Functional diversity of brain networks supports consciousness and verbal intelligence

How are the myriad stimuli arriving at our senses transformed into conscious thought? To address this question, in a series of studies, we asked whether a common mechanism underlies loss of information processing in unconscious states across different conditions, which could shed light on the brain mechanisms of conscious cognition. With a novel approach, we brought together for the first time, data from the same paradigm—a highly engaging auditory-only narrative—in three independent domains: anesthesia-induced unconsciousness, unconsciousness after brain injury, and individual differences in intellectual abilities during conscious cognition. During external stimulation in the unconscious state, the functional differentiation between the auditory and fronto-parietal systems decreased significantly relatively to the conscious state. Conversely, we found that stronger functional differentiation between these systems in response to external stimulation predicted higher intellectual abilities during conscious cognition, in particular higher verbal acuity scores in independent cognitive testing battery. These convergent findings suggest that the responsivity of sensory and higher-order brain systems to external stimulation, especially through the diversification of their functional responses is an essential feature of conscious cognition and verbal intelligence.

Understanding the brain mechanisms of conscious cognition is one of the great frontiers of cognitive neuroscience. A much-researched yet unresolved question is how the myriad sensory inputs arriving at our senses become integrated into meaningful representations that inform cognitive performance and give rise to individual differences in intellectual abilities. In the conscious brain, cognition is thought to arise from iterative interactions among brain regions of graded functional specialization. These include sensory-driven, e.g., auditory and visual, regions on one end of the functional hierarchy, and supramodal regions in frontal and parietal lobes that carry out higher-order cognition, such as executive function, on the other [1][2][3] . However, to fully understand how the interactions of these widespread brain systems give rise to conscious information processing, it is necessary to factor out brain processes that are not intrinsic to consciousness 4 . To this end, functional neuroimaging of individuals rendered unconscious under deep anesthesia or after severe brain injury provides a unique window for demarcating unconscious processes, and conversely, shedding light on brain mechanisms that are essential for conscious information processing and cognition in the healthy brain.
In a series of studies, we asked whether a common mechanism underlies loss of information processing in unconscious states across different conditions, which could shed light on the brain mechanisms of conscious cognition. To address this question we brought together, for the first time, data from the same paradigm-a highly engaging auditory-only narrative-in three independent domains: anesthesia-induced unconsciousness, unconsciousness after brain injury, and individual differences in intellectual abilities during conscious cognition.
Despite a growing number of anesthesia studies, it remains unknown how loss of consciousness affects synthesis of information across sensory and higher-order brain systems. To date, the majority of functional Magnetic Resonance Imaging (fMRI) studies of anesthesia have investigated the brain during a task-and stimulus-free

Results
Information processing under deep anesthesia. To measure information processing during the story, we adopted a previously established method using the same audio story 30 , where we showed that the extent of stimulus-driven cross-subject correlation provided a measure of regional stimulus-driven information processing ( Fig. 1A-C). In the wakeful condition of the anesthesia study, we observed widespread and significant (p < 0.05; FWE cor) cross-subject correlation between healthy participants within sensory-driven (primary and association) auditory cortex, as well as higher-order frontal and parietal regions (Fig. 1D), consistent with Naci et al. 30 . By contrast, in deep anesthesia, the significant (p < 0.05; FWE cor) cross-subject correlation was limited to the auditory cortex, with the exception of two small clusters in left prefrontal and right parietal cortex (Fig. 1E), suggesting that the processing of sensory information was preserved in the sensory, but almost entirely abolished in fronto-parietal regions.
