Early psychological descriptions of consciousness closely associated it with attention and emphasised its role in serial processing as a ‘limited capacity channel’ [6]. Later historical models of consciousness identified conscious contents with working memory contents [7] and executive control [8].

The most prominent psychological theory of consciousness, ‘global workspace theory’ [9, 10], echoes these earlier perspectives. The theory proposes that there is an unconscious competitive process among local units, in order to win access to a ‘global workspace’ of widely available content. The global broadcasting of these contents to multiple receiving modules establishes their presence in consciousness. This theory, with elements both of working memory and attention, provides some explanation for the unified and serial nature of consciousness, but fails adequately to explain the phenomenology of awareness.

The question of the association between consciousness and attention has been a source of vigorous recent debate [11–14]. At the turn of the twenty-first century, it was widely assumed that attention and consciousness were at the very least very closely related [15, 16]. However, over the last decade, there have been claims that consciousness and attention are entirely dissociable and are, in fact, independent processes, with attention neither being necessary nor sufficient for consciousness [14].

Although such positions are entirely dependent on exact definitions of consciousness and attention, there is nevertheless good evidence that attentional effects can occur without the attended stimuli reaching consciousness [17–21]. For example, the response times to a visible target were modulated by the congruency of invisible prime locations that had just been attended to [19]. Although such studies do indeed demonstrate that attention can be effectively deployed on stimuli that are not consciously perceived, it does not necessarily follow that consciousness is dissociated from attention. It is possible, for instance, that attention towards an unconscious object increases the probability that the object will be consciously perceived. Such an interpretation would be consistent with other studies that have demonstrated that attention increases conscious detection rates [22] and enhances the contrast of a consciously perceived stimulus [23].

Evidence for the opposite situation, of consciousness without attention, is far weaker and more controversial. For instance, natural scene perception, or ‘gist’, where general semantic details of a visual scene are consciously available, even when presentation times are extremely brief [24], is used to argue that consciousness can occur without attention. The argument goes that if meaningful consciousness is possible with a presentation time of 150 ms, then consciousness does not require attention, since this timeframe is too short for attention to be established [14]. However, it seems entirely possible that selective attention, which requires about 300 ms to occur [25], can easily act on iconic memory, which persists for longer than the duration of stimulus presentation. In line with this, if attention is effectively withdrawn from a briefly presented visual scene, then visibility of the gist of the scene is also largely abolished [26, 27]. Indeed, for gist or any other claimed example of consciousness without top-down attention, either bottom-up attention is most likely present or, when experiments have actively removed attentional resources, consciousness for a given stimulus has also been reduced or removed [11]. Therefore, the weight of evidence suggests that while attention can occur without consciousness, the opposite is not the case.

Attention is therefore probably a necessary component of consciousness, acting as the gateway for conscious contents [1, 11, 28]. Many lines of evidence support the important role of attention for consciousness. Inattentional blindness is a well-studied paradigm where attention is drawn away from an otherwise striking and perfectly visible object (such as a gorilla walking across a scene), causing the subject to fail to consciously perceive it [29, 30]. Change blindness is a similar paradigm, where in its usual form there is an obvious change between two otherwise identical visual images, which alternate, interleaved by a brief blank screen. Again, if attention isn’t directed at this clear difference, the subject fails consciously to detect the change [31]. Both these paradigms demonstrate the surprising limits of conscious contents, where only a small subset of available objects can be simultaneously perceived, presumably constrained by short-term memory capacity limits and therefore only allowing a maximum of approximately 3–4 items [32]. However, these items are usually only consciously perceived as fully formed, high-level mental objects, integrated across modalities and carrying all relevant conceptual and linguistic content.

Further links between attention and consciousness are demonstrated by the neurological condition of hemispatial neglect, which usually results from damage to the right parietal or prefrontal cortex [33, 34]. Although classically assumed to be an attentional deficit, neglect is just as easily characterised as a reduction of awareness of one side of space, most frequently on the left. Neglect is neither a condition of sensory impairment per se nor is it limited to the visual modality. For instance, touch-based neglect has been reported [35].

In addition to links with attention and working memory, consciousness has been associated generally with effortful cognitive processing [36] and has been shown to be required for most complex or novel forms of thought, including understanding cause and effect, any nontrivial mathematical tasks, most logical operations, sequential information, as well as processing and acting on social knowledge [37].

In line with consciousness being intimately associated with complex information processing, it also appears to be heavily modulated by prior expectations [38, 39]. This view of consciousness as an attentionally gated, highly limited mental workspace suitable for fully processed mental items, where novel and complex processes are carried out, suggests a candidate evolutionary role for consciousness: it may provide a mental space where disparate forms of information can be compared and combined, in order to discover innovative solutions to novel or complex problems [1, 11]. In this way, more advanced strategies can be formulated to reach otherwise intractable biological goals and avoid otherwise probable dangers.

