Mechanisms Underlying Disorders of Consciousness: Bridging Gaps to Move Toward an Integrated Translational Science

Disorders of consciousness are characterized by a wide array of symptoms and etiologies. Consequently, a number of classification schemes across different dimensions have been proposed that are based on aspects such as cognitive functions [10], awareness [11] and sensory, motor, and arousal behavioral functions [12]. With the acknowledgment that any one framework cannot fully capture all the constructs relevant to DOC, we propose to contextualize our recommendations within the framework provided by Sanders et al. [13], which distinguishes between consciousness, environmental connectedness, and responsiveness (C-EC-R), as outlined below.

Lack of responsiveness remains the clinical criterion for DOC. However, there is a clear conceptual distinction between responsiveness (specifically, nonreflexive behavioral responses) and consciousness (presence of subjective experience, regardless of what the experience is about [13]; note that here we will use this term to also encompass awareness). The possibility for a dissociation between these two dimensions is sharply illustrated by the (rare) case of patients experiencing intraoperative awareness during anesthesia, whose unresponsiveness due to paralysis is mistaken for evidence of unconsciousness [14‐16]. Likewise, independent studies during sleep and anesthesia (in which retrospective reports can be obtained on awakening) demonstrate that subjective experience frequently occurs even in the absence of any behavioral responsiveness [17, 18]. Within DOC, routine assessment based on overt responsiveness may be missing covert consciousness in about ~15–20% of patients [3, 4], a fact that has practical and ethical implications for the management of these patients.

In addition to consciousness and responsiveness, a third relevant dimension is environmental connectedness (as originally proposed in the context of anesthesia [13]). Environmental connectedness corresponds to consciousness of the external environment so that what happens in the environment influences the contents of one’s consciousness (unlike, e.g., during some dream states). On the basis of this framework of C-EC-R, patients with CMD can be characterized as being conscious and environmentally connected but not (overtly) responsive. An intriguingly similar example of dissociation between consciousness and responsiveness is found in patients studied under anesthesia by using the isolated forearm technique, whereby an inflated cuff on the arm prevents it from being paralyzed during general anesthesia [13]. Despite the absence of spontaneous behaviors, such patients can perform simple commands, such as squeezing the physician’s hand (although the potential confound of arousing nociception due to the inflated cuff should also be considered). Examples of conscious but environmentally disconnected states include dreaming and dissociative states induced by ketamine; examples of states that are conscious and environmentally connected but unresponsive are awareness during general anesthesia and sleep paralysis, whereby the paralysis that naturally occurs during rapid eye movement (REM) sleep persists for a short time after awakening [19]. Of course, it is also important to consider the notion of behavioral arousal; although likely neither sufficient for consciousness (given the presence of sleep–wake cycles in patients with VS/UWS) nor necessary for it (given the possibility of dreams during deep non-REM sleep), arousal may nevertheless be a background prerequisite for responsiveness and possibly also for full-fledged environmental connectedness (Fig. 2).

Correctly characterizing each patient in terms of the three dimensions described in the C-EC-R framework may be valuable for improving diagnosis and determining the prognosis and appropriate standards of care; for instance, it is especially pressing to establish communication with unresponsive but connected patients and not only identify reliable biomarkers for conscious contents, such as emotional distress or pain (i.e., the conscious processing of nociceptive information), but also identify whether some environmental stimuli trigger them (environmental connectedness). Additionally, by capitalizing on the availability of subjective reports after awakening from sleep or anesthesia in the laboratory environment, it will be crucial to identify ways to detect environmental connectedness based on brain function alone.

By identifying the neural mechanisms that determine transitions between the dimensions of the C-EC-R framework, it may be feasible to devise targeted treatment strategies aimed at restoring each in their own right and to identify which personalized avenues may be the most advantageous for a given patient. Specifically, efforts should be made not only to investigate whether specific mechanisms identified from studies of anesthesia and sleep [13] could also apply to patients with DOC (e.g., to what extent could the isolated forearm scenario constitute a good model for CMD?) but also to investigate how these different states differ from one another (anesthesia is reversible on a short time scale, but DOC may not always be). Additionally, investigations should seek to identify possible dissociations between each of these aspects in subgroups of patients with DOC.