Subsequently, we investigated the impaired brain mechanism underlying loss of information processing in these higher-order regions. Current theories of consciousness [48][49][50] , such as the Integrated Information Theory (IIT), propose that conscious cognition relies on the brain's capacity to efficiently integrate information across different specialized systems 48 , suggesting that both interconnectedness and functional differentiation of brain systems are important for information processing. However, different putative mechanisms are consistent with our results, including reduced/abolished connectivity among distinct brain systems 49 and loss of functional differentiation 48 (i.e., homogeneous connectivity across them). To directly investigate the underlying mechanism, we distinguished four possible impairment patterns consonant with theories of consciousness that could explain impaired information processing in deep anesthesia: 1) a loss of long-range connectivity between auditory and fronto-parietal networks ( Fig. 2A); (2) a loss of connectivity between areas within each network, e.g., between frontal and parietal regions (   The global effect of anesthesia on brain networks' connectivity. Initially, we investigated how deep propofol anesthesia perturbed the patterns of global connectivity. During the audio story, a two-way ANOVA with factors connectivity type (within, between) and state (wakeful, deep anesthesia) showed that connectivity across networks increased significantly (main effect of state: F(16) = 8.57; p = 0.01) (Fig. 3A,B) in deep anesthesia relative to wakefulness. Connectivity between increased more than within networks (interaction effect, state x connectivity type: F(15) = 5.58; p = 0.03) (Fig. 3E), driven by a significant increase in the between network connectivity (t(15) = 3.82, p = 0.002) and no overall change in the within connectivity (see SI for complete results; Figures S3,  S4). By contrast, during the resting state, deep anesthesia showed the opposite effect on between and within network connectivity, with a larger impact on the within relative to between network connectivity (interaction effect, state x connectivity type: F(15) = 5.4; p = 0.03) (Fig. 3C,D). Connectivity within was significantly reduced, but no changes were observed in deep anesthesia in the between network connectivity (see SI for complete results). A direct comparison between the audio story and resting state confirmed that anesthesia affected connectivity in the two conditions in opposite directions. A two-way ANOVA with factors condition (audio story, resting state) and state (wakeful, deep anesthesia) showed a condition x state interaction [F(15) = 7.01; p = 0.02] that was driven by an overall connectivity reduction during the resting state and connectivity increase during the audio story in deep anesthesia. The effects were the same when functional differentiation was measured as the ratio of between-to within-network connectivity.
These suggest that, when the brain is at rest, reduced connectivity within brain networks rather than loss of functional differentiation between them, characterizes the unconscious state. By contrast, when the brain is exposed to complex naturalistic stimuli from the environment, reduced functional differentiation between brain networks leads to loss of information processing in the unconscious state. However, these results must be interpreted with caution, in light of the consistent block order in deep sedation.
The effect of anesthesia on auditory and fronto-parietal networks' connectivity. Next, we asked specifically whether reduced functional differentiation between the auditory and fronto-parietal networks drove the loss of information processing in the fronto-parietal regions during the story. Consistent with effects at the whole-brain level, we found a significant increase in the AUD-DAN and AUD-ECN connectivity [t(15) = 2.6, p = 0.02; t(15) = 4.98, p = 0.0002, respectively] ( Fig. 4A-D), or a significant reduction of the functional differentiation between the AUD and DAN, ECN in deep anesthesia relative to wakefulness. By contrast, in the resting state, connectivity between these networks was not affected by sedation (Fig. 3). These results suggested that reduced functional differentiation between the auditory and fronto-parietal networks leads to loss of external information processing in the unconscious state. Conversely, they suggested that the functional differentiation between the auditory and fronto-parietal networks underlies conscious processing of complex auditory information.
To test specifically whether functional differentiation in the conscious state would be driven by the complex features of the audio story, including its narrative, rather than merely the presence of external stimulation, we compared the AUD-DAN and AUD-ECN pairwise connectivity in wakeful individuals during the audio story with those in the two baseline conditions, a scrambled version of the story that retained the sensory features but was devoid of the narrative, and the resting state. During the scrambled story, functional connectivity between the AUD and DAN, but not ECN, was significantly lower than in the resting state The effect of severe brain injury on the auditory and fronto-parietal networks' connectivity.
Results from both conscious and unconscious conditions in the previous study suggested that functional differentiation between the auditory and fronto-parietal networks underlined conscious processing of complex auditory information. In the next study, we further tested this claim in severe brain-injury, which served as an independent manipulation of consciousness.
