Tononi koch-08

The Neural Correlates of Consciousness
An Update
GIULIO TONONIa AND CHRISTOF KOCHb
a
Department of Psychiatry, University of Wisconsin, Madison, Wisconsin, USA
b
Division of Biology and the Division of Engineering and Applied Science, California
Institute of Technology, Pasadena, California, USA

This review examines recent advances in the study of brain correlates of consciousness. First, we
briefly discuss some useful distinctions between consciousness and other brain functions. We then
examine what has been learned by studying global changes in the level of consciousness, such as
sleep, anesthesia, and seizures. Next we consider some of the most common paradigms used to
study the neural correlates for specific conscious percepts and examine what recent findings say
about the role of different brain regions in giving rise to consciousness for that percept. Then we
discuss dynamic aspects of neural activity, such as sustained versus phasic activity, feedforward
versus reentrant activity, and the role of neural synchronization. Finally, we briefly consider how
a theoretical analysis of the fundamental properties of consciousness can usefully complement
neurobiological studies.

Key words: consciousness; brain; information

Introduction consciousness is, how it can be generated by a physi-
cal system, and how it can be measured and then to
There are several strategies by which scientists can test the predictions of the theory against data from
approach the relationship between the brain and con- neuroscience (Tononi 2004b).
sciousness. One is to wait and hope that, once we know This review focuses on recent advances in the study
all there is to know about neuroanatomy and neuro- of the NCC. First, we briefly discuss some useful dis-
physiology (or maybe when we can model in detail the tinctions between consciousness and other brain func-
working of the entire brain; Markram 2006), the an- tions. We then examine what has been learned about
swer will somehow pop out. A more practical approach the NCC by studying global changes in the level of con-
is to use the tools of neuroscience that are available sciousness, such as sleep, anesthesia, and generalized
now to shed light on the neural structures and activ- seizures. Next, we consider some of the most com-
ity patterns that underlie consciousness. For example, mon paradigms used to study the NCC for specific
one can examine how brain activity changes when, conscious percepts and examine what recent findings
everything else being as equal as possible, a stimulus say about the role of different brain regions in giving
is experienced or not. In this way, one can attempt to rise to consciousness for that percept. Finally, we dis-
identify, more and more precisely, the neural correlates cuss the relation to the NCC of dynamic aspects of
of consciousness (NCC), defined as the minimal neuronal neural activity, such as sustained versus phasic activity,
mechanisms jointly sufficient for any one specific conscious percept feedforward versus reentrant activity, and the role of
(Koch 2003). A complementary strategy is to consider neural synchronization. We end with a few remarks on
conditions in which consciousness is globally dimin- the importance of a combined theoretical and neuro-
ished, such as deep sleep or anesthesia, and ask what biological approach.
has changed in the brain. Finally, another approach is
to develop a theoretical framework that clarifies what
Consciousness and Other Brain
Functions
Address for correspondence: Christof Koch, Division of Biology and the
Division of Engineering and Applied Science, 216-76, California Institute
Many suggestions have been ventured in the hope of
of Technology, Pasadena, CA 91125. Fax: +1-626-796-8876. alleviating the puzzle of subjective experience. Perhaps
koch@klab.caltech.edu consciousness emerges somehow when an organism is
Ann. N.Y. Acad. Sci. 1124: 239–261 (2008). C 2008 New York Academy of Sciences.
doi: 10.1196/annals.1440.004 239

240 Annals of the New York Academy of Sciences

immersed in some complex sensorimotor loop that in- et al. 1998). Patients with the locked-in syndrome can
cludes the environment. Another common idea is that be almost completely paralyzed, and yet they are just
consciousness may somehow arise when one part of as conscious as healthy subjects (Laureys et al. 2005)
the brain, acting as the “subject” (typically the front), and can compose eloquent accounts of their condition
looks upon another part as its “object” (typically the (Bauby 1997). A transient form of paralysis is one of the
back), and evaluates or reflects upon its activity. It is of- characteristic features of narcolepsy. Severe cataleptic
ten thought that in the end consciousness may reduce attacks can last for minutes and leave the patient col-
to attention and its brain mechanisms, since we are lapsed on the floor, utterly unable to move or to sig-
usually conscious of what we attend. Much could be nal, but fully aware of her surroundings (Guilleminault
said about each of these suggestions. Here, we briefly 1976; Siegel 2000). Or consider the Californian drug
consider some recent results (and some very old evi- addicts known as the frozen addicts who acquired some of
dence) indicating that consciousness—in the sense of the symptoms of severe, late-stage Parkinson’s disease,
having an experience—does not require sensorimotor fully conscious, yet unable to move or speak (Langston
loops involving the body and the world, does not re- & Palfreman 1995). All six had previously taken syn-
quire self-reflection (or language), and does not reduce thetic heroin tainted with MPTP, which selectively and
to attention. permanently destroyed dopamine-producing neurons
in their basal ganglia. Thus, consciousness here and
Consciousness and Sensory Input/Motor now seems to depend on what certain parts of the brain
Output are doing, without requiring any obligatory interaction
We are usually conscious of what goes on around with the environment or the body.
us, and occasionally of what goes on within our body.
So it is only natural to think that consciousness may Consciousness and Self-Reflection
be tightly linked to the ongoing interaction we main- Consciousness is usually evaluated by verbal reports,
tain with the world and the body. However, there are and questions about consciousness (“Did you see any-
many examples to the contrary. We are conscious of thing on the screen?”) are answered by “looking in-
our thoughts, which do not seem to correspond to any- side” retrospectively and reporting what one has just
thing out there; we can also imagine things that are experienced. So it is perhaps natural to suggest that
not out there. When we do so, sensory areas can be consciousness may arise through the ability to reflect
activated from the inside (Kosslyn et al. 2001), though on our own perceptions: our brain would form a scene
there are some differences (Amedi et al. 2005). Also, of what it sees, but we would become conscious of it—
stimulus-independent consciousness is associated with experience it subjectively—only when we, as a subject
its own patterns of activation within cortex and tha- of experience, watch that scene from the inside. This
lamus (Mason et al. 2007). During dreams, we are suggestion is often framed in a neurobiological context
virtually disconnected from the environment (Hobson by assuming that patterns of activity corresponding
et al. 2000)—hardly anything of what happens around to “unconscious” or “subconscious” percepts form in
us enters consciousness, and our muscles are paralyzed posterior regions of the cerebral cortex involved in the
(except for eye muscles and diaphragm). Nevertheless, categorization/association of sensory stimuli. These
we are vividly conscious: all that seems to matter is percepts then become conscious when mainly ante-
that the thalamocortical system continues to function rior prefrontal and cingulate regions involved in self-
more or less as in wakefulness, as shown by unit record- representations interact with posterior cortex, perhaps
ing, EEG, and neuroimaging studies performed dur- by reading in signals through forward connections and
ing rapid eye movement (REM) sleep, when dreams selectively amplifying them through back connections
are most intense (Maquet et al. 1996). Interestingly, (more on this later).
certain regions of the thalamocortical systems, such There is of course no doubt that the brain catego-
as dorsolateral prefrontal cortex, are deactivated in rizes its own patterns of activity in the sense that neu-
REM sleep, which likely accounts for some peculiari- rons respond mainly to the activity of other neurons,
ties of dreaming experiences, such as the reduction of so the brain is constantly “looking at itself.” However,
voluntary control. this is not necessarily in terms of a “subject” (the front)
Neurological evidence also indicates that neither looking at an “object” represented in sensory cortices
sensory inputs nor motor outputs are needed to gener- (the back). Leaving aside the mystery of why reflecting
ate consciousness. For instance, retinally blind people on something should make it conscious, this scenario
can both imagine and dream visually if they become is made less plausible by a common observation: when
blind after 6–7 years of age or so (Hollins 1985; Buchel we become absorbed in some intense perceptual task,

