Main

The search for the neural correlate (or correlates) of consciousness1 has resulted in two important distinctions being made. The first is that the neural correlates of the level of consciousness (for example, awake, asleep, attentive or drowsy) should be distinguished from the neural correlates of specific phenomenal content (such as a green apple versus an orange). This distinction reflects our everyday experience of the phenomenal distinction between being conscious (as opposed to being unconscious) and being conscious of X (as opposed to not being conscious of X). The level of consciousness can be thought of as an enabling factor (like neuronal arousal mediated by the ASCENDING RETICULAR ACTIVATING SYSTEM2, or even heartbeat) that is required for awareness but does not directly reflect specific conscious experiences.

The second distinction is between neural activity that correlates directly with a single conscious experience and activity that reflects unconscious perception (or action) associated with that experience. This reflects the common observation that many of the cognitive processes that underpin our conscious experience are inaccessible to introspection. Similarly, stimuli that bypass awareness can influence behaviour not only in pathological conditions (such as NEGLECT3), but also in normal subjects4. To identify neural correlates of consciousness, investigators must use experimental designs that dissociate the contributions of conscious and unconscious processes to sensory experience, typically by manipulating them independently5.

A neural correlate of consciousness is a specific pattern of brain activity that correlates with particular conscious experiences. It is not clear how any physical process, such as neural activity, can give rise to a subjective phenomenon such as awareness. Indeed, even the possibility of such a causal relationship remains controversial among philosophers6,7,8. The search for the neural correlates of consciousness is therefore an empirical investigation that remains initially neutral on issues of causality, seeking instead to identify and characterize patterns of neural activity that specifically correlate with conscious experience, rather than with unconscious perception or action.

In this review, we focus on the neural correlates of awareness of specific phenomenal content in the human visual system9,10. The visual system is the best-studied sensory system, and a wealth of anatomical and physiological data from non-human primates complements psychophysical, electrophysiological and functional neuroimaging investigations in humans.

Awareness and primary visual cortex

When a stimulus is presented visually, successive hierarchical levels of the visual system are rapidly activated through a cascade of feedforward connections (see below for a discussion of feedback connections). Anatomically, the primary visual cortex, V1, is the lowest cortical visual area. Patients with damage to V1 typically report no conscious visual experience in their blind visual field, even though some residual capacity for unconscious visual discrimination might remain — a phenomenon known as BLINDSIGHT11,12. This shows that V1 is necessary for normal visual perception. However, by itself, it does not reveal whether activity in V1 gives rise directly to conscious visual experience, or whether further processing in extrastriate areas is necessary. This situation is similar to that of the retina. Activity in retinal ganglion cells is necessary for normal vision. But most observers would agree that the receptive field properties of these cells are too different from the properties of visual perception for them to give rise directly to conscious vision.

In 1995, Crick and Koch13 proposed that a key function of consciousness is to produce the best current interpretation of the visual scene in a compact form and to make this information available to the planning stages of the brain. As the cognitive processes involved are associated with the frontal lobes14, and as V1 does not project directly to premotor or prefrontal areas, Crick and Koch surmised that the activity of V1 neurons, although necessary for most forms of normal vision, does not constitute part of the neural correlates of consciousness.

Psychophysical evidence is broadly consistent with the proposal that information represented in V1 is not available to consciousness. For example, we cannot perform utrocular discrimination (identifying the eye to which a stimulus is presented), even though V1 contains neurons that receive input from one eye only, and so represent this information15. Similarly, orientation-dependent aftereffects, which are thought to arise from the activity of cells in the primary visual cortex, can be induced by stimuli that cannot be consciously perceived. For example, the orientation of a grating is rendered invisible and indiscriminable by crowding when it is presented with four similar gratings positioned above and below it. However, adaptation to such a grating results in an orientation-dependent aftereffect that is indistinguishable from that produced when it is presented alone with its orientation clearly (and consciously) perceived16. In addition, very high-frequency gratings that are perceptually indistinguishable from a uniform field nevertheless produce robust orientation-dependent aftereffects17. The presence of aftereffects from stimuli that are not consciously perceived18 indicates that stimulus properties must be represented outside awareness. So, activity in V1 often does not correlate with awareness.

