1.1,1.2,…,4 (Eq. 1)

where, di is the new set of connectivity strengths, D is the set of connectivity strengths (A-matrix) from the average DCM, P is the set of PEB coefficients, C is a vector where Ck = 1 if P k has > 95% of differing between disconnected and connected consciousness and is 0 otherwise, and ai is the current PEB scaling factor. The ∘ operator indicates the elementwise multiplication of P and C, such that only the coefficients with > 95% confidence of difference will have non-zero values added to D. ERPs were simulated using SPM’s ‘spm_dcm_simulate’ function. In essence, the DCM was iteratively pushed the direction of disconnected consciousness using PEB derived parameter scaling to observe the impact on simulated ERP waveforms. The choice of 1 to 4 was largely arbitrary and simply served as a way to find a range where all connections demonstrated linear scaling activity, making them easier to compare.

Fig. 4.

Fig. 4.

Open in a new tab

PEB Changes Associated with Sensory Disconnection. A) Final PEB model results (corresponding to Fig. 1 C top level) for the A-matrix, modeling the difference in response to standard tones in disconnection compared to baseline. Negative values indicate a greater response in baseline while positive values indicate greater response with disconnection. Parameters estimated as different from 0 with >95% confidence are color filled yellow if negative or purple if positive. B) Model architecture of M30 with connections colored to match the A-matrix PEB results seen in A. As before, purple arrows indicated connections greater in disconnection compared to baseline (>95% confidence), while yellow arrows indicate connections greater in baseline compared to disconnection (>95% confidence). C) Same as A but illustrating results from the B-matrix, representing the difference in oddball vs standard effect in disconnection compared to baseline. D) Same as B but results correspond to the B-matrix. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article).

To test the impact of each connection on ERP morphology, one parameter was fixed at its average value while the other 8 were scaled by their PEB coefficients. The ERPs for each fixed parameter were compared to the ERPs generated when all 9 parameters were scaled to determine the impact of that parameter on P3 scaling. P3 amplitude was quantified by tracking the mean amplitude between 300 and 400 ms for each scaling factor. Due to the observation of consistent linear scaling between scaling factors of 1–3 (Fig. 5 A, Supplementary Fig. 3), P3 scaling was quantified as the slope of the P3 amplitude over scaling factor (from 1 to 3), calculated using linear regression. Lastly connections that showed an effect on simulated P3 scaling when they were fixed were evaluated in combination to determine which parameters were necessary or sufficient to produce the same degree of scaling observed when all 9 parameters were modulated.

Fig. 5.

Fig. 5.

Open in a new tab

Impact of DCM Connectivity Strength on P3 Scaling. A) (Left) Simulated ERPs at electrode Cpz generated from an averaged DCM model with connectivity strengths modulated based on PEB coefficients for the 9 parameters with >95% confidence of differing between baseline and disconnection. (Right) P3 amplitude (averaged between 300 and 400 ms) vs the scaling factor the PEB coefficients were multiplied by. Black dashed line shows the regression line calculated from scaling factors between 1 and 3. The slope of this line is used as the Reference P3 Scaling. B) P3 scaling observed when 8 parameters are scaled and 1 is fixed at the average. Data are normalized to the reference (shown in A). Values less than 1 indicate a reduction in P3 scaling when that parameter is not scaled. Bars color coded by the type of connection: feedforward (red), feedback (green-blue), lateral (purple). C) P3 scaling, relative to reference, when only the x-axis parameters are scaled by their PEB coefficients. All other coefficients are fixed at their averages. Values less than 1 indicate a given parameter combination is insufficient for producing the full P3 scaling effect. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article).

3. Results

At baseline, 31 instances of connected consciousness (24.6% of all reports) were reported during the roving oddball paradigm, including reports by all 18 subjects. All baseline reports were of connected consciousness. 43 instances of disconnected consciousness (34.1% of all reports) were reported after sedation began, including reports from 17/18 subjects (i.e. one subject did not report any instances of disconnected consciousness during the sedation). Reports of unconsciousness (no experience) and reports that could not be confidently labeled, due to internally conflicting responses or loss of responsiveness, were not included in the analyses.

At baseline, oddball tones elicit pronounced N1 (~100 ms), N2 (~175 ms), and P3 (~250 ms) waveforms (Fig. 2 A top, B top-left). In contrast, standard tones only reproducibly evoke an N1 component which is smaller in magnitude than that observed from oddball tones. The increased negative amplitude between 100–200 ms and positive amplitude between 200–300 ms observed in oddballs relative to standards produces a stereotyped mismatch negativity and P3 in the difference waveform respectively (Fig. 2 B top-left black trace), consistent with prior literature (Garrido et al., 2007; Boly, 2011). Both the 100–200 ms (N2) and 200–300 ms (P3) periods show significant differences between oddballs and standards in the baseline condition (N2 oddballs vs standards: tmax = 4.54, p = 0.001, P3 oddballs vs standards tmax = 6.32, p < 0.001).