The structural profiles and full behavioral description of the convenience sample of brain-injured patients (N = 11) are shown in Figure S1 and Tables S2, S3. Patients who showed willful brain responses in the independent command-following assessment 22,28 were considered covertly aware and labeled DoC+ (N = 6), and those who showed no signs of conscious awareness were labeled as DoC− (N = 5) (Fig. 6), for subsequent analyses. Similarly to conscious individuals (Fig. 5), we expected DoC+ patient group to show a heightened differentiation/down-regulation of the AUD and DAN, ECN pairwise connectivity during the audio story relative to resting state baseline connectivity. By contrast, we did not expect a down-regulation of the connectivity between these networks during the audio story in DoC-patient group.
The DoC+ group showed a significant down-regulation of the auditory and fronto-parietal networks connectivity in the audio story relative to the resting state conscious cognition. In the third study, we further tested this claim directly, by asking whether it predicted individual differences in cognitive performance. We assessed the cognitive performance of a participant subset (14/16) from the anesthesia study, who came back to the laboratory weeks later, with a battery comprising 12 cognitive tests 51 that measured short-term memory, reasoning, and verbal acuity (SI , Table S4). Based on converging results from studies 1 and 2, we expected stronger differentiation between the auditory and fronto-parietal networks during the audio story to predict stronger cognitive performance, or, a negative relationship between the AUD and DAN, ECN connectivity and the independently measured cognitive performance in the same individuals.

Auditory Processing
Object localization: reaching (asleep during scanning) Figure 6. Summary of DoC patients' clinical and fMRI assessment data. Auditory processing. In the fMRI assessment, three patients clinically diagnosed to be in a VS did not show evidence of auditory processing. The other eight patients who showed evidence of auditory processing, two patients clinically diagnosed as VS did not show evidence of brain-based command-following, and the other six, including two diagnosed as VS, showed evidence of brain-based command-following, and thus, of covert awareness. Command-following. 6/11 patients followed task commands by willfully modulating their brain activity as requested, and thus, provided evidence of conscious awareness. Two of these (P2, P5) presented a CMD profile, or a behavioral diagnosis of VS that was inconsistent with their positive fMRI results. 5/11 patients showed no evidence of willful responses in the fMRI command-following task, and, thus, provided no neuroimaging evidence of awareness. One (P7) showed no neuroimaging evidence of awareness despite an MCS diagnosis, due to falling asleep in the scanner for the entirely of the session (Materials and Methods).
The individuals' AUD-DAN connectivity during the audio story was significantly negatively correlated (r = −0.66; p < 0.007) with their cognitive performance in the verbal acuity (Fig. 8A,B) component of the battery, which accounted for the variance of tasks that used verbal stimuli (i.e., digit span, verbal reasoning, color-word remapping; Supporting Information). The AUD-DAN connectivity did not predict performance in the other two components, and the AUD-ECN connectivity did not predict performance in any of the three (Fig. 8A,C). Pairwise connectivity between these networks in the resting state did not predict cognitive performance in any of   the domains (Fig. 8A). Further, we found no relationship between the connectivity of the AUD and default mode network (DMN), included as a control high-order network, and cognitive performance. In summary, the results of the third study converged with the other two, and suggested that the extent to which the functional responses of the auditory and fronto-parietal networks to complex auditory stimuli dissociated from one another predicted independent cognitive performance in the verbal domain, and thus, may be a determining factor in individual differences in verbal acuity.

Discussion
In a series of studies, we asked whether a common mechanism underlies loss of information processing in unconscious states across different conditions, which could shed light on the brain mechanisms of conscious cognition. To this end, for the first time, we brought together two very disparate conditions where consciousness is lostdeep anesthesia and severe brain injury-to investigate the modulation of functional connectivity between the auditory and fronto-parietal networks by identical complex stimulation in identical paradigms. Subsequently, we tested whether findings from these studies predicted individual differences in intellectual abilities during conscious information processing.