Tononi and Koch: The Neural Correlates of Consciousness 241

for example watching an engrossing movie, playing a and social setting. Although displaying many of the
fast-paced video game, or rushing through the woods typical frontal lobe behavioral disturbances, he never
at high speed, we are vividly conscious—we are im- complained of loss of sensory perception, nor did he
mersed in the rapid flow of experience—without any show visual or other deficits (Mataro et al. 2001). An-
need for reflection or introspection. Often, we become other case is that of a 27-year-old woman with mas-
so immersed in such flow that we may lose the sense sive bilateral prefrontal damage of unclear etiology
of self, the inner voice. Perhaps the habit of thinking (Markowitsch & Kessler 2000). While manifesting
about consciousness has distracted the scholars who grossly deficient scores in frontal lobe sensitive tests,
write upon such matters to devalue the unreflective she showed no abnormal perceptual abilities (that is
nature of much of experience. not to say that such patients do not suffer from subtle
A recent neuroimaging study by Malach and col- visual deficits; Barcelo et al. 2000).
laborators (Hasson et al. 2004) throws some inter-
esting light on these old observations. Subjects were Consciousness and Attention
scanned with fMRI in three conditions. In the slow Consciousness and attention have resisted clear and
categorization task, subjects were asked to catego- compelling definitions. Few would dispute that the re-
rize pictures into animal/no-animal categories. Dur- lationship between the two is an intimate one. When
ing the introspective task, subjects viewed the images subjects pay attention to an object, they become con-
and then introspected about their emotional response scious of its various attributes; when the focus of at-
(strong/neutral). Finally, the fast categorization task tention shifts away, the object fades from conscious-
was identical to the “slow” condition but at triple ness. Indeed, more than a century of research efforts
the stimulation rate. Thus, “slow” and “introspection” have quantified the ample benefits accrued to attended
conditions were identical in terms of sensory stimuli and consciously perceived events (Pashler 1998; Braun
and motor output but differed in the cognitive task. et al. 2001). This has prompted many to posit that the
On the other hand, “slow” and “rapid” conditions two processes are inextricably interwoven, if not iden-
were similar in the cognitive task but differed in the tical (Posner 1994; Merikle & Joordens 1997; Chun &
sensorimotor processing and attentional loads. Behav- Wolfe 2000; O’Regan & Noe 2000). Others, however,
ioral measurements confirmed that self-awareness was going back to the 19th century, have argued that atten-
high during the introspection task, and virtually abol- tion and consciousness are distinct phenomena, with
ished during rapid categorization. The neuroimaging distinct functions and neuronal mechanisms (Iwasaki
results were clear: during introspection there was an 1993; Hardcastle 1997; Lamme 2003; Baars 2005;
activation of prefrontal regions, whereas sensory cor- Block 2005; Dehaene et al. 2006; Koch & Tsuchiya
tex was strongly activated during rapid categorization. 2007). Recent psychophysical and neurophysiological
Crucially, during introspection self-related cortex was evidence argues in favor of a dissociation between selec-
deactivated below the rest condition. This deactiva- tive attention and consciousness, suggesting that events
tion of prefrontal regions was thus the neural corre- or objects can be attended to without being consciously
late of “losing oneself ” in the task. To the extent that perceived. Conversely, an event or object can be con-
these prefrontal regions were indeed involved in self- sciously perceived in the near absence of top-down
representation, these findings suggest that their activa- attentional processing.
tion is not necessary for the emergence of perceptual
consciousness, but only to reflect upon it and report it Attention without Consciousness
to others. Indeed, it appears that self-related activity Consider that subjects can attend to a location for
is actually shut off during highly demanding sensory many seconds and yet fail to see one or more attributes
tasks. of an object at that location. In lateral masking (visual
Lesion studies also seem to support the notion that crowding), the orientation of a peripherally presented
consciousness does not require prefrontal cortex and, grating is hidden from conscious sight but remains
by inference, the functions it performs. Though pa- sufficiently potent to induce an orientation-dependent
tients with widespread, bilateral damage to prefrontal aftereffect (He et al. 1996) Montaser-Kouhsari and
cortex are rare, two recent clinical studies provide Rajimehr (2004) showed that an aftereffect induced
some intriguing evidence. A man who, at the age by an invisible illusory contour required focal atten-
of 21, had fallen on an iron spike that completely tion, even though the object at the center of attention
penetrated through both of his frontal lobes, never- was invisible. Naccache and colleagues (2002) elicited
theless went on to live a stable life—marrying and priming for invisible words (suppressed by a combi-
raising two children—in an appropriate professional nation of forward and backward masking) but only if


the subject was attending to the invisible prime-target (or a vehicle), while being unable to distinguish a red-
pair; without attention, the same word failed to elicit green bisected disk from a green-red one (Li et al.
priming. That is, the subject needs to attend to the 2002). Likewise, subjects can tell male from female
masked stimulus for priming to occur, but the actual faces or even distinguish a famous from a nonfamous
stimulus is not seen (with an associated d = 0). In face (Reddy et al. 2004; Reddy & Koch 2006) but are
another experiment, male/female nudes attracted at- frustrated by tasks that are computationally much sim-
tention and induced involuntary eye movements when pler (e.g., discriminating a rotated letter “L” from a
they were rendered completely invisible by continuous rotated “T”). This is quite remarkable. Thus, while we
flash suppression (Jiang et al. 2006). When subjects had cannot be sure that observers do not deploy some lim-
to discriminate the location of the masked nude from ited amount of top-down attention in these dual-task
the location of a masked shuffled nude, they were at experiments that require training and concentration
chance; without the intraocular masking, the images (i.e., high arousal), it remains true that subjects can
are clearly visible. fMRI evidence confirms attentional perform certain discriminations but not others in the
modulation of invisible images in primary visual cor- near absence of top-down attention. And they are not
tex (Bahrami et al. 2007). In conclusion, attentional guessing. They can be quite confident of their choices
selection does not necessarily engender phenomenal and “see,” albeit often indistinctly, what they can
sensations, although it may often do so. discriminate.
The existence of such dissociations—attention
Consciousness in the Absence of Attention without consciousness and consciousness without
When focusing intensely on one event, the world attention—should not be surprising when considering
is not reduced to a tunnel, with everything outside their different functions. Attention is a set of mecha-
the focus of attention gone: we are always aware of nisms whereby the brain selects a subset of the incom-
some aspects of the world surrounding us, such as its ing sensory information for higher level processing,
gist. Indeed, gist is immune from inattentional blind- while the nonattended portions of the input are ana-
ness (Mack & Rock 1998)—when a photograph cover- lyzed at a lower band width. For example, in primates,
ing the entire background is briefly and unexpectedly about one million fibers leave each eye and carry on
flashed onto the screen, subjects can accurately report the order of one megabyte per second of raw infor-
a summary of its content. In the 30 ms necessary to ap- mation. One way to deal with this deluge of data is to
prehend the gist of a scene, top-down attention cannot select a small fraction and process this reduced input in
play much of a role (because gist is a property asso- real time, while the nonattended data suffer from be-
ciated with the entire image, any process that locally nign neglect. Attention can be directed by bottom-up,
enhances features is going to be of limited use). exogenous cues or by top-down endogenous features
Or take the perception of a single object (say a bar) in and can be applied to a spatially restricted part of the
an otherwise empty display, a nonecological but com- image (focal, spotlight of attention), an attribute (e.g.,
mon arrangement in many animal and human experi- all red objects), or to an entire object. By contrast, con-
ments. Here, what function would top-down, selective sciousness appears to be involved in providing a kind
attention need to perform without any competing item of “executive summary” of the current situation that
in or around fixation? Indeed, the most popular neu- is useful for decision making, planning, and learning
ronal model of attention, biased competition (Desimone (Baars).
& Duncan 1995), predicts that in the absence of com-
petition, no or little attentional enhancement occurs,
yet we are perfectly aware of the object and its back- Changes in the Level of Consciousness
ground.
In a dual-task paradigms, the subject’s attention is In neurology, a distinction is traditionally made be-
drawn to a demanding central task, while at the same tween the level of consciousness and the content of
time a secondary stimulus is flashed somewhere in the consciousness. When you fall asleep, for example, the
periphery. Using the identical retinal layout, the sub- level of consciousness decreases to the point that you
ject either performs the central task, the peripheral become virtually unconscious—the degree to which
task, or both simultaneously (Sperling & Dosher 1986; you are conscious (of anything) becomes progressively
Braun & Sagi 1990; Braun & Julesz 1998). With fo- less and less. The content of consciousness, instead,
cal attention busy at the center, the subject can still refers to the particular experience you’re having at any
distinguish a natural scene containing an animal (or given time, for the same level of consciousness. It is use-
a vehicle) from one that does not include an animal ful to maintain this distinction when considering the