Fig. 2.

Fig. 2.

Open in a new tab

Scalp Event Related Potentials of Standards and Oddballs. A) Topographical view of scalp potentials over time (0–400ms) in response to standard and oddball tones in baseline and disconnected data. Plots represent grand averages across all subjects. B) ERPs over time at electrode Cpz (inset shows the location of this electrode on the scalp). Subplots show response to standards and oddballs at baseline (top-left) and during disconnection (top-right) as well as comparisons of baseline and disconnection response to standards (bottom-left) and oddballs (bottom-right). Blue highlighted regions indicate times where significant differences (p < 0.05) were observed between the contrasted conditions or tones based on permutation tests. The statistical contrasts were: top-left, baseline oddball vs baseline standard; top-right, disconnected oddball vs disconnected standard; bottom-left, disconnected standard vs baseline standard; bottom-right, disconnected oddball vs baseline oddball. N1, N2, and P3 components are labeled in the baseline condition. Note that “baseline ” always refers to the pre-sedation data, not the pre-stimulus period of the ERP. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article).

In contrast to the baseline ERPs, the disconnected consciousness data showed smaller amplitude responses to oddball tones (Fig. 2 A bottom, B top-right). These oddball amplitudes were significantly smaller in disconnection compared to baseline during both the N2 (tmax = 3.39, p = 0.036) and P3 (tmax = 3.68, p = 0.020) periods (Fig. 2 B bottom-right). Standard tones also showed a visual reduction in amplitude in disconnection compared to baseline at N1, though this was non-significant (tmax = 2.29, p = 0.624, 75–125 ms), and a significantly increased amplitude for the P3 (tmax = 3.55, p = 0.038) component (Fig. 2 B bottom-left). Notably, the addition of a P3 in response to standards during disconnection produced an ERP that resembled the oddball response during connected consciousness. In addition, oddballs and standards during periods of disconnected consciousness showed no significant differences (tmax = 1.38, p = 1.00, 0–400 ms) (Fig. 2 B top-right). The loss of significance from baseline to disconnection cannot be explained away as a limitation of statistical power as we collected more disconnection data than baseline data and thus have greater power to detect a difference in this condition, making the absence of difference a conspicuous finding. Exploratory analyses evaluating the spatial distribution of these effects can be found in Supplementary Fig. 1. These results demonstrate that the observed differences between oddballs and standards during baseline generalize across a wide range of centrally oriented electrodes (not just Cpz). Similarly, the observed lack of difference between standards and oddballs in disconnection also generalized, with no significant differences being found at any electrode.

To investigate the neural mechanisms underpinning the observed alterations in ERPs, we employed DCM to explore changes in effective connectivity (EC) between connected and disconnected consciousness. To establish the optimum model, we applied Bayesian model selection (BMS). BMS identified the fully connected model (M30) as having the greatest model evidence in both the baseline and disconnected data separately, as well as when they were pooled together (Fig. 3 B, left column). Similarly, all subdivisions of data showed that models with all 6 regions of interest (ROIs) outperform model families with fewer regions, despite being more complex and thus more heavily penalized (Fig. 3 B, middle column). Finally, models containing FF, FB, and intrinsic self-modulation outperformed those with no connections, only FF connections, only FB connections, and FF and FB connections (without self-modulation), in all subdivisions (Fig. 3 B right column).

Due to M30 outperforming all other models by a large margin, we selected this model for further analysis of individual parameters using PEB. Interestingly, all FB connections showed increased connectivity in response to standard tones in disconnection relative to baseline (Fig. 4A–B). We also observed two connections with a higher response to standard tones in the baseline condition: a FF connection between L-STG and L-IFG and a lateral connection from R-STG to L-STG. The modulation of connection strength by oddball tones showed a predominance of connections with greater connectivity during sensory disconnection, including all but one of the FF connections but none of the feedback connections (Fig. 4 C–D). Again, two connections showed higher oddball modulation in the baseline condition: a FB connection from R-IFG to R-STG and a lateral connection from L-STG to R-STG.

In order to understand how these PEB results relate to the prior scalp ERP findings, namely the emergence of a P3 in response to standard tones during disconnection, we simulated scalp ERPs for standard tones, at Cpz, using an average DCM (averaged across subjects). When DCM connectivity strengths were modulated by adding multiples of the PEB coefficients (Fig. 4A–B, only yellow and purple connections), we observed increased amplitudes in the P3 range (300–400 ms) (Fig. 5 A left). This scaling of simulated ERPs as a function of PEB scaling factor was used as the reference for the subsequent simulation analyses.