These inferences are complemented by direct electrophysiological measurements of neuronal activity in V1 in macaque monkeys that show that perceptual report (presumably reflecting awareness) and neuronal activity can change completely independently. For example, disparity-selective cells in V1 differentiate between local depth cues even when these do not give rise to an overall depth percept19. Similarly, V1 cells can respond in an identical manner to two depth cues that yield very different global depth percepts19. This indicates that single V1 cells that are sensitive to binocular disparity might represent a first crucial stage for generation of stereo cues, but that conscious perception of depth is probably generated further upstream. A similar lack of correlation between perceptual experience and the response properties of V1 cells can be seen during blinks. Blinking produces profound and frequent interruptions in retinal stimulation, but barely impinges on our visual experience (which remains continuous). The responses of cells in the supragranular layers of V1 to visual stimulation decrease substantially during reflex blinks20, and this reduction is significantly more pronounced than when there is an equally long gap in visual stimulation. Similarly, MICROSACCADES, which occur continually, do not alter our impression of the world, but are reflected in the activity of neurons in V1 (Ref. 21). Finally, activity in striate cortex neurons faithfully follows rapid chromatic flickering of a coloured grating even when the flickering is too fast for humans to resolve the individual colours22. So, individual cells in primary visual cortex show response patterns that are quite unlike conscious visual experience.

Changes in conscious perception can also occur in the absence of concomitant variations in neuronal activity in V1. Binocular rivalry has proved to be a powerful paradigm with which to study the neural correlates of conscious visual experience23. When two dissimilar images are presented to corresponding areas of the two eyes, they compete for perceptual dominance. Each image is visible in turn for a few seconds, while the other is suppressed. Perceptual transitions between each monocular view occur without any corresponding change in the physical stimulus, allowing experimental dissociation of the neural correlates of consciousness from changes that are attributable to stimulus characteristics. Monkeys can be trained to report their percept during rivalry, and their behaviour is similar to that of humans24. In monkeys, the firing of most cells in and around V1 primarily reflects stimulus properties rather than the conscious percept reported by the animal25, which is consistent with the evidence presented above. However, several neuroimaging studies in humans have presented evidence that argues for a stronger role of V1 in binocular rivalry and hence, by implication, visual awareness. Averaged over the whole of human V1, there are reliable fluctuations in blood oxygen level dependent (BOLD) contrast activity when the conscious percept changes26. These fluctuations are about half the size of those evoked by non-rivalrous stimuli, and even larger in the part of V1 that is associated with the blind-spot representation27 (which contains cells that receive only monocular input from the other eye). The reasons for the discrepancy between human and monkey data (and between the blind-spot representation and the rest of V1) are not yet clear. Areas of human V1 such as the blind-spot representation might have special properties or, alternatively, the discrepancy might lie in the relationship between single-cell activity and BOLD contrast measurements (Box 1).

Other studies in humans show a poor correlation between V1 activity and visual awareness. During dreaming, intensely visual experiences are associated with suppression of activity in V1 (Box 2). Activity in V1 in the absence of awareness is observed in patients with damage to the right parietal cortex and VISUAL EXTINCTION28,29,30. Patients with visual extinction show deficient awareness for contralesional visual stimuli, particularly when a competing stimulus is also present ipsilesionally3. An extinguished left visual field stimulus of which the patient is unaware nevertheless evokes activity in V1 that is very similar to that evoked by a stimulus in the left visual field that is seen28,29,31 (Fig. 1). This indicates that the presence of activity in primary visual cortex per se is insufficient to support visual awareness, at least after damage to the right parietal cortex.

Figure 1: Activity in V1 does not correlate with awareness.
figure 1

a | A sagittal slice through the right hemisphere including the calcarine sulcus of patient G.K., who suffered a right parietal stroke that caused visual extinction. The image shows the locus in V1 that is activated by an unseen and extinguished left visual field stimulus (as measured by blood oxygen level dependent (BOLD) functional magnetic resonance imaging (fMRI), superimposed). b | Peristimulus time histogram of BOLD contrast activity evoked by three different types of visual stimulus, taken from the activated right V1 locus in panel a. Left visual field stimuli that are either seen (unilateral left trials) or unseen (bilateral extinguished trials) evoke similar patterns of activity, whereas a right visual field stimulus that is seen (unilateral right) results in little positive response in right V1. c | An axial image through V1 of another patient with right parietal stroke and visual extinction. Superimposed on the image is the BOLD contrast activity evoked in right V1 by an unseen extinguished left visual field stimulus. d | Peristimulus time histogram of BOLD contrast activity evoked by four types of visual stimulation, taken from the activated right V1 locus in c. Seen bilateral, unseen and extinguished bilateral, and seen left unilateral stimulation all produce similar patterns of activity, whereas seen right unilateral stimulation evokes little response in ispilateral visual cortex. Parts a and b modified with permission from Ref. 31 © 2000 Oxford University Press. Parts c and d modified with permission from Ref. 29 © 2001 National Academy of Sciences, USA.