NCC, and begin by considering changes in the level of An intriguing possibility is that changes in the level
consciousness. of consciousness during sleep may be related to the de-
gree of bistability of thalamocortical networks (Tononi
Sleep 2004b; Massimini et al. 2007). In a recent study, tran-
Sleep offers a commonplace, daily demonstration scranial magnetic stimulation (TMS) was used in con-
that the level of consciousness can change dramati- junction with high-density EEG to record cortical
cally. In the laboratory, a subject is awakened during evoked responses to direct stimulation in waking and
different stages of sleep and asked to report “anything sleep (Massimini et al. 2005; Massimini et al. 2007).
that was going through your mind just before waking During wakefulness, TMS induced a sustained re-
up.” What is most noteworthy for the present pur- sponse made of rapidly changing patterns of activity
poses is that a number of awakenings from non-REM that persisted until 300 ms and involved the sequen-
(NREM) sleep, especially early in the night when EEG tial activation of specific brain areas, depending on
slow waves are prevalent, can yield no report what- the precise site of stimulation. During early NREM
soever. Thus, early slow-wave sleep is the only phase sleep, however, the brain response to TMS changed
of adult life during which healthy human subjects may markedly. When applied to lateral cortical regions,
deny that they were experiencing anything at all. When the activity evoked by TMS remained localized to the
present, reports from NREM sleep early in the night site of stimulation, without activating connected brain
are often short and thought-like. However, especially regions, and lasted for less than 150 ms (Massimini
later in the night, reports from NREM sleep can be et al. 2005). This finding indicates that during early
longer, more hallucinatory and, generally speaking, NREM sleep, when the level of consciousness is re-
more dream-like. On the other hand, awakenings dur- duced, effective connectivity among cortical regions
ing REM sleep almost always yield dreams—vivid con- breaks down, implying a corresponding breakdown of
scious experiences, sometimes organized within a com- cortical integration. Computer simulations suggest that
plex narrative structure (Hobson et al. 2000; Hobson this breakdown of effective connectivity may be due to
& Pace-Schott 2002). the induction of a local down state (Esser & Tononi, in
What are the processes underlying the fading of the preparation). When applied over centromedian pari-
level of consciousness during early slow-wave sleep? etal regions, each TMS pulse triggered a stereotypical
Though metabolic rates decrease in many cortical ar- response implying the induction of a global down state:
eas, especially frontal and parietal areas, the thalam- a full-fledged, high-amplitude slow wave (Massimini et
ocortical system is far from shutting down. Instead, al. 2007) that closely resembled spontaneous ones and
triggered by a decrease in acetylcholine and other that traveled through much of the cortex (Massimini
modulators, cortical and thalamic neurons undergo et al. 2004). Such stereotypical responses could be in-
slow oscillations (1 Hz or less) between an up and duced even when, for the preceding seconds, there
a down state (Steriade et al. 2001). During the up were no slow waves in the spontaneous EEG, indicat-
state, cortical cells remain depolarized at waking levels ing that perturbations can reveal the potential bista-
for around a second and fire at waking rates, often bility of a system irrespective of its observed state. Al-
in the gamma range (Destexhe et al. 2007). How- together, these TMS–EEG measurements suggest that
ever, the up state of NREM sleep is not stable as the sleeping brain, despite being active and reactive,
in wakefulness and REM sleep, but it is inherently becomes inherently bistable: it either breaks down in
bistable. The longer neurons remain depolarized, the causally independent modules or bursts into a global,
more likely they become to precipitate into a hyper- stereotypical response. By contrast, during REM sleep
polarized down state—a complete cessation of synap- late in the night, when dreams become long and vivid
tic activity that can last for a tenth of a second or and consciousness returns to levels close to those of
more—after which they revert to another up state. wakefulness, the responses to TMS also recover and
The transition from up to down states appears to be come to resemble more closely those observed dur-
due to depolarization-dependent potassium currents ing wakefulness: evoked patterns of activity become
and to short-term synaptic depression, both of which more complex and spatially differentiated, although
increase with the amount of prior activation (Steri- some late components are still missing (Massimini &
ade 2003). The slow oscillation is found in virtually Tononi, unpublished).
every cortical neuron and is synchronized across the
cortical mantle by cortico-cortical connections, which Anesthesia
is why the EEG records high-voltage, low-frequency The most common among exogenous manip-
waves. ulations of the level of consciousness is general


anesthesia. Anesthetics come in two main classes: intra- of action of several anesthetics is the posterior cingu-
venous agents used for induction, such as propofol and late/medial parietal cortical areas, as well as the me-
ketamine, which are generally administered together dial basal forebrain). However, both PET and fMRI
with sedatives such as midazolam and dexmedeto- signals mostly reflect synaptic activity rather than cel-
midine; and inhaled agents, such as isoflurane, lular firing, and the thalamus receives a massive in-
sevoflurane, and desflurane or the gases xenon and nervation from cortex. Moreover, the relative reduc-
nitrous oxide. The doses of inhaled anesthetics are tion in thalamic activity occurs on a background of a
usually referred to their minimum alveolar concen- marked decrease in global metabolism (30–60%) that
tration (MAC): a MAC value of 1 is the dose that involves many cortical regions. Thus, thalamic activity
prevents movement in 50% of subjects in response as recorded by neuroimaging may represent an es-
to a painful surgical stimulation. At low MAC values pecially sensitive, localized readout of the extent of
(0.1–0.2) anesthetics produce amnesia, first explicit and widespread cortical deactivation, rather than the final
then implicit. Frequently there are distortions of time common pathway of unconsciousness (Ori et al. 1986).
perception, such as slowing down and fragmentation, In fact, spontaneous thalamic firing in animal models
and a feeling of disconnection from the environment. of anesthesia is mostly driven by corticothalamic feed-
Also, at low MAC values anesthetics produce increas- back (Vahle-Hinz et al. 2007), and the metabolic effects
ing sleepiness and make arousal progressively more dif- of enflurane on the thalamus can be abolished by an
ficult, suggesting that to some extent they can mimic ipsilateral cortical ablation (Nakakimura et al. 1988),
neurophysiological events underlying sleep. At around suggesting that the switch in thalamic unit activity is
0.3 MAC people experience a decrease in the level driven primarily through a reduction in afferent cor-
of consciousness, also described as a “shrinking” in ticothalamic feedback more than by a direct effect of
the field of consciousness, as if they were kept on the anesthesia on thalamic neurons. Recently, thalamic ac-
verge of falling asleep. MAC-awake, usually around tivity was recorded using depth electrodes in a patient
0.3–0.4 MAC, is the point at which response to ver- undergoing anesthesia for the implant of a deep brain
bal command is lost in 50% of patients and is con- stimulator (Velly et al. 2007). With either propofol or
sidered the point at which consciousness is lost (LOC). sevoflurane, when the patient lost consciousness, the
The transition to unconsciousness (LOC) appears to be cortical EEG changed dramatically. However, there
rather brusque, not unlike the collapse of muscle tone was little change in the thalamic EEG until almost
that usually accompanies it, suggesting that neural pro- 10 min later. This result implies that the deactivation
cesses underlying consciousness change in a nonlinear of cortex alone is sufficient for loss of consciousness,
manner. At concentrations above LOC, movements and conversely that thalamic activity alone is insuffi-
are still possible, especially partially coordinated re- cient to maintain it.
sponses to painful stimuli, suggesting that some degree These findings do not imply that the thalamus is
of “unconscious” processing is still possible. Complete irrelevant, but rather suggest that its effects on con-
unresponsiveness is usually obtained just above MAC sciousness are largely accounted for by its effects on
of 1.0. cortical circuits, most notably by providing a tonic fa-
At the cellular level, many anesthetics have mixed ef- cilitatory action that can help to restore consciousness,
fects, but the overall result is a decrease in neuronal ex- or cause its collapse when it fails. A dramatic demon-
citability by either increasing inhibition or decreasing stration is provided by a recent study, where rats kept
excitation. Most anesthetics act by enhancing GABA under anesthetic concentration of sevoflurane could be
inhibition or by hyperpolarizing cells through an in- awakened by a minute injection of nicotinic agents in
crease of potassium-leak currents. They can also inter- an intralaminar thalamic nucleus (Alkire et al. 2007).
fere with glutamatergic transmission and antagonize Conversely, GABA agonists infused into the same side
acetylcholine at nicotinic receptors (Campagna et al. caused an immediate loss of consciousness (Miller et al.
2003; Franks 2006). But what are critical circuits me- 1989; Miller & Ferrendelli 1990) (though nicotinic an-
diating the LOC induced by anesthetics? tagonists did not).
A considerable number of neuroimaging studies in
humans have recently shed some light on this issue Coma and Vegetative States
(Alkire & Miller 2005). The most consistent effects While consciousness may nearly fade during cer-
produced by most anesthetics at LOC is a reduction tain phases of sleep and be kept at very low lev-
of thalamic metabolism and blood flow, suggesting the els for a prescribed period during general anesthe-
possibility that the thalamus may serve as a conscious- sia, coma and vegetative states are characterized by
ness switch (Alkire et al. 2000). (Another common site a loss of consciousness that is hard or impossible to