To determine which model connections contributed to the observed P3 emergence, we repeated the same PEB scaling simulations with a different parameter being fixed at its average value each time. Changes in P3 morphology (shape and amplitude) were observed when R-A1 to R-STG, R-IFG to R-STG, and R-STG to R-A1 were fixed at their average values (Supplementary Fig. 2). The most notable changes being observed when R-STG to A1 was fixed and P3 scaling was greatly reduced. These changes in P3 scaling were quantified by examining changes in the P3 scaling slope (Supplementary Fig. 3) relative to the reference (Fig. 5 A right). When fixed to their average values, we observed reductions in P3 scaling, relative to reference, of 22.6%, 32.9%, and 77.7% for R-A1 to R-STG, R-IFG to R-STG, and R-STG to R-A1, respectively (Fig. 5 B). Lastly, we scaled the DCM connectivity strengths using all combinations of these three connections, but fixing all others to their average, to determine which connection(s) would be necessary or sufficient to produce the reference P3 scaling (Supplementary Figs. 4 and 5). Only when all three connections were modulated did we observe the full scaling effect, 102% of reference (Fig. 5 C). Notably scaling only R-STG to R-A1 was sufficient to produce 55.3% of the reference scaling, while scaling R-A1 to R-STG and R-IFG to R-STG together (without R-STG to R-A1) only produced 22.1% of reference scaling.

4. Discussion

The aim of this work was to assess perturbations in predictive processing within auditory pathways that may explain how the conscious brain handles sensory input when connected to the environment vs disconnected from the environment. Our results demonstrate that disconnected consciousness is associated with altered responses to both expected (standard tones) and unexpected (oddball tones) stimuli and that these differences may be explained by changes in connectivity within auditory networks. This study advances our understanding of the mechanisms the brain may use to achieve a state of disconnection from the environment.

Our scalp level ERP data show that disconnected consciousness is associated with a loss of discriminability between standard and oddball tones. Notably, the response to standards during disconnection looks remarkably similar to the oddball response. This similarity arises from the addition of a P3 component in response to standard tones (Fig. 2 B bottom-left) and a loss of N2 amplitude in response to oddball tones (Fig. 2 B bottom-right). This finding is quite striking as the P3 is typically associated with stimulus perception and involuntary attention recruitment (Fitzgerald and Todd, 2020; Friedman et al., 2001; Patel and Azzam, 2005). This interpretation of the P3 does not align with the current sensory disconnection data where the subjects, by definition, did not consciously perceive or attend to the tones. It should also be noted that the P3 response has been shown to correlate with how unexpected the stimulus is (Ravden and Polich, 1998; Romero and Polich, 1996). Viewing the P3 as an index of expectation, the finding of a P3 in response to standard tones could be interpreted as a loss of neural habituation in response to repeated stimuli. In essence, the similar neural response to standards and oddballs suggests that in a state of disconnected consciousness, every stimulus is surprising due to a mismatch in internally generated predictions and externally generated stimuli.

It has been proposed that the auditory N2 consists of overlapping, but distinct, ERP components, namely the N2a and N2b (Fitzgerald and Todd, 2020; Patel and Azzam, 2005). Both components are triggered by deviant stimuli but the N2a does not require conscious perception of deviance while the N2b does (Fitzgerald and Todd, 2020; Sams et al., 1985). The observed loss of N2 amplitude during disconnection is thus consistent with loss of the N2b. We have previously hypothesized that higher order thalamus (e.g. the pulvinar) plays a key role in mediating sensory disconnection via control of precision weighting of error signals (Sanders et al., 2021). We suggest that reduced precision weighting, and thus reduced cortical gain, in response to sensory signals could explain the loss of N2b. Lower precision weighting of sensory signals is also consistent with the view that standards and oddballs become similarly surprising. Low precision information should not be given undue weight when updating one’s model predictions about the world as it is not very trustworthy. Loss of model updating would result in a loss of habituation to repeated tones, thus even tones that have been heard many times before would remain surprising. However, the stage of information processing within the cortical hierarchy at which perturbations (potentially through precision weighting) are impacting the ERPs is not clear based on scalp level analysis alone.

Our use of DCM begins to address the question of where in the processing hierarchy information transfer is being perturbed. The BMS results show that both sensory connected and disconnected states are best represented by a 6-ROI model containing FF, FB, lateral, and self-connections. This implies that, though signaling within the network may be altered, there is not a complete breakdown of its fundamental functional architecture during disconnection. The family-level analysis further highlights that no class of connection types (FF, FB, or Self) is systematically lost in disconnected consciousness. Though there is a reduced exceedance probability for the FF/FB/Self family relative to FF/FB in disconnection compared to baseline, caution should be taken in interpreting this result as model evidence is only comparable when the models have been fitted to the same data (Stephan et al., 2010).

