678 nature neuroscience • volume 4 no 7 • july 2001

provide a possible adjunct therapy to anti-viral drugs for reducing HIV-associateddementia in patients with AIDS.

Finally, a number of unanswered ques-tions remain about this signaling chain.Why does the [Ca2+]i rise evoked by PgE2lead to glutamate release, whereas thatevoked initially by SDF-1 does not? Doesthis reflect subcellular compartmentaliza-tion of calcium, or just a larger rise of[Ca2+]i evoked by PgE2 and a non-lineardependence of release on [Ca2+]i? Howdoes TNFα generate rapid Ca2+-indepen-dent release of prostaglandin E2—is rapidactivation, rather than time-consuminginduction, of a phospholipase A2 or COX-2 involved? Why is the TNFα stepneeded, when in other cells ERK can acti-vate phospholipase A2—does this implythat TNFα is needed to activate COX-2rather than phospholipase A2? In othercells, TNFα can activate ERK, so is therean amplifying loop at the ERK → TNFα

595–599 (1998).

4. Ma, Q. et al. Proc. Natl. Acad. Sci. USA 95,9548–9553 (1998).

5. Tanabe, S. et al. J. Immunol. 159, 905–911(1997).

6. Meucci, O. et al. Proc. Natl. Acad. Sci. USA 95,14500–14505 (1998).

7. Limatola, C. et al. Eur. J. Neurosci. 12,2497–2504 (2000).

8. Davis, C. B. et al. J. Exp. Med. 186, 1793–1798(1997).

9. Kaul, M., Garden, G. A. & Lipton, S. A. Nature410, 988–994 (2001).

10. Parpura, V. et al. Nature 369, 744–747 (1994).

11. Araque, A., Sanzgiri, R. P., Parpura, V. &Haydon, P. G. Can. J. Physiol. Pharmacol. 77,699–706 (1999).

12. Newman, E. A. & Zahs, K. R. J. Neurosci. 18,4022–4028 (1998).

13. Lazarini, F. et al. Eur. J. Neurosci. 12, 117–125(2000).

14. Komuro, H. & Rakic, P. Science 260, 95–97(1993).

15. Wallach, D. et al. Annu. Rev. Immunol. 17,331–367 (1999).

stage of Fig. 1, which would promoteTNFα and glutamate release? Is the poten-tiation of this pathway by microglia dueentirely to microglial TNFα release, orcould the presence of microglia upregulatecomponents of the astrocyte signalingpathway between CXCR4 and TNFαrelease in Fig. 1? Why do gp120 and SDF-1 produce the same signaling in astrocyteshere, whereas in an earlier study13 thesame variant of gp120 did not produce theactivation of ERK seen with SDF-1? Fur-ther work will be needed to answer thesequestions, but the potential applicabilityof this signaling pathway to a range of neu-rodegenerative disorders should ensurethat the answers are rapidly forthcoming.

1. Bezzi, P. et al. Nat. Neurosci. 4, 702–710 (2001).

2. Asensio, V. C. & Campbell, I. L. TrendsNeurosci. 11, 504–512 (1999).

3. Zou, Y.-R., Cottman, A. H., Kuroda, M.,Taniuchi, I. & Littman, D. R. Nature 393,

news and views

Seeing is not perceivingGeraint Rees

A new study reports unconscious activation of occipital andfrontal cortex by masked visual stimuli, showing thatactivity in these areas may be insufficient for awareness

When does neural activity give rise toconscious visual experience? One way toaddress this question is to study the neur-al correlates of unconscious visual expe-rience, and thus identify patterns ofneural activity that are not sufficient forconsciousness. Much psychologicalresearch has supported the idea that visu-al stimuli can affect responses even whenthey are not consciously perceived. Forexample, in the 1970s, the British psy-chologist Anthony Marcel showed thatlexical decisions on letter strings are facil-itated (or ‘primed’) when they are pre-ceded by words that are masked so thatsubjects are unaware of them1. Whateverpatterns of brain activity are produced bymasked words in such a protocol musttherefore be insufficient to support aware-ness. In an exciting new report, Dehaeneand colleagues2 have identified such pat-

terns of brain activity associated withunseen masked words in areas of ventralvisual and prefrontal cortex.