reverse. Coma—an enduring sleep-like state of immo- cortices (Laureys et al. 2000a). A recent study by Niko
bility with eyes closed from which the patient cannot be Schiff and colleagues shows the role of the thalamus
aroused—represents the paradigmatic form of patho- even more dramatically: bilateral deep brain electri-
logical loss of consciousness. Typically, coma is caused cal stimulation of the central thalamus restored a de-
by a suppression of thalamocortical function by drugs, gree of behavioral responsiveness in a patient who had
toxins, or internal metabolic derangements. Other remained in a minimally conscious state for 6 years
causes of coma are head trauma, strokes, or hypoxia following brain trauma (Schiff et al. 2007).
due to heart failure, which again cause widespread de- In vegetative states, brain metabolism is globally re-
struction of thalamocortical circuits. Smaller lesions of duced by 50 to 60%, most notably in regions such as the
the reticular activating system can also cause uncon- posterior cingulate cortex and the precuneus (Laureys
sciousness, presumably by deactivating the thalamo- et al. 2004; Schiff 2006a). These are also the areas that
cortical system indirectly. reactivate most reliably if a patient regains conscious-
Patients who survive a coma may recover, while ness. A recent case study reported an extraordinary re-
others enter the so-called vegetative state, in which covery of verbal communication and motor function in
eyes reopen, giving the appearance of wakefulness, but a patient who had remained in a minimally conscious
unresponsiveness persists. Soon, regular sleep/waking state for 19 years (Voss et al. 2006). Diffusion tensor
cycles follow. Respiration, other autonomic functions MRI showed increased fractional anisotropy (assumed
and brainstem functions are relatively preserved, and to reflect myelinated fiber density) in posteromedial
stereotypic, reflex-like responses can occur, including cortices, encompassing cuneus and precuneus. These
yawning and grunting, but no purposeful behavior. Pa- same areas showed increased glucose metabolism as
tients may remain vegetative for a long time or may studied by PET scanning, likely reflecting the neu-
emerge to a “minimally conscious state.” This is dis- ronal regrowth paralleling the patient’s clinical recov-
tinguished from the vegetative state by the occasional ery. This and other neuroimaging studies of vegetative
occurrence of some behaviors that do not appear to be or minimally conscious patients are demonstrating ever
purely reflexive, suggesting that some degree of con- more clearly that a purely clinical diagnosis of persis-
sciousness may be present. tent loss of consciousness may at times be dramatically
Post-mortem analysis in vegetative patients reveals inaccurate. It was already known that patients with
that the brainstem and hypothalamus, and specifi- locked-in syndromes due to pontine lesions that left
cally the reticular activating system, are largely spared, them almost completely paralyzed may seem unre-
which explains why patients look awake. Usually the sponsive, but if they preserve even a minimal ability
vegetative state is due to widespread lesions of gray to signal (typically by moving an eye or eyelid up or
matter in neocortex and thalamus, to widespread down) they can communicate a rich subjective expe-
white-matter damage and diffuse axonal injury, or to rience. What about patients who show no voluntary
bilateral thalamic lesions, especially of the paramedian movement at all? Several imaging studies have now
thalamic nuclei. It is not clear whether thalamic dam- shown that, even in completely unresponsive patients,
age is largely a reflection of diffuse cortical damage as long as significant portions of the thalamocortical
(as in metabolic studies of anesthesia, changes in the system are preserved, cognitive stimuli can induce pat-
thalamus are much more concentrated and therefore terns of activation similar to those seen in healthy sub-
easier to document than changes in cortex) or whether jects (Laureys et al. 2004; Schiff 2006b). As we have
even isolated thalamic damage can cause persistent seen in a previous section, stimuli that are not per-
unconsciousness. Matrix, calbindin-positive thalamic ceived consciously can still activate appropriate brain
cells, which are most numerous within intralaminar areas. Thus, inferring the presence of consciousness
nuclei, project diffusely to supragranular layers of cor- may be unwarranted. However, in a recent, thought-
tex (Jones 1998) where they facilitate coherent high- provoking study, a clinically vegetative, seemingly un-
frequency oscillations (Joliot et al. 1994). Such cells may responsive patient was put in the scanner and asked
be especially important in enabling effective interac- to imagine playing tennis or navigating through her
tions among many cortical regions: if their activating room. Remarkably, the patient showed fMRI activa-
function is lost, patients may remain vegetative even tion patterns of the appropriate cortical regions, exactly
though cortical tissue may be relatively intact. Indeed, like healthy subjects. Obviously, these activations could
Steven Laureys and colleagues showed that recovery not be due to unconscious processing of stimuli (Owen
from a vegetative state was associated with the restora- et al. 2006). Of note, this patient had widespread
tion of functional connectivity between intralaminar frontal lesions, while posterior cortex was largely
thalamic nuclei and prefrontal and anterior cingulate preserved.


Seizures tivity less focally than initially thought (Blumenfeld &
The abnormal, hypersynchronous discharge of neu- Taylor 2003; Blumenfeld et al. 2003; Blumenfeld et al.
rons is a frequent cause of short-lasting impairments of 2004; Blumenfeld 2005). In fact, it now appears that
consciousness. Consciousness is lost or severely affected all seizures causing an impairment of consciousness
in so-called generalized seizures, such as absence and are associated with changes in activity in three sets of
tonic-clonic seizures, and to a lesser extent in complex brain areas, namely: i) increased activity in the upper
partial seizures. Absence seizures, which are more fre- brainstem and medial thalamus; ii) decreased activity
quent in children, are momentary lapses of conscious- in the anterior and posterior cingulate, medial frontal
ness during which a child stops what she was doing cortex, and precuneus; iii) altered activity in the lateral
and stares straight ahead blankly. Absence seizures are and orbital frontal cortex and in the lateral parietal
accompanied by spike-and-wave complexes at around cortex. In tonic-clonic seizures, fronto-parietal associ-
3 Hz in the EEG, reflecting cycles of synchronous fir- ation areas show increased activity, while the pattern is
ing and silence of large numbers of neurons. There more complex and variable in absence seizures, though
is great variability in the degree of unresponsiveness parietal areas are usually deactivated. Complex partial
both across subjects and, within the same subject, be- seizures show decreased activity in frontal and parietal
tween seizures. Sometimes, simple behaviors such as association cortex, which is associated with the onset
repetitive tapping or counting can proceed unimpaired of slow waves similar to those of sleep or anesthesia
during the seizures, but more complex tasks come to a (rather than to epileptiform discharges).
halt. At this stage, it is not clear which of these three sets
Generalized convulsive seizures usually comprise a of areas, alone or in combination, are crucial for the
tonic phase of muscle stiffening, followed by a clonic loss of consciousness. However, two things are clear:
phase with jerking of the arms and legs. After the con- first, the areas involved in the loss of consciousness
vulsion the person may be lethargic or in a state of associated with seizures correspond to those affected
confusion for minutes up to hours. During the tonic in sleep, anesthesia, and the vegetative state, pointing
phase of a convulsive seizure neural activity is greatly to a common substrate for the most common forms of
increased, as indicated by high frequency activity in loss of consciousness; and second, especially during the
the EEG. The clonic phase is accompanied by syn- tonic phase of convulsive seizures, it would seem that
chronous spikes and waves in the EEG, correspond- consciousness is lost when neurons are excessively and
ing to millions of neurons alternately firing in strong synchronously active, rather than inactive.
bursts and turning silent. Loss of consciousness during
the tonic phase of generalized seizures is noteworthy
because it occurs at times when neuronal activity is Changes in the Content of
extremely high and synchronous. Consciousness—-Localizing the NCC
Partial complex seizures often begin with strange
abdominal sensations, fears, premonitions, or auto- As we have seen, much can be learned by studying
matic gestures. The person progressively loses contact how brain activity changes with changes in the overall
with the environment, exhibits a fixed stare, and is level of consciousness. A complementary approach is
unable to respond adequately to questions or com- to examine how brain activity changes when a specific
mands. Stereotyped, automatic movements are com- content of consciousness changes—for example, a vi-
mon. Complex partial seizures usually last from 15 sec sual stimulus becomes visible or invisible—while every-
to 3 min. Seizure activity is usually localized to the thing else, including the overall level of consciousness
medial temporal lobe. as well as the sensory input, remains as constant as pos-
The diagnosis of seizures is made through clini- sible. Recording spike trains from individual neurons
cal observation and EEG recordings. In absence and or measuring hemodynamic signals while the subject
tonic-clonic seizures the scalp EEG shows diffuse ab- is looking at a stimulus in an MR scanner allows re-
normalities, suggesting a generalized involvement of searchers to investigate what kind of neuronal activity
brain networks, whereas in partial complex seizures is correlated with the conscious perception of the stim-
the abnormalities are confined to a medial temporal fo- ulus rather than with the mere presentation of the
cus on one side. However, neuroimaging studies using stimulus. The goal is to follow the footprints of con-
SPECT, PET, and fMRI, and depth EEG recordings sciousness in the brain by ultimately identifying the
in humans and animals have revealed that general- neural correlates of consciousness (NCC)—the mini-
ized seizures do not affect all brain areas indiscrimi- mal neuronal mechanisms that are jointly sufficient for
nately, whereas complex partial seizures alter brain ac- any one specific conscious percept (Crick & Koch 1995,