While BMS informs us about the general model architecture, PEB gives us deeper insight into the ‘between-conditions’ differences within an established model. Rather than observing a loss of EC, as one might expect for a state defined by loss of sensory connectedness, we observe a predominance of connections with increased parameter estimates in the disconnected consciousness data. In particular, we observed increased EC in all FB connections in disconnection relative to baseline (Fig. 4 A–B). This finding is of particular note in the context of the previously mentioned P3 response to standard tones. Previous modeling work has demonstrated that the late components of the oddball ERP are dependent on FB connections (Garrido et al., 2007). Therefore, the increased FB connectivity appears a likely explanation for the emergence of a standard P3 during disconnected consciousness. The evaluation of this possibility through ERP simulations is discussed further below.

The modulation of connectivity by oddball tones shows perturbation across a bilateral frontotemporal network for which R-STG appears to be a key hub (Fig. 4 D). We observed increased connectivity from R-STG to R-IFG both directly and indirectly (R-STG to L-STG, L-STG to L-IFG, and L-IFG to R-IFG) during disconnected consciousness. At the same time, we observed reductions in FB connectivity from R-IFG to R-STG and lateral input from L-STG to R-STG. Together these changes point to a shift in the balance of information flow at the level of the R-STG towards more outflow and less inflow. The exact implications of this altered frontotemporal network during disconnected consciousness are not obvious but merit further investigation as it may represent a physiological substrate of sensory disconnection.

The above PEB findings seemed a likely explanation for the observed P3 response to standard tones during disconnection, however, based on these two separate analyses, it is unclear which changed connection(s) may contribute to the effect. Our simulation analyses link our scalp ERP and PEB findings by demonstrating that a subset of connections con- tribute to P3 amplitude while the majority do not. Namely, FF connectivity from R-A1 to R-STG, and FB connectivity from R-STG to R-A1 and from R-IFG to R-STG play important roles in the increased P3 amplitude observed when DCM EC strengths were scaled by their disconnection PEB parameters (Fig. 5 B). Furthermore, these three connections together are sufficient to produce comparable P3 scaling as a model where all 9 connections are modulated (Fig. 5 C). Conversely, leaving out any one of these three parameters results in reduced P3 scaling compared to the reference. These three connections do not contribute equally, however. FB from R-STG to R-A1 was sufficient to produce 55.3% of the reference scaling, but R-A1 to R-STG and R-IFG to R-STG together could only produce 22.1% of reference scaling. This suggests that while FB from R-STG to R-A1 is the primary driver of P3 emergence, R-A1 to R-STG and R-IFG to R-STG play synergistic roles, resulting in a P3 larger than the sum of their parts.

We previously proposed a model in which sensory disconnection is defined by a reduction in FF connectivity with maintenance of FB connectivity (Sanders et al., 2021). The relative excess of FB over FF signaling, we argued, would explain the phenomenological state of dreaming (disconnected consciousness), where conscious experience is internally driven and not related to the external world. While we hypothesized that sensory disconnection would be associated with a reduction of FF connectivity, our data indicated that this was generally not the case. In fact, only one connection showed lower FF connectivity in disconnection compared to baseline: the standard response from R-STG to R-IFG. Rather, our findings suggest that sensory disconnection is associated with an increase in FB connectivity in response to standard stimuli. Interestingly, the end result of increased FB connectivity, with preserved FF connectivity, could be qualitatively similar to a loss of FF connectivity with preserved FB connectivity. In both cases, a state where FB signaling dominates is achieved. We suggest that new information is still being passed forward to the cortex (through FF connections), however, the brain’s internally generated experience (communicated through FB connections) predominates through a relative enhancement of FB precision compared to FF precision; preserving the integrity of the endogenous state. Hence, each standard tone is a surprising event as it is mis- matched to the internal model of the world (the dream), which we argue is supported by the finding of a P3 in response to standard tones. This interpretation also appears consistent with the recently proposed ‘apical drive’ hypothesis of dreaming (Aru et al., 2020). This hypothesis states that dreaming (disconnected consciousness) occurs when input to the apical dendrites, carrying feedback signals, of layer-5 pyramidal neurons are able to directly drive spiking activity at the cell body. This in contrast to normal cellular behavior found during wake (connected consciousness), where apical inputs only selectively amplify the neuronal response to somatic inputs carrying information about the environment. The observed increased FB connectivity may represent a state of apical drive. Whether all brain regions respond in a homogenous manner how-ever remains to be determined.