The authors used a typical backwardvisual masking procedure to study brainactivity with functional MRI and event-related potentials (ERPs). Single wordswere presented briefly in close spatial andtemporal proximity to a patterned mask,so that subjects were unable to detect,name or subsequently remember them.Despite not being perceived, maskedwords (versus blanks) reliably activated alarge area of left fusiform gyrus and leftprecentral sulcus. ERP measurementsrevealed an early positive occipital wave-form in response to masked words, fol-lowed by two subsequent negativeleft-lateralized components. The ERPcomponents plausibly correspond to theactivations observed with fMRI infusiform cortex and precentral gyrus.These areas are part of a distributed net-work known to be associated with read-ing. Thus, components of the readingsystem for words were activated by visu-al stimuli of which the subjects were

unaware. This is an exciting result, as itsuggests that despite the usual close cor-relation between activity in ventral visu-al cortex and visual consciousness, ventralvisual cortex activation can also occurwithout conscious reportability.

These data show that words may acti-vate the reading system without reachingawareness, but do not tell us anythingabout the specificity of the activation. Asmasked words were compared to blanks,activation might conceivably represent anonspecific burst of activity associatedwith any type of visual stimulus, and notbe specifically related to word identity.However, in a second experiment,Dehaene and colleagues2 show thatunconscious activation of visual cortex isindeed word-specific and independent ofthe physical characteristics of the word.They took advantage of a well-studiedresponse called ‘repetition suppression’,a reduction in brain activity with repeat-ed stimulus presentations3, by measuringbrain activity produced by a visible tar-get word that was preceded by an invisi-ble prime word. The two words could bethe same or different, and could be writ-ten in the same or different letter case(Fig. 1). Activity was reduced in extras-triate, fusiform and precentral cortexwhen prime and target were the same,compared to when they were different. Inright extrastriate cortex, repetition sup-pression was specific to prime and targetwords that shared both identity and case,but in left fusiform gyrus and precentral

Geraint Rees is at the Institute of CognitiveNeuroscience, University College London,Alexandra House, 17 Queen Square, LondonWC1N 3AR, UK. e-mail: g.rees@fil.ion.ucl.ac.uk

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

© 2001 Nature Publishing Group http://neurosci.nature.com

nature neuroscience • volume 4 no 7 • july 2001 679

cortex, activity was reduced even whenthe case of the prime differed from thatof the target. These data demonstrate per-sistence of the effect even for invisiblemasked primes, showing that informa-tion independent of letter case (and thusprobably specific to word identity) wasextracted and encoded unconsciously inthese brain regions.

The authors’ main claim is that activi-ty in extrastriate visual cortex detected byfMRI and ERP can be insufficient forawareness. Such a conclusion is stronglysupported by other recent human neu-roimaging data. In visual extinction fol-lowing right parietal damage, patients candetect isolated visual stimuli on eitherside, yet they often miss contralesional(typically, left) stimuli during bilateralstimulation. Despite being unseen, such‘extinguished’ left visual field stimuli nev-ertheless evoke activity in primary visualcortex and category-specific areas of theventral visual cortex, reflecting uncon-scious identification of stimulus catego-ry4,5. Extrastriate visual cortex can also beactivated without awareness followingstimulation of the blind hemifield inpatients with damage to primary visualcortex6. Finally, in normal subjects, stim-ulus changes that the subject fails to detect(in so-called ‘change blindness’) never-theless evoke activity in ventral visual cor-tex7. Taken together, these studies suggestthat for a wide variety of stimulus cate-gories, in both patients and normal sub-jects, the mere presence of activity inventral visual cortex is insufficient forconscious awareness.