1998, 2003). Thus, every phenomenal, subjective state A Test Case: The NCC in Visual Cortex
will have associated NCC: one for seeing a red patch, Many experiments exploit perceptual illusions in
another one for seeing grandmother, yet a third one which the physical stimulus remains fixed while the per-
for hearing a siren, another one for feeling free to act cept fluctuates. The best known example is the Necker
one way or another and so on. Perturbing or inactivat- cube, whose 12 lines can be perceived in one of two
ing the NCC for any one specific conscious experience different ways in depth. Another perceptual illusion
will affect the percept or cause it to disappear. If the that is easy to vary parametrically is binocular rivalry
NCC could be induced artificially, for instance by cor- (Blake & Logothetis 2002): a small image, for example
tical microstimulation in a prosthetic device or during a horizontal grating, is presented to the left eye and an-
neurosurgery, the subject will experience the associated other image, for example a vertical grating, is shown
percept. This definition of the NCC stresses the word to the corresponding location in the right eye. In spite
“minimal,” because the question of interest is which of the constant visual stimulus, observers consciously
subcomponents of the brain are actually needed. For see the horizontal grating alternate every few seconds
instance, while neural activity in the cerebellum may with the vertical one. The brain does not allow for the
ultimately affect some behaviors elicited by the presen- simultaneous perception of both images.
tation of a stimulus (such as eye movements), it is likely Macaque monkeys can be trained to report whether
that such activity has little to do with the generation of they see the left or the right image. The distribution of
the conscious percept of that stimulus, and thus is not the switching times and the way in which changing the
part of the NCC. contrast in one eye affects the reports leaves little doubt
Many questions can be asked: What characterizes that monkeys and humans experience the same basic
the NCC for a specific experience? What do the NCC phenomenon. In a series of elegant experiments, Lo-
for different experiences have in common? Do the gothetis and colleagues (Leopold & Logothetis 1996;
NCC involve just the thalamocortical system, or do Logothetis 1998) recorded from a variety of visual cor-
they extend to structures such as the hippocampus, tical areas in the awake macaque monkey while the
the claustrum, or the basal ganglia? Do they involve all animal performed a binocular rivalry task. In primary
pyramidal neurons in cortex at any given point in time? visual cortex (V1), only a small fraction of cells weakly
Or only a subset of cells in frontal lobes that project to modulated their response as a function of the percept of
the sensory cortices in the back? Only layer 5 cortical the monkey. The majority of cells responded to one or
cells? Neurons that fire in a phasic, synchronous, or the other retinal stimulus with little regard to what the
oscillatory manner? Or neurons that are included in animal perceived at the time. In contrast, in a high-
reentrant loops? These are some of the proposals that level cortical area such as the inferior temporal (IT)
have been advanced over the past years (Chalmers cortex along the ventral pathway, almost all neurons
2000). responded only to the perceptual dominant stimulus,
Much of the contemporary work aimed at char- that is, to the stimulus that was being reported. For
acterizing the NCC has concentrated on changes in example, when a face and a more abstract design were
specific visual contents of consciousness (or awareness, presented, one of these to each eye, a “face” cell fired
which is here used interchangeably), in part because only when the animal indicated by its performance
visual experience is easy to manipulate experimentally, that it saw the face and not the design presented to
and in part because more is known about the anatomy the other eye. This result implies that the NCC involve
and function of visual areas in primates than about activity in neurons in inferior temporal cortex (but not,
any other brain regions. Visual psychologists have per- of course, that the NCC are local to IT).
fected a number of techniques—masking, binocular In a related perceptual phenomenon, flash suppres-
rivalry, continuous flash suppression, motion-induced sion (Wolfe 1984), the percept associated with an image
blindness, change blindness, inattentional blindness— projected into one eye is suppressed by flashing another
in which the seemingly simple and unambiguous rela- image into the other eye (while the original image re-
tionship between a physical stimulus in the world, the mains). A methodological advantage over binocular
resulting neural activity, and its associated percept is rivalry is that the timing of the perceptual transition
disrupted (Kim & Blake 2005). We first consider what is determined by an external trigger rather than by
such approaches have revealed about the brain areas an internal event. The majority of responsive cells in
that underlie the NCC, focusing specifically on the vi- inferior temporal cortex and in the superior temporal
sual system. In the following sections we ask whether sulcus follow the monkey’s behavior—and therefore its
the NCC for visual percepts corresponds to some spe- percept. That is, when the animal perceives a cell’s
cial kind of neural activity. preferred stimulus, the neuron fires; when the stimulus


is present on the retina but is perceptually suppressed, tually invisible line drawings can still be modulated
the cell falls silent, even though legions of V1 neu- by selective visual attention. Also, Haynes and Rees
rons fire vigorously to the same stimulus (Sheinberg (2005) exploited multivariate decoding techniques to
& Logothetis 1997). Single neuron recordings in the read out perceptually suppressed information (the ori-
medial temporal lobe of epileptic patients during flash entation of a masked stimulus) from V1 BOLD ac-
suppression likewise demonstrate abolition of their re- tivity, even though the stimulus orientation was so effi-
sponses when their preferred stimulus is present on the ciently masked that subjects performed at chance levels
retina but not seen (Kreiman et al. 2002). when trying to guess the orientation. That is, although
In a powerful combination of binocular rivalry and subjects did not give any behavioral indication that
flash suppression, a stationary image in one eye can be they saw the orientation of the stimulus, its slant could
suppressed for minutes on end by continuously flash- be predicted on a single-trial basis with better-than-
ing different images into the other eye (continuous flash chance odds from the V1 (but not from V2 or V3)
suppression; Tsuchiya & Koch 2005). This paradigm BOLD signal.
lends itself naturally to further investigation of the rela- In conclusion, information pertaining to visual stim-
tionship between neural activity—whether assayed at ulus can be registered in primary visual cortex that may
the single neuron or at the brain voxel level—and con- or may not be accessible to the subject. It may be mod-
scious perception. Imaging experiments with such per- ulated by attention, but it does not appear to correlate
ceptual illusions demonstrate quite conclusively that directly with the subject’s percept. In that sense much
BOLD activity in the upper stages of the ventral path- of the neural activity in V1 probably does not belong
way (e.g., the fusiform face area and the parahippocam- to the NCC. Moreover, as suggested by PET exper-
pal place area) follow the percept and not simply the iments in patients in a persistent vegetative state, in
retinal stimulus (Tong et al. 1998; Rees & Frith 2007). whom stimuli evoked strong but localized activity in
There is a lively debate about the extent to which primary auditory and primary somatosensory cortices
neurons in primary visual cortex simply encode the (Laureys et al. 2000b; Laureys et al. 2002), it may well
visual stimulus or are directly contributing to the sub- be that none of the primary sensory cortices contributes
ject’s conscious percept (Tong 2003). That is, is V1 directly to sensory consciousness.
part of the NCC (Crick & Koch 1995)? It is clear that
retinal neurons are not part of the NCC for visual ex-
periences. While retinal neurons often correlate with Consciousness and Neural Dynamics
visual experience, the spiking activity of retinal gan-
glion cells does not accord with visual experience (e.g., The studies discussed above indicate that, even
there are no photoreceptors at the blind spot; yet no within the cerebral cortex, changes in neural activity do
hole in the field of view is apparent, in dreams vivid im- not necessarily correlate with changes in conscious ex-
agery occurs despite closed eyes, and so on). A number perience. Also, we saw earlier that most of the cortex is
of compelling observations link perception with fMRI active during early NREM sleep and anesthesia, not to
BOLD activity in human V1 and even in the lateral mention during generalized seizures, but subjects have
geniculate nucleus (LGN) (Tong et al. 1998; Lee et little conscious content to report. Thus, it is natural to
al. 2005). These data are at odds with single-neuron suggest that some additional dynamic feature of neural
recordings from the monkey. This discrepancy is prob- activity must be present to generate conscious content.
ably explained by the often tight relationship between Here we consider the roles of i) sustained versus phasic
consciousness and attention that was mentioned above activity, ii) reentrant versus feedforward activity, and
(see section “Consciousness and Attention”). Unless at- iii) synchronous or oscillatory activity.
tentional effects are carefully controlled for, their neu-
ral correlates cannot be untangled from those of con- Sustained versus Phasic Activity
sciousness (Huk et al. 2001; Tse et al. 2005). This has A plausible idea is that neural activity may con-
now been achieved in an elegant study of Lee, Blake, tribute to consciousness only if it is sustained for a
and Heeger (2007). Using a dual-task paradigm they minimum period of time, perhaps around a few hun-
found that hemodynamic BOLD activity in human V1 dred ms. At the phenomenological level, there is no
reflects attentional processes but does not directly cor- doubt that the “now” of experience unfolds at a time
relate with the conscious percept of the subject. The scale comprised between tens and several hundred ms
experiment by Bahrami and colleagues (Bahrami et (Bachmann 2000) and in some aspects may even stretch
al. 2007) comes to a similar conclusion—they show to one or two seconds (Poppel & Artin 1988). Other
that the fMRI signal in V1 associated with percep- experiments have made use of the attentional blink