Pioneering early studies to examine the neural correlates of conscious vision used single-unit recording in monkeys as they performed a binocular rivalry task, where one image was presented to the left eye and a different image was presented to the right [40, 41]. Although these images didn’t change, the percept semi-regularly switched, as it does in humans. Logothetis and colleagues found that whereas only about 20 % of the primary and secondary visual cortex neurons tracked changing perceptions, almost all inferotemporal cortex neurons were excited by the percept itself. This leads to the suggestion that late ventral stream object recognition neurons were somehow responsible for consciousness. However, since this technique only sampled the visual stream itself, it was unclear at the time whether other brain regions were equally as involved in conscious processes. Indeed, in an extension of this work, again using monkey single-unit recording and a version of binocular rivalry, it was found that an equally high proportion of lateral prefrontal cortex neurons responded to the percept itself, indicating that this region is also heavily involved in consciousness [42]. This result highlights the predominant finding in consciousness research, which is the involvement in conscious processing not just of sensory specific regions but also of association cortex. From many different paradigms, the most robust observation in the literature has been the link between consciousness and the lateral prefrontal cortex, along with the posterior parietal cortex, often referred to together as the prefrontal parietal network (PPN). Human fMRI studies that have used bistable percepts, such as binocular rivalry or ambiguous figures (e.g. the Necker cube), have consistently observed PPN activations when percepts have switched from one stimulus viewpoint to the other [28, 43, 44].

Another way of examining the neural correlates of conscious content is to manipulate conscious detection of the same type of stimuli, for instance, by masking. In one study, using visual words, undetected stimuli were largely associated with the visual word form area. However, when the stimuli were visible, visual word form area activity was significantly raised, and new activity was observed in the PPN [45]. Analogous results have been observed in the auditory [46] and tactile domains [47], where large-scale PPN activity, usually bilaterally, is only associated with awareness of the stimulus. One issue with such studies is that they could be confounding subjective visibility with objective performance. To address this issue, Lau and Passingham used a metacontrast masking study, where objective performance was matched, whether the stimulus was visible or not [48]. Using this paradigm, they observed that only the dorsolateral prefrontal cortex (DLPFC) activity was linked with visible stimuli. Although these studies fail to provide a causal role for the PPN in conscious processing, there is a wealth of data from other sources to indicate that the PPN is indeed necessary for consciousness. For instance, unilateral prefrontal cortex lesion patients were less likely than controls to consciously detect briefly presented numbers, when controlling for objective performance [49]. In addition, a rare bilateral prefrontal lesion patient was reported to be awake, but almost completely unresponsive to stimuli [50].

Deficits in consciousness have also been reported following posterior parietal cortex lesions. For instance, bilateral posterior parietal patients showed reduced subjective experience in a recollection task, when compared with controls, despite matched objective performance [51]. The PPN has also been shown to be necessary for consciousness in normal subjects, via transcranial magnetic stimulation (TMS). For instance, TMS either applied to the DLPFC or the posterior parietal cortex impairs conscious change detection [52, 53]. In addition, Rounis and colleagues reported impaired visibility ratings, when controlling for objective performance, after TMS was applied to the DLPFC [54]. In a similar study, which also controlled for objective performance, Fleming and colleagues found that subjective ratings of awareness of the stimulus correlated with the anterior lateral prefrontal grey matter density [55]. In a follow-up study, the authors demonstrated a correlation between activity in this region and reported confidence [56].

Structural changes in the parietal cortex have also been linked with consciousness. For instance, the individual differences in perceptual alternation rates for bistable moving images correlated with posterior parietal grey matter density, cortical thickness, and white matter integrity [57]. The links between consciousness level and the PPN are equally robust. Bilateral damage to parietal and prefrontal white matter is associated with coma or VS [58]. Furthermore, disrupted backward connectivity between the prefrontal and temporal cortices was found in VS, as compared with controls [59]. Similarly, intact functional connectivity between the prefrontal cortex and the thalamus is predictive of recovery from vegetative state [60]. If the PPN is critically involved in conscious processes, one would also assume that PPN activity would reduce as conscious levels drop. And this has indeed been observed in a range of studies, for instance, comparing normal awake subjects with those in a slow-wave sleep [61] or sedated following the administration of general anaesthesia [62, 63]. In addition, VS patients have specific reductions in PPN activity compared with controls [64].

The DoC field, in addition to the PPN, also highlights the critical role of the thalamus. For instance, thalamic abnormalities are the most common neurophysiological dysfunction in vegetative state patients [65]. Reinforcing this point, Fernandez-Espejo and colleagues found that lack of white matter integrity in the thalamus was the main diagnostic marker for vegetative state, compared with minimally conscious state [66]. In addition, Schiff and colleagues showed that thalamic deep brain stimulation in a minimally conscious state patient significantly improved conscious levels and general functional outcomes [67].

