Consciousness, a multifaceted phenomenon, is conventionally defined by awareness (content) and wakefulness (level). However, a multidimensional model is needed to capture its complexity. This study addresses a critical knowledge gap by identifying the neural substrates of consciousness dimensions and linking them to the brain's neurofunctional properties. Traditional functional localization approaches that focus on discrete brain regions or networks fail to adequately address the complexity of consciousness. This research proposes that the dimensions of consciousness are encoded in multiple neurofunctional dimensions of the brain, represented by cortical gradients—continua that span from unimodal sensory areas to transmodal association cortices. The interplay between the spatial organization of these gradients and the temporal dynamics of brain activity is explored as a potential key to understanding the neural basis of consciousness. The authors hypothesize that: 1) Different neurofunctional dimensions of consciousness are encoded by different cortical gradients; 2) Cortical gradients form a multidimensional space where altered states are associated with both common and state-specific changes; and 3) Brain network functional geometry disruptions correlate with altered dynamic brain state transitions. These hypotheses are tested by comparing cortical gradients across groups with normal wakefulness, various anesthetic states, and individuals with psychiatric or neurological conditions.

The paper draws on previous research defining consciousness as a multidimensional phenomenon, often characterized by awareness and wakefulness. Studies of unresponsive wakefulness syndrome (UWS), where patients exhibit wakefulness without awareness, are cited. The limitations of traditional functional localization focusing on specific brain regions are discussed, highlighting the need for a more holistic approach considering the brain's intrinsic functional geometry. The authors refer to recent advances in neuroimaging allowing for the study of cortical gradients, continua representing the brain's overarching functional geometry spanning unimodal to transmodal systems. The importance of considering both spatial and temporal properties of brain activity is emphasized, referencing studies showing cortex-wide fMRI co-activations propagating as waves along cortical gradients. The paper also cites studies exploring dynamic brain states and their temporal circuits.

The study uses five fMRI datasets from three independent research sites: 1) Propofol deep sedation (PDS), 2) Propofol general anesthesia (PGA), 3) Ketamine anesthesia (KA), 4) Unresponsive wakefulness syndrome (UWS) patients, and 5) Patients with schizophrenia (SCHZ), bipolar disorder (BD), and attention deficit hyperactivity disorder (ADHD). The cortical gradient analysis employs a nonlinear diffusion space mapping method to decompose the functional similarity structure of fMRI data into embedding components (gradients). fMRI time courses were extracted from 400 cortical areas. A 400x400 functional connectivity matrix was calculated for each participant and condition, transformed into a normalized cosine angle affinity matrix, and then analyzed using a diffusion map embedding algorithm to identify cortical gradient components. The first three gradients explained approximately 37% of the variance and represented: 1) unimodal to transmodal, 2) visual to somatomotor, and 3) visual/default-mode to multiple-demand systems. Statistical analyses included Bayesian paired and independent samples t-tests to compare gradient ranges (minimum to maximum eigenvector values) and global dispersion (sum squared Euclidean distance of all regions to the global centroid) between baseline and altered states. Network eccentricity (distance between a network's centroid and the global centroid) and pairwise network distances were also calculated for seven predefined functional networks. Co-activation pattern (CAP) analysis tracked large-scale co-activations by calculating similarity between fMRI volume signals and CAP centroids, quantifying the occurrence rates of CAPs for DMN, DAN, and VAN networks. Bayesian Kendall correlations analyzed the relationship between network distance and CAP occurrence rates. Data preprocessing included slice timing correction, motion correction, coregistration, spatial normalization, band-pass filtering, and global signal regression.

The study revealed that disruptions of consciousness are associated with degradation (less functional differentiation) of one or more major cortical gradients. Specifically: PDS and UWS showed decisive evidence for reduction in Gradient-1 (unimodal to transmodal); PGA showed decisive evidence for reduction in Gradient-1 and Gradient-3 (visual/default-mode to multiple-demand); KA showed decisive evidence for reduction in Gradient-2 (visual to somatomotor); and SCHZ showed decisive evidence for reduction in Gradient-2. Global dispersion was significantly reduced in PGA, UWS, KA, and SCHZ. Analysis of network eccentricity showed that the ventral attention/salience network (VAN) was consistently vulnerable across depressed states of consciousness. Pairwise network distance analysis revealed shortened distances between VAN-DMN consistently across depressed states; and other shortened distances in PGA, KA, and UWS. Finally, the study showed a correlation between network distance (VAN-DMN-DAN) and the occurrence rates of corresponding CAPs. In depressed states, co-activations shifted from DMN and DAN to VAN.

The findings support a multidimensional model of consciousness, where specific combinations of cortical gradient changes characterize different altered states. Gradient-1 (unimodal to transmodal) appears crucial for awareness, Gradient-2 (visual to somatomotor) for sensory organization, and Gradient-3 (visual/default-mode to multiple-demand) for arousal. The consistent vulnerability of VAN and DMN across states suggests their crucial role in conscious processing, potentially as an interface for salient information prioritization and transfer to transmodal areas for abstract representation. The covariation between the brain's functional geometry and its temporal dynamics highlights the integrated nature of consciousness. The dissociable effects of propofol and ketamine on cortical gradients suggest distinct mechanisms of action. The results have implications for understanding disorders of consciousness and psychiatric disorders, particularly schizophrenia, emphasizing the importance of multidimensional assessment beyond behavioral responsiveness.

This study provides strong evidence supporting a multidimensional neurofunctional model of consciousness based on cortical gradients. The identified gradients—unimodal to transmodal, visual to somatomotor, and visual/default-mode to multiple-demand—represent distinct dimensions potentially related to awareness, sensory organization, and cortical arousal. Disruptions of consciousness are linked to gradient degradation. The vulnerability of VAN and DMN networks and the interplay of spatial and temporal dynamics offer valuable insights into the neural mechanisms of consciousness. Future research should explore the role of subcortical regions, refine gradient analysis methods for brain-injured patients, and investigate the causal relationship between gradients and specific molecular pathways. Applying this framework to other neurological and psychiatric conditions may greatly advance our understanding of consciousness.

The study's limitations include the focus on only the first three well-documented cortical gradients; the exclusion of subcortical gradients due to methodological constraints; a limited number of UWS patients; and the use of behavioral unresponsiveness as a surrogate measure for depressed states of consciousness. Further research should address these points and investigate the causal relationships between cortical gradients, subcortical regions, and specific molecular mechanisms.