Physiologically, dexmedetomidine acts as an α2-adrenergic receptor agonist which inhibits noradrenaline release from the locus coeruleus (LC), producing a state similar to physiological sleep (Huupponen et al., 2008; Purdon et al., 2015). LC activity can be tonic or phasic, with tonic activity tracking general level of arousal (and causally impacting sensory evoked awakenings (Hayat et al., 2020)) and phasic activity occurring in response to unexpected stimuli (Hayat et al., 2020; Rajkowski et al., 1994; Aston-Jones and Bloom, 1981). Indeed, phasic activity in the LC has been suggested to code for prediction errors (Ferreira-Santos, 2016) and such phasic activity can still be elicited during sleep, when tonic activity is low (Hayat, Regev, and Matosevich, 2020). Multiple models have been proposed in which phasic LC activity plays a key role in facilitating internal prediction updates in response to new information (Aston-Jones and Cohen, 2005; Bouret and Sara, 2005; Sales et al., 2019). Finally, phasic LC activity has been previously linked to P3 generation (Nieuwenhuis et al., 2005). Together, these findings are consistent with a model of sensory disconnection as a state of low tonic but evocable phasic LC activity. This phasic activity, however, is no longer coupled to successful prediction updates. This hypothesis could be further tested in animal models, sedated with an α2-adrenergic receptor agonist, by recording LC activity in response to an oddball paradigm. If this model is correct, we would expect to find similar phasic LC activity in response to standards and oddballs in the sedated animals.

We note that while we interpret the disconnected standard P3 as evidence of the stimulus becoming surprising and that the increase in FB connectivity during disconnection could explain why the stimulus is surprising, not all FB connections appear to contribute to the P3. FB from R-IFG to R-STG and from R-STG to R-A1 were critical for P3 emergence in our simulation analyses, but the left hemisphere connections were not. FF from R-A1 to R-STG also played a role, though this connection seems to have the smallest contribution of the three. The observation that only right hemispheric connections contribute to P3 scaling is not necessarily a surprising one, as many prior studies have demonstrated right lateralized dominance in auditory processing (Zatorre et al., 2002; Boemio et al., 2005; Abrams et al., 2008; Abrams et al., 2009; Telkemeyer et al., 2009; Doelling et al., 2014). This appears to be especially common for tasks that don’t require much language processing demand or that do require precise frequency monitoring (Zatorre et al., 2002; Assaneo et al., 2019), both of which are qualities of the roving oddball paradigm. As such, it seems reasonable that P3 generation under this paradigm would also be right lateralized.

As with all research, the current study has several limitations. One major limitation in this field is the lack of accurate objective markers of sensory awareness vs consciousness. Because of this, we relied on subjective reports from each subject as the ground truth of their experience. While subjective reports have limitations due to the possibility of human error in assessing one’s own mental state, using such methods are currently the only way to separate out disconnected consciousness and unconsciousness, which is critical for this research. In a related vein, we are limited by our inability to experimentally assign subjects to a disconnected consciousness condition. Using pharmacological intervention, we can bias subjects to states of either disconnected consciousness or unconsciousness, but we cannot directly control which state they will enter. Because of this, different subjects will report being in a state of disconnected consciousness more or less frequently than others, resulting in unequal amounts of disconnection data for each subject. Fortunately, the Bayesian framework that DCM is built upon mitigates the consequences of this issue. Each subject’s DCM contains not only the prior estimate of each parameter, but also the full posterior distribution. This allows group level analyses, like PEB, to weight each model based on the confidence of parameter estimates when estimating the whole group effect. We also acknowledge that while we propose thalamic controlled precision weighting as a plausible mechanism to explain our findings, imaging modalities with better spatial accuracy, such as fMRI, would be better suited to testing this hypothesis as EEG is known to have poor accuracy and sensitivity when estimating source activity in deep subcortical structures (Piastra et al., 2021; Attal and Schwartz, 2013; Hillebrand and Barnes, 2002).

In the current work, we experimentally isolated sensory connect-edness by controlling for conscious experience in our contrasting conditions (participants were conscious in both groups). By capitalizing on recent advances in the generative modeling technique of DCM, we demonstrate that sensory disconnection is associated with an increase of FB connectivity within the auditory processing network in response to standard (predictable) tones, while oddball (unpredictable) tones elicit altered information flow within a frontotemporal subnetwork consisting of bilateral STG and IFG. We also demonstrate that a subset of connections showing altered response to standard tones during disconnection can explain the emergence of a P3 (surprise) response observed in the scalp ERP data. We interpret these DCM results together with our ERP findings as evidence that disconnected consciousness is associated with a breakdown in predictive coding processes, resulting in all incoming stimuli being ‘surprising’ from a neuronal perspective. These findings provide useful targets for future study on the mechanisms of sensory disconnection and may develop into clinically relevant biomarkers of awareness for use in anesthetic monitoring.

Supplementary Material

Supplementary Information

Acknowledgments

We are grateful to advice from Prof Giulio Tononi, Dr. Brady Reidner, Dr David Plante and Dr. Melanie Boly (University of Wisconsin, USA) when setting up this project and for loan of the EEG equipment.

Funding

This work was supported by the Department of Anesthesiology at the University of Wisconsin and by NIH NIND 1R01NS117901–01.

Footnotes

Declaration of Competing Interest

The authors have no competing interests to declare.