Full size image

We conclude that the current data are compatible with the hypothesis that neurons in V1, although necessary for normal visual experience, do not give rise, by themselves, to conscious visual perception — in other words, activity in V1 cells is not sufficient to explain conscious perception.

Awareness and ventral visual cortex

Beyond V1, the primate visual system is organized in a distributed fashion, with different aspects of the visual scene being analysed in different cortical areas30. These areas are densely interconnected in non-human primates, although in humans the equivalent anatomical connectivity is less well established (Fig. 2). On the basis of anatomical criteria, different areas form a rough hierarchy that fits well with the physiological hierarchy in receptive field size and complexity of tuning properties in these areas. Damage to cortical areas that contain neurons that are tuned to specific features of the visual environment leads to a corresponding deficit in consciousness of these features. So, bilateral damage to the human homologue of area V5/MT (middle temporal area), which contains neurons that respond specifically to moving visual stimuli, leads to AKINETOPSIA32. Similarly, damage to colour-selective areas in human ventral occipitotemporal cortex33 can result in ACHROMATOPSIA34. Finally, damage to areas of the fusiform gyrus that contain face-responsive neurons (the 'fusiform face area' or FFA35,36) can cause PROSOPAGNOSIA37. In contrast to damage to V1, which gives rise to a general deficit in awareness for all stimuli presented in a particular region of space, damage to ventral visual cortex causes specific deficits in awareness for particular types of stimulus or object feature. These findings indicate that activity in a functionally specialized cortical area — termed an 'essential node' by Zeki38 — is required to evoke consciousness of the attribute that is analysed in that area.

Figure 2: Anatomical hierarchy of visual areas in human and non-human primate.
figure 2

a | A summary of the layout and demonstrated anatomical connectivity of macaque visual cortex. Thirty-two visual cortical areas, two subcortical visual stages and several non-visual areas are shown, connected by 187 anatomically demonstrated links, most of which are reciprocal. Modified with permission from Ref. 30 © 1991 Oxford University Press. b | Identified visual areas, demonstrated connectivity and putative hierarchical organization of human visual cortex. There has been little direct study of the pattern of connections between human visual areas, but several characteristics of the macaque model (such as selective connections within V1 and between V1 and V2, plus the range of backprojections from higher-order visual areas) are present. Modified with permission from Ref. 125 © 1997 Plenum Press. M, blob and i-blob are subdivisions of V1, characterized by cytochrome oxidase staining. a-stripe, b-stripe and i-stripe are subdivisions of V2, identified using cytochrome oxidase staining. AIT, anterior inferotemporal cortex; BA, Brodmann area; CIT, central inferotemporal cortex; d, dorsal; DP, dorsal prelunate area; ER, entorhinal cortex; FEF, frontal eye fields; FST, floor of superior temporal cortex; HC, hippocampus; LGN, lateral geniculate nucleus; LIP, lateral intraparietal area; M, magnocellular regions; MDP, mediodorsal parietal area; MIP, medial intraparietal area; MSTd, dorsal part of the medial superior parietal area; MSTi, inferior part of the medial superior parietal area; MT, middle temporal cortex (visual area 5); P, parvocellular regions; P-B, parvo-blob; P-I, parvo-interblob; PIP, posterior intraparietal area; PIT, posterior inferotemporal cortex; PO, parieto-occipital area (visual area 6); RGC, retinal ganglion cells; STPa, anterior superior temporal polysensory cortex; STPp, posterior temporal polysensory cotex; TF–TH, temporal areas; v, ventral; V1–V4t, visual areas; VIP, ventral intraparietal area; VOT, visual occipitotemporal cortex; VP, ventroposterior visual area.