Common mechanism for loss of information processing in unconsciousness during anesthesia and after severe brain injury. We use a novel approach [30][31][32] , to measure external information processing in response to richly evocative stimulation portraying real-world events, during deep anesthesia and severe brain injury. In the anesthesia study, we found that the processing of the story information was preserved in auditory cortex, but almost entirely abolished in fronto-parietal regions. Deep anesthesia led to a significant reduction in the functional differentiation of several networks across the brain, and specifically, between the auditory and fronto-parietal networks, during the story condition. These results suggested that anesthesia impaired the processing of complex external information in fronto-parietal regions by eroding their functional differentiation from sensory (e.g., auditory) systems, and not by impairing connections between or within them. Propofol was used here as a common anesthetic agent, and future studies that employ the same paradigm across different agents will help elucidate whether specific agents vary in their effect on connectivity during complex stimulation. Our results are consistent with previous findings from resting state studies, suggesting that anesthesia reduces the repertoire of discriminable brain states 52,53 , and that during loss of consciousness global synchrony impairs information processing by leading to a breakdown of causal interactions between brain areas [54][55][56] . Further, they are consistent with resting state studies using sleep-induced altered states of consciousness, which show that hyper-synchrony perturbs the feed-forward propagation of auditory information 57 , as well as feedback projections 58 , and more broadly, the stable patterns of causal interactions in response to external stimulation across the brain 59 . While these previous resting state studies suggest that global synchrony breaks down causal interactions, the investigation of causal cortico-cortical interactions was outside the scope of this work. We did not find an effect of deep sedation on thalamo-cortical connectivity in any of the five brain networks (SI), and while outside of our scope here, a potential causal role of thalamic inputs to cortico-cortical connectivity in deep sedation remains to be investigated further.
Our findings from the resting state condition in deep anesthesia manipulation agree with a previously reported reduction of brain connectivity in deep propofol anesthesia during the resting state 49,54,60 , in particular with a reduction of connectivity within the default-mode 60-62 ; but, see 53,63 ), and the executive control networks 60,64 . Although consistency with previous studies suggests otherwise, we note that differences between the story and resting state conditions in deep sedation must be interpreted with caution, in light of the consistent block order. Nevertheless, the results from the sensory stimulation condition reveal a different mechanism underlying the loss of external information processing than suggested by resting state studies. First, in the audio-story condition, the connectivity within networks was affected by sedation in the opposite direction to the resting state. Second, in the resting state condition, we observed no effect of deep anesthesia on connectivity between distinct networks, which, by contrast, increased significantly during the auditory stimulation condition suggesting loss of functional differentiation across the cortex. Another type of stimulation-transcranial magnetic stimulationhas previously been used to directly perturb the cortex in unconscious states and demonstrate that responses across the cortex become undifferentiated from one another 55,65 . In summary, these results suggest that deep anesthesia affects the brain differently when it is exposed to complex external stimulation relative to rest, with the stimulus-evoked feed-forward processing cascade being echoed undifferentiated throughout the brain, thus overcoming the inhibitory effect of propofol on neural connectivity that has been reported in resting state studies 66 .
Similarly to deeply anesthetized unconscious individuals, severely brain-injured patients who were not consciously aware during the study showed significantly reduced differentiation between the auditory and fronto-parietal networks during the story relative to their resting baseline. Conversely, similarly to healthy wakeful individuals, severely brain-injured patients who were covertly aware showed the opposite effect: significantly enhanced differentiation between the auditory and fronto-parietal networks during the story relative to their resting baseline. The modulation of the sensory to higher-order networks' relationship by environmental stimuli in severely brain-damaged (albeit conscious) patients suggests this is a fundamental feature of the conscious brain, which is resilient to substantial metabolic dysfunction following brain injury 67 . We caution that our results do not suggest that each DoC+ patient understood or processed the story similarly as healthy individuals. Foremost, DoC patients who retain covert awareness vary widely in their arousal level throughout the day. Further, although the individual DoC+ patients discussed here retained the functional brain architecture to support covert conscious awareness, the absence of a sensory baseline and lack of individual-level statistics, render it impossible to ascertain the extent of conscious processing of the story or its understanding in individual patients.