Credit authorship contribution statement

Cameron P. Casey: Writing – original draft, Writing – review & editing, Conceptualization, Formal analysis, Investigation, Visualization, Data curation. Sean Tanabe: Investigation. Zahra Farahbakhsh: Investigation. Margaret Parker: Investigation. Amber Bo: Investigation. Marissa White: Investigation. Tyler Ballweg: Investigation. Andrew Mcintosh: Investigation. William Filbey: Investigation. Matthew I. Banks: Supervision, Writing – review & editing. Yuri B. Saalmann: Supervision, Writing – review & editing. Robert A. Pearce: Project administration, Supervision, Writing – review & editing. Robert D. Sanders: Conceptualization, Funding acquisition, Project administration, Supervision, Investigation, Writing – review & editing, Data curation.

Supplementary materials

Supplementary material associated with this article can be found, in the online version, at doi: 10.1016/j.neuroimage.2022.119657.

Data and code availability

Data may be made available via a request to the authors and pending IRB review. A data sharing agreement may be required before access is granted. The software used in this work has been cited within the main text.

Data availability

Data will be made available on request.

References

  1. Abrams DA, Nicol T, Zecker S, Kraus N, 2008. Right-hemisphere auditory cortex is dominant for coding syllable patterns in speech. In: J Neurosci, 28. Society for Neuroscience, pp. 3958–3965. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Abrams DA, Nicol T, Zecker S, Kraus N, 2009. Abnormal cortical processing of the syllable rate of speech in poor readers. In: J Neurosci, 29. Society for Neuroscience, pp. 7686–7693. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Aru J, Siclari F, Phillips WA, Storm JF., 2020. Apical drive —a cellular mechanism of dreaming?. In: Neurosci Biobehav Rev, 119. Elsevier Ltd, pp. 440–455. [DOI] [PubMed] [Google Scholar]
  4. Assaneo MF, Rimmele JM, Orpella J, Ripollés P, de Diego-Balaguer R, Poeppel D, 2019. The lateralization of speech-brain coupling is differentially modulated by intrinsic auditory and top-down mechanisms. In: Front Integr Neurosci, 13. Frontiers Media S.A., p. 28. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Aston-Jones G, Bloom FE., 1981. Activity of norepinephrine-containing locus coeruleus neurons in behaving rats anticipates fluctuations in the sleep-waking cycle. J. Neurosci. Soc. Neurosci 1, 876–886. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Aston-Jones G, Cohen JD., 2005. An integrative theory of locus coeruleus-norepinephrine function: adaptive gain and optimal performance. Annu. Rev. Neurosci 28, 403–450. [DOI] [PubMed] [Google Scholar]
  7. Attal Y, Schwartz D, 2013. Assessment of subcortical source localization using deep brain activity imaging model with minimum norm operators: a MEG study. In: PLoS One, 8. Public Library of Science, p. e59856. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Bell AJ, Sejnowski TJ., 1995. An information-maximization approach to blind separation and blind deconvolution. Neural Comput. 7, 1129–1159. [DOI] [PubMed] [Google Scholar]
  9. Benjamini Y, Hochberg Y, 1995. Controlling the false discovery rate: a practical and powerful approach to multiple testing. In: J R Stat Soc Ser B, 57. John Wiley & Sons, Ltd, pp. 289–300. [Google Scholar]
  10. Blain-Moraes S, Tarnal V, Vanini G, et al. , 2015. Neurophysiological correlates of sevoflurane-induced unconsciousness. In: Anesthesiology, 122. NIH Public Access, p. 307. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Boemio A, Fromm S, Braun A, Poeppel D, 2005. Hierarchical and asymmetric temporal sensitivity in human auditory cortices. In: Nat Neurosci 2005 83, 8. Nature Publishing Group, pp. 389–395. [DOI] [PubMed] [Google Scholar]
  12. Boly M, 2011. Preserved feedforward but impaired top-down processes in the vegetative state. Science 46, 1–25. [DOI] [PubMed] [Google Scholar]
  13. Bouret S, Sara SJ., 2005. Network reset: a simplified overarching theory of locus coeruleus noradrenaline function. In: Trends Neurosci, 28. Elsevier Current Trends, pp. 574–582. [DOI] [PubMed] [Google Scholar]
  14. Casey CP, Tanabe S, Farahbakhsh Z, et al. , 2022. Distinct EEG signatures differentiate unconsciousness and disconnection during anaesthesia and sleep. British Journal of Anaesthesia. Elsevier. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. David O, Harrison L, Friston KJ., 2005. Modelling event-related responses in the brain. In: Neuroimage, 25. Neuroimage, pp. 756–770. [DOI] [PubMed] [Google Scholar]