Full size image

Data from neuroimaging studies are consistent with this idea. Patients with schizophrenia show activity in modality-specific cortex during visual or auditory hallucinations39, even though no external stimulus is presented. During visual hallucinations that have a specific form, activity in visual cortex reflects the perceived content of the hallucinatory episode40. In normal subjects, contingent aftereffects based on colour41,42,43 or motion44,45,46 result in activation of corresponding functionally specialized cortices, and the time course of such activation reflects reported experience44. Perception of illusory47 or implied48 motion in a static visual stimulus results in activation of area V5/MT, whereas perception of illusory contours activates EXTRASTRIATE CORTEX49,50,51. Areas that are associated with processing the meaning of flashed words are activated when the words are consciously perceived at the focus of attention, but not when attention is withdrawn and the words are no longer consciously perceived52. In these experiments, the subjects' phenomenal experience changes without corresponding changes in the physical stimulus5. Concomitant with these changes in perceptual experience, altered activity is observed in areas of the brain that contain neurons with stimulus specificities that encompass the attribute being represented in consciousness.

Binocular rivalry and the related phenomenon of FLASH SUPPRESSION53 have also been used to probe the involvement of areas beyond V1 in visual awareness. Classic studies in monkeys have shown that the proportion of neurons with activity that reflects the monkey's perceptual report (rather than the stimulus properties) increases from V1 along the ventral visual pathway54. In monkey inferior temporal cortex, most of the relevant neuronal responses reflect conscious perception rather than the retinal stimulus25. These data are compatible with more recent functional magnetic resonance imaging (fMRI) observations in humans experiencing binocular rivalry26,55,56 (Fig. 3). During rivalry, fMRI responses that are recorded in the human FFA to face stimuli (and in the parahippocampal gyrus to images of places) are large and equal in magnitude to responses that are evoked by non-rivalrous stimuli56.

Figure 3: Single-neuron activity in human temporal cortex correlates with awareness.
figure 3

a,b | Electrode placement. The trajectory of an electrode placed in the hippocampus is depicted in axial (a) and coronal (detail, b) structural magnetic resonance images. Post-operative computerized tomography (CT) and magnetic resonance imaging (MRI) were used to confirm the location of the electrode. The CT was co-registered with MRI structural information for anatomical verification. The distal end of the electrode included platinum–iridium microwires from which single neurons were recorded. The microwires extended about 4 mm from the tip, lying on a cone with an opening angle of less than 45°. c | Individual responses of a single neuron during vision. Two images were shown separately for 1 s each, with five repetitions per image, indicated by horizontal green and blue bars. After each picture, subjects pressed a button to indicate whether or not the picture was a human face. The continuous red line shows the spike density function. d | After ten visual presentations, subjects closed their eyes and imagined one picture on hearing a high tone and the other picture on hearing a low tone. Tones were alternated every 3 s. Data shown are from the same neuron during visual imagery. This neuron showed a similar pattern of firing during visual presentation and visual imagery, with increased firing rate in response to the image of the ball, but not that of a human face. Modified with permission from Refs 57,59 © 2000 Macmillan Magazines Ltd.

Full size image

The activity of single neurons can be recorded in even more anterior regions of the medial temporal lobe in epileptic patients in whom electrodes have been implanted for presurgical mapping. Such rare electrophysiological investigations in humans have found many neurons that are tuned for specific categories of visual stimulus57. Most of these neurons fire selectively when their preferred stimulus is perceived — but not when it is perceptually suppressed (and invisible) — during binocular rivalry or flash suppression58. Furthermore, single neurons that show category-specific responses to visual stimulation are activated to a similar degree during visual imagery or passive presentation of a visual stimulus59 (Fig. 4). These single-cell findings complement neuroimaging studies that show activation of common brain areas during both visual processing and recall during imagery of specific types of stimulus60,61. Moreover, they indicate that activity at the level of neuronal populations, as seen with fMRI in humans, might be reflected at the level of single neurons (Box 1).

