Introduction

The coupling between conscious awareness and its external motor manifestation is so pervasive that it is difficult to comprehend the devastating state of fully conscious patients who are unable to respond. Severe brain injury can lead to such cases, termed the “locked-in” syndrome (LIS). As a result of motor disconnection, it is challenging to differentiate such cases from those in which awareness itself is disrupted – termed vegetative state (VS) or minimally conscious state (MCS). The differential diagnosis between VS and MCS is even more challenging and up to 40% misdiagnosis has been reported [1], [2], [3], [4]. The method of choice for diagnosis of conscious status has been careful bedside observations, which are challenging due to fluctuation in arousal, motor deficits and other deficits attributed to the injury, such as aphasia. This method, due to its subjective nature could partly contribute to the misdiagnosis rate [5]. Recent studies have demonstrated that fMRI may provide some DOC patients with a means for communication through blood oxygen level dependent (BOLD) signals evoked by mental imagery, even in the complete absence of motor outputs [6], [7]. However, this method relies on patient cooperation as well as attentional capacity and may not be suitable for the general patient population. Even more problematic is prognosis, the ability to predict which patients have better chances of recovery. These challenges highlight the urgent need for an objective physiological measure complementing current evaluation tools.

Recently, a series of studies uncovered a robust phenomenon that offers exciting potential for a complementary diagnosis of unresponsive patients. Even in the absence of intentional sensory-motor tasks, the human cortex manifests high-amplitude ultra-slow (<0.1 Hz) fluctuations in its BOLD signals that reflect distinct functional systems [8], [9], [10]. These spontaneous fluctuations show anatomical specificity in that correlations (also termed functional-connectivity) are more pronounced between the functionally related cortical regions (e.g. right and left auditory cortices) than between functionally unrelated cortical regions (e.g. ‘Extrinsic’/‘task-positive’ and ‘default-mode’ networks [11]). A particularly striking and consistent aspect of this connectivity is the correlation across homotopic sites in the two hemispheres [9], [12], [13], [14]. More recently, a likely neuronal correlate of these spontaneous BOLD fluctuations was found in ultra-slow modulations of neuronal firing rates and gamma power in local field potentials [12], [15], [16].

Although the functional role of the ultra-slow spontaneous fluctuations remains unclear, they could potentially aid clinical diagnosis. Indeed, such fluctuations and their network correlations were shown to be altered in several neurological and neuropsychiatric disorders [10], [17], [18], [19], [20]. The spontaneous nature of ultra-slow fluctuations, emerging without the need for intentional cooperation, makes them ideally suited as a complementary diagnostic tool in DOC. It has been recently shown that connectivity within the default-network [21], a subset of regions that are deactivated during externally oriented tasks, is negatively correlated with the degree of clinical consciousness impairment [22], see also [23]. In these studies resting–state connectivity within the default mode network (DMN) was assessed using probabilistic independent component analysis in DOC patients and an alteration in the spatial extent of the DMN was found at the group level [22] and at the individual level [23].

Although it has been suggested that the DMN is associated with basic functions related to consciousness [24], such as self-related processes [25], [26], it is not clear whether the reduction in connectivity is restricted to the default-mode network or rather extends into externally oriented regions. Furthermore, in order to use connectivity analyses for the diagnosis of these patients, it will have to be established how reliable these measures are on a subject-by-subject basis.

Here, we examined resting-state inter-hemispheric connectivity in three homotopic regions of interest, which were easily identified based on anatomical landmarks, and were part of the externally oriented network. We found reduction in inter-hemispheric functional connectivity in impaired awareness subjects as compared to intact awareness subjects. In addition, functional connectivity was correlated with the level of consciousness and was found to deviate from the healthy sample in in 5/8 patients using a single case statistical test. Importantly, of the three patients whose connectivity indices were comparable to the healthy sample, one was diagnosed as locked-in. These results suggest that resting-state functional connectivity might prove beneficial in the future as a complementary measure in the diagnosis of DOC patients.