Previous studies that have compared anesthetized and unconscious brain-injured patients have highlighted that, similarly to the effect of common anesthetic agents including propofol 60,68 , brain dysfunction in this SCIeNTIFIC REPORTS | (2018) 8:13259 | DOI:10.1038/s41598-018-31525-z population is prevalent within the fronto-parietal network 69,70 . They have indicated preserved sensory processing (e.g., responses to noxious stimulation, auditory or speech perception) in the absence of higher-order components (e.g., neural evidence of pain perception, language comprehension) 5 , and suggested that disconnection between sensory and fronto-parietal systems is common to both populations. By contrast, our findings suggest that, when the brain is exposed to complex external information, these systems do not disconnect from one another in these unconscious states, as previously suggested by aforementioned resting state studies. Rather, our findings demonstrate that the erosion of functional differentiation among these systems underlies impaired information processing when consciousness is lost. Conversely, the significant increase of functional differentiation between these systems by low-level sensory stimulation relative to the resting state baseline in different conscious populations (i.e., healthy and brain-injured individuals) suggests that functional responsivity to external stimulation is a robust feature of the conscious brain.
Although loss of consciousness is common to both deeply anesthetized and some severe brain-injured patients, these two populations differ greatly. In the former no structural changes occur, and the functional brain response is altered pharmacologically. In the latter, an array of structural damage, greatly varying across patients, is present and affects altered brain responses, leading to complete functional loss in some domains and potential functional re-organization and preservation in others. Given the large differences between these two, the similarity of the functional response to previously validated targeted stimulation 30 across these populations provides strong evidence for a common mechanism underlying loss of information processing in the unconscious state. These results are consistent with current theories of consciousness, which suggest that it requires both differentiation and integration of information in neural circuits 48,54,71 , and elucidate the underlying brain mechanism by showing the critical role of functional differentiation between sensory and higher-order systems when information processing is required.
Mechanism for conscious information processing and cognition. The third study further confirmed the role of the functional differentiation between the sensory and higher-order systems in conscious cognition. Individuals who showed higher differentiation between the auditory and dorsal attention network (DAN) in response to the audio story had higher verbal acuity scores than individuals who showed lower differentiation. The story elicited a range of cognitive processes such as the orientation and modulation of attention to the saliency of incoming auditory inputs-a function subserved primarily by the DAN 72 -and language perception and comprehension, which corresponded to those engaged by the verbal acuity tasks of the cognitive battery. The functional relationship between the auditory and executive control network (ECN) was not predictive of cognitive performance, which is likely accounted for by the nature of the stimulus and fMRI paradigm which did not require behavioral response planning or monitoring-a function sub-served primarily by the ECN 73 . There was no relationship between the auditory and fronto-parietal connectivity in response to the story and performance in the short-term memory or reasoning components of the cognitive battery, likely due to the story's cognitive demands low loading on these components. We note that the verbal component of the cognitive battery, which comprised 12 tasks, accounted for the majority of variance in a subset of different tasks that used verbal stimuli (digit span, verbal reasoning, color-word remapping; full description in SI). Thus, in capturing a cross-section of processes employed in these different tasks, the verbal component represented a robust example of varied domain-specific processes, which are abstracted away from the demands of particular tasks. Therefore, although these results suggested that the relationship between brain connectivity and intelligence is domain-specific, future studies are required to further test the sensory-higher-order networks' relationship and other cognitive domains/ processes.