  16. Delorme A, Makeig S, 2004. EEGLAB: An open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. In: J Neurosci Methods, 134. Elsevier, pp. 9–21. [DOI] [PubMed] [Google Scholar]
  17. Doelling KB, Arnal LH, Ghitza O, Poeppel D, 2014. Acoustic landmarks drive delta—theta oscillations to enable speech comprehension by facilitating perceptual parsing. In: Neuroimage, 85. Academic Press, pp. 761–768. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Ferreira-Santos F, 2016. The role of arousal in predictive coding. In: Behav Brain Sci, 39. Cambridge University Press, p. e207. [DOI] [PubMed] [Google Scholar]
  19. Fitzgerald K, Todd J, 2020. Making sense of mismatch negativity. In: Front Psychiatry, 11. Frontiers Media S.A., p. 468. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Friedman D, Cycowicz YM, Gaeta H, 2001. The novelty P3: an event-related brain potential (ERP) sign of the brain’s evaluation of novelty. In: Neurosci Biobehav Rev, 25. Pergamon, pp. 355–373. [DOI] [PubMed] [Google Scholar]
  21. Friston KJ, Harrison L, Penny W, 2003. Dynamic causal modelling. In: Neuroimage, 19. Academic Press Inc., pp. 1273–1302. [DOI] [PubMed] [Google Scholar]
  22. Friston KJ, Litvak V, Oswal A, et al. , 2016. Bayesian model reduction and empirical Bayes for group (DCM) studies. In: Neuroimage, 128. Academic Press, pp. 413–431. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Garrido MI, Friston KJ, Kiebel SJ, Stephan KE, Baldeweg T, Kilner JM., 2008. The functional anatomy of the MMN: a DCM study of the roving paradigm. Neuroimage 42, 936–944. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Garrido MI, Kilner JM, Kiebel SJ, Friston KJ., 2007. Evoked brain responses are generated by feedback loops. Proc. Natl. Acad. Sci. USA 104, 20961–20966. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Garrido MI, Kilner JM, Kiebel SJ, Stephan KE, Friston KJ., 2007. Dynamic causal modelling of evoked potentials: a reproducibility study. Neuroimage 36, 571–580. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Groppe DM, Urbach TP, Kutas M, 2011. Mass univariate analysis of event-related brain potentials/fields I: a critical tutorial review. In: Psychophysiology, 48. John Wiley & Sons, Ltd, pp. 1711–1725 10.1111. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Hayat H, Regev N, Matosevich N, et al. , 2020. Locus coeruleus norepinephrine activity mediates sensory-evoked awakenings from sleep. In: Science Advances. American Association for the Advancement of Science, p. 6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Hillebrand A, Barnes GR., 2002. A quantitative assessment of the sensitivity of whole– head meg to activity in the adult human cortex. In: Neuroimage, 16. Academic Press, pp. 638–650. [DOI] [PubMed] [Google Scholar]
  29. Huupponen E, Maksimow A, Lapinlampi P, et al. , 2008. Electroencephalogram spindle activity during dexmedetomidine sedation and physiological sleep. In: Acta Anaesthesiol Scand, 52. John Wiley & Sons, Ltd, pp. 289–294. [DOI] [PubMed] [Google Scholar]
  30. Kiebel SJ, Garrido MI, Moran RJ, Friston KJ., 2008. Dynamic causal modelling for EEG and MEG. In: Cogn Neurodyn, 2. Springer, pp. 121–136. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Ku S-W, Lee U, Noh G-J, Jun I-G, Mashour GA. Preferential inhibition of frontal-to-parietal feedback connectivity is a neurophysiologic correlate of general anesthesia in surgical patients. Ward LM, editor. PLoS One Public Library of Science; 2011; 6: e25155. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Lee U, Ku S, Noh G, Baek S, Choi B, Mashour GA., 2013. Disruption of frontal-parietal communication by ketamine, propofol, and sevoflurane. In: Anesthesiology, 118. The American Society of Anesthesiologists, pp. 1264–1275. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Lennertz R, Pryor KO, Raz A, et al. , 2022. Connected consciousness after tracheal intubation in young adults: an international multicentre cohort study. British Journal of Anaesthesia. Elsevier. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Meara JG, Leather AJM, Hagander L, et al. , 2015. Global Surgery 2030: Evidence and solutions for achieving health, welfare, and economic development. In: Lancet. Lancet Publishing Group, pp. 569–624. [DOI] [PubMed] [Google Scholar]
  35. Nieuwenhuis S, Aston-Jones G, Cohen JD., 2005. Decision making, the P3, and the locus coeruleus-norepinephrine system. Psychol. Bull 131, 510–532. [DOI] [PubMed] [Google Scholar]
  36. Noreika V, Jylhänkangas L, Móró L, et al. , 2011. Consciousness lost and found: subjective experiences in an unresponsive state. In: Brain Cogn, 77. Elsevier Inc., pp. 327–334. [DOI] [PubMed] [Google Scholar]
