In parallel to event-related EEG potential recordings and active paradigms, it would be useful to develop more sensitive methods to detect recovery of consciousness that do not depend on the integrity of sensory and motor pathways nor on the subject's ability to comprehend or carry out instructions. Theoretical considerations (Tononi, 2004; Laureys, 2005; Dehaene et al., 2006; Tononi and Koch, 2008; Seth et al., 2008) as well as experimental data (Del Cul et al., 2007; Alkire et al., 2008) suggest that a basic requirement for consciousness is that multiple, specialized areas of the thalamocortical system can engage in rapid causal interactions (effective connectivity). One way to gauge effective connectivity among thalamocortical modules involves perturbing directly a subset of cortical neurons with transcranial magnetic stimulation (TMS) and recording the reaction of the rest of the brain with millisecond resolution by means of EEG (Ilmoniemi et al., 1997; Litvak et al., 2007; Morishima et al., 2009; Akaishi et al., 2010; Casali et al., 2010). When consciousness is present, the thalamocortical system should respond to TMS with a complex pattern of activation, involving different cortical areas at different times; conversely, it should react with a simple response that remains localized to the stimulated area if consciousness is reduced (Alkire et al., 2008; Massimini et al., 2009a). In a recent series of experiments these hypotheses were tested during wakefulness, deep sleep and anaesthesia. In healthy awake subjects, TMS induced a sustained EEG response involving the sequential activation of different brain areas and affecting much of the cortex (Massimini et al., 2005). In contrast, after loss of consciousness induced by general anaesthesia (Ferrarelli et al., 2010), TMS pulses invariably produced a simple response that remained localized to the site of stimulation, indicating a breakdown of effective interactions among thalamocortical modules. A similar breakdown of effective connectivity was observed during slow wave sleep early in the night (Massimini et al., 2005), when subjects report little or no conscious content upon awakening (Hobson et al., 2000). Importantly, during rapid eye movement sleep, when subjects are unresponsive to sensory stimuli and virtually paralysed but report vivid dreams upon awakening, the cortical response to TMS recovered its complexity and became similar to that observed during wakefulness (Massimini et al., 2010).

As in previous studies (Massimini et al., 2005, 2010; Ferrarelli et al., 2010), in order to probe the ability of distributed thalamocortical modules to interact, we stimulated a subset of cortical neurons with TMS and performed EEG source modelling to detect, on a millisecond time-scale, the chain of effects triggered in the rest of the brain by this initial perturbation. Compared to methods based on the observation of resting brain activity, this perturb-and-measure approach (Paus, 2005) readily dissociates functional connectivity (temporal correlations) from effective connectivity (causal interactions), which is defined as the ability of a subset of neurons to causally affect the activity of other groups of neurons (Friston, 2002; Lee et al., 2003). Recent studies have shown that by employing TMS/EEG and source modelling it is possible to detect patterns of effective connectivity that are generally predicted by main anatomical pathways (Ilmoniemi et al., 1997; Litvak et al., 2007; Morishima et al., 2009; Casali et al., 2010). On the other hand, since TMS tends to activate a large set of cortical axons in a way that is difficult to control fully (Wagner et al., 2007), this technique is more likely to provide a coarse rather than a fine-grained estimation of effective connectivity. In the present context, a broader estimation of effective connectivity may constitute an advantage, since theoretical works (Tononi, 2004; Tononi and Koch, 2008), experimental data (Maandag et al., 2007; Alkire et al., 2008; Shulman et al., 2009) and clinical evidence (Markowitsch and Kessler, 2000; Mataro et al., 2001; Schiff, 2010) suggest that consciousness depends not so much on some specific circuits, but rather on the capacity of distributed regions of the brain to interact through divergent cortico–cortical and cortico–thalamo–cortical connections. Indeed, as demonstrated by previous experiments, TMS/EEG measures of effective connectivity can distinguish readily between conditions in which consciousness is present (alert wakefulness, dreaming) (Massimini et al., 2005, 2010) and conditions in which consciousness is reduced, or lost (sleep and anaesthesia) (Massimini et al., 2005; Ferrarelli et al., 2010).