Further, these results agree with a previous proposal that the relationship between brain connectivity and intelligence is context specific 74 . In contrast to the a-priory predicted relationship between these networks' connectivity and intelligence during complex sensory stimulation, we found no relationship between them in the resting state. Notably, these results were predicted from two different populations where loss of from information in unconsciousness suggested a common mechanism for information processing during conscious cognition. Consistent with a recent emerging view in the field 75 , they suggested that individual differences in intellectual abilities rely on the dynamic reconfigurations of connectivity in response to incoming sensory information 76 , within a widespread system comprising sensory-specific and extra-modal cortices in fronto-parietal cortex.
In summary, findings herein suggest that the dissolution of functional differentiation is a common basis for loss of information processing across widely different conditions where consciousness is lost. Conversely, they suggest that the responsivity of sensory and higher-order brain systems to external stimulation, especially through the diversification of their functional responses is an essential feature of conscious cognition and domain-specific intelligence.

Participants. Ethical approval was obtained from the Health Sciences Research Ethics Board and Psychology
Research Ethics Board of Western University. All experiments were performed in accordance with the relevant guidelines and regulations set out by the research ethics boards. All healthy volunteers were right-handed, native English speakers, and had no history of neurological disorders. The respective substitute decision makers gave informed written consent for each patient's participation. They signed informed consent before participating and were remunerated for their time. 19 (18-40 years; 13 males) healthy volunteers, 11 (19-55 years; 5 males) DoC patients, and 14 (18-40 years; 12 males) healthy volunteers participated in study 1, 2, and 3, respectively. Three volunteers (1 male) were excluded from data analyses of study 1, due to headphone malfunction or physiological impediments to reaching deep anesthesia in the scanner. Stimuli and Design. In study 1, a plot-driven audio story (5 minutes) was presented in the fMRI scanner to healthy volunteers and they were asked to simply listen with their eyes closed. A resting state scan (8 minutes) was also acquired, during which volunteers were asked to relax with their eyes closed and not fall asleep. A novel re-analysis of data from the scrambled story condition from Naci et al. 30 (SI) was performed, as a baseline condition with the intact audio story. In study 2, severely brain-injured patients were scanned as they listened to the same audio story as healthy volunteers, and also during the resting state. In study 3, 14/16 of volunteers from the anesthesia study completed a cognitive battery comprising 12 tasks based on classical cognitive psychology paradigms (www.CambridgeBrainSciences.com) (SI). The stimuli and design for each were reported in Hampshire et al. 51 .
Sedation procedure. fMRI data was acquired during the audio story and resting state conditions while participants were awake (non-sedated) and deeply anesthetized with propofol (Ramsay score 5) 77 . Prior to acquiring fMRI data for the wakeful and deeply anesthetized states, 3 independent assessors (two anesthesiologists and one anesthesia nurse) evaluated each participant's Ramsay level by communicating with them in person inside the fMRI scanner room, as follows. Awake Non-sedated. Volunteers were fully awake, alert and communicated appropriately. For the wakeful session, they were not scored on the Ramsay sedation scale, which is intended for patients in critical care settings or patients requiring sedation. During the wakeful audio story and resting state conditions, wakefulness was monitored with an infrared camera placed inside the scanner. Deep anesthesia. Intravenous propofol was administered with a Baxter AS 50 (Singapore). We used an effect-site/plasma steering algorithm in combination with the computer-controlled infusion pump to achieve step-wise increments in the sedative effect of propofol. The infusion pump was manually adjusted to achieve desired levels of sedation, guided by targeted concentrations of propofol, as predicted by the TIVA Trainer (the European Society for Intravenous Aneaesthesia, 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 78 compartment algorithm for each participant, as incorporated in the TIVA Trainer software. Propofol infusion commenced with a target effect-site concentration of 0.6 µg/ml and oxygen was titrated to maintain SpO2 above 96%. If Ramsay level was lower than 5, 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 5. Once participants stopped responding to verbal commands, were unable to engage in conversation, and were rousable only to physical stimulation they were considered to be at Ramsay level 5. The mean estimated effect-site propofol concentration was 2.48 (1.82-3.14) µg/ml, and the mean estimated plasma propofol concentration was 2.68 (1.92-3.44) µg/ml. Mean total mass of propofol administered was 486.58 (373.30-599.86) mg. The variability of these concentrations and doses is typical for studies of the pharmacokinetics and pharmacodynamics of propofol (SI). For both sessions, prior to the scanning, volunteers were asked to perform a basic verbal recall memory test and a computerized (4 minute) auditory target detection task (SI), which further assessed each individual's wakefulness/deep anesthesia level independently of the anesthesia team. Scanning commenced only once the agreement among the 3 anesthesia assessors on the Ramsey level 5 was consistent with the lack of response in both verbal and computerized behavioral tests.