The emerging picture from all these neurophysiological studies is that content-specific regions provide the conscious details (e.g. the inferotemporal cortex for visual objects) but that for all manner of conscious contents, the PPN in concert with the thalamus are required for consciousness to occur. Why should these regions be so closely associated with consciousness? The PPN is generally implicated in novel and complex processing [11, 68] and is particularly robustly activated when performing high-level strategic processing, even compared with more difficult versions of the same task [69–71]. The thalamus is thought to act as a multisensory information relay hub and therefore could be seen to facilitate the formation of multimodal mental objects, primarily within the PPN. These functional roles are quite consistent with the view outlined in the psychological section above, where a potential evolutionary purpose for consciousness might be to discover innovative solutions to otherwise intractable problems.

Given the relatively consistent and mature empirical picture, the main current theories of consciousness are broadly in agreement over certain key features of how consciousness could be physically instantiated. One assumption common to theories is that feedforward activity alone is not sufficient for consciousness to occur; instead, recurrent processing is required, preferably in a global way that encompasses the PPN [72]. Current theories also emphasise that network architecture is key to supporting consciousness, with central, highly interconnected regions better able to generate consciousness and those more peripheral, isolated areas only providing a support role at best, possibly by supplying specific conscious contents [28, 73]. This view is consistent with various anatomical features relevant to consciousness, such as that the PPN is a substantial portion of a particularly densely interconnected ‘inner core’ of brain networks, as measured by monkey white matter tracing [74]. In addition, the cerebellum, despite having considerably more neurons than the entire cortex, is not necessary for consciousness, possibly because it is largely composed of small independent, isolated modules [73].

The two most popular current physical models of consciousness, global neuronal workspace theory (GNW) and integrated information theory (IIT), also both emphasise two features of a network that are reflective of consciousness: global integration of information in a network and the ability of that network to represent a large repertoire of different functional states [75, 28, 76]. GNW diverges from IIT, however, in being more closely rooted to the human neurophysiological data. Consequently, in GNW, which is derived from the global workspace theory of Baars [10], domain-specific local processing is the source of conscious contents, which only enter provide conscious access when they are integrated into a global workspace, largely involving the PPN. Under this theory, approximately 300 ms is needed to build up an ‘ignition’ of widespread conscious activity in these cortical areas, in an all or none fashion, with long-range cortical synchronisation occurring in the high gamma band [28, 77].

IIT is a more ambitious, and more controversial, theory that more generally relates network architecture and activity to consciousness. The network can just as easily be a silicon one, if the architecture is appropriate, as a biological one. IIT doesn’t just associate integrated information with quantity of consciousness, but equates it to the amount of experience. Integrated information under IIT is mathematically formalised and corresponds to that information specified by an entire system (such as a brain) that is more than the sum of its constituent parts. Each experience is uniquely differentiated from all others, and the amount of differentiation possible in a system is also related to the amount of consciousness it can support. Under the latest version of the theory (3.0), new emphasis has been given to conceptual structure and its role in consciousness [76, 73]. Although the specific details of IIT are disputed [78, 79], and its mathematical formalism is unsuitable for practical calculations, many researchers are sympathetic to its strong emphasis on integrated information as a hallmark of consciousness. This view sits well with the behavioural links between consciousness and complex processing, as well as the close association between consciousness and the PPN, given that the PPN is associated with high-level information-based tasks, such as chunking [71, 11].

Active, task-based, DoC imaging paradigms have pioneered research-based assessment of DoC patients, where behavioural clinical assessment alone has proven very unreliable. For instance, Monti and colleagues showed that a patient previously diagnosed as VS could not only perform mental imagery tasks but could also use such tasks to communicate, via fMRI-based brain activity [80, 81]. However, such active tasks rely on the willing participation of the patient, which cannot always be guaranteed in such an ill patient group. Passive paradigms, where the patient has no task to perform, sidestep this issue and may ultimately be more useful clinically. Some of the most promising latest developments in using passive imaging paradigms to assess DoC patients have arisen out of a strong association between consciousness and complex information processing and have been inspired by information-based consciousness theories, such as IIT [5]. For instance, Chennu and colleagues analysed EEG signals in DoC patients and controls using graph theory and found that the patients’ networks were impaired at long-range integrated information [82]. Another recent study, based directly on IIT, examined the EEG response to TMS pulses and converted this response signature to a measure of informational complexity, a ‘perturbational complexity index’ (PCI), by analysing how compressible the EEG signal was [5]. This PCI measure was able to discriminate, at the single subject level, between normal wakefulness, sleep, and general anaesthesia, as well as between different DoC patient groups.