Crucially, the limitations of our present ability to discriminate between elements of the C-EC-R framework in clinical practice are symptomatic of deeper gaps that need to be addressed to obtain a mechanistic understanding of DOC. DOC can arise from a variety of causes, highlighting the need for a more comprehensive understanding of the intricate interactions between structure and function and of their temporary vs. permanent nature. How brain function in turn determines C-EC-R also needs to be clarified, with clear diagnostic utility and prognostic value for recovery. Likewise, a complete understanding of the mechanisms underlying the presence vs. absence of C-EC-R will need to span multiple levels of analysis [20] and biological detail: from the cellular and molecular microscale to macroscale systems and networks. Reconciling the distinct levels of analysis will require the concerted interaction of data-driven approaches combining multimodal data from large-scale studies with theoretical approaches able to integrate their findings into a coherent framework, as well as synthesis through computational modeling. Thus, against the backdrop of the clinical need to discriminate between consciousness, connectedness, and responsiveness in patients with DOC, we envision these complementary approaches as establishing a virtuous cycle to drive forward both scientific understanding and clinical practice.

Advancing the science of DOC requires a precise mapping of the heterogenous structural and functional brain alterations observed in DOC to clinically relevant dimensions (e.g., C-EC-R) across a variety of temporal and spatial scales. Although it is sometimes possible to precisely map neurological dysfunction to a specific location of damaged tissue, it is important to emphasize that brain regions are intricately interconnected, such that local structural or functional changes may well have far-reaching or even global repercussions on other components of the network (diaschisis [46]). Classical lesion-based methods have found success focusing on specific regions of interest (ROI); however, future work should attempt an integration of these approaches with the notion that there exists a many-to-many mapping between brain structure and functional brain states [47]. A full mapping between structural and functional correlates of DOC will require the leveraging of multimodal neuroimaging and neurophysiological techniques, combined with novel analytical methods for integrating them, including the emerging approach of whole-brain computational modeling (Box 2).

Historically, circumscribed lesions and changes in activity within damaged brains have been used to identify ROI and model simplified circuits that may be relevant to DOC symptoms. For instance, the influential mesocircuit model proposes that because of pathological changes following severe brain injuries, a reduction of thalamocortical and thalamostriatal outflow withdraws drive to the frontal cortex and striatum, thus implicating basal ganglia–thalamocortical circuits in the symptoms of DOC [10]. Compared to network or computational perspectives, these focal lesion-based models have the benefit of providing clear targets for treatment (e.g., via deep brain stimulation, low-intensity-focused ultrasound, pharmacological). For instance, emphasis on the role of the thalamus in DOC has compelled the development of techniques for its stimulation, which have been associated with improved behavioral responsiveness in a subset of patients with DOC [9, 48]. Yet the precise functional roles of key ROI, such as the thalamus, remain far from fully characterized; indeed, it remains unclear why nearly full ablation of the thalamus does not appear to result in a loss of consciousness in rodent models [49] despite its consistent association with DOC. Thus, focal characterization of structure–function relationships retains unique value but must continue to be refined.

For instance, although the original mesocircuit model emphasizes pallido-thalamo-cortical communication, this model has been updated by the recent discovery of direct structural connections between the globus pallidus and the cortex, first in rodents [50] and then in humans [51] by using DTI. In parallel, DCM (see Box 1) was applied to the mesocircuit and specifically implicated pallidocortical communication in the transition between states of consciousness under anesthesia [52]. This example illustrates the kind of multidisciplinary workflow that should inform the placement of ROI within increasingly complex frameworks (network, computational) for understanding the structural and functional relationships underlying DOC.

Compared to ROI approaches, the structural correlates of DOC may be further detailed by employing mass-univariate, voxel-wise analysis, which can produce mappings of behavioral symptoms to structural alterations at the millimeter scale (e.g., voxel-level symptom mapping [53, 54]; also see voxel-level shape analysis [55, 56]). Similar approaches have been applied to the structural connectome derived from DTI (known as connectome-based lesion symptoms mapping [57, 58]). Although these methods allow for a more spatially precise connection between large-scale functions (e.g., behavioral arousal) and structural damage, they do not capture the perhaps more elusive reorganizations in functional networks that often follow structural insult and that ultimately produce changes in behavior (e.g., diaschisis) as well as inadequate or maladaptive compensatory mechanisms, which may also produce DOC symptoms. Indeed, regions that are only secondarily affected may be particularly promising treatment targets because of their relative retained structural integrity, which may hold greater potential to reach preinjury levels of functioning. To capture the relationship between gray matter atrophy, white matter disconnection, and functional interactions, there is a need to better integrate structural correlates with the full range of functional modalities available to us instead of behavioral symptoms alone.