If stimulus-related activity in ventralvisual cortex does not necessarily result inthat stimulus reaching awareness, whatextra ingredient is required? Dehaene andcolleagues2 compared activation producedby seen unmasked words to unseenmasked words and observed three impor-tant differences. Most obviously, wordsthat were seen produced much strongeractivation in the areas of ventral visualcortex that were less robustly activated byunseen masked words. The overall level ofactivity in visual cortex might thereforedetermine awareness, with only higher

tions between different perceptual statesduring binocular rivalry and perceptionof ambiguous figures9–11 and awarenessof stereo ‘pop-out’12, and associated withconscious detection of change duringchange blindness7. Moreover, covariationof activity between a distributed networkof dorsal cortical areas and the ventralvisual pathway is also observed duringbinocular rivalry11, and during awareness(compared to extinction) of left visualfield stimuli following right parietal dam-age5. Taking these results together, it istempting to speculate that awarenessdepends on activity in a distributed cor-tical network encompassing stimulus-spe-cific representations in ventral visualcortex and stimulus-independent activi-ty in dorsal frontoparietal cortex.

An important challenge for the futureis how to map such physiological descrip-tions of distributed neural activity associ-ated with visual awareness onto afunctional description of the psychologi-cal processes involved. One possibility isthat activity in such a distributed networkmight reflect stimulus representationsgaining access to a ‘global workspace’ thatconstitutes consciousness13. According tosuch an account, representations in visu-al cortex become conscious via a networkof multiple distributed brain areas thatinteract in a coordinated manner. Howev-er, other functional alternatives are equal-ly plausible, and the current data do notdiscriminate among them. For example,some have suggested that involvement offrontal areas is crucial for report of visualstimuli14, or that activity in frontal andparietal cortex might be the functionalcorrelate of a ‘binding’ process necessaryfor the integration of different features ofan object15. It will be important for futurework to investigate how such functionalinterpretations can be constrained by thedata, and to identify the key empirical pre-dictions on which competing functionalaccounts can be distinguished.

Empirical investigation of the neuralactivity associated with conscious andunconscious visual experience now rep-resents an important and rapidly growingarea in neuroimaging. We are a long wayfrom explaining the hard problem: whyphysical activity in the brain gives rise tosubjective experience in the first place.Nevertheless, by showing unconsciousactivation of visual cortex by maskedwords, Dehaene and colleagues have con-tributed an important step in under-standing the necessary and sufficientconditions for neural activity to give riseto visual awareness.

levels leading to awareness. It should benoted that because fMRI and ERP mea-sure the summed activity of large popu-lations of neurons, it is not possible todistinguish whether the enhanced activityin response to seen words compared tounseen masked words reflects enhancedfiring in the same, or different, popula-tions of neurons. Similarly, the extent towhich neural processes such as syn-chrony, as well as changes in firing rate,may contribute to the enhanced fMRI andERP signal is not clear.

A second important differencebetween seen and unseen words was acti-vation of prefrontal and parietal cortex.These areas, outside the ventral visualpathway, were not activated by maskedwords. Finally, stronger covariation ofactivity between ventral visual cortex andthese non-visual areas of parietal andprefrontal cortex was observed whenwords were consciously perceived. Thesefindings suggest that awareness of stim-ulus properties encoded in activity in theventral visual pathway may require anadditional input from, or interactionwith, dorsal frontoparietal cortex. How-ever, an important limitation in inter-preting these data is that subjects wereable to name consciously perceived butnot masked words. Thus, differencesbetween masked and unmasked condi-tions encompass not just awareness ofword identity but also performance of anaming task. Some differences in thebrain activation observed may thereforerepresent implicit (that is, unconscious)aspects of task performance, rather thansubjective awareness alone. It is impor-tant to realize that implicit processes maycontribute not only to sensory process-ing but also to motor output; carefulexperimental design will be needed totease apart these different contributions8.

Despite these methodological issues,Dehaene’s observation that awareness ofvisual stimuli is associated with enhance-ment of activity in dorsal frontal and pari-etal cortex (as well as ventral visual cortex)converges strikingly with other recentwork. For example, activity in frontal andparietal cortex is time-locked to transi-

news and views

FACE face

271 ms 29 ms 29 ms 29 ms 500 ms

Fig. 1. Case-independent repetition priming. A lexical decision on a word (‘face’) is facilitatedwhen it is preceded by the same word in different case (‘FACE’) presented briefly in close proxim-ity to pattern masks that render it invisible. Dehaene and colleagues2 found neural correlates ofthis unconscious priming effect in left fusiform and precentral gyrus, suggesting that these areasunconsciously extract the identity of the masked word.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