Scanning took place in a research not hospital setting, thus, breathing in the deeply anesthetized individuals could not be protected by intubation and was kept under spontaneous individual control. Therefore, although individuals were monitored closely by two anesthesiologists, airway security was at risk during scanning and time inside the scanner was kept at the minimum to ensure return to normal breathing. Thus, safety concerns for the deeply anesthetized individuals dictated that the novel condition of the naturalistic audio story be presented first. The baseline condition of the resting state was considered of secondary importance, as it has been reported previously in deep sedation condition of clinical studies. Therefore, this condition was acquired after the story condition across participants. However, the mean estimated effect-site propofol concentration and the mean estimated plasma propofol concentrations were kept stable by the pharmacokinetic model delivered via the TIVA Trainer infusion pump throughout the deep sedation session, and the lack of significant differences in the frame-wise movement parameters (assessed according to Power et al.) 79 between the story and the resting state conditions further suggested no difference in the level of sedation between the two conditions. For similar safety reasons, data on the meaningless baseline (scrambled version of the audio story) that was designed to clarify processing mechanisms in wakeful individuals, was not collected in deeply anesthetized individuals. Throughout the deep sedation scanning session, the participant's behavioral profile was monitored inside the scanner room by the anesthesia nurse and one of the anesthesiologists and outside from the scanner control room, with an infrared camera that displayed the participant's face. No movement, fluctuations of sedation, or any other state change, was observed during the deep sedation scanning for any of the participants included in the study.
Patients. The severely brain-injured patients were selected based on their clinical diagnoses (i.e., VS/MCS/ LIS, at the time of fMRI data acquisition) to form a convenience sample of the disorders of consciousness (DoC) population. No previous fMRI data was available for any of the patients at the time of scanning. Prior to commencing the scanning sessions, all VS/MCS patients were tested behaviorally at their bedside (outside of the scanner) with the Comma Recovery Scale-Revised (CRS-R) 80 . At the bedside behavioral testing, six patients met the recognized criteria for the vegetative state (VS), four for the minimally conscious state (MCS), and one for the locked-in syndrome (LIS) (full description of behavioral scores in Table S3). LIS describes an individual who, as a result of acute injury to the brain stem, has (almost) entirely lost the ability to produce motor actions, apart for small, but reproducible eye movements that confirm the presence of consciousness 81 . The patients' demographic and clinical data are summarized in Tables S2, S3, and the structural, functional MRI assessment data in Figs 6 and S1, S5, S6. Inside the scanner, each patient underwent a previously established fMRI-based protocol for assessing auditory perception and detecting covert awareness 22,28 (Figs 6 and S5), in the same visit as the audio story scan to help establish the genuine status of consciousness. Prior to assessing command-following, we assessed auditory perception to ensure that it could not have been a limiting factor to producing willful brain responses. Patients had complex underlying medical states, including head flexion and overall muscle rigidity, tracheal tubes for assisted feeding and suctioning, etc., and the highly physically constraining scanning environment compromised their comfort. Some could not lie flat for long periods, others needed frequent suctioning, and other still became agitated after an initial brief period in the scanner. Therefore, to limit patient discomfort, time in the scanner was kept at a minimum and data on the meaningless baseline (scrambled version of the audio story) was not collected.
