Signatures of access to consciousness (1)

Prefrontal, parietal and cingulate network. The contrast between conscious and non-conscious stimuli on MRI reveals a distributed yet restricted network that isolates the mostly bilateral regions of dorsolateral prefrontal, inferior parietal and anterior cingulate cortex (Dehaene et al., 2001). Lau and Passingham (2006) introduce an important control by equalizing cognitive performance in both conditions, relying on the possibility of outperforming chance without awareness. They observe a specific correlation of access to consciousness with activity in the left dorsolateral prefrontal cortex. Similarly, in a recent experiment, Van Gaal et al (2010) contrasted conscious and non-conscious performance on a go-no go inhibition task. Small specialized regions of frontal cortex (pre-SMA, anterior insula) are likely to be activated non-consciously, but awareness is accompanied by an amplification of these networks and the extension of activation to large bilateral prefrontal and inferior parietal regions.

Decoding conscious content. John-Dylan Haynes (2009) points out that, in order to isolate the cerebral mechanisms of consciousness in brain imaging, it is not enough to detect a signal that is present during conscious trials and absent during non-conscious trials. We also need to show that this signal constitutes a neural code of the subject's state of consciousness, such that we can, in principle, design a decoding algorithm which, starting from the observed activation, precisely reconstructs the subject's state of consciousness. The observed code must (1) present the same degree of invariance as the subject's conscious content, (2) preserve subjective similarity relations between conscious contents, (3) not encode physical differences that the subject does not perceive, but (4) encode the perceived object, even when it is an illusory perception. Only a few recent MRI experiments meet this demanding program in the visual modality (Schurger, Pereira, Treisman, & Cohen, 2009; Sterzer, Haynes, & Rees, 2008; Williams, Dang, & Kanwisher, 2007). Their results suggest that the earliest retinotopic areas do not always covary with subjective content, but that such covariation is observed in more advanced areas such as the lateral occipital cortex or the fusiform region. The results are compatible with the existence of a neural code partially activated by invisible stimuli, but amplified, spatially extended and made more reproducible from trial to trial during access to consciousness. The absence of whole-brain analysis, however, limits the significance of this work.

In the auditory modality, an interesting MRI experiment focuses on the McGurk illusion, in which the perception of a syllable is influenced by the vision of a speaker (Hasson, Skipper, Nusbaum & Small, 2007). Remarkably, the brain activity that correlates with perceived content is not derived from early auditory regions, but from more advanced, multimodal processing that jointly mobilizes temporal and prefrontal regions.

Time course of access to consciousness. Evoked potentials, magnetoencephalography and electrophysiology in animals enable us to pinpoint the point of temporal divergence between conscious and non-conscious stimuli. From the earliest work, it has been clear that the first stages of cortical processing can be perfectly preserved during non-conscious visual processing (work by Schiller and Chorover and Bridgeman). In animals, in primary visual cortex as well as in inferotemporal cortex and oculogyric frontal fields, late amplification of neuronal discharges is associated with stimulus reportability and is interrupted during masking (Kovacs, Vogels, & Orban, 1995; Super, Spekreijse, & Lamme, 2001; Thompson & Schall, 1999). However, only human studies can both record whole-brain electro-magnetic activity and correlate it with an authentic subjective report of visibility rather than a behavioral measure. Evoked potentials isolate a cascade of successive stages of sensory evidence accumulation, but only the late activity, beyond 270 ms (P3 wave), shows a response profile closely correlated with that of the perception reported by subjects (Del Cul, Baillet, & Dehaene 2007). Cortical source modeling shows that the late non-linearity of the P3 wave is associated with sudden distributed activation, particularly of the inferior frontal region, in conjunction with a second wave of late activity in visual areas. It should be noted that, even when the stimulus is unilateral, when consciously perceived, the activity it evokes invades a wide range of bilateral cortical regions. These results replicate very precisely earlier work carried out on the attentional blink experimental condition, which also has the advantage of presenting rigorously identical stimuli but perceived sometimes consciously, sometimes unconsciously (Sergent, Baillet & Dehaene, 2005). In this study, sensory activation is identical up to ~ 200 ms, then abruptly diverges in a manner compatible with the hypothesis of a sudden, self-amplifying activation of distant prefrontal and parietal regions, as well as reactivation of the sensory cortices initially activated. The sudden and virtually all-or-nothing nature of the activity evoked by a conscious stimulus is confirmed by recordings of single neurons in the medial temporal cortex of people with epilepsy (Quiroga, Mukamel, Isham, Malach & Fried, 2008).

Four electrophysiological signatures. In recent work from the laboratory, Gaillard et al (2009) compare the course of intracranial recordings of local potentials in response to a conscious or non-conscious word in ten epileptic patients. Their results show that, during masking, a whole series of early stages continue to unfold without consciousness. At around 270-300 ms, a strong correlation begins to emerge between brain activity (P3) and stimulus visibility. In the same time window, the amplification of oscillations in the gamma band (> 30 Hz), the synchronization of distant regions in the beta band (13-30 Hz), and the establishment of a bidirectional causal dialogue between these regions, measurable by Granger causality, all appear to be signatures of a synchronous and distributed cortical state that appears exclusively when the threshold of consciousness is crossed (see also Fisch et al., 2009; Gross et al., 2004). Access to consciousness is therefore a late, global, sudden and synchronous cortical phenomenon. Particularly interesting is the fact that almost 70% of recorded cortical sites show signal modulation linked to the presence of a conscious word (versus 24.4% in the non-conscious case). This observation is compatible with the hypothesis that access to consciousness is accompanied by a diffusion of information to very large sectors of the cortex(broadcasting proposed by Baars 1989, Dehaene and Changeux, 1998).