© 2001 Nature Publishing Group http://neurosci.nature.com

680 nature neuroscience • volume 4 no 7 • july 2001

1. Marcel, A. J. Cognit. Psychol. 15, 197–237(1983).

2. Dehaene, S. et al. Nat. Neurosci. 4, 752–758(2001).

3. Desimone, R. Proc. Natl. Acad. Sci. USA 93,13494–13499 (1996).

4. Rees, G. et al. Brain 123, 1624–1633 (2000).

5. Vuilleumier, P. et al. Proc. Natl. Acad. Sci. USA98, 3495–3500 (2001).

6. Goebel, R., Muckli, L., Zanella, F. E., Singer, W. &Stoerig, P. Vision Res. 41, 1459–1474 (2001).

USA 96, 1669–1673 (1999).

12. Portas, C. M., Strange, B. A., Friston, K. J.,Dolan, R. J. & Frith, C. D. Proc. R. Soc. Lond. BBiol. Sci. 267, 845–850 (2000).

13. Baars, B. A Cognitive Theory of Consciousness(Cambridge Univ. Press, Cambridge, 1988).

14. Crick, F. & Koch, C. Nature 375, 121–123(1995).

15. Treisman, A. M. & Gelade, G. Cognit. Psychol.12, 97–136 (1980).

7. Beck, D., Rees, G., Frith, C. D. & Lavie, N. Nat.Neurosci. 4, 645–650 (2001).

8. Frith, C. D., Perry, R. & Lumer, E. TrendsCogn. Sci. 3, 105–114 (1999).

9. Kleinschmidt, A., Buechel, C., Zeki, S. &Frackowiak, R. S. J. Proc. R. Soc. Lond. B Biol.Sci. 265, 2427–2433 (1998).

10. Lumer, E. D., Friston, K. J. & Rees, G. Science280, 1930–1934 (1998).

11. Lumer, E. D. & Rees, G. E. Proc. Natl. Acad. Sci.

news and views

Boundaries on motion integrationVisual perceptual processes occur with an accuracy and ease that belies the difficulty of the problems they solve, and the perceptionof object motion is no exception. On page 745 of this issue, Jean Lorenceau and David Alais shed new light on how we perceiveobject motion, showing that an object’s shape exerts a powerful influence on motion integration, the process by which ambiguouslocal motion measurements are combined to determine the object’s motion. A simple diamond, translating horizontally (left),illustrates the problem. Receptive fields of neurons in the visual cortex sample small regions of space, and those situated on a movingobject’s edge cannot determine the object’s true motion. Although the diamond moves horizontally, an edge viewed in isolation(through the circle) appears to move diagonally, and is consistent with a wide range of object motions. Yet even when theunambiguous motions of the corners are covered with occluding surfaces (center), leaving only the edges, the diamond still appearsto move rightward. The only way to derive rightward motion from this stimulus is to combine motion information across edges. Eachedge alone is ambiguous, but together, they yield the correct motion direction. Much research now supports the idea that localmotion measurements are integrated across space in this manner.

The integration process is tricky, however. Most scenes contain more than one object, and if edge motions are combined across objects,the resulting motion estimates will be wrong. The visual system must somehow restrict motion integration to measurements made on thesame object. Lorenceau and Alais investigate this process by permuting the contours in their stimuli. In the example on the right, the top andbottom contours of the diamond have simply been swapped. The motion information is thus unchanged, but the new contours form adifferent global shape. This simple manipulation has a drastic effect on motion perception. Whereas the original diamond configuration isperceived to move as a single solid object, the contours in the new configuration are almost always seen to move independently. Thus thevisual system opts not to integrate the edge motions even though they are perfectly consistent with a single global motion. What accountsfor the difference? One critical variable seems to be contour closure. The authors found that contour configurations that did not form aclosed shape were almost never seen to cohere. The unclosed shapes remained incoherent even after subjects were shown the full,unoccluded shape (in this case, a bow-tie). The visual system thus seems to have a built-in bias favoring integration for closed shapes. Giventhat objects in the world tend to have closed bounding contours, these influences of shape on motion interpretation may serve to avoidintegrating across different objects.

Josh McDermott

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

© 2001 Nature Publishing Group http://neurosci.nature.com