A jumping-off point for multimodal integration in DOC may be to overlay the structural correlates of DOC with the healthy human connectome (both structural and functional) to derive likely locations for a diaschisis effect, a method that avoids the often challenging process of multimodal data collection within patients themselves [59, 60]. Such an approach could be used to build whole-brain computational models, including the known structural correlates of DOC and known large-scale functional correlates (Box 2).

However, multimodal data collection in patients with DOC—and new methods for merging modalities with minimal information loss [60]—will be necessary to account for the inevitable translational gaps between healthy patients, computational models, and real patients with DOC. Some recent studies have pioneered these approaches; for instance, EEG markers have been linked to subcortical damage in acute [71] and chronic [72] DOC, and recent evidence indicates that preserved fractal (self-similar) character of structural brain networks is associated with covert consciousness on the basis of fMRI response to mental imagery tasks [73]. Inspiration for future joint investigations of structure and function may be drawn from other models of disrupted consciousness, such as anesthesia, in which links between structural and functional networks have recently been identified [24, 38].

To achieve clinical relevance, it will be crucial not only to collect increasingly abundant and multimodal data but also to distill from them the biomarkers that are most predictive of consciousness, connectedness, and responsiveness, either in terms of brain states or information structures, which may better explain how functions such as C-EC-R emerge instead of only from where [62]. For instance, the framework of connectome harmonic decomposition [47, 74] allows for functional brain activity to be decomposed in terms of different contributions of the human structural connectome [47]. Such connectome harmonics are wave-like patterns of spatial oscillations that represent how information spreads over the structural connectivity of the human brain. Through this approach, a common neural signature was recently identified between reduced consciousness in DOC and under anesthesia [74] and thus may represent a general signature of consciousness that can inform theoretical interpretations.

Of course, a plethora of other measures have emerged in recent years to describe network properties, including graph-theoretic and information-theory-based measures. The perturbational complexity index (PCI), which measures the complexity of EEG signals following causal perturbation by pulses of transcranial magnetic stimulation (TMS), appears particularly adept at detecting covert consciousness during sleep and anesthesia, as well as in patients with DOC, without the need for behavioral response [6]. Other measures of this kind, although perhaps less accurate for diagnosis, have also shown significant prognostic value [41, 75]. Future approaches should seek relationships and convergence between metrics derived from information theory, graph theory, and dynamical systems theory and strive to connect them with structural measures. Contemporary examples include the recent link found between PCI and subcortical atrophy [76] and the recent association between focal cortical lesions and the generation of pathological slow waves, disconnection, and lost complexity [77, 78].

Alongside sleep or anesthesia, seizures present yet another way to interrogate the generalizability of brain states to the C-EC-R framework in unique contexts (e.g., presence of fast-spiking high metabolism in seizures compared with anesthesia and DOC). This is especially relevant given that specific deficits in C-EC-R can all present during seizures in patients with chronic epilepsy [79‐81]. The loss of informational complexity typically observed during seizures [82] and the restored behavioral arousal following subcortical (pons, thalamus) stimulation during focal seizures in rats [83] suggests that generalizability is likely to be found; however, this area is ripe for more investigation.

Importantly, the first steps toward the integration of structure and function in DOC are largely taking place by experimental, analytic, and computational integration of the various neuroimaging methods that probe macroscale phenomena. However, a full mapping of structure–function relationships must inevitably dive deeper into the microscale neural substrate on which macroscale networks arise. Indeed, modeling the biological correlates of DOC across all spatial scales is likely to improve the location of individual patients within the heterogeneous space of DOC manifestations (e.g., as defined by C-EC-R). Thus, in the next section, we detail the gaps that exist in our understanding of microscale neural underpinnings of DOC and coma and how macro and micro levels of understanding may be bridged.