  37. Patel SH, Azzam PN., 2005. Characterization of N200 and P300: selected studies of the event-related potential. In: Int J Med Sci, 2. Ivyspring International Publisher, p. 147. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Piastra MC, Nüßing A, Vorwerk J, Clerc M, Engwer C, Wolters CH., 2021. A comprehensive study on electroencephalography and magnetoencephalography sensitivity to cortical and subcortical sources. In: Hum Brain Mapp, 42. Wiley-Blackwell, p. 978. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Purdon PL, Pierce ET, Mukamel EA, et al. , 2013. Electroencephalogram signatures of loss and recovery of consciousness from propofol. In: Proc Natl Acad Sci, 110. National Academy of Sciences, pp. E1142–E1151. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Purdon PL, Sampson A, Pavone KJ, Brown EN., 2015. Clinical electroencephalography for anesthesiologists. In: Anesthesiology. Lippincott Williams and Wilkins, pp. 937–960. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Rajkowski J, Kubiak P, Aston-Jones G, 1994. Locus coeruleus activity in monkey: phasic and tonic changes are associated with altered vigilance. In: Brain Res Bull, 35. Elsevier, pp. 607–616. [DOI] [PubMed] [Google Scholar]
  42. Ravden D, Polich J, 1998. Habituation of P300 from visual stimuli. In: Int J Psychophysiol, 30. Elsevier, pp. 359–365. [DOI] [PubMed] [Google Scholar]
  43. Romero R, Polich J, 1996. P3(00) habituation from auditory and visual stimuli. In: Physiol Behav, 59. Elsevier, pp. 517–522. [DOI] [PubMed] [Google Scholar]
  44. Rosch RE, Auksztulewicz R, Leung PD, Friston KJ, Baldeweg T, 2019. Selective pre- frontal disinhibition in a roving auditory oddball paradigm under N-methyl-D-aspartate receptor blockade. In: Biol Psychiatry Cogn Neurosci Neuroimaging, 4. Elsevier, p. 140. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Sales AC, Friston KJ, Jones MW, Pickering AE, Moran RJ., 2019. Locus Coeruleus tracking of prediction errors optimises cognitive flexibility: an Active Inference model. PLoS Comput Biol, 15. Public Library of Science. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Sams M, Paavilainen P, Alho K, Näätänen R, 1985. Auditory frequency discrimination and event-related potentials. In: Electroencephalogr Clin Neurophysiol Potentials Sect, 62. Elsevier, pp. 437–448. [DOI] [PubMed] [Google Scholar]
  47. Sanders RD, Casey C, Saalmann YB., 2021. Predictive coding as a model of sensory disconnection: relevance to anaesthetic mechanisms. In: Br J Anaesth, 126. Elsevier Ltd, pp. 37–40. [DOI] [PubMed] [Google Scholar]
  48. Sanders RD, Gaskell A, Raz A, et al. , 2017. Incidence of connected consciousness after tracheal intubation: a prospective, international, multicenter cohort study of the isolated forearm technique. In: Anesthesiology, 126. Lippincott Williams and Wilkins, pp. 214–222. [DOI] [PubMed] [Google Scholar]
  49. Sanders RD, Tononi G, Laureys S, Sleigh JW., 2012. Unresponsiveness ≠ unconsciousness. Anesthesiology 116, 946–959. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Stephan KE, Penny WD, Moran RJ, den Ouden HEM, Daunizeau J, Friston KJ., 2010. Ten simple rules for dynamic causal modeling. In: Neuroimage. Elsevier, pp. 3099–3109. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Telkemeyer S, Rossi S, Koch SP, et al. , 2009. Sensitivity of newborn auditory cortex to the temporal structure of sounds. In: J Neurosci, 29. Society for Neuroscience, pp. 14726–14733. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Weiser TG, Haynes AB, Molina G, et al. , 2015. Estimate of the global volume of surgery in 2012: an assessment supporting improved health outcomes. In: Lancet, 385. Elsevier, p. S11. [DOI] [PubMed] [Google Scholar]
  53. Zatorre RJ, Belin P, Penhune VB., 2002. Structure and function of auditory cortex: music and speech. In: Trends Cogn Sci, 6. Elsevier Current Trends, pp. 37–46. [DOI] [PubMed] [Google Scholar]
  54. Zeidman P, Jafarian A, Seghier ML, et al. , 2019. A guide to group effective connectivity analysis, part 2: Second level analysis with PEB. In: Neuroimage, 200. Academic Press, pp. 12–25. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Information
NIHMS1879500-supplement-Supplementary_Information.docx (2.4MB, docx)

Data Availability Statement

Data may be made available via a request to the authors and pending IRB review. A data sharing agreement may be required before access is granted. The software used in this work has been cited within the main text.

Data will be made available on request.

RESOURCES