Cognitive auditory evoked potentials in coma: can you hear me?

Abstract

This scientific commentary refers to ‘Neural detection of complex sound sequences in the absence of consciousness’, by Tzovara et al. (doi:10.1093/brain/awv041).

This scientific commentary refers to ‘Neural detection of complex sound sequences in the absence of consciousness’, by Tzovara et al. (doi:10.1093/brain/awv041).

Event related potentials (ERPs), classically measured by EEG, are the electrophysiological brain responses to a stimulus. Somatosensory and auditory ERPs have been used as a non-invasive tool for assessing brain functions and predicting outcomes in disorders of consciousness and coma (for review see Bruno et al., 2011). While early ERPs (such as the absence of cortical responses on somatosensory-evoked potentials) predict poor outcomes, cognitive ERPs may be indicative of recovery of consciousness after coma (Vanhaudenhuyse et al., 2008). Auditory cognitive ERPs permit assessment of residual higher-order processing, such as echoic memory (e.g. using mismatch negativity potentials), acoustic and semantic discrimination (e.g. P3 or P300 evoked potentials), and incongruent language detection (e.g. N400 potentials). Classically, these ERP paradigms are based on the detection of a stimulus in violation of an auditory regularity, composed of two neural events, each characterized at the EEG level by a stereotypical morphology and latency: a mismatch negativity followed by a later complex named P300 further divided into an early (P300a) and a late component (P300b). The mismatch negativity is thought to reflect unconscious processing (Daltrozzo et al., 2009), whereas the P300b has been linked to consciousness (Bekinschtein et al., 2009). Neither of these ERP components are specific to auditory stimuli and both can be elicited by other sensory modalities. However, auditory cognitive ERPs have been used most extensively in the field of coma because they are easy to deliver and—in contrast to visual ERPs—can be easily acquired in eyes closed conditions. Given the difficulties inherent in the differentiation between P300a and P300b, Bekinschtein and Naccache et al. (2009) developed an elegant ERP paradigm (local-global paradigm) composed of two embedded levels of auditory regularity, one local (at the single trial level) and one global (across trials). Applying such a paradigm, they demonstrated that, while violations of local regularities elicited measurable ERPs both in conscious and unconscious conditions, responses to violations of global regularities are reliably detectable only in the presence of conscious perception—in line with the ‘global neuronal workspace’ hypothesis of consciousness (Noirhomme et al., 2010). In this issue of Brain, Tzovara and colleagues use the local-global auditory ERP technique to challenge the notion that detection of global violations is linked to conscious processing, showing for the first time that responses to such violations can be elicited even in comatose patients (Tzovara et al., 2015).

The local-global paradigm relies on the delivery of five brief sounds in a row, the first four identical, while the fifth can be either identical to the previous or deviant—the nature of the fifth sound establishing the local regularity (or its violation). The global regularity within a block of trials is established by selecting one of the two series as the global standard and the other as the global deviant; this is achieved by delivering the two kinds of trials in a pseudorandom manner within each block (i.e. 80% of trials being global standards and 20% deviants). At variance with the original method developed by Bekinschtein et al. (2009), who manipulated the pitch of the stimuli to create deviants, in Tzovara et al. (2015), regularity violations are obtained by varying stimulus duration. Using this paradigm, the authors investigated differential EEG responses to global standard versus global deviant sounds in 21 healthy controls and 24 comatose patients with anoxic-ischaemic coma. For each patient two EEG recordings were performed, the first under therapeutic hypothermia and the second in normothermic conditions within 48 h of coma onset. The age-matched control group was divided into an ‘active group’ (asked to count the number of global deviant series) and a ‘passive group’ (instructed to let their minds wander while listening to the sounds).

A multivariate decoding analysis (Tzovara et al., 2012, 2013) modelled voltage topographies of single-trial EEG activity for each recording. The model estimation was performed on a portion of the trials (the training data set) and then used to classify the category of sounds (standard/deviant) in a separate portion of trials (the validation data set). A significant decoding performance here implied a significant response to regularity violations. This approach does not focus on the detection of a specific EEG component but rather on a data-driven estimation of the most discriminative time-windows within the trial. It aims to overcome the difficulties inherent in the detection of a P300b component, which in patients may not be detectable at its classical scalp location and timing because: (i) clinical ERP recordings may be contaminated by physiological and environmental artefacts (especially those obtained in intensive care settings); (ii) patients in coma often show severe brain damage which may alter cortical ERP topography; and (iii) white matter impairments and cortical processing dysfunction may cause temporal delays and intertrial variability in patients in coma. Using their statistical decoding analysis, Tzovara and colleagues found a significant global discrimination for 10/24 patients in coma, five during hypothermia and seven during normothermia (two patients showed a significant effect in both experimental conditions). The average decoding performance was comparable in hypothermia and normothermia. All the patients showing a global effect were behaviourally diagnosed as unconscious (i.e. comatose) and their clinical condition was indistinguishable from that of the non-responsive patients. The authors also demonstrated that while the decoding performance per se was not predictive of a patient’s chances of survival, its progression in time was of prognostic value. Indeed, the majority (78%) of patients showing an improvement in global auditory discrimination in the second recording when compared to the first, subsequently showed a good outcome.

We remain somehow puzzled by the results obtained by Tzovara et al. (2015) in healthy controls. They report that the global discrimination in the active control group was significant in 4/11 controls (37%). This proportion seems rather low, as we would expect that the vast majority of healthy subjects instructed to pay attention to global violations of auditory regularities would show a significant response. Previous studies (Bekinschtein et al., 2009; Faugeras et al., 2012) indeed reported a global discrimination using classical ERP analysis in 100% of active healthy controls. Similar results were also reported by King et al. (2013) who, using a multivariate approach conceptually similar to that used by Tzovara et al. (2015), found significant decoding performance in 9/10 attentive healthy controls. It remains to be shown whether this discrepancy might be ascribable to the use of time deviants instead of pitch deviants or to other methodological differences.

The detection of auditory ERP global effects in comatose patients corroborates the hypothesis that such complex cerebral patterns can be detected even in the absence of consciousness. This finding seems in contrast to previous evidence that a global discrimination is subtended by the subjects’ awareness of the presented auditory patterns (Beckinschtein et al., 2009). It also contrasts with results obtained in post-comatose states where the percentage of patients presenting a global response increased depending on the level of consciousness (i.e. 14% in vegetative state/unresponsive wakefulness syndrome (Laureys et al., 2010), 31% in minimally conscious state and 52% in patients who recovered consciousness; King et al., 2013). The observation of Tzovara et al. that a progression of the global auditory discrimination between the first 2 days of coma seems linked to patients’ chances of survival, offers a possible reconciliation with the abovementioned findings. Tzovara et al. suggest that global discrimination relies on neural processes (and we would add, neural circuitries), which in patients with a negative clinical evolution are gradually impaired over time, so that while these processes and circuitries could still be preserved at the very onset of coma, they gradually deteriorate and eventually become totally impaired. Global discrimination ability could indeed be a marker of preservation of specific neural processes rather than a marker of consciousness. To verify this hypothesis a longitudinal study on cognitive auditory ERPs in coma survivors could be conducted, systematically probing patients’ global discrimination ability from coma onset and during the clinical evolution over the following weeks and months, be it vegetative state/unresponsive wakefulness syndrome, minimally conscious state, confusional state or full recovery of consciousness.