Figure 3 summarizes the results obtained after applying TMS in all 17 patients and shows that it is possible to discriminate reliably between a vegetative and minimally conscious state, at the single-subject level. Crucially, this discrimination was achieved in a way that is completely independent on the patient's ability to exchange information with the surrounding environment. The fact that TMS/EEG detected a clear-cut difference between vegetative state and minimally conscious state (unconsciousness versus low-level of consciousness) but not between minimally conscious state, emergence from minimally conscious state and locked-in syndrome (lower versus higher levels of consciousness) suggests that the availability of effective interactions among thalamocortical modules may be a critical mechanism that correlates closely with the presence/absence of a minimal level of consciousness. This aspect is particularly relevant if one considers that the most challenging task at the bedside is distinguishing between patients in a vegetative state and non-communicating patients in a minimally conscious state (Majerus et al., 2005). As an example, in the present work, TMS/EEG detected the resumption of rapid, effective intracortical interactions in the brain of a patient (Patient 15) who (during Session 2) had temporarily slipped back into a clinically vegetative state, possibly due to transient fluctuations in her ability to interact with the environment; this patient was reassessed clinically as minimally conscious state and then emerged from minimally conscious state.

Clearly, validating an objective marker of consciousness that can be applied to patients that are unable to interact with the external environment is challenging by definition, since, in these cases, there is no behavioural reference to assess the presence of consciousness. In an attempt to overcome this circularity, we have previously tested TMS/EEG measures in states in which consciousness is unambiguously present [alert wakefulness (Massimini et al., 2005), dreaming (Massimini et al., 2010)] or unambiguously reduced [early slow wave sleep (Massimini et al., 2005), general anaesthesia (Ferrarelli et al., 2010)]. Here, we demonstrate that TMS/EEG measures are reliable when they are applied to brain-injured patients with a stable clinical diagnosis (Group I) and that they are sensitive in detecting a clear-cut resurgence of cortical effective connectivity in the brains of individual patients who gradually recover consciousness and functional communication (Group II). In future works, the same approach should be further tested, in a back-and-forth process, both in definite and in ambiguous clinical conditions, such as the one of Patient 15. It will be equally important to directly compare the ability of TMS/EEG to discriminate between vegetative and minimally conscious states at the individual level with the diagnostic capacity of other neurophysiological methods, such as peripherally evoked potentials (Kotchoubey et al., 2005; Bekinschtein et al., 2009; Fischer et al., 2010; Boly et al., 2011) and long-term EEG recordings (Landsness et al., 2011). The lack of a direct comparison with other techniques represents a clear limitation of the present study and is due to logistical and time constraints (in each patient, we stimulated from two to four cortical sites) in the intensive care unit. For now, we can only compare the present results to the current literature and, in particular, to a number of works in which the mismatch negativity was evaluated systematically in patients in a vegetative state and patients in a minimally conscious state. Altogether, this body of literature suggests that, while the mismatch negativity may differ significantly between vegetative state and minimally conscious state at the group level, it does not discriminate reliably between these two conditions at the individual patient's level; in fact, this late component may be undetectable in a large proportion (up to 60%) of patients who are behaviourally in a minimally conscious state (Kotchoubey et al., 2005; Fischer et al., 2010; Holler et al., 2011). Since in the present study we found consistent TMS/EEG results across sites of stimulation, in future work it will be feasible to directly compare the EEG response to TMS of a single cortical area with a battery of sensory evoked potentials (N20, mismatch negativity and P3b) recorded in the same patient, on the same day. These joint measurements will be crucial to precisely quantify the relative diagnostic power of complementary neurophysiological techniques that may enter the routine evaluation of severely brain-injured patients. To this regard, the present experiments show that, like peripheral evoked potentials, TMS-evoked potentials can be recorded at the patient's bedside, in the intensive care unit. A technical disadvantage of TMS/EEG is that it requires a more complex set-up, which includes a TMS main unit, a TMS-compatible EEG amplifier and, ideally, a navigation system in order to precisely target TMS on the cerebral cortex. Navigating TMS based on prior anatomical knowledge (CT or MRI scan) may be especially important in the assessment of brain-injured patients for two reasons. First, because it allows avoiding obvious cortical lesions and stimulating the cortical surface at supra-threshold intensity (see ‘Materials and methods’ section and Casali et al., 2010) and second (and most important) because it ensures high test-retest reproducibility when TMS-evoked potentials are performed longitudinally (Lioumis et al., 2009; Casarotto et al., 2010). Hardware solutions aside, developing TMS/EEG towards routine clinical applications may require the implementation of a standard, fast data analysis procedure to calculate the spatial-temporal complexity of the cortical response to TMS.