Figure 4: Neural correlates of awareness during binocular rivalry.
figure 4

a | Stimuli that elicit binocular rivalry consist of dissimilar images that are presented separately to each eye (the stimulus shown is used in conjunction with red/green filters to achieve this). Subjects typically report spontaneous perceptual alternations between each monocular view every few seconds, as illustrated below (percept). b | Neuroimaging studies of binocular rivalry in humans have typically compared the activity associated with switches between alternate percepts during binocular rivalry with that associated with yoked physical alternations between monocularly presented stimuli, using the temporal sequence derived from the perceptual report of an earlier period of binocular rivalry (yoked stimulus). c | Activity, assessed by magnetic resonance imaging (MRI), in the fusiform face area (FFA) and parahippocampal place area (PPA) that is associated with perceptual alternations between the rivalrous house and face stimuli shown in a. When a switch occurs during rivalry from house to face, activity in the FFA increases and that in the PPA decreases; the reverse pattern is seen for switches from face to house (top panels). So, activity in these functionally specialized areas of visual cortex closely correlates with the content of reported perceptual experience. The strength of these modulations of activity is very similar to that produced by physical alternation between stimuli (bottom panels). Modified with permission from Ref. 56 © 1998 Elsevier Science. d | Similar patterns of activity can be seen when activity is recorded from single neurons in monkey area V4 during binocular rivalry in which the rivalrous stimuli have different orientations, one of which is the preferred orientation of the cell. The upper panel shows activity recorded from a cell that responded strongly when the monkey reported perception of the preferred orientation (preferred), but not when the non-preferred orientation was dominant (null). The lower panel shows a neuron that fired strongly when its preferred orientation was suppressed (null), but not when its preferred orientation was perceived (preferred). Modified with permission from Ref. 54 © 1996 Macmillan Magazines Ltd.

Full size image

The evidence reviewed above indicates that the spiking activity of individual neurons in ventral visual cortex, in contrast to those in V1, generally correlates well with the visual contents of consciousness. However, the correlation is not perfect and other studies using fMRI in humans indicate that activity in ventral visual cortex might not always be correlated with visual awareness (Fig. 1). Stimulation of the blind hemifield of patients with damage to V1 results in activation of extrastriate cortex in the ventral visual pathway by the unseen visual stimulus62,63. In visual extinction resulting from damage to the right parietal cortex, an unseen and extinguished face stimulus can evoke face-specific activity in extrastriate visual cortex and more anteriorly in the FFA28,29,31. These studies in patients with cortical damage are complemented by investigations of stimulus masking in normal subjects. Awareness of a single word is profoundly degraded if it is presented very briefly and followed by a pattern mask. When subjects are unable to report the presence or absence of a masked word, it can nevertheless evoke activity in extrastriate visual cortex and the fusiform gyrus, leading to identity-specific priming, independently of the physical characteristics of the stimulus64. Similarly, semantically anomalous words presented during the ATTENTIONAL BLINK produce an N400 EVOKED POTENTIAL (thought to have its basis in processing in the ventral visual pathway65), even though they do not reach awareness66.

So, although single-cell activity at some levels of the ventral visual pathway correlates strongly with conscious vision, evidence from fMRI indicates that the presence of some level of BOLD activity by itself might not be sufficient to produce a conscious experience. BOLD contrast activity evoked by unseen stimuli in ventral visual cortex is typically much smaller than for seen stimuli. For example, there is approximately a tenfold difference in the level of activation in ventral visual cortex between masked words that are consciously perceived and those that are not64. It is not yet clear whether this is because neural activity has to exceed a certain threshold to reach consciousness, or because consciousness is related in a continually graded way to activation. For example, the observation that object recognition performance correlates strongly with the overall level of activity in occipitotemporal visual cortex67 indicates that there is a graded relationship between brain activity and visual awareness. Resolving this issue will be an important focus for future research, and depends in part on understanding the relationship between BOLD contrast and neuronal activity (Box 1).

We conclude that the current data show that activity in functionally specialized regions of ventral visual cortex is necessary for conscious experience of specific features, and that there is generally a strong correlation between activity in such areas and conscious experience. However, as with V1, although activity in ventral visual cortex is necessary, it might not be sufficient for conscious visual experience.

The timing of neural activity

The data reviewed above have addressed the hypothesis that there might be anatomically localized neurons or neural pathways with a level of spiking activity that mediates awareness. However, the overall level of spiking in a specific area need not be the relevant neural correlate. Another hypothesis is that awareness might be associated with specific types of neural activity, such as oscillatory or synchronized discharges. These two hypotheses are not mutually exclusive, and a combination of synchronous activity and neural spike rates might be an important attribute of the neural correlates of consciousness.

The surface electroencephalogram (EEG) is recorded from the scalp and reflects electrical activity in the human brain. It shows complex patterns of activity in different frequency ranges68. The dominant frequency components change according to the level of consciousness (Box 2). It has been suggested that synchronized oscillations in the high-frequency range (40–150 Hz) might underlie feature integration (and so be a candidate solution to the so-called BINDING PROBLEM69) and form a substrate for visual awareness1,70, although this idea remains controversial71. In support of this theory, cognitive processes that are thought to be closely associated with awareness72 can modulate synchrony. For example, selective attention synchronizes evoked activity in visual73 and somatosensory74 cortex. Moreover, spontaneous fluctuations in perceptual state during binocular rivalry are accompanied by changes in synchrony among V1 neurons in STRABISMIC CATS75. On a larger spatial scale, STEADY-STATE NEUROMAGNETIC RESPONSES have shown a correlation between perception of a visual stimulus during binocular rivalry and both interhemispheric and intrahemispheric coherence76. Viewing ambiguous visual stimuli that can be perceived as either faces or meaningless shapes leads to a long-distance pattern of synchronization in the scalp EEG that is specific for face perception and corresponds to the moment of conscious perception itself77. Although these findings are highly suggestive of a role for synchronous processes in conscious vision, a synthesis has not yet emerged. There is no clear evidence in primates that disrupting synchrony causes any perceptual impairment, unlike the situation for the insect olfactory system78, so any causal role for synchrony in perceptual awareness remains open to debate. This is a very difficult question to address empirically.

Another possibility is that the specific time at which neural activity occurs (relative to stimulus presentation) might be important. Neurons in V1 change their tuning dynamically over the course of a response to a visual stimulus79. Similarly, complex properties of facial stimuli such as facial identity seem to be represented only in the later part of the responses of neurons in inferior temporal cortex80. Neural responses are also modulated by signals that originate outside their CLASSICAL RECEPTIVE FIELDS. This modulation reflects several properties that are relevant to visual perception, such as texture segregation and grouping81, as well as higher-order processes such as working memory81. It has been proposed that these higher-order properties result from feedback connections from higher areas and might be required for awareness82,83. In humans, perception of moving phosphenes caused by stimulation of area V5 with transcranial magnetic stimulation (TMS) can be diminished by stimulation of V1, but only if the stimulation of V1 is applied after the TMS pulse to V5 (Ref. 84). This preliminary evidence indicates that conscious perception of moving phosphenes might be contingent on feedback connections to V1 that are disrupted by the second pulse.

Parietal and prefrontal cortex

Recent neuroimaging investigations into the neural correlates of bistable perception have indicated that activity in the parietal and prefrontal cortices might be associated with visual awareness in normal subjects85,86,87. These studies measured brain activity time-locked to spontaneous fluctuations in awareness both during binocular rivalry85,86 and for other bistable figures87. Cortical regions with activity that reflects perceptual transitions include ventral extrastriate cortex, and parietal and frontal regions that have been implicated previously in the control of spatial attention. Although extrastriate areas are equally engaged by non-rivalrous perceptual changes, areas in frontal and parietal cortex show significantly greater activation time-locked to the perceptual alternations during binocular rivalry. During rivalrous fluctuations in perceptual experience, the time course of haemodynamic activity in V1 shows little correlation with that recorded elsewhere in the brain86. However, the time course of activity recorded from areas in extrastriate cortex that are thought to be involved in binocular rivalry correlates strongly with activity in areas of frontal and parietal cortex that have been implicated in rivalrous transitions. Similar parietal and frontal regions are active during perceptual transitions that occur while viewing bistable figures87 (such as the Necker cube and Rubin's face/vase figure). These data indicate that activity in frontal and parietal cortex might be causally associated with the generation of transitions between different percepts. This idea is supported by the clinical observation that patients with prefrontal cortex lesions typically have abnormal transitions in bistable perception88,89,90. However, the correlation of activity in frontal and parietal cortex with visual awareness is not restricted to bistable perception. When subjects become consciously aware of a change in a visual scene, compared with when they are blind to that change, there is reliable parietal and dorsolateral prefrontal activation91. Similar cortical loci (Fig. 5) are activated by STEREO POP-OUT but not with stable viewing92, and during perceptual binding of colour and motion in object recognition93.

Figure 5: Parietal and prefrontal correlates of consciousness.
figure 5

Areas of parietal and prefrontal cortex that show activation correlated with changes in visual awareness (Refs 85–87,91,92) are plotted on a standardized brain. Each coloured circle is placed at the centre of a cluster of activation, with different colours representing different studies; overlapping loci from the same study are omitted for clarity. There is prominent clustering of activations in superior parietal and dorsolateral prefrontal cortex, highlighted by large, light circles.

Full size image

So, although activity in ventral visual cortex is a consistent neural correlate of consciousness, it might be insufficient to produce awareness without an additional contribution from parietal and prefrontal loci. The nature of the cognitive processes involved in such a contribution and the anatomical connectivity that might mediate such interactions (Fig. 2) are not yet clear. The loci of activation in parietal and frontal areas that are associated with visual awareness generally overlap with areas that have been associated with spatial and non-spatial attention94,95,96. This is consistent with our intuition that awareness and attention are intimately linked, and it is reflected in the common claim that attention helps to modulate visual awareness97 by biasing competition within the ventral visual pathway. The anatomical overlap further strengthens the proposed link between attention and awareness by providing evidence that some of the mechanisms that subserve the two processes might share a common neural substrate98,99,100. Future work should address more precisely the degree of anatomical overlap and the nature of the cognitive relationship between these two processes.

Consistent with these data, disturbances of visual attention and visual awareness in humans are seen not only after damage to ventral visual cortex, but also after damage to the parietal or frontal cortex. For example, visual extinction is a common component of the neglect syndrome that can follow parietal damage, in which patients show deficient awareness of contralesional visual stimuli (see above)3. In animals, much larger lesions have more profound effects. In monkeys, a massive cortical ablation that spares most of the modality-specific visual cortex, but not parietal and frontal areas, leads to chronic blindness101,102. Similarly, removal of frontoparietal cortex in cats produces as much or more decrement in visual discrimination than does removal of temporal cortex103. Such syndromes illustrate that visual awareness for particular regions of visual space can be lost even when visual cortex and its afferent inputs are intact. Nevertheless, awareness of ipsilesional visual stimulation is possible following parietal damage, even with profound spatial neglect. Similarly, although bilateral damage to human parietal cortex can give rise to BALINT'S SYNDROME or SIMULTANAGNOSIA, patients still see something, even if it is only a single object104. Therefore, although neglect or extinction reveals a profound inability to process the world beyond the immediate focus of attention, visual awareness (in some form) seems to be present. Activity in the localized cortical areas that are damaged in these syndromes might therefore be crucial in normal vision, but it is not strictly necessary for some conscious visual experiences. Whether the same can be said for activity in prefrontal cortex remains uncertain.

Future directions

The focus on the identification of reliable neural correlates of consciousness in vision has led to a general consensus on the types of experiment that are likely to prove informative, mainly those that explicitly dissociate conscious and unconscious neural processes. Most of the single-cell evidence from monkeys and fMRI data in humans are compatible with the hypothesis that activity in V1, although necessary for many forms of vision, does not correspond to visual perception. Other experiments have been interpreted to suggest that some aspects of V1 activity do relate to conscious perception. Most observers agree that the neural correlates of consciousness are associated with functionally specialized areas in the ventral visual pathway interacting with specific areas of prefrontal and parietal cortex. However, a fully developed empirical synthesis has yet to emerge. In part, this might be due to the present state of theoretical models of consciousness, which offer plausible frameworks for understanding empirical findings, but few specific predictions that can be tested with current technology.

Two areas that might provide fertile ground for future study emerge from this review. For the first challenge, within functionally specialized areas of visual cortex it will be important to characterize more closely the nature of the neural correlates of consciousness. For example, is awareness a function of the intensity of neural activity within such an area, or does it relate more closely to the timing of that activity? Do the neural correlates of consciousness within a visual area represent activity only in a particular fraction of the neuronal population? How are those cells individuated anatomically and physiologically? However, as we have emphasized, understanding the neural correlates of consciousness in functionally specialized visual areas might be insufficient to answer these questions, as activity in prefrontal and/or parietal cortex also correlates with changes in the contents of consciousness. A second challenge is to explore the role and nature of the interactions between these dorsal areas and the ventral visual pathway. Cognitive processes such as working memory and visual attention are thought to be associated with such dorsal cortical areas. It will therefore be important to characterize how these cognitive processes interact with signals in the visual cortex, exploring the anatomical and physiological substrates that allow such large-scale integration and result in phenomenal awareness in the human brain.