phenomenon: when an observer detects a target in a the target, though they may deny having seen it con-
rapid stream of visual stimuli, there is a brief period of sciously. Thus, consciousness would seem to require
time during which the detection of subsequent targets that neural activity in appropriate brain structures lasts
is impaired. Remarkably, targets that directly follow the for a minimum amount of time, perhaps as much as
first target are less impaired than those that follow after is needed to guarantee interactions among multiple
200 to 400 ms (Raymond et al. 1992). By manipulat- areas.
ing attention, identical visual stimuli can be made con- On the other hand, other data would seem to sug-
scious or unconscious. In such studies, event-related gest that it may actually be the phasic, onset, or off-
potentials reflecting early sensory processing (the P1 set discharge of neurons that correlates with experi-
and N1 components) were identical for seen and un- ence. The most stringent data come again from studies
seen stimuli but quickly diverged around 270 ms, sug- of visual masking (for a review see Macknik 2006).
gesting that stimuli only become visible when a sus- To be effective, masking stimuli must either precede
tained wave of activation spreads through a distributed (forward masking) or follow (backward masking) tar-
network of cortical association areas (Vogel et al. 1998; get stimuli at appropriate time intervals and usually
Sergent et al. 2005). need to be spatially contiguous. Macknik, Martinez-
More generally, the requirement for sustained dis- Conde, and collaborators showed, using a combina-
charge might account for why fast, reflex responses tion of psychophysics, unit recordings in animals, and
of the kind mediated by, say, the spinal cord, do not neuroimaging in humans, that for simple, unattended
seem to contribute to consciousness. Fast, reflex-like target stimuli, masking stimuli suppress visibility if their
responses also take place in the cerebral cortex. For ex- “spatiotemporal edges” overlap with the spatiotempo-
ample, the action-oriented dorsal visual stream rapidly ral edges of the targets, that is, if they begin or end
adjusts movements that remain outside of conscious- when target stimuli begin or end, in space and time.
ness (Goodale & Milner 2005). Perhaps these reflex- They confirmed that such spatiotemporal edges cor-
like adjustments to fast-changing aspects of the envi- respond to transient bursts of spikes in primary visual
ronment that must be tracked online are incompatible cortex. If these bursts are inhibited, for example if the
with the development of sustained discharge patterns. offset discharge elicited by the target stimulus is oblit-
By contrast, such sustained patterns may be necessary erated by the onset of the mask, the target becomes
for the ventral stream to build a stable representation invisible. Most likely, these spatiotemporal edges of in-
of a visual scene. By the same token, if the ventral creased firing are both generated (target stimuli) and
stream could be forced to behave in a reflex-like man- suppressed (masking stimuli) by mechanisms of lateral
ner, it should cease to be part of the NCC. Indeed, inhibition, which are ubiquitous in sensory systems
this can be achieved by pushing the ventral system to (Macknik 2006).
perform ultra-rapid categorizations such as deciding Additional evidence for the importance of phasic,
whether a natural image contains an animal or not transient activation of neurons in determining the visi-
(Thorpe et al. 1996; Bacon-Mace et al. 2005; Kirch- bility of stimuli comes from studies of microsaccades—
ner & Thorpe 2006). In such cases, a sweep of activity the small, involuntary movements that our eyes make
travels from the retina through several stages of feedfor- continually (Martinez-Conde et al. 2004). If microsac-
ward connections along the hierarchy of ventral visual cades are counteracted by image stabilization on the
areas, until it elicits an appropriate categorization re- retina, stationary objects fade and become completely
sponse. This process takes as little as 150 ms, which invisible. In a recent study (Martinez-Conde et al.
leaves only about 10 ms of processing per stage. Thus, 2006), subjects were asked to fixate a central dot (which
only a few spikes can be fired before the next stage tends to reduce microsaccades) while attending to a
produces its output, yet they are sufficient to spec- surrounding circle. Soon, the circle fades and merges
ify selective responses for orientation, motion, depth, into the background (the Troxler illusion). It was found
color, shape, and even animals, faces, or places con- that before a fading period, the probability, rate, and
veying most of the relevant information about the magnitude of microsaccades decreased. Before transi-
stimulus (Hung et al. 2005). While this fast feedfor- tions toward visibility, the probability, rate, and mag-
ward sweep within the ventral system is sufficient for nitude of microsaccades increased, compatible with
the near-automatic categorization of stimuli and a be- the hypothesis that microsaccades are indeed neces-
havioral response, it seems insufficient to generate a sary for visibility. Importantly, in macaque monkeys,
conscious percept (VanRullen & Koch 2003). For ex- when an optimally oriented line was centered over
ample, if another image (the mask) is flashed soon after the receptive field of cells in V1, the cells’ activity in-
the target image, subjects are still able to categorize creased after microsaccades and tended to emit bursts


(Martinez-Conde et al. 2002), suggesting again that (Hebb 1949), implementing hypothesis testing through
phasic activity may be crucial for visibility. resonances (Grossberg 1999), linking present with past
One should remember, however, that the impor- (Edelman & Mountcastle 1978; Edelman 1989), and
tance of phasic discharges for stimulus awareness has subject with object (Damasio 1999).
only been demonstrated for early visual cortex; but, A more concrete reason why reentrant activity is
as we saw above, the NCC are more likely to lie an attractive candidate for the NCC is that it travels
elsewhere. Psychophysically, masking is similarly effec- through back connections, of which the cerebral cortex
tive when masking stimuli are presented monoptically is extraordinarily rich (Felleman & Van Essen 1991).
(through the same eye as the target) and dichoptically In primates, feedforward connections originate mainly
(through the other eye). The analysis of fMRI data in supragranular layers and terminate in layer 4. Feed-
in humans shows that a correlate of monoptic visual back connections instead originate in both superficial
masking can be found in all retinotopic visual areas, and deep layers and usually terminate outside of layer
whereas dichoptic masking is only seen in retinotopic 4. The sheer abundance of back connections in sensory
areas downstream of V2 within the occipital lobe (Tse regions suggests that they ought to serve some impor-
et al. 2005), suggesting an anatomical lower bound for tant purpose, and giving rise to a conscious percept
the NCC. Thus, it could be that phasic onset and off- might just fit the bill. However, back connections and
set discharges are important not in and of themselves, associated reentrant volleys are just as numerous be-
but because they are particularly effective in activating tween V1 and visual thalamus, which is usually denied
downstream areas that directly support the NCC. In any direct contribution to awareness. Also, there does
these downstream areas, perhaps, the NCC may ac- not seem to be any lack of back connections within
tually require sustained firing. Indeed, the duration of the dorsal stream. It should be emphasized that the
the activation of face-selective neurons in IT is strongly strength and termination pattern of back connections
correlated with the visibility of masked faces (Rolls et seem more suited to a modulatory/synchronizing role
al. 1999). than to driving their target neurons. For example, the
focal inactivation of area 18 can slightly increase or
Reentrant versus Feedforward Activity decrease discharge rates of units in area 17, but it
Another possibility is that it is not so much sustained does not change their feature selectivity for location
firing that triggers the awareness of a stimulus, but and orientation (Martinez-Conde et al. 1999). Also,
rather the occurrence of a “reentrant” wave of activity the numerosity of backward connections is a natural
(also described as recurrent, recursive, or reverberant) consequence of the hierarchical organization of feed-
from higher to lower cortical areas. In this view, when a forward ones. For instance, cells in the LGN are not
stimulus evokes a feedforward sweep of activity, it is not oriented, while cells in area 17 are. To be unbiased,
seen consciously, but it becomes so when the feedfor- feedback to any one LGN cell should come from area
ward sweep is joined by a reentrant sweep (Lamme & 17 cells of all orientations, which requires many con-
Roelfsema 2000). This logic could apply both to early nections; on the other hand, since at any given time
sensory areas such as V1 and to higher areas such as IT. only area 17 cells corresponding to a given orientation
For example, when face neurons in the fusiform face would be active, feedback effects would not be strong.
area are first activated, we would not see a face con- If they were, properties of area 17 cells, such as orienta-
sciously, although we could turn our eyes toward it or tion selectivity, would be transferred upon LGN cells,
press a button to indicate our unconscious categoriza- which they are not (to a first approximation). Back-
tion as a face. However, when face-selective neurons ward signals certainly play a role in sensory function:
receive a backward volley from some higher area, for for instance, they can mediate some extraclassical re-
example frontal cortex, the face would become visible. ceptive field effects, provide a natural substrate for both
This view is based on several considerations. An im- attentional modulation and imagery, and can perhaps
portant one, though rarely confessed, is that a mere dynamically route feedforward processing according
sequence of feedforward processing steps seems far to prior expectations. But are backward signals really
too “straightforward” and mechanical to offer a sub- critical for consciousness?
strate for subjective experience. Reentrant processes, The most intriguing data in support of a role for
by “closing the loop” between past and present activ- reentrant activity in conscious perception have come
ity, or between predicted and actual versions of the from neurophysiological experiments. In awake mon-
input, would seem to provide a more fertile substrate keys trained to signal whether or not they see a salient
for giving rise to reverberations (Lorente de N´ 1938),
o figure on a background, the early, feedforward response
generating emergent properties through cell assemblies of V1 neurons was the same, no matter whether or not