Materials and Methods

The study was approved by the ethics committee of the Faculty of Medicine at the University of Liège, Belgium. Written informed consent for healthy volunteers and patients was obtained from all subjects and legal guardians, respectively.

Results

We examined correlations between spontaneous BOLD activity in selected cortical regions of interest (ROIs) of the “Extrinsic”- externally oriented, task-positive network [11], [44] and all other cortical regions. Subjects were tentatively grouped into intact and impaired awareness (see methods for further details). Figure 1 shows the correlation map computed for a “seed” ROI in the right preCG. In the intact awareness group (Figure 1A), activity in the preCG significantly correlated with neighboring somato-sensory cortex, and with the homotopic “mirror” region in the left hemisphere. Activity in the cingulate sulcus was also significantly correlated with that of the preCG (Figure S1). Negative correlations were observed in the posterior cingulate cortex, lateral temporal cortex, and inferior parietal lobule, which are commonly referred to as the default-mode network [45], [46] or “intrinsic” network [11], [25]. In the impaired awareness group, significant correlations were restricted to immediately neighboring cortex, while inter-hemispheric correlations were largely absent (Figure 1B). This drastic reduction of inter-hemispheric correlations was confirmed by statistically comparing the group maps (Figure S2) taking into account the different sample sizes. Highly similar results were observed for “seed” regions in the right intra-parietal sulcus (IPS, Figure S3) and the right posterior central gyrus (postCG, Figure S4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Correlations between spontaneous BOLD fluctuations in right pre-central gyrus and other cortical regions.

Group correlation maps between a “seed” region in the pre-central gyrus (preCG) and all other cortical voxels, projected on inflated left (LH) and right (RH) hemispheres (lateral view). (a) Correlations of spontaneous activity in the intact awareness group (n = 12). (b) Correlations in the impaired awareness group (n = 7). Red arrow, “seed” region location. Blue arrow, homotopic “mirror” regions in the left hemisphere. Note that inter-hemispheric correlations are largely absent in the impaired awareness group. Abbreviations: CS, central sulcus; LS, lateral sulcus; IPS, intra-parietal sulcus.

https://doi.org/10.1371/journal.pone.0037238.g001

To obtain a quantitative measure of the inter-hemispheric correlations in each subject, we introduced an inter-hemispheric correlation index (ICD). This index represents the average inter-hemispheric correlations between the pre-defined ROIs (for details see materials and methods). We found that ICD values were decreased in the majority of DOC patients (Figure 2). Individual ICD scores were found to significantly correlate with the degree of consciousness, ranging from brain dead, coma, VS, MCS, LIS, to healthy controls (Spearman's correlation coefficient r = .61, p = .0057). Since a decrease of resting-state connectivity with age has been reported for the DMN [[47]; but see [48]], we confirmed that lower ICD values in the patient group did not reflect age differences. Indeed, the correlation between age and ICD value for all participants was not statistically significant. (Spearman's correlation coefficient r = −.36, p = .128). In addition, when using age as a control variable in a partial correlation analysis, the correlation between ICD and the level of consciousness remained significant (r = .52, p = .027). We further addressed this concern by analyzing data from additional 11 healthy controls that were matched for age (53.54±15.97 years). Resting-state data and structural scans were downloaded from a freely available online source (http://fcon_1000.projects.nitrc.org/indi/pro/nki.html), and identical preprocessing and analysis was performed (see Figure S5, Table S2 and Material S2 for further details). The results show that when individual ICD values of DOC patients were tested against the normative sample of age-matched controls (see below), highly similar statistical results were obtained as found in the younger control group.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Inter-hemispheric Correlation Index (ICD) in individual subjects.