fMRI Acquisition and Analysis. Healthy individuals. Functional images were acquired on a 3 Tesla Siemens Tim Trio system, with a 32-channel head coil. Standard preprocessing procedures and data analyses were performed with SPM8 and the AA pipeline software 82 . In the processing pipeline, a temporal high-pass filter with a cut-off of 1/128 Hz was applied and movement was accounted for by regressing out the 6 motion parameters (x, y, z, roll, pitch, yaw). Additionally, frame-wise movement parameters according to Power et al. 79 were computed. 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, a confounding effect previously reported by Murphy and others 83,84 , we performed no global signal regression. Group-level correlational analyses explored, for each voxel, the inter-subject correlation in brain activity, by measuring the correlation of each subject's time-course with the mean time-course of all other subjects. Significant clusters/voxels survived the p < 0.05 threshold, corrected for multiple comparisons with the family wise error (FWE). Functional connectivity (FC) was measured by computing via Pearson correlation the similarity of the fMRI time-courses of regions of interest (ROI)-based on well-established landmark ROIs from the resting state literature 85 (Table S1)-within and between different networks 86 (SI). As this measure of connectivity reflected the degree of similarity between the networks' functional time-courses, an increase/ up-regulation of connectivity indicated more similar time-courses between networks, and thus a loss of functional differentiation. Thus, 'differentiation' in this context is measured as the inverse of the Pearson correlation value and must not be confused with measures used in other approaches 48 . We note that Pearson correlation is a simple FC measure that, while advantageous for its minimal assumptions regarding the true nature of brain interactions and breath of its use in the neuroscientific literature, does not directly imply causal relations between neural regions. However, it is an adequate measure of FC for our purposes, because the time-course and spatial extent of the auditory and fronto-parietal networks encompassed a vast swath of the hierarchical processing cascade and, thus, many regions of cause-effect space were triggered by the stimulus. Their FC, as measured through Pearson correlation, reflected their interactions over the several minutes and the resulting computations on the information content of the auditory inputs. Future studies will also investigate the connectivity between these regions by using direct measures of causal relationships 87,88 . T-tests used to explore effects of interest between functional connectivity and cognitive performance were Bonferroni corrected for multiple comparisons. Severely brain-injured patients. Patient scanning was performed using the same 3 Tesla Siemens Tim Trio system, 32-channel head coil, and data acquisition parameters as for the healthy participants. The same data preprocessing and analyses procedures as for healthy participants were applied to patient data. The patients' spontaneous arousal during the audio story condition was monitored with an infrared camera placed inside the scanner. One patient (P7) fell asleep in the scanner for the entirely of the session and thus, showed no neuroimaging evidence of awareness despite an MCS diagnosis. The extent of information processing in individual patients (Fig. S6) was investigated with a novel technique developed by Naci and colleagues 30,32 . This approach did not involve normalization to a healthy template, nor did it constrain the patient's expected brain activity based on the localization of the effect in healthy controls. Instead, the time-course of brain activity in healthy controls served to build a strong prediction for the temporal evolution of brain activity in the patients. The precise location of a patient's brain activity was expected to deviate from that of the healthy controls' . Not only is this naturally the case for individual healthy participants, but also, importantly, it is to be expected in brain-injured patients as a result of structural and concomitant functional re-organization of the brain. Nevertheless, a spatial heuristic based on the controls' data informed the interpretation of the patients' results, helping to infer the nature of the underlying residual brain function. In summary, drawing comparisons in the temporal domain enabled direct relation of the healthy controls' activation to that of brain-injured patients, while avoiding stringent spatial constraints on the patients' functional anatomy (Fig. S6). By contrast, for the analysis of functional connectivity based on a set of network nodes pre-defined in the healthy literature in the MNI standard neurological space, each patient's brain was normalized to the healthy template. We reasoned that any damage within the regions of interest in each patient's brain would add noise to the brain activity measurement and reduce the power to detect an effect. Therefore, any results in brain injured patients, that aligned with a-priory hypotheses based on the anesthesia study were highly unlikely given the heterogeneous structural preservation and would present a conservative estimate of the underlying effect.

Data Availability Statement
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.