the monkey saw the figure (Super et al. 2001). How- stimuli that are processed in visual cortex. If, around
ever, a later response component was suppressed when 100 ms later, a TMS pulse was applied to V1, the stim-
the monkey did not see the figure. Light anesthesia also ulus became perceptually invisible, although on forced
eliminated this later component without affecting the choice subjects could still discriminate orientation or
initial response. Late components are thought to reflect color. This result indicates that, without any overt par-
reentrant volleys from higher areas, although other ticipation of V1, stimuli can reach extrastriate areas
studies have disputed this claim (Rossi et al. 2001). without eliciting a conscious percept, just as in blind-
The late response component crucial for the visibil- sight patients. It also suggests that, since the forward
ity of stimuli under backward masking might also be sweep reaches V1 after just 30 or 40 ms, the TMS pulse
due to a reentrant volley (Lamme & Roelfsema 2000). may abolish the awareness of the stimulus not so much
In this case the timing of maximal backward mask- by blocking feedforward transmission as by interfer-
ing should be independent of target duration, since it ing with the backward volley (see also Ro et al. 2003).
would be determined exclusively by the time needed On the other hand, it cannot be ruled out that the
for the early component to travel to higher areas and TMS pulse may act instead by triggering a cortical–
return to primary visual cortex. Instead, the timing of thalamo–cortical volley that interferes with the offset
maximal masking depends on the timing of target off- discharge triggered by the stimulus, as may indeed be
set, suggesting that the component that is obliterated the case in backward masking.
is a feedforward offset discharge, not a reentrant one Yet another study, this time using fMRI, examined
(Macknik 2006). Moreover, the late component can be the neural correlates of brightness (perceptual light-
dissociated from a behavioral response simply by rais- ness) using backward masking. The psychometric vis-
ing the decision criterion (Super et al. 2001), and it ibility function was not correlated with the stimulated
can occur in the absence of report during change and portion of V1, but with downstream visual regions, in-
inattentional blindness (Scholte et al. 2006). Perhaps in cluding fusiform cortex, parietal–functional areas, and
such cases subjective experience is present but, due to with the sectors of V1 responding to the unstimulated
an insufficient involvement of frontal areas, it cannot be surround (Haynes et al. 2005). Remarkably, visibility
reported (Block 2005; Tsuchiya & Koch 2005; Lamme was also correlated with the amount of coupling (ef-
2006). But then, by the same token, why should we fective connectivity) between fusiform cortex and the
rule out subjective experience during the feedforward portion of V1 that responded to the surround. Once
sweep? again, this result could be explained by the activation
Another main source of evidence for a role of reen- of reentrant connections, though fMRI cannot distin-
trant volleys in consciousness comes from experiments guish between forward and backward influences.
using TMS. In an early experiment, Pascual-Leone
and Walsh applied TMS to V5 to elicit large, mov- Synchronization and Oscillations
ing phosphenes (Pascual-Leone & Walsh 2001). They Another influential idea is that consciousness may
then applied another, subthreshold TMS pulse to a require the synchronization, at a fine temporal scale,
corresponding location in V1. When TMS to V1 was of large populations of neurons distributed over many
delivered after TMS to V5 (+5 to +45 ms), subjects cortical areas, in particular via rhythmic discharges
often did not see the V5 phosphene, and when they in the gamma range (30–70 Hz and beyond) (Crick
saw one, it was not moving. Their interpretation was & Koch 1990). The emphasis on synchrony ties in
that disruption of activity in V1 at the time of arrival well with the common assumption that consciousness
of a reentrant volley from V5 interferes with the ex- requires the “binding” together of a multitude of at-
perience of attributes encoded by V5. In a subsequent tributes within a single experience, as when we see
study (Silvanto et al. 2005), it was shown that, when a a rich visual scene containing multiple objects and
subthreshold pulse was applied over V5, followed 10 attributes that is nevertheless perceived as a unified
− 40 ms later by a suprathreshold pulse over V1, sub- whole (Singer 1999). According to this view, the NCC
jects reported a V5-like phosphene (large and moving), of such an experience would include two aspects: first,
rather than a V1 phosphene (small and stationary). the underlying activation pattern (groups of neurons
Their interpretation was that activity in V5 that, on that have increased their firing rates) would be widely
its own, is insufficient to induce a moving percept can distributed across different areas of the cerebral cor-
produce such a percept if the level of induced activity tex, each specialized in signaling a different object
in V1 is high enough. or attribute within the scene; and second, the firing
In another study (Boyer et al. 2005), subjects where of these activated neurons would be synchronized on
shown either an oriented bar or a colored patch— a fast time scale to signal their binding into a single


percept. In this respect, synchrony seems ideally suited consciously perceived words induced theta oscillations
to signal relatedness: for example, in the presence of a in multiple cortical regions until the test stimulus was
red square, some neurons would respond to the pres- presented and a decision reached, while a burst of
ence of a square, while others would respond to the gamma activity occurred over central and frontal leads
presence of the color red. If they synchronize at a fast just prior to and during the presentation of the test
time scale, they would indicate to other groups of neu- stimulus. Importantly, the earliest event distinguishing
rons that there is a red square, “binding” the two fea- conscious and unconscious processing was not visi-
tures together (Tononi et al. 1992). By contrast, signal- ble in the power changes of oscillations but in their
ing relatedness by increased firing rates alone would be phase locking. About 180 ms after presentation of stim-
more cumbersome and probably slower (Singer 1999). uli that were consciously perceived, induced gamma
Also, fine-temporal-scale synchrony has the welcome oscillations recorded from a large number of regions
property of disambiguating among multiple objects— exhibited precise phase locking both within and across
if a green cross were present simultaneously with the hemispheres for around 100 ms. These results have
red square, cross and green neurons would also be ac- been interpreted to indicate that a transient event of
tive, but false conjunctions would be avoided by precise gamma synchrony resets multiple parallel processes to
phase locking (Tononi et al. 1992). Moreover, compu- a common time frame. The global theta rhythm that
tational models predict that, for the same level of firing, follows after this trigger event could provide the time
synchronous input is more effective on target neurons frame for allowing a global integration of information
than asynchronous input (Abeles 1991; Tononi et al. provided by sensory inputs and internal sources. Or
1992), and indeed synchrony makes a difference to the perhaps beta–gamma synchrony could enable the in-
outputs to the rest of the brain (Brecht et al. 1999; tegration of activity patterns within local cortical areas,
Schoffelen et al. 2005). Finally, large-scale models pre- while theta synchrony would permit the integration of
dict that synchrony in the gamma range occurs due to more globally distributed patterns: the more global the
the reciprocal connectivity and loops within the thala- representation, the longer the time scale for the inte-
mocortical system (Lumer et al. 1997a, b), and indeed gration of distributed information.
phase alignment between distant groups of neurons in It would be premature to conclude, however, that
the gamma range anticipates by a few ms an increase synchrony in one or another frequency band is neces-
in gamma-band power (Womelsdorf et al. 2007). In sarily a marker of NCC. As revealed by an increasing
this respect, oscillatory activity, even when subthresh- number of EEG, MEG, electrocorticography, and mul-
old, could further facilitate synchronous interactions tiunit recordings, cognitive tasks are associated with
by biasing neurons to discharge within the same time complex modes of synchronous coupling among pop-
frame (Engel et al. 2001). ulations of neurons that shift rapidly both within and
Experimental evidence concerning the role of syn- across different frequency bands, within and between
chrony/synchronous oscillations in perceptual oper- areas, and in relation to the timing of the task (Wom-
ations was initially obtained in primary visual areas elsdorf et al. 2007). It would rather seem that, given
of anesthetized animals (for a review see Singer & the brain’s remarkable connectivity, synchrony is an in-
Gray 1995). For example, in primary visual cortex, evitable accompaniment of neural activity, and that it
neurons have been found that respond to a coher- is bound to change just as activity patterns change,
ent object by synchronizing their firing in the gamma depending on the precise experimental conditions.
range. Stimulus-specific, gamma-range synchroniza- Whether a particular kind of synchrony, for example
tion is greatly facilitated when the EEG is activated in the gamma range, is uniquely associated with con-
by stimulating the mesencephalic reticular formation sciousness is still unclear. For example, high beta and
(Munk et al. 1996; Herculano-Houzel et al. 1999). It is gamma synchrony can be found in virtually every brain
similarly facilitated by attention (Roelfsema et al. 1997) region investigated, including some that are unlikely to
and by increases in a dominant stimulus under binoc- contribute directly to consciousness. Also, it is not yet
ular rivalry even though firing rates may not change clear whether synchrony in the gamma range vanishes
(Fries et al. 1997, 2001). during early NREM sleep, during anesthesia, or even
Other evidence has come from EEG studies of phase during seizures. Human studies are still inconsistent
locking in humans. A recent study compared the neu- on this point, whereas animal studies show a paradox-
ral correlates of words that were consciously perceived ical increase in gamma synchrony when rats lose the
with those of masked words that had been processed righting reflex (thought to correspond more or less to
and semantically decoded but had remained uncon- the loss of consciousness in humans; see Imas et al.
scious (Melloni et al. 2007). The results showed that 2005a, b). It would seem that there can be synchrony