Subjects are separated on the x-axis depending on their clinical state (patients in red and healthy controls in black). The solid line represents the mean ICD value in the healthy controls group and the dashed line represents the mean-2*standard deviation. Abbreviations: H, healthy; L, locked-in syndrome; V, vegetative state; M, minimally conscious state; C, coma; B, brain death.

https://doi.org/10.1371/journal.pone.0037238.g002

To test whether individual ICD scores from patients significantly deviate from the healthy subjects (used as normative sample), we used a t-test specifically developed for single case studies [43]. As shown in Table 1, individual ICD values from 5/8 patients were significantly different (p<.05) from the average ICD value of the healthy controls. In addition to the ICD measure, the disruption of inter-hemispheric correlations in DOC patients could also be discerned to a large extent in individual BOLD time courses (Figure 3) and single-subject maps (Figure S6).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Correlations between spontaneous BOLD signal time-courses across hemispheres.

(a) Anatomical locations of regions-of-interest (ROIs) in the right pre-central gyrus (red arrow) and the homotopic ROI in the left hemisphere (blue arrow). (b) Time-courses in a healthy subject exhibit high correlation (r = 0.77). (c) Time-courses in the vegetative patient who recovered consciousness exhibit high correlation (r = 0.73). (d) Time-courses in an impaired awareness patient exhibit low correlation (r = −0.02). Red and blue time-courses denote signals from the right and left hemispheres, respectively normalized to percent signal change. Abbreviations: CS, central sulcus; LS, lateral sulcus; IPS, intra-parietal sulcus.

https://doi.org/10.1371/journal.pone.0037238.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. ICD values and Crawford and Howell test results.

https://doi.org/10.1371/journal.pone.0037238.t001

Taken together, these results demonstrate that impaired awareness was associated with reduced inter-hemispheric correlations and was largely evident at the single subject level. The three patients with non-significant difference from the healthy control group were the LIS patient (which was expected to have normal ICD since his consciousness level is identical to a healthy control), one Coma patient (which showed a trend towards significance, p = .0517) and one VS patient (which would be expected to have decreased correlation values). The ICD value of this VS patient, studied seven days after anoxic brain damage was 0.65, well within the range of healthy controls (0.56±0.1; mean ± SD). Importantly, shortly after the brain imaging (13 days), this patient progressed to a state of MCS and later (40 days post scan), recovered consciousness, reaching a state of “moderate disability” according to the Glasgow Outcome Scale [31]. In our sample, this was the only patient to improve in diagnosis, (see Glasgow Outcome Scale for all patients in Table S1), one patient remained in a MCS and the rest of the patients did not survive.

Importantly, inter-operator variability was low; control analysis based on ROIs drawn by a separate fMRI researcher yielded a significant correlation (r = 0.65, p<0.01, Figure S7) between the ICD scores obtained by these two independent investigators, supporting of the future potential of this method for clinical use (see Table S3 and Material S1 for further information).

Discussion

In this study, we show reduced inter-hemispheric connectivity between homologous cortical regions within the “Extrinsic”, task positive network using resting-state fMRI in DOC patients. The observed decrease in connectivity was significantly correlated with the degree of consciousness impairment, and was evident to a large extent at the single subject level.

Inter-hemispheric correlations, or symmetry, between homologous regions is one of the prominent characteristics of resting-state fMRI and have been demonstrated in healthy populations using various techniques, such as region of interest based analysis [9], [44], independent component analysis (ICA) [14], [36], and whole brain approaches [49], [50]. Inter-hemispheric symmetry appears to be a ubiquitous characteristic of brain anatomy [51] and function [52]. Inter-hemispheric connections, in analogous fashion to within-hemisphere connections, play a role in the integration of information and coordination between the two hemispheres. Not surprisingly, alteration in inter-hemispheric correlation has been demonstrated in various diseases and behavioral impairments [53], [54], [55], [56], [57] illustrating the importance of intactness of communication/synchronization for the normal functioning of the brain. Although the link between reduction in connectivity within the DMN and DOC has been shown previously [22], [23], it has remained unclear whether a reduction in connectivity is restricted to this network alone. Our results demonstrate a significant reduction in connectivity between homotopic regions belonging to the “Extrinsic”, task positive network. This finding raises the question of how widespread the reduction of connectivity is in DOC patients. Since reduction in connectivity is not restricted to one network, it seems, however, that DOC might reflect a more global impairment in functional connectivity and in the integrity of different circuits. In other words, it may be that reduced connectivity within specific cortical networks may affect specific behaviors but is not sufficient to affect the overall level of consciousness. Along this line, it has been shown in stroke patients that a reduction in connectivity that is specific to the attention network is reflected behaviorally as neglect symptoms, but not in the level of consciousness [58]. Indeed, reduction in connectivity within one network might not be a sufficient marker for the diagnosis of DOC patients, and a whole brain analysis might be better suited to test more global impairments.

The exact source of the reduction in connectivity is still not fully understood. However, it is beyond the scope of our paper to investigate whether the striking disruption in the inter-hemispheric correlations observed in DOC patients is due to a widespread cortical, subcortical or white matter damage. Although structural damage will evidently lead to loss of connectivity, especially in the immediate time following injury [59], but see [60] and [61], such reduction may be linked to synaptic changes well below the resolution of brain imaging and could also be influenced by plasticity changes following the injury through sub-cortical connections [60]. In addition, reduction in functional connectivity following structural damage has been reported for areas that appear structurally intact [58], [62]. Recently, Bruno et al. reported a case of “functional hemispherectomy” in two DOC patients, with near-normal DMN components in one hemisphere, given a structural and metabolic deficit in the other hemisphere [63]. This result highlights the need for a multimodal neuroimaging approach, as one of the challenges related to connectivity measures in the population of DOC patients will be to conceive a quantification of the heterogeneous damage typically observed in this population, taking into account asymmetric structural damage which poses a challenge for the ICD measure proposed in our study.

On a more general scope, the inference one can make about the cognitive level from spontaneous fMRI fluctuations remains controversial. On the one hand, spontaneous fluctuations can appear in the absence of any task or intentional activity, are detectable during anesthesia [64], [65], and even accentuated during sleep [12], [66]. On the other hand, other evidence implicates a contribution of ultra-slow fluctuations in perceptual decision making [67], [68] and motor control [69]. Furthermore, the fact that the spatial organization of spontaneous ultra-slow activity replicates task-related activity of functional networks [8], [9], [14] suggests that their presence may reflect a hebbian co-activation process [70], and conversely, their disruption may thus be due to a reduction in network functionality.

As to the clinical significance of our findings, the results point to the potential usefulness of the ICD in diagnosing individual cases of impaired awareness. In order to validate such measure for future clinical use, a larger population of DOC patients and healthy subjects need to be tested, thus also allowing for an estimation of the ICD measure's specificity and sensitivity. The fact that we did not observe the lowest ICD value in the brain dead patient further emphasizes the need for a larger sample to separate “true” ICD values from ICD values generated by spurious noise. In a recent paper by Boly and colleagues, using a seed based approach in the default mode network, a brain dead patient failed to show any significant correlations in a whole brain map [32].

The unexpected recovery of the VS patient showing a normal-level ICD, points to the promising possibility that the ICD index may serve, at least in specific cases, as a prognostic measure for recovery from DOC. However, this single observation is of course far from providing a conclusive demonstration and should be taken at this stage only as a catalyst for a wide-scope search for additional similar cases.

To conclude, we propose the ICD index as a measure of symmetry in functional connectivity that can be used as a diagnostic marker in DOC. This measure has the advantage of relying on spontaneous fMRI signal fluctuations and thus does not depend on patient cooperation, which is often absent in DOC patients.

Supporting Information

Acknowledgments

The authors would like to thank Avital Hahamy who accepted to draw the regions of interest used for the inter-operator analyses, and Dr. Daniel Margulies for his insightful comments on an earlier version of the manuscript.