without consciousness, though perhaps not conscious- are organized in a radically different way from human
ness without synchrony, at least in mammalian brains. brains?
Finally, there are some difficult conceptual prob-
lems in characterizing synchrony as an essential ingre- Consciousness as Integrated Information
dient that unifies perceptual states and thereby makes To address these questions, it would seem that we
them conscious. While many experiences do indeed need a theoretical approach that tries to establish, at the
involve several different elements and attributes that fundamental level, what consciousness is, how it can be
are “bound” together into a unified percept, there are measured, and what requisites a physical system must
many other experiences, equally conscious, that do not satisfy in order to generate it. The integrated information
seem to require much binding at all: for instance, an theory of consciousness represents such an approach
experience of pure darkness or of pure blue, or a loud (Tononi 2004b).
sound that briefly occupies consciousness and has no According to the theory, the most important prop-
obvious internal structure, would merely seem to re- erty of consciousness is that is extraordinarily informa-
quire the strong activation of the relevant neurons, with tive. This is because, whenever you experience a par-
no need for signaling relatedness to other elements, ticular conscious state, it rules out a huge number of
and thus no need for synchrony. Also, the idea that alternative experiences. Classically, the reduction of
the NCC of a given conscious experience are given by uncertainty among a number of alternatives consti-
active neurons bound by synchrony discounts the im- tutes information. For example, you lie in bed with
portance of inactive ones: information specifying that eyes open and experience pure darkness and silence.
particular unified experience must be conveyed both This is one of the simplest experiences you might have,
by which neurons are active and which are not, yet for one that may not be thought as conveying much in-
inactive neurons there does not seem to be anything formation. One should realize, however, that the in-
to bind. On the other hand, if most neurons in the formativeness of what you just experienced lies not in
cortex were to become active hypersynchronously, as how complicated it is to describe, but in how many
is the case in generalized seizures, they should result in alternatives were ruled out when you experienced it:
maximal “binding,” but consciousness vanishes rather you could have experienced any one frame from any of
than become more vivid.. innumerable movies, or the smoke and flames of your
room burning, or any other possible scene, but you did
not—instead, you experienced darkness and silence.
Consciousness from a Theoretical This means that when you experienced darkness and
Perspective silence, whether you think or not of what was ruled out
(and you typically don’t), you actually gained access to
Progress in the study of the NCC will hopefully a large amount of information. This point is so simple
lead to a better understanding of what distinguishes that its importance has been overlooked.
neural structures or processes that are associated with It is essential to realize, however, that the informa-
consciousness from those that are not. But even if we tion associated with the occurrence of a conscious state
come closer to this goal, we still need to understand is integrated information. When you experience a par-
why certain structures and processes have a privileged ticular conscious state, that conscious state is an inte-
relationship with subjective experience. For example, grated whole—it cannot be subdivided into compo-
why is it that neurons in thalamocortical circuits are nents that are experienced independently. For exam-
essential for conscious experience, whereas cerebellar ple, the conscious experience of the particular phrase
neurons, despite their huge numbers, are not? And you are reading now cannot be experienced as subdi-
what is wrong with many cortical circuits, including vided into, say, the conscious experience of how the
those in V1, that makes them unsuitable to yield subjec- words look independently of the conscious experience
tive experience? Or why is it that consciousness wanes of how they sound in your mind. Similarly, you cannot
during slow-wave sleep early in the night, despite lev- experience visual shapes independently of their color,
els of neural firing in the thalamocortical system that or perceive the left half of the visual field of view inde-
are comparable to those in quiet wakefulness? Other pendently of the right half.
questions are even more difficult to address in the ab- Based on these and other considerations, the theory
sence of a theory. For example, is consciousness present claims that the level of consciousness of a physical system is
in animals that have a nervous system considerably related to the repertoire of causal states (information) available to
different from ours? And what about computerized the system as a whole (integration). That is, whenever a sys-
robots or other artifacts that behave intelligently but tem enters a particular state though causal interactions


among its elements, it is conscious in proportion to how that bear some structural resemblance to different parts
many system states it has thereby ruled out, provided of the brain (Tononi 2004b, 2005). For example, by us-
these are states of the system as a whole, not decom- ing computer simulations, it is possible to show that
posable into states of causally independent parts. More high requires networks that conjoin functional spe-
precisely, the theory introduces a measure of integrated cialization (due to its specialized connectivity, each ele-
information called , quantifying the reduction of un- ment has a unique functional role within the network)
certainty (i.e., the information) that is generated when a with functional integration (there are many pathways
system enters a particular state through causal interac- for interactions among the elements.). In very rough
tions among its parts, above and beyond the informa- terms, this kind of architecture is characteristic of the
tion that is generated independently within the parts mammalian thalamo-cortical system: different parts of
themselves (hence integrated information). The parts the cerebral cortex are specialized for different func-
should be chosen in such a way that they can account tions, yet a vast network of connections allows these
for as much nonintegrated (independent) information parts to interact profusely. Indeed, the thalamo-cortical
as possible. system is precisely the part of the brain which cannot
If a system has a positive value of (and it is not be severely impaired without loss of consciousness.
included within a larger subset having higher ) it is Conversely, is low for systems that are made up of
called a complex. For a complex, and only for a complex, small, quasi-independent modules. This may be why
it is appropriate to say that when it enters a particular the cerebellum, despite its large number of neurons,
state, it generates an amount of integrated information does not contribute much to consciousness: its synaptic
corresponding to . Since integrated information can organization is such that individual patches of cerebel-
only be generated within a complex and not outside lar cortex tend to be activated independently of one
its boundaries, consciousness is necessarily subjective, another, with little interaction between distant patches
private, and related to a single point of view or per- (Cohen & Yarom 1998; Bower 2002).
spective (Tononi & Edelman 1998). Some properties Computer simulations also show that units along
of complexes are worth pointing out. A given physical multiple, segregated incoming or outgoing pathways
system, such as a brain, is likely to contain more than are not incorporated within the repertoire of the main
one complex, many small ones with low values, and complex. This may be why neural activity in afferent
perhaps a few large ones. We suspect that in the brain pathways (perhaps as far as V1), though crucial for
there is at any given time a complex of comparatively triggering this or that conscious experience, does not
much higher , which we call the main complex. Also, contribute directly to conscious experience; nor does
a complex can be causally connected to elements that activity in efferent pathways (perhaps starting with pri-
are not part of it through ports-in and ports-out. In that mary motor cortex), though it is crucial for reporting
case, elements that are part of the complex contribute each different experience.
to its conscious experience, while elements that are not The addition of many parallel cycles also generally
part of it do not, even though they may be connected does not change the composition of the main complex,
to it and exchange information with it through ports-in although values can be altered. Instead, cortical
and ports-out. One should also note that the value and subcortical cycles or loops implement specialized
of a complex is dependent on both spatial and tem- subroutines that are capable of influencing the states
poral scales that determine what counts as a state of of the main thalamocortical complex without joining
the underlying system. In general, the relevant spatial it. Such informationally insulated cortico-subcortical
and temporal scales are those that jointly maximize loops could constitute the neural substrates for many
(Tononi 2004b). In the case of the brain, the spa- unconscious processes that can affect and be affected
tial elements and time scales that maximize may by conscious experience (Baars 1988; Tononi 2004a,
be local collections of neurons such as minicolumns b), such as those that enable object recognition, lan-
and periods of time comprised between tens and hun- guage parsing, or translating our vague intentions into
dreds of milliseconds, respectively, though at this stage the right words. At this stage, however, it is hard to say
it is difficult to adjudicate between minicolumns and precisely which cortical circuits may be informationally
individual neurons. insulated. Are primary sensory cortices organized like
massive afferent pathways to a main complex “higher
Accounting for Neurobiological up” in the cortical hierarchy? Is much of prefrontal
Observations cortex organized like a massive efferent pathway? Do
Measuring and finding complexes is not easy for certain cortical areas, such as those belonging to the
realistic systems, but it can be done for simple networks dorsal visual stream, remain partly segregated from

Tononi koch-08

Recommandé

Recommandé

Contenu connexe

Tendances

Tendances (20)

En vedette

En vedette (15)

Similaire à Tononi koch-08

Similaire à Tononi koch-08 (20